Mitigating Power Fluctuations from Renewable Energy Sources Sean Leavey Supervisor: Dr Stefan Hild March 2012 Abstract Government legislation, new technology and efforts to reduce carbon emissions will result in renewable energy sources making up a higher proportion of the National Grid’s electricity supply in the future. A grid using a higher proportion of renewable energy sources will suffer from larger supply fluctuations than presently experienced due to the unpredictable natures of the underlying phenomena from which renewable sources derive their energy. Current mitigation techniques are outlined, then a model is created to demonstrate a grid powered solely from renewable energy sources. This model is used to predict the fluctuations that could be experienced in such a grid. With this information in mind, a prototype of the world’s first smart charging laptop is built and demonstrated. This involves sensing the grid frequency in real time and controlling the laptop’s charger accordingly. This system is tested with two different laptop computers, and it is found that rates of charge and discharge and battery life are the dominant factors in the computers’ performances. Further insight is gained from simulations of battery levels. Finally, the large scale effect of the smart charging technique is modelled. 1 Contents 1 Introduction 1.1 Legislation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 The National Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Supply and Demand Pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 4 4 2 Power Fluctuations 2.1 Supply Fluctuations from Renewable 2.1.1 Periods of Low Wind . . . . . 2.2 Demand Fluctuations . . . . . . . . 2.3 Dealing with Power Fluctuations . . . . . . 4 5 5 6 6 3 Characterising Supply/Demand Mismatch 3.1 Grid Frequency Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 7 4 Current Methods for Mitigating 4.1 Spinning Reserve . . . . . . . . 4.2 Short Term Operating Reserve 4.3 Frequency Response . . . . . . 7 7 8 8 Energy Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Mitigating Power Fluctuations in the Future 5.1 A 100% Renewable Energy Grid . . . . . . . 5.2 Demand-Side Management . . . . . . . . . . 5.2.1 Smart Electric Cars . . . . . . . . . . 5.2.2 Smart Fridges . . . . . . . . . . . . . . 5.2.3 Consumer Incentives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Project Aims 8 8 8 9 9 9 10 7 Prototyping, Testing and Characterisation of a Smart 7.1 Measuring the Grid Frequency . . . . . . . . . . . . . . 7.2 Frequency Measurement Technique . . . . . . . . . . . . 7.3 Verifying the Frequency Measurement Technique . . . . 7.4 Laptop Battery Performance . . . . . . . . . . . . . . . 7.5 Laptop Battery Simulations . . . . . . . . . . . . . . . . 8 Large Scale Impact of Smart Charging Laptop 8.1 Quantifying UK Grid Demand . . . . . . . . . . . . . . 8.2 Entirely Renewable Grid . . . . . . . . . . . . . . . . . . 8.2.1 Demand Model . . . . . . . . . . . . . . . . . . . 8.2.2 Wind Model . . . . . . . . . . . . . . . . . . . . 8.2.3 Solar Model . . . . . . . . . . . . . . . . . . . . . 8.2.4 The Combined Model . . . . . . . . . . . . . . . 8.3 Entirely Renewable Grid with Dynamic Demand Laptop 8.3.1 Large Scale Laptop Model Assumptions . . . . . 8.3.2 Simulation Mechanism . . . . . . . . . . . . . . . 9 Discussion 9.1 Grid Models and Simulations . . . . . . . . . 9.1.1 Wind and Solar Models . . . . . . . . 9.2 Laptop Controller Experiment . . . . . . . . 9.2.1 Long Term Effect on Laptop Batteries 2 . . . . . . . . . . . . . . . . . . . . . . . . Charging Laptop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 10 10 11 12 13 . . . . . . . . . 15 15 15 15 16 16 17 18 19 19 . . . . 20 20 20 21 21 9.3 9.4 Project as a Whole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 10 Conclusions 22 References 22 Appendices 25 A Interconnector Power Imports/Exports Sample 25 B National Grid Website Logger Program 25 B.1 Initial Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 B.2 Refined Frequency Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 C Battery Logging Shell Scripts 26 C.1 Ubuntu Linux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 C.2 Mac OS X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 D Laptop Grid Model Assumptions 27 List of Figures 1 2 3 4 5 6 7 8 9 10 11 13 12 14 16 15 17 18 19 20 21 22 23 24 Average UK weekly demand 2011 . . . . . . . . . . . . Typical UK weekday demand pattern . . . . . . . . . Irish wind power fluctuations, Apr-May 2011 . . . . . Example of a future UK power production scenario . . A typical demand TV pickup . . . . . . . . . . . . . . Grid frequency spectrogram . . . . . . . . . . . . . . . A dynamic demand refrigerator . . . . . . . . . . . . . Control system output and grid frequency comparison Dynamic demand laptop system flowchart . . . . . . . Control system input/output signal comparison . . . . Dell laptop charge time series . . . . . . . . . . . . . . Apple laptop charge time series . . . . . . . . . . . . . Full Dell smart charging system . . . . . . . . . . . . . Real and simulated laptop charge comparison . . . . . Twenty-one day laptop charge simulation . . . . . . . Laptop charge simulation histograms . . . . . . . . . . Three day solar model simulation . . . . . . . . . . . . Power surplus/deficit using renewable grid model . . . Histogram of renewable grid model’s periods of deficit A rare period of very long deficit . . . . . . . . . . . . Power surplus/deficit using renewable grid model, with Example of power mitigation using 20 million laptops Interconnector imports/exports sample . . . . . . . . . Histograms of laptop grid model assumptions . . . . . 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mitigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 5 5 5 6 7 9 10 12 12 13 13 14 14 15 16 17 18 18 18 19 20 25 27 1 Introduction independent body responsible for maintaining the grid on the government’s behalf, are tasked Increasing the use of renewable energy sources with keeping the supply and demand of elecwill bring about a reduction in the reliability of tricity in balance. electricity generation being able to meet electricity demand, due to the inherent variability 1.3 Supply and Demand Pattern of the physical phenomena renewable energy sources derive their power from. It is therefore Electricity demand in the UK follows a reaimportant to find ways in which to mitigate sonably predictable pattern. National Grid these fluctuations to prevent them from caus- PLC forecast demand using a number of meaing shortcomings in supply during periods of surements such as weather patterns, time of low prevalence of the underlying physical pro- year, day of the week and time of day. Naturally, demand for electricity is greatest in cesses. the colder months (see Figure 1) when people need more heat, and particularly in the day1.1 Legislation light hours when offices are lit and heated, and The UK Government is working towards pro- when people use productive equipment such as viding a greater proportion of the National computers (see Figure 2). The whole counGrid’s electricity via renewable energy sources try’s demand throughout the day will tend to in order to reduce the UK’s greenhouse gas fluctuate only gradually, as localised occasions emissions from fossil fuel power plants. Un- where power is required tend to be balanced der the Climate Change Act signed into law by other simultaneous occasions where power in 2008, the government set out targets to cut is no longer required, such as a kettle being CO2 equivalent emissions by 34% compared to switched on in one household as another house1990 levels by 2020 [1]. In Scotland, the gov- hold’s kettle switches off having finished boiling ernment has been considering plans ”to meet some water [4]. This is known as diversity of an equivalent of 100% demand for electricity demand. from renewable energy by 2020” [2]. The EU also has a directive which legislates that the UK should increase production of electricity from renewable energy sources by 15% over the same period [3]. 1.2 The National Grid Electricity is supplied to homes and businesses in the UK from the National Grid, a series of interconnected power stations, substations and other infrastructure connected to homes and businesses. Instantaneously, there is a specific demand (typically of the order 35 GW) created by the electrical devices connected to the grid, and a specific power being supplied by the various interconnected generators from power plants. Keeping these two factors in balance is extremely important because the electricity grid does not have the ability to store any energy, so the demand must match the supply to a high degree at all times otherwise devices connected to the grid across the country will receive too little or too much electricity. National Grid PLC (Publicly Limited Company), the Figure 1: Average weekly demand in the UK for 2011 (data sourced from National Grid website). 2 Power Fluctuations Power fluctuations, where the supply does not match the demand, need to be dealt with properly to ensure that electrical devices connected to the grid don’t receive too little or too much power. These fluctuations can arise from variations in supply, such as when a power plant stops producing output, or from variations in demand, when consumers require more or less power than is being supplied. 4 Over the period of a week at the start of June 2011, wind turbines in the Republic of Ireland (for which there is information readily available) produced an average output of 315 MW [5]. The average UK power demand over a typical week is about 35 GW [6]. If the UK were to use only wind turbines to power the grid, the power output of the wind turbines would need to match the power demand. In addition, there would have to be enough of a buffer to cover periods where there is not enough wind power to match demand. This buffer can be formed from energy storage facilities and methods of reducing demand, as outlined later in Section 4. Figure 2: Typical UK weekday demand pattern. Demand is greatest around dinner time, and lowest during the night (data from demand on 23/10/11, sourced from National Grid website). 2.1 Supply Fluctuations from Renewable Energy Sources Renewable energy sources rely on highly complex physical processes which are hard to predict to a high degree of accuracy. Electricity generation from wind, for instance, relies on physical processes involving the interactions of innumerable air molecules, the Coriolis force, the Moon’s gravitational pull and thermodynamic quantities. As such, providing a predictable power output from these sources is extremely tricky (an example of the fluctuations can be seen in Figure 3). Instead, it is necessary to evolve processes to cope with the natural fluctuations from these renewable energy sources. Figure 4: A simulation of UK power production over the course of a day for a grid powered solely from wind turbines, and a typical UK power demand profile for the same day (where the average power generation matches the average power demand). Unscaled data taken from Republic of Ireland wind power production on 23/10/11. Figure 4 shows a potential power production profile for the UK, using a grid powered solely from wind turbines. Weather fluctuations result in a varying power output, whereas power demand remains reasonably stable. In this instance, a buffer of about 15 GW (equivalent to the output of about 30 typical large nuclear power reactors1 ) would be required to cover for Figure 3: Fluctuations of the power output of wind the worst case scenario of the lowest wind pefarms in Ireland in April and May 2011. riod coinciding with the highest demand (occurring around 9am). In reality, the buffer figure would need to be even higher in order to cover longer periods 2.1.1 Periods of Low Wind without wind, higher winter demands, periods When wind power production takes place of low wind coinciding with peak demand, and across a large area, localised instances of low to allow for failures in wind farm transmission or high wind speeds somewhat average out. 1 Nuclear power plants usually have two or more reHowever, there can still occasionally be peri- actors, meaning their output level is typically around ods of low average production when large scale 1 GW. Coal fired power plants, of which there are 18 weather patterns have an impact. in the UK, have a typical output of about 1.5 GW. 5 devices and other parts of the National Grid. UK demand is also likely to increase in the future as consumers move to using electric transport, heating and other uses of electricity as dwindling oil and gas reserves push these fuel prices higher. a television demand pickup of the order 2 GW, the three large pumped storage hydroelectric facilities in the UK have a combined output just large enough to cover this [9]. Occasionally it is necessary to purchase power from France and the Netherlands during this period to cover the gap in supply [10] (see Appendix A). An indirect way of dealing with a demand spike from one part of the grid would be to decrease the demand in another part of the grid to compensate. The National Grid have a program called Frequency Control Demand Management which allows large consumers of electricity such as industrial works to take part in a system where they will scale back or halt their electricity use by at least 3 MW for a period of 30 minutes during times of high demand [11]. The consumers are paid a premium [12] for the time they spend on a reduced consumption. In September 2011, the cost to the National Grid in utilising commercial frequency control demand management contracts amounted to £7.32M for a reduction of 369 GWh (£19.82/ MWh) [13]. Figure 5: A typical TV demand pickup on Thursday 20/10/11. The left bump occurred after EastEnders finished at 8pm (a 400 MW pickup) and the right bump occurred after Coronation Street finished at 9pm (a 200 MW pickup). 2.2 Demand Fluctuations Other than fluctuations in supply, there can be fluctuations in demand in addition to the reasonably predictable day to day demand variation. Occasionally, demand will undergo a sudden change when the level of electricity consumption across the country is somewhat synchronised. So-called demand pickups can occur after popular television programmes such as soaps or football cup final matches finish, where demand increases by as much as 2.8 GW [7–9] as the viewers switch their kettles on for a post-programme cup of tea. The effect can be seen in Figure 5 for a typical Thursday evening. 3 Characterising Mismatch Supply/Demand The mismatch between supply and demand on the grid can be characterised by the frequency of the AC (alternating current) voltage it provides. In the UK and most of Europe the grid is designed to run at 50 Hz [14,15]. When there is too much power on the grid, the frequency will be higher than this value, and when there is too little power on the grid the frequency will be lower [6, 12]. Typically, the frequency will deviate within ±0.2 Hz of that value, which is a fluctuation limit deemed small enough so as to not damage electrical equipment plugged in to the grid [6]. Dangers arise when the frequency drops lower than 49.5 Hz or increases above 50.5 Hz, as this suggests that substantially too much or too little power is being transmitted over the grid. National Grid PLC has statutory obligations to maintain the frequency within 50.0 ± 0.5 Hz [6, 14] in the absence of extraneous circumstances. Figure 6 shows the grid frequency as a spectrogram over a 10 hour period. The grid fre- 2.3 Dealing with Power Fluctuations In order to make effective use of renewable energy sources, technologies must be developed which can mitigate their power fluctuations. One of the main ways the National Grid controllers can directly deal with a power fluctuation is by changing output to meet the demand, but this can be costly and of limited use and economy for shorter periods of increased or decreased demand. Power stations can raise or lower their electricity production temporarily, but this conflicts with their ideal output level and production efficiencies can suffer [7]. For 6 than the ideal 50 Hz, there is too much energy in the system, and when it is lower than 50 Hz, there is too little energy in the system. It is on this basis that the frequency can be used as a guideline as to the current balance of supply and demand on the grid. The grid frequency is regulated by power plants measuring the current grid frequency and having their generators try to match that frequency. For a coal plant, this might involve regulating a steam inlet valve so as to converge upon the current grid frequency; whereas for a hydroelectric plant this might involve regulating the flow of water through the turbines. In the event of a power plant going offline or a large surge in demand creating a supply deficit, power plants will initially provide the extra power required by using their flywheels’ kinetic energy. Once this is no longer sustainable, the power plant’s frequency will drop unless measures to compensate for the supply deficit have been put in place by grid controllers. An analogy to this is with a car using cruise control starting to go up a hill: initially the car will match the speed it was travelling at before the hill, but eventually it will have to slow down if its torque cannot match the force required to oppose gravity. Figure 6: A spectrogram of the mains frequency as measured at the University of Glasgow, between 00:00 and 10:00 on 01/11/2011. quency varies around the 50 Hz mark, keeping within the National Grid’s own target of 50 ± 0.2 Hz for the large majority of the time in this instance. 3.1 Grid Frequency Mechanics The mechanics of the voltage frequency produced by a power plant arises from the angular velocity at which its generator’s rotor is spinning [15]. When the demand is higher than the supply, the deficit is met by extracting kinetic energy from the rotational inertia attributed to the generator [12]. Typically, for a power plant class generator, the demand can be met in this way for a period of 2 to 8 seconds [16]. After this period, additional power generators will have had to have been brought online to cover the demand, or the demand will have had to have been reduced to compensate. All of the power plants synchronised to the grid at any given moment can been empirically represented as one generator with a flywheel with output power and frequency equivalent to that of the grid as a whole [17]. The flywheel’s moment of inertia can be described as I, and so if one knows a value for the inertia, one can estimate the electrical energy available on the grid through the use of the highly generalised Equation 1. 1 E = Iω 2 2 ω = 2πf 4 Current Methods for Mitigating Power Fluctuations Current methods for mitigating power fluctuations primarily involve the use of additional so-called ’peaking’ power plants – fast start (typically gas turbine) generators which can be turned on to meet short term spikes in demand – and by increasing the output power provided by power plants already in operation. In addition, the grid can purchase power from France, Northern Ireland and the Netherlands via underwater cables [18]. (1) 4.1 Spinning Reserve (2) A number of power plants generating electricity for the National Grid are running at only From this we can see that the kinetic en- a fraction of their total capacity. This extra ergy is proportional to the square of the fre- capacity is kept available in order to use when quency f , for a flywheel with moment of iner- needed, to cover supply shortages that the base tia I. Therefore when the frequency is higher load cannot handle. However, running power 7 plants at less than 100% capacity typically results in losses of efficiency, as generators are usually optimised to run at full load. A typical coal plant might have a thermal efficiency of 40% at full output, but only 35% at half load [7]. The full efficiency is usually only achievable after a long period of warming up (around 8 hours), so this method is not ideal for handling shorter term fluctuations. electricity in its entirety from renewable energy sources. 5.1 A 100% Renewable Energy Grid If the UK were to move towards a state where the power supplied to the National Grid was solely provided by renewable energy sources such as wind, wave and solar, a great deal of care would need to be given towards handling the power fluctuations these renewable forms would produce. The diversity of supply from renewable energy in the UK might mean that a lack of wind in Scotland might not necessarily impact greatly on the overall power supply from wind in the UK, as the geographical diversity of other wind farms in the UK such as those in Wales would ensure that adequate power is supplied from the wind [9]. However, there can be periods where weather patterns contribute to low wind and low temperature across the whole country, such as in winter. In this situation, there can be little power generation from wind farms and high energy consumption from domestic heating, which would create a problem in a grid designed to maximise wind power generation. Using weather reports from January 2005, a theoretical installed capacity of 25 GW of wind turbines (just under half the peak demand in the UK) would have placed demand on other sources of power such as fossil fuel plants at between 6 and 56 GW at various points throughout the month [19]. Clearly, there is scope for a form of power fluctuation mitigation which involves a reduction in demand rather than an increase in supply from conventional power plants. Over shorter periods of time, this reduction can be met through the use of smart household appliances and a temporary decrease in the power consumption of industrial consumers. Over longer periods of time it would be necessary for behaviours to change with respect to power consumption when there is a lack of power from renewable energy. A study by Oswald et al [19] discusses these issues. 4.2 Short Term Operating Reserve Certain types of electricity generators such as gas turbine plants or hydroelectric dams (termed ’pumped storage’) are very quick to produce electricity when needed. The National Grid keeps a selection of these plants on standby for use at short notice. The use of gas turbine plants is expensive, as they substitute fuel efficiency for a quick reaction time, a bit like a supercharger in a race car. A gas turbine plant may only have a thermal efficiency of around 30% [7]. A hydroelectric dam has an operation time limited to the amount of water available to produce electricity from, and so is only available after a suitable resupply period after its last use, and for a limited amount of time. 4.3 Frequency Response The National Grid has commercial agreements with large electricity consumers to limit their use in times of high demand or low supply where spinning reserves are unable to cope [12]. Consumers willing to take part in this are most likely to have electricity use patterns which aren’t adversely affected by short periods of loss of supply, such as steel works (where the furnace can take days to heat up with an induction heater) or cold storage facilities. 5 Mitigating Power Fluctuations in the Future The use of fossil fuels for providing power is ultimately unsustainable as increasing populations and energy usage put more strain on nat5.2 Demand-Side Management ural resources. This fact, along with government legislation, means that there might come Methods for reducing demand involve the use a time when the National Grid must supply of buffers in order to temporarily prevent 8 devices that consume electricity from taking power from the grid, instead continuing their operation by using its own internal supply of energy. Such buffers can take the form of flywheels (such as those used in kinetic energy recovery systems in turbines), pumped storage using gravitational potential energy and also even batteries in laptops – an application which will be investigated later. Devices with buffers can be designed to use grid electricity only when there is a supply surplus, and not when there is a deficit. Managing demand in this way is called demand-side management. A number of demand-side management techniques exist, such as: grid frequency, such that periods of grid deficit increase the temperature at which they start to cool the contents. Typically, these fridges can maintain the internal temperature within safe limits for up to an hour before being forced to switch back on again regardless of the state of the grid. This window gives the National Grid controllers some time in which to get reserve power systems online. If every fridge in the UK (estimated to be around 40 million [23]) were to use this kind of technology, savings of between 500 MW [24] (roughly equivalent to a hydroelectric power plant) and 1200 MW [23] could be achieved for up to an hour at short notice. Figure 7 shows the possible effect this deferral could have. • Smart appliances, such as air conditioners [20], which either monitor the grid frequency for an instantaneous measurement of the supply-demand balance or receive information as to the current grid load via radio, internet or other means. They use this information to limit their power use in times of high demand or low supply. • Flexible tariffs, smart electric cars and fridges, as outlined in the following sections. 5.2.1 Figure 7: An example of the effect that dynamic demand management in fridges could have on the electricity demand over the course of a few hours. During a period of high demand, fridge energy consumption is deferred until a later time. Image taken from Stabilization of Grid Frequency Through Dynamic Demand Control (Short et al. [12]). Smart Electric Cars Various proposals exist for the charging of electric cars using systems which sense the availability of electricity on the grid [21]. Some systems even propose using electric cars as electricity supplies, releasing power back into the grid when required, leaving enough stored for 5.2.3 the owner’s own usage [22]. 5.2.2 Consumer Incentives A number of demand-based tariffs have been trialled around the world, from a lower ’Economy7’ tariff used in Europe for 7 hours during the night [25] to an hour-by-hour rate used in parts of Florida [26]. In France, the high proportion of nuclear power plants, which are best utilised in providing a constant output over long periods of time, has led to incentives for consumers to use more of their electricity during the night. A higher proportion of French homes are heated with electricity than gas, using night time storage heaters, due to the incentive created by Smart Fridges Refrigerators with built-in controllers monitoring the grid frequency in order to decide when to initiate refrigeration have been tested and have shown positive results [12]. Fridges typically have two threshold temperatures: one temperature above which the fridge will turn on its compressor to cool the contents, and another temperature below which the fridge will turn off its compressor to stop cooling the contents too much. Smart fridges scale these threshold temperatures by the inverse of the 9 cheaper night time electricity [9]. This helps to move some of the peak load during the day into the night. A study conducted in a German town [22,27] showed that electricity tariffs which incentivise use of electricity outside of peak times can have an impact in reducing peak load during the day, but failed to fully mitigate large spikes in demand. In the study, a larger than normal spike occurred around the border between a low tariff and a medium tariff time, which had the opposite effect of mitigation in that instance; though the use of electricity during the day when demand is provided by the base load was lower. its internal battery, it must know the state of the grid, as reflected in its frequency. A real time digital control system running a Linux operating system was used to make this decision. The mains alternating current was measured by using an AC transformer, which stepped the voltage down from 230 V to 9 V. This allowed the mains input to be safely measured by the control system. A set of frequency time series values collected over the course of 24 hours using the system was compared against the corresponding 24 hours worth of frequency measurements as reported by the National Grid website, using data obtained by the Java program written primarily to log the grid demand earlier. The data matched up well, meaning that the mains 6 Project Aims signal received by the digital control system was unperturbed by other equipment nearby This project will focus on the development of and the data was being logged correctly and concepts for mitigating power fluctuations usinstantaneously. The results can be seen in ing demand-side techniques, to allow for a reFigure 8. duction in the amount of standby fossil fuel power plant capacity required. To investigate methods for mitigation, a system to control the charging of a laptop which is responsive to changes in grid frequency will be investigated, using mains frequency sensitive hardware and software methods. The reason for this is that laptops are widespread already and so would not need major capital investment, and the potential impact is significant. Figure 8: A comparison between the frequency as To assist in quantifying the effect that such measured by Fourier transforms conducted on data a system might have on a future grid, a model collected by the digital control system and the frewill be developed to provide information about quency as reported by the National Grid website. the forms and magnitudes of the fluctuations The National Grid website reports frequency values that might be present in such an environment. which may be up to three minutes out of date, so a systematic correction has been applied to this data in order for it to overlay the corresponding control system measurements correctly. 7 Prototyping, Testing and Characterisation of a Smart Charging Laptop 7.2 Frequency Measurement Technique The frequency of the input signal could have been measured directly using Fourier transforms (like that which has been done in Figure 8). However, these operations are computationally expensive and require a lot of memory. For the purposes of this project, it is 7.1 Measuring the Grid Frequency also true that it is only necessary to know In order for a laptop to effectively judge when whether the input frequency is above or below to use power from the mains and when to use the threshold in question, 50 Hz. Therefore, it The simple case was considered of a laptop which charges when the grid frequency is above 50 Hz and uses its battery when the grid frequency is below 50 Hz. 10 was more practical to use a differential band pass filter method in order to obtain such information. changing sign too quickly when the grid frequency is quickly fluctuating above and below 50 Hz • The signal is input into a comparator, which outputs from the control system a high signal (in this case, 4 V) if the signal from the previous step is negative, or a low signal (in this case, 0 V) if the signal is positive The Method Used in order to Measure the State of the Signal • The voltage signal from the National Grid is stepped down from 230 V to 9 V • This signal is passed through an analogueto-digital converter, then passed into the digital control system, where the signal is sampled a frequency of 4096 Hz • The signal is passed through a band pass filter with range 45-55 Hz to eliminate any frequencies well outside the possible range2 , such as higher frequency harmonics and background noise • The signal is split into two copies, with the first copy passing through a band pass filter with a range of 48-50 Hz, and the second copy passing through a band pass filter with a range of 50-52 Hz • Both signals are saturated so as to become square waves • The signals undergo a half wave rectification and the absolute value is taken, so as to result in the signal being directly proportional to the signal the bandpass filters pick up • The signals are passed through a low pass filter with corner frequency 1 Hz, so as to remove any fast transients • The signal that passed through the higher bound band pass filter is then subtracted from the signal that passed through the lower bound band pass filter, thereby giving a value which, if positive, represents an underlying frequency below 50 Hz, and, if negative, represents an underlying frequency above 50 Hz • The signal is passed through a low pass filter of 0.1 Hz to prevent the signal from 2 The National Grid frequency very rarely falls below the 49.5 Hz or rises above 50.5 Hz, so this limit is very lenient. • This output control signal is amplified by a factor of 6 • The amplified signal is input into a Form A (’normally open’) relay requiring a control signal potential difference of 20 V to switch on, which drives the laptop charger circuit Notes: The frequency filters used are digital versions; and the output signal from the control system has to be amplified because of constraints on the ability of the control system to output larger voltage signals. The differential band pass filter method is effectively comparing a signal with frequency components below 50 Hz with a signal with frequency components above 50 Hz. At any given time, depending on the frequency of the grid, one of these signals should be zero. This means that, upon subtracting one signal from the other, the polarity of the resulting signal will represent whether the frequency of the grid is above or below 50 Hz. A flowchart of the process can be seen in Figure 9. Characterisation measurements conducted with a calibrated frequency generator showed that this method has a resolution of about ±1 mHz. 7.3 Verifying the Frequency Measurement Technique In order to check that the frequency was being correctly measured, the input signal and the output control signal were measured for a period of 24 hours. A frequency spectrogram was then produced using Fourier analysis performed on the input signal data, in order to show the frequency components present in the signal. In order to retrieve the underlying grid frequency from this spectrogram, the frequency 11 harmonic with the strongest amplitudes was chosen (in this case 250 Hz) and the maximum value at each time interval was obtained. This maximum value represents the grid frequency signal, or in this case, the 5th harmonic. The frequency was obtained simply by dividing the harmonic by 5. Figure 10 shows the input and output signals from the control system alongside the frequency of the input signal as produced from performing spectrogram analysis. It is clear to see that the control system output the correct control signal for each input measurement. Figure 10: A comparison between the control system’s input signal frequency (as inferred from Fourier analysis) and the control system’s output signal. It can be observed that when the input signal frequency is above 50 Hz, the output signal is high, and when the input signal frequency is below 50 Hz, the output signal is low. 7.4 Laptop Battery Performance Figure 9: The control flowchart for the dynamic demand laptop system. The signal is taken from the National Grid, stepped down from a voltage of 230 V to 9 V, converted to digital form and analysed by the control system. A control signal is then output by the control system which in turn operates the relay. The relay then governs the charging of the laptop from the mains. The area shaded blue is part of the control system and the area shaded red is the National Grid. The system would be useless unless it was able to allow the laptop of interest (we have arbitrarily chosen a Dell Latitude E5510 provided by the university) to be used in a way it would have been used without the dynamic demand system. It was therefore necessary to measure the laptop’s battery charge as a function of time when using the smart charging technique. In order to make such a measurement, a shell script was written to continuously run using a terminal within the Ubuntu Linux operating system installed on the laptop, logging the state of the battery obtained from a special battery monitoring program to a text file. The results of the 24 hour experiment can be seen in Figure 11. The laptop managed to retain some level of charge for 23 hours, until a particularly long period of grid frequency below the 50 Hz threshold caused the laptop’s 12 battery life to eventually drain. A full comparison of the whole system in operation can be seen in Figure 12. For the sake of comparison, the same test was also performed on an Apple laptop for a period of 7 hours using a modified terminal script for Mac OS X. It was reasoned that the Apple would give a useful comparison as its battery life is much longer than that of the first laptop. The results from this experiment can be seen in Figure 13. Clearly, the Apple laptop’s ability to charge much more quickly than it discharges and its larger buffer size helped it remain on throughout the test, but a longer period of measurement would have proved these outcomes more conclusively. Both shell scripts can be seen in Appendix C. Figure 11: The charge level of a Dell laptop during a 24 hour experiment using the dynamic demand control system. The laptop manages to keep some charge for 23 hours before a long period of low grid frequency results in the battery draining. Figure 13: The charge level of an Apple Mac laptop during a 7 hour experiment using the dynamic demand control system. The laptop manages to remain operational throughout the test, though the test might not have been long enough to conclusively prove that this will always remain so. Rather than comparing the two laptop simulations in terms of their rates of charge and discharge percentage points per second, which would inherently rely on each laptop’s battery capacity, a more reliable figure for comparisons is the ratio between the relative rates of charge and discharge. This value denotes how quickly the laptop charges with respect to how quickly it discharges (see Equation 3). R= Rate of Charge Rate of Discharge (3) The Apple laptop charges more than twice as quickly as it discharges, whereas the Dell’s rates of charge and discharge are lower. This means that, during the experiment, the Dell was more susceptible to running out of charge during prolonged periods of below-threshold frequency. It is clear that the value chosen as a threshold frequency has a big impact on the lifetime of a laptop plugged into the control system. While a threshold of 50 Hz should theoretically allow for the same amount of time spent charging as time spent discharging in the control system (so long as the National Grid controllers manage to maintain an average frequency of 50 Hz), the individual periods in which the grid frequency is above or below 50 Hz can last longer than the battery lifetime of a laptop like the Dell. In a situation where a long period of frequency lower than the threshold occurs, this can result in the battery draining and the laptop becoming non-operational. For a laptop that is similar to the Dell, it is reasonable to make a compromise by lowering the threshold used to make the decision of whether to charge the laptop or not. A lower threshold frequency means that, in principle, the controller will output a high signal for longer periods than it will output a low signal. This in turn means that the laptop should spend more time charging than it does discharging, ensuring that it retains a high level of buffer capacity. In contrast, laptops which can charge much faster than they discharge, like the Apple Mac, will probably not need to have the threshold lowered as far (or at all). 7.5 Laptop Battery Simulations In order to find a value for the threshold which is adequate in terms of keeping the laptop functional but which is also still high enough to provide a reasonable amount of mitigation, it was easier to simulate the charge level of the laptops. This prevents having to run experiments 13 Figure 12: Combined version of Figures 10 and 11, related to the smart charging system using the Dell. The top graph shows the grid frequency during the 24 hour test, the middle graph shows the corresponding control signal output and the bottom graph shows the laptop’s charge level for the same time series. The laptop manages to remain operational until 23 hours into the experiment. Laptop Dell Mac Ratio 1.8522 2.2857 reached this threshold in order to help compensate for this effect. Finally, the charge level simulation was undertaken using the same set of frequency Table 1: A list of the rates (percentage points data collected in the experiment. The per second) of charge, RCharge , and discharge, charge/discharge rates were perturbed until RDischarge , and the ratio between the two. the simulated charge levels closely matched the real levels. The result can be seen in Figure 14. with different thresholds, which, if to be conducted thoroughly, would have taken weeks. It also allows for better comparability of results, as the same frequency measurement time series can be used for many different variations on the same experiment. The first step in creating a simulation was to measure the rates of charge and discharge of each laptop, by approximating the gradients of the charge rates from the logs obtained in Figure 14: A comparison between the real and simSection 7.4. These approximations are listed ulated charge levels during the 24 hour test. As can be seen, the simulation successfully recreates in Table 1. the charge levels experienced in reality to a good However, it can be seen from the charge degree. The points at which the simulation over graph that the Dell laptop appears to slow or underestimates the charge can be attributed to down its rate of charging as the charge level slightly incorrect charge/discharge gradients which surpasses 90-95%. This might have been due might arise from effects governed by the real batto a mechanism built into the battery itself, tery’s microcontroller. or from a software controller, in order to preserve long term battery performance. RegardThree weeks of frequency data was then obless, this change affected the naive assumption tained using a purpose built Java program, of linear charge and discharge rates. A correc- developed by modifying the previous demand tion factor was applied to charge rates as they logger program to instead log frequency data RCharge 0.0213 0.0128 RDischarge 0.0115 0.0056 14 to higher (15 second) resolution (see Appendix B.2). This data was used in the simulations so as to provide a conclusive result as to whether or not a particular threshold frequency would allow a laptop to stay operational at all times. Simulations using different threshold frequencies below 50 Hz were then undertaken. Histograms of five of these thresholds are shown in Figure 15. From these histograms it is easy to see which threshold frequencies will allow the laptop to remain operational, and details of the range of expected charge percentages are also readily apparent. For instance, a threshold frequency of 49.955 Hz always keeps the laptop in the charge range 60-100%. The best compromise with the Dell seemed to be to use a threshold of 49.975 Hz. Figure 16 shows a time series of the charge level in the laptop, simulated over a period of 21 days using this particular threshold. It is clear to see that the laptop charge does not fall below 40%, and typically stays within the 70-100% charge level, allowing the user to disconnect from the mains if required and still have an adequate charge level for laptop use on the move. data obtained using this method was useful for quantifying the demand on the scale of a week (the length of time the program was operational), but it was not ideal for measuring the effects of longer demand variations due to the need to log data in real time. It was therefore favourable to obtain historical data so as to be able to demonstrate the long term variations in demand. At the beginning of 2012, the National Grid made available demand data to half-hour resolution for all of 2011. A visualisation of this data can be seen in Figure 1. From this it was possible to then obtain a figure for the average power demand over the course of the year, which could be used to calibrate future models. Figure 16: A time series of the simulated charge levels over a period of 21 days using a threshold frequency of 49.975 Hz. The charge level does not drop below 40% and typically stays in the 70-100% region. 8.2.1 8 Large Scale Impact of Smart Charging Laptop 8.1 Quantifying UK Grid Demand Initially, in order to quantify the UK demand, a computer program was written which logs the UK demand to five minute resolution (and UK grid frequency to 3 minute resolution) as reported by the National Grid website. The 8.2 Entirely Renewable Grid Due to the lack of substantial, high resolution data available regarding the fluctuations that would be present in a grid powered solely from renewable energy sources in the UK, such fluctuations can only be simulated. In order to simulate this future grid, some assumptions must be made as to the form such a grid would take. The arbitrary assumption was made that the grid would be powered solely from wind and solar energy sources, and that the split would be equal between the two, such that the average production from each source equalled half of the average demand. Demand Model A model of the demand in 2011 was produced using the data obtained from the National Grid, with a resolution of 30 minutes. It was possible to use 2011 data as long as it can be assumed that demand will not differ greatly in 2020 from that in 2011. From Figures 1 and 2 it is clear to see that demand is greatest during the day around dinner time and lowest during the early hours of the morning. Throughout the year, the demand is greatest in winter and least in summer. On average, demand for 2011 was 35 GW. At its peak it was 55 GW, and at its lowest it was 19 GW. 15 Figure 15: A set of histograms of laptop charge simulations over 21 days, each with a different frequency threshold. This shows the amount of time spent in each charge percentage range for each threshold value, in steps of 10 percentage points. It is clear to see that a threshold of 49.975 Hz or less would ensure that the connected laptop keeps at least 40% of its battery charge. 8.2.2 Wind Model 8.2.3 Information with regards to the wind power production in the United Kingdom was not readily available. The Irish electricity grid operators, EirGrid, however, provide details of Irish power production from wind on their website and so it was possible to obtain from them an entire year’s worth of data to 15 minute resolution for 2011. Making the assumption that weather patterns affecting wind, particularly from the Atlantic Ocean, are broadly similar in the UK and Ireland, it was possible to scale the wind power production values obtained from EirGrid to a level at which they would represent the power production required to meet 50% of the average demand in the UK grid (i.e. 17.5 GW). This is equivalent to about 65 GW3 of installed capacity. This assumption is likely to represent the worst case scenario, as one would expect better averaging of wind fluctuations due to the larger land mass in the UK with respect to Ireland. 3 This is equivalent to about 10000 of the largest current wind turbine designs. Solar Model It became apparent that comprehensive solar production data was not available for the UK, and so it was necessary to develop a simulation of such data. Data was obtained for the times of sunrise and sunset on a daily basis for 2011. From this information it was possible to calculate the time of solar noon—the time at which the Sun is highest in the sky directly above the line of longitude on Earth in question—for each day. Additionally, information about the angle the Sun made with respect to Edinburgh on a daily basis was also used in order to correct for the fact that the latitude being considered was not on the equator (and thus receiving the greatest power density), and to account for the seasonal variation of this angle. Both sets of data were obtained from the UK Hydrographic Office [28], and the Sun angle data included corrections for atmospheric extinction. The solar intensity for a given day was then modelled using a Gaussian curve centred around the solar noon. The full width at 5% of maximum was calculated such that it coincided with sunrise and sunset, resulting in the intensity being very low during the night, as 16 intended. It was assumed that sunset and sunrise provide less than 5% of the maximum solar noon value, with a small amount of light coming from atmospheric refraction, moonlight, artificial light and other sources. The effect of clouds was then modelled by introducing ten intensity correction factors of varying frequency, which had the effect of reducing the intensity of light incident on the ground on the scale of minutes, hours, days and weeks. Each ’cloud’ had an associated intensity correction factor between 0 and 1, which is multiplied with the intensity value associated with the solar model for a given time step to produce the actual intensity of light through the cloud. Each cloud had a different correction factor, had an effect for different lengths of time and recurred on a different frequency. The result was that it was clear to see significant drops in intensity throughout the day as these ’clouds’ covered the Sun in the sky, which allowed the model to better represent reality. Finally, the actual power generated P from a square meter A of solar panels was calculated using Equation 4. This used the model’s calculation for solar intensity for a given time, I, along with the power of sunlight incident on the equator (�, the solar constant, roughly 1.3×104 kW/m2 ) and the efficiency of the solar cells, η, which was assumed to be 40%4 . The area of panels was scaled in the same way as the wind model so that the average power produced in a year was equal to half of the average demand for a year. P = η�IA (4) As can be expected, the intensity of the Sun as reported by the solar model is greatest during the summer months, when the Northern Hemisphere is tilted towards the Sun, and greatest during solar noon each day when the sunlight is as close to perpendicular as it can be to the ground it is incident upon. In any particular day, clouds affect the power production to different degrees. A typical three day period modelled for September 2011 can be seen in Figure 17. 4 Current commercial solar panels only offer an efficiency of about 10%, but prototypes have been shown to demonstrate an efficiency upwards of 40% [29], which is why this figure was used for the 2020 model. Figure 17: Solar model of three days in September 2011 using data regarding the angle of the Sun with respect to Edinburgh, with the effect of different clouds factored in. The intensity axis is with respect to the intensity at the equator. 8.2.4 The Combined Model The wind and solar models combined would provide, on average, enough power to exactly meet demand for a year like 2011. In order to combine the two energy sources in the renewable grid model, and to then have a discrete value for the surplus or deficit supply power for each time interval throughout 2011, it was necessary that each set of data used the same resolution. The solar model was created in such a way that it could generate a data set with any resolution necessary as long as the set of cloud frequencies were recalibrated to remain realistic. The wind data available from EirGrid was of 15 minute resolution, and the demand data from National Grid was of 30 minute resolution. Both the demand and wind models could be interpolated so as to fit the 15 or 30 minute resolution of the other, but it was decided that it would be better to use as high a time resolution as possible5 . It was also reasoned that artificially increasing the resolution of the demand model would result in less loss of information than in decreasing the resolution of the wind model by averaging neighbouring values, due to the diversity of demand, and because the various supply ’spikes’ in the wind supply ’signal’ appeared to be of higher frequency than the spikes in the demand ’signal’. This is to say, the wind supply model is more stochastic than the demand model. The demand model was interpolated by a 5 A 15 minute resolution would allow more short term fluctuations to be seen than a 30 minute resolution would. 17 factor of 2 in order to achieve the required resolution. The time vector used to specify the 30 minute intervals for the demand model had extra time values inserted in between the existing values. A linear interpolation was then used to calculate the respective demand values for the new time intervals. The resulting set of wind, solar and demand models were then combined into one model, with the wind and solar supplies being subtracted from the demand for each time interval. This provided a new set of power values denoting the deficit or surplus of power supply for each time interval. This new model can be seen for a particular day in Figure 18. ure 20. In this situation, low winter solar power production coincided with a period of low pressure weather, causing a lack of wind along with cold temperatures, creating a huge deficit in supply. Figure 19: A histogram of the periods the renewable grid model spends in deficit (i.e. not enough supply to cover demand). For the most part, these periods are of the order an hour, but there are some periods of significantly longer deficit which must be considered as part of the solution to the problem of mitigating power fluctuations in a renewable grid. Figure 18: An example of the power surplus and deficit throughout a typical day in February 2011 using the combined renewable grid model. The green shaded areas are times when there is more production than demand, and the red shaded areas are times when there is too little production for demand. The right hand side of the image shows a power axis in terms of conventional 500 MW power plants, thus around 7:30pm on this day, the deficit would require 40 of these power plants in the ab- Figure 20: An example of a rare period where the sence of mitigation techniques. renewable grid model is at a deficit for a period greater than around 5 hours. The deficit here lasts The largest single deficit in the model was 47 hours, from around 2pm on the first day to 61.9 GW which occurred on 21st Jan 2011 at around 1pm on the third. The deficit region in 5pm. The wind and solar power production at question is shaded red. A few hours after the deficit period ends, the renewable grid model is back in that time was 0.87 GW and 0.08 GW, respec- deficit again as solar noon comes and goes, which tively. Occasions like this where peak demand would leave little time for buffers to replenish. coincides with darkness and lack of wind are common so long as sunset occurs before peak demand. On this particular day, sunset occurred at 4:22pm. A histogram of the periods of deficit is shown in Figure 19. This shows that deficits typically last less than 2 hours. However, there are some longer periods, and there are two occasions in the year where the period of deficit lasts significantly longer. The longest deficit lasted 46h45m, occurring between January 16th and January 18th 2011. This can be seen in Fig- 8.3 Entirely Renewable Grid with Dynamic Demand Laptop Charging In order to quantify the effect that the dynamic demand laptop control system would have on an entirely renewable grid as a whole, it was necessary to simulate the effects of a large number of laptops using this system. To achieve this, it was necessary to combine the parts of the project as presented in Sections 18 8.2 and 7. Ideally, in a situation like the one shown in Figure 18, the area under the curve during periods of supply surplus would cancel the area under the curve during supply deficit, as shown in Figure 21. Figure 21: An example of the situation shown in Figure 18, but with some form of mitigation being used to cancel the surplus and deficit, just like in Figure 7. 8.3.1 Large Scale Laptop Model Assumptions 8.3.2 Simulation Mechanism In the model, 1000 entities represent 20000 laptops each, so as to simulate 20 million. The model was designed in this way for performance reasons6 . For each time step, which took the form of each 15 minute period in 2011 as defined by the renewable grid model, each of the 1000 laptops had its battery charge level calculated based on its previous value and the state of the grid (i.e. whether or not the frequency was below or above 50 Hz). If the battery had to charge, then the power the laptop used in doing so was added to the renewable grid model’s total demand for the next time step. Otherwise, the value was subtracted (representing an occasion where the laptop uses its buffer). Finally, the new grid model was obtained by running this model using the desired data, either for the whole year or for individual days. Figure 22 shows the effect that the laptop grid model has on a particular day, in terms of the power successfully mitigated from periods of supply deficit. At about 6:30am, the grid was running at a surplus and so this extra power was used to charge the 20 million laptops’ batteries. This resulted in a reduction in the amount of surplus being produced, as of course the laptops had used some of this power. After about 45 minutes, the grid stopped running at a surplus and instead entered a period of power deficit. The 20 million laptops then started using their batteries, removing their power demand from the grid, and so the total amount of power deficit was reduced. This whole process repeated three more times later in the day, and to a larger degree. The slopes at the top of the mitigation peaks are due to some of the 20 million laptops running out of charge. Clearly, longer periods of deficit would result in almost all of the 20 million laptops running out of charge, so this technique cannot mitigate all occurrences of grid deficits, particularly those lasting longer than about 4 hours. Notice in Figure 22 that the power scales on the vertical axes are of different orders of magnitude. Clearly, the effect of 20 million laptops In order to create a large scale laptop use model, some assumptions had to be made. It was assumed that there were around 20 million laptops connected to the mains and capable of using this system in the United Kingdom, and that each laptop had a certain usage pattern throughout the day. Laptops have different charge and discharge rates, and different battery lives, which was also taken into account. The assumption was also made that at the start of the simulation, laptops will have a random level of charge between 0 and 100%. The charge and discharge rates, battery lives and laptop use times were calibrated using pseudorandom numbers which, on average, form a Gaussian distribution. The Gaussian waist and standard deviation for the laptop use times were defined so as to have a mean of 6 hours, but with a range of time lengths between 3 and 9 hours falling within one standard deviation, to represent the different ways in which people use laptops. Similarly, for the operational times, the time half way through the laptop’s use in one day was again defined using a Gaussian distribution, with the midtime falling around midday. The charge rates 6 The size of the matrix representing the model is were distributed around the value for the Dell about 1 × 108 (about 800 megabytes of double precision laptop. Histograms of these calibration factors floating point variables) as opposed to about 2 × 1012 can be seen in Appendix D. (about 15 terabytes) by doing this! 19 Figure 22: Power mitigated using the laptop grid model during a day in July. Periods of grid surplus are used to charge the laptops’ batteries (regions shown in green), and periods of grid deficit are mitigated by making the laptops use their batteries instead of the mains (regions in red). ’unplugging’ is small in comparison to the entire grid. However, the amount of power mitigated in this case was still significant—about 500 MW at its peak—equivalent to a large power plant. Over the whole year, the greatest effect of mitigation was 764 MW, equivalent to about 1.5 power plants. The average power mitigated when the grid was in a supply deficit was 237 MW, equivalent to about half a power plant. 9 Discussion 9.1 Grid Models and Simulations 9.1.1 Wind and Solar Models The assumption of 50% production from wind power in the wind production model is equivalent to about 65 GW of installed capacity. David MacKay [30] estimates the power per unit land area of a wind turbine to be about 2.2 W/m2 . Therefore, to achieve this installed capacity, the area of land required would be about 29000 km2 , or about 12% of the total land area of the UK (244000 km2 ). Whether this figure sounds reasonable or not is up for debate. While land containing wind turbines can be used for other purposes, such as farming, there are objections along the lines of noise, aesthetics and other reasons. A discus- sion of these issues can be found in the MacKay book. The area of solar panels required to produce the 50% equivalent supply would total about 693 km2 . This figure is reliant upon the cloud simulation characteristics, and is thus only an empirical estimate. The model was created with the assumption that the solar panels would have an efficiency of 40%, which is equivalent to the performance of the most cutting edge prototypes today, and not the standard mass production designs in 2012. A more realistic figure might be about 10%, meaning the area required for the same production would be quadruple. However, there is little to stop the use of solar panels on the same land as wind turbines, except to remove the ability to use the land for farming and other means. Wind turbines cast little shadow and solar panels don’t affect the wind above a few metres off the ground. The MacKay book estimates the area of south facing roof space in the UK available for solar panels to be about 10 m2 per person, or about 600 km2 in total. Therefore, covering every roof in the country and using some dedicated spaces for solar panels would theoretically allow for this production. Again, whether this is plausible is up for debate. In the solar model, the difference between the average production in summer and winter is huge—about 40 GW, equivalent to the aver- 20 age demand in the year. The purpose of this project was to look at ways of mitigating power fluctuations without having to resort to methods such as those outlined in Section 4. If the long term difference between summer and winter output from solar necessitates the use of a set of standby power plants to make up the difference, this would counteract the effect that the mitigation techniques would have in reducing the need for these power plants. Clearly a renewable grid would need to be carefully designed in order to prevent the need to resort to using the ’old ways’ for parts of the year, and relying so much on solar power is probably not the way forward. The two renewable energy source models – the wind and solar models – were produced using some assumptions as to how they would perform as an energy source. These assumptions are accurate enough to obtain ’ball park figures’ for the kind of outputs these sources might produce throughout a year, but they are not accurate enough to be used for any sort of precision modelling. The models can be improved in many respects by using higher resolution underlying data, more accurate calibration factors such as cloud effects by basing them on real observations, and by factoring in regional geographical variations which affect the ability to produce electricity. 9.2 Laptop Controller Experiment The differential band pass filter method used to obtain the state of the grid is very effective, with accuracy to within about 1 mHz. Noise present in the signal is also suppressed in the stage which subtracts one copy from the other. Although the method provides information as to whether the frequency is above or below the chosen threshold, it does not give an estimate as to how far away the frequency is from this point. Such information would be useful for differentiating between slight and major shortfalls in demand. If this information was available, it would be possible to vary the threshold at which the laptop system would stop taking power from the grid. For instance, if the laptop’s battery was below, say, 50% full, the control system could decide to only stop the laptop from taking power from the grid when the threshold fell below a lower threshold, ef- fectively helping the grid only when the grid really needs the help. The results regarding the laptop controller are proofs of concept. The best frequency threshold for a particular laptop depends on a number of variables, including the battery life, rates of charge and discharge, and whether or not the laptop has a high CPU load when in use. A desirable situation would be where the user decides the threshold they wish to use, taking account of their personal needs. A laptop that stays ’plugged in’ to a wall socket at all times can safely use a high threshold, whereas a laptop that is frequently used on the move will need to retain a high level of battery charge level at all times for such circumstances. Similarly, a fail safe mechanism would need to be built in to end user systems to prevent long periods of frequency below threshold from draining the laptop battery entirely, with the potential for data corruption and loss of productivity. 9.2.1 Long Term Effect on Laptop Batteries An issue that has not been considered in this study is the long term effect frequent charge and discharge cycles can have on laptop batteries. When a lithium-ion cell charges, typical of those used in modern laptops, deposits are formed inside the electrolyte, which increases the internal resistance of the battery and decreases its ability to store energy over time. This means that frequent charging will limit the life of the battery. Tests of a lithium-ion cell showed that the battery life is influenced mainly by the conditions in which the battery is charged, and not so much the discharge conditions [31]. In particular, over-charging, where the battery is charged beyond its design capacity, is a large factor in the degradation [32]. This means that it would be reasonable to build some logic into the controller which takes into account the conditions in which the laptop would be charging. These conditions could be elements such as the ambient temperature, current charge level (for instance, whether or not to charge if the battery is already almost full), or other factors. 21 9.3 Project as a Whole For certain periods of power deficit of the order a couple of hours long, it is not inconceivable to suggest that this kind of deficit could be dealt with through some forms of demand-side management such as the laptop control system outlined as part of this study. The particular laptop system developed in this study was able to mitigate an average of 237 MW of demand when used by 20 million devices, which represents about 0.6% of the total average UK demand. If all of our electronic devices, such as lights, heating and cooling systems, electric cars, ovens, desktop computers, televisions and so on implemented a similar form of demandside management, the power available for mitigation at any one time would become even greater. A combination of these devices shutting down in tandem might be enough to mitigate some of the fluctuations that a grid powered from renewable energy sources might create, and thus there would no longer be the need to build so many cost intensive ’peaking’ power plants7 . However, demand-side management techniques are not quite so useful for longer periods of supply deficit, and in a grid powered from a high proportion of volatile renewable energy sources such as wind, such periods of supply deficit will occasionally occur. Again, if the grid is to involve a greater amount of energy sources such as wind, models must be in place to deal with the loss of production during periods of low output. In particular, it is worth focusing on the scenarios where long periods of low wind coincide with high periods of demand, such as in winter and in the evening. 9.4 Future Work crocontroller could also be used in other applications. The models developed as part of this study can be improved as outlined in the discussion. With higher resolution data, it could be possible to undertake different forms of analysis, such as looking at the fluctuations from a frequency domain perspective. This method would have the potential to provide information about the magnitude and frequency of different fluctuations present in the renewable grid model, which would be useful for better coordinating mitigation techniques. 10 Conclusions The concept for power fluctuation mitigation using laptops as outlined in this study shows promise. Although the simulations by no means fully represent reality, which is vastly more complex than the basic assumptions made in the model allow, the concept has been clearly demonstrated to have a positive effect. This technique, along with some of the conventional methods of mitigation, could stand the National Grid in good stead to face the challenges of the future, in terms of quality of electricity service, environmental impact (the potential for a reduction in traditional gas and coal fired plants) and cost (not having to pay upwards of £7M a month for industries to temporarily cease electricity consumption). In the future, the concepts developed as part of this study can be applied to other forms of buffers, which, working in tandem, can provide a valuable means of demand reduction during times of supply deficit. References In the future, it would be worth exploring ways in which to miniaturise the laptop control system. The system developed in this study involved the use of a large server class computer, which is obviously useless for this application in practice. The best method for laptops would likely be a microcontroller build in to the charger’s transformer box. Such a mi7 A typical new 500 MW nuclear power plant costs of the order £1.5B to build [33]. 22 [1] Parliament of the United Kingdom. Climate Change Act, 2008. [2] Scottish Government. 2020 Routemap for Renewable Energy in Scotland. http://www.scotland.gov. uk/Publications/2011/08/04110353/2. [3] The European Parliament. 2009/28/EC, April 2009. Directive [4] IEEE. IEEE Recommended Practice for [16] P. Kundur, N.J. Balu, and M.G. Lauby. Electric Power Distribution for Industrial Power System Stability and Control. The EPRI Power System Engineering Series. Plants. ANSI/IEEE Std 141-1986, 1986. McGraw-Hill, 1994. [5] EirGrid. Wind generation system performance data. http://www.eirgrid.com/ [17] O.I. Elgerd. Electric Energy Systems Theoperations/systemperformancedata/ ory: An Introduction. McGraw-Hill, 2007. windgeneration/. [18] National Grid. National Grid: Inter[6] National Grid. Electricity connectors. http://www.nationalgrid. Real Time Operational Data. com/uk/interconnectors/. http://www.nationalgrid.com/uk/ Electricity/Data/Realtime/, 2011. [19] James Oswald, Mike Raine, and Hezlin Ashraf-Ball. Will British weather pro[7] Janet Ramage Godfrey Boyle, Bob Evvide reliable electricity? Energy Policy, erett, editor. Energy Systems and Sustain36(8):3212 – 3225, 2008. ability: Power for the Sustainable Future. Oxford University Press, 2003. [20] Tai-Lang Jong Chi-Min Chu. A Novel Direct Air-Conditioning Load Control [8] National Grid. ForecastMethod. Power Systems, IEEE Transacing Demand. http://www. tions on, 23(3):1356 – 1363, August 2008. nationalgrid.com/NR/rdonlyres/ 1C4B1304-ED58-4631-8A84-3859FB8B4B38/ [21] Chang-Yu Huang, J.T. Boys, G.A. Covic, 17136/demand.pdf. J.R. Lee, and R.V. Stebbing. Implementation and Evaluation of an IPT Bat[9] Godfrey Boyle. Renewable Energy. Oxford tery Charging System in assisting Grid University Press, 2004. Frequency Stabilisation through Dynamic [10] I. Harrison and A. Marr. Britain from Demand Control. In Vehicle Power Above. Pavilion, 2009. and Propulsion Conference (VPPC), 2010 IEEE, pages 1 – 6, September 2010. [11] National Grid. Frequency Control by Demand Management (FCDM). [22] M. Ifland, N. Exner, and D. Westermann. http://www.nationalgrid.com/uk/ Appliance of Direct and Indirect Demand Electricity/Balancing/services/ Side Management. In Energytech, 2011 frequencyresponse/fcdm/, 2011. IEEE, pages 1 – 6, May 2011. [12] J.A. Short, D.G. Infield, and L.L. Freris. [23] Department of Energy and Climate Stabilization of Grid Frequency Through Change. The Potential for Dynamic DeDynamic Demand Control. Power Sysmand. Technical report, Department of tems, IEEE Transactions on, 22(3):1284 Energy and Climate Change, November – 1293, August 2007. 2008. URN 08/1453. [13] National Grid. Monthly Balancing Services Summary, September 2011. [24] National Grid. Operating the Electricity Transmission Networks in 2020. Technical http://www.nationalgrid.com/uk/ report, National Grid PLC, June 2009. Electricity/Balancing/Summary/, 2011. [25] J. Platts. Electrical Load Management: The British Experience. IEEE Spectrum, [14] Parliament of the United Kingdom. Elec16(2), February 1979. tricity Act, 1989. [15] Nick Jelley John Andrews. Energy Science [26] M. A. Fischetti. Electric Utilities: Poisted - Principles, Technologies, and Impacts. For Deregulation? IEEE Spectrum, 23(5), Oxford University Press, 2007. May 1986. 23 [27] Residens Project. http://www. tu-ilmenau.de/fakmn/9999+ M54099f70862.0.html. [28] UK Hydrographic Office. Her Majesty’s Nautical Almanac Office. http://astro. ukho.gov.uk/. [29] Martin A. Green, Keith Emery, Yoshihiro Hishikawa, Wilhelm Warta, and Ewan D. Dunlop. Solar cell efficiency tables (version 39). Progress in Photovoltaics: Research and Applications, 20(1):12–20, 2012. [30] D.J.C. MacKay. Sustainable Energy Without the Hot Air. UIT, 2009. [31] Soo Seok Choi and Hong S Lim. Factors that affect cycle-life and possible degradation mechanisms of a Li-ion cell based on LiCoO2. Journal of Power Sources, 111(1):130 – 136, 2002. [32] L. S. Kanevskii and V. S. Dubasova. Degradation of Lithium-Ion batteries and how to fight it: A review. Russian Journal of Electrochemistry, 41:1–16, 2005. [33] Nuclear Engineering International. How much? http://www.neimagazine.com/ story.asp?storyCode=2047917. 24 Appendices A Interconnector Power Imports/Exports Sample The UK grid has interconnectors with neighbouring grid networks. Within the British Isles, there are interconnectors between Northern Ireland and Scotland, Northern Ireland and the Republic of Ireland, the Isle of Man and England, Scotland and England, and North England and South England. Outwith the UK, there are interconnectors between the UK and France (2 GW), and the UK and the Netherlands (1 GW). There are also preliminary plans to create an interconnector between the UK and Norway (see http: //en.wikipedia.org/wiki/Scotland-Norway_interconnector and http://en.wikipedia. org/wiki/HVDC_Norway-Great_Britain) by 2020. Figure 23 shows the French and Dutch interconnectors in use. Figure 23: National Grid interconnector imports and exports on 24th October 2011. Throughout the day the UK buys power from France and the Netherlands, except for a period in the morning when the UK sells power in the other direction. Interestingly, there is a period during the day when the UK grid imports power from the Netherlands and exports power to France at the same time. B National Grid Website Logger Program B.1 Initial Program Real time data is published on the National Grid’s website, on this page: http://www. nationalgrid.com/ngrealtime/realtime/systemdata.aspx. The data available is demand, frequency, and system transfers (electricity export/import) between Scotland/England and North England to South England, and transfers from the interconnectors (Northern Ireland, France and the Netherlands). This data is of limited resolution. The program8 developed uses regular expressions in order to extract the figures for demand and frequency from the HTML source code behind this web page. 8 SVN Path: /trunk/GridAnalyticsProgram/ 25 B.2 Refined Frequency Program High resolution frequency data, along with a wealth of other information, is published by the National Grid on their dedicated statistics website, BM Reports (http://www.bmreports. com/). The frequency data is published in both XML and CVS format, and can be found in the ’Electricity Data Summary’ page within the ’General’ tab. The program9 developed downloads an XML file periodically and adds the data it doesn’t already have to its logs. C Battery Logging Shell Scripts C.1 Ubuntu Linux Install the battery-charge-graph package from the official Ubuntu repository, either using the Software Centre or by typing the following in a terminal: sudo apt-get install battery-charge-graph Then start the logging by typing the following in a terminal: sudo battery-stats-collector -o batterylog.txt Where batterylog.txt is the name of the file you wish to use for the log, in the current working directory. Leave the terminal running the command open for as long as you wish to log the battery charge level. C.2 Mac OS X Create a script file such as batterylog.sh and type the following into it: #!/bin/bash cd ~/Desktop/ date >> battery.txt; while true; do ioreg -w0 -l | grep Capacity >> battery.txt sleep 15 done Save the file and make sure it has execute permissions, either by altering its properties in the GUI or by running the following terminal command: chmod +x batterylog.sh Where batterylog.sh is the name of the script file you created. The script will log some information regarding battery state from the Mac OS X kernel to a file called battery.txt. Dividing the value for ’current’ with the value for ’capacity’ will yield the current battery charge level between 0 and 1. Note that the program will sleep for 15 seconds during each iteration. During subsequent analysis, it will be necessary to add a correction factor to this 15 second interval due to the small but significant time it takes to execute the program on each iteration. During the analysis in this experiment, the correction factor was found to be about 0.5 s. 9 SVN Path: /trunk/FrequencyLoggerProgram/ 26 D Laptop Grid Model Assumptions (a) Charge Rates (b) Initial Charge Levels (c) Times of Mid-Use (d) Operation Times Figure 24: Histograms of laptop grid model assumptions 27