1 Inverter performance prediction L. Kaci1, A. Hadj Arab2, R. Zirmi1, S. Semaoui2 and S. Boulahchiche2 1 Centre de Développement des Energies Renouvelables, CDER, BP 62 Route de l'Observatoire, Algiers Algeria 2 Université Mouloud Mammeri, Tizi Ouzou, Algeria Abstract— based on the outdoor PV system of the CDER at the height of Algiers the inverter efficiency under field conditions has been analyzed and compared to manufacturer data. A model is developed that expresses the inverter efficiency as a function of the inverter AC output power. Model parameters are calculated using field data. The model is used to predict inverter efficiencies for different periods of time and compared to the measured ones. Index Terms—Inverter electrical conversion efficiency, PV system I. INTRODUCTION T he cost reduction in photovoltaic technology coupled to environmental concerns, led to an important increase in the installed PV systems worldwide, it shows the PV technology will play an important role in the energy mix. The additional installed photovoltaic PV capacity worldwide (98 GW) an increase of up to 29 % compared to 2016, bringing the cumulative installed capacity to 402.5 GW [1]. The estimation of the payback time and the profitability of grid-connected photovoltaic (PV) systems require information about performance of PV modules and inverters. For this reason, the attention is strongly paid on the efficiency. The performance of a PV system depends on the performances of its components namely, PV modules and inverters. Energy losses are present in all solar energy conversion process. Specifically, losses can be categorized into pre-module losses, module losses and system losses. A PV system comprises three main units: the PV generator unit, the string combiner unit and the power conditioning system. The PV generator unit is the PV array. The string combiner unit includes the connections and wiring between strings as well as the fuses/block diodes, which are employed for string overcurrent protection. Finally, the main parts of the power conditioning system are the inverters, AC connections and wiring. All three units are associated with operational losses [2] Increasing the performance of PV systems is one of the main goals of research [3]; to increase the amount of electricity produced per Wp installed PV capacity. To achieve such goal we need to understand power losses that occur in the installed PV systems. Losses due to the energy conversion by PV cell need a technology breakthrough in the PV components to be reduced, other losses are unavoidable such us shading. Inverters are key components in a grid connected PV Submission date April 5th 2020 system; they influence system performance to a great extent, a small improvement in their efficiency will reduce the overall cost of the system significantly. PV inverters consists of variety of components such as power semiconductors, sensors, resistors, magnetics and control circuits those components introduce an amount of power loss dissipated as heat in the inverter. Figure 1 shows classification of the power losses of the power converter [2]. All of the inverters manufacturers still use the calculation method of the EURO efficiency in their datasheets [4], which is a weighted conversion efficiency and originally were taken for the location of North Germany and still represent a so called standard on the international level. The Present work will deal with inverter electrical efficiency measurement and modeling. A model is developed; its parameters calculated using field data and applied to data collected under real operation conditions. Fig. 1: Classification of the power losses of the power converter. II. PV SYSTEM DESCRIPTION AND DATA COLLECTION The PV system is located in the Centre de Développement des Energies Renouvelables (CDER), Algeria. The main purpose of the PV system facility is to learn how different modules perform with a variety of inverters and to learn about the performance of grid-tied photovoltaic systems. The PV system is formed by 90 PV modules (Isofoton 106Wp-12 at STC) divided in three sub-generators of 3 kWp each. The sub-generators are formed by two parallel strings of 15 PV modules in series. Each sub-generator is connected to a single phase inverter of 3 kW (SMA/ Sunny Boy 3000TL see table I for the specifications) that injects the generated energy into a phase of the public low voltage distribution network of the National Company (SONELGAZ). The block diagram of this PV system is shown in Figure. 2. 2 TABLE I SMA/ SUNNY BOY 3000TL SPECIFICATIONS DC input Fig. 2: Schematic diagram of the CDER grid connected PV system. The CDER PV system is fully instrumented which allow performance analysis of both PV array and inverters, it is monitored at five minute intervals, five-minute average performance data was recorded for the CDER PV system for up to one year of operation. III. INVERTER EFFICIENCY MODELS Inverter efficiency calculation and modeling is very important for the following reasons: Optimum yield of expensive PV energy Criteria for purchase decision Indicator for innovation capability of the inverter manufacturer It can be used in conjunction with a photovoltaic array performance to calculate expected system performance. Inverter manufacturer’s published data generally lists the conversion efficiencies under optimal conditions; figure 3 shows measured inverter efficiencies versus output power at three different input voltages VDC of the SMA/ Sunny Boy 5000TL PV inverter (Based on manufacturers data). Conversion efficiencies under optimal conditions can be misleading as the optimal efficiencies are not maintained over the whole range of operation. Inverter performance models have been proposed to continuously monitor efficiencies that may indicate need for repair or maintenance. Maximum DC power at cos φ = 1 Maximum DC voltage* MPP voltage range at AC nominal power DC nominal voltage Minimum DC voltage Start voltage, adjustable Maximum input current Maximum input current per string Number of MPP trackers Strings per MPP tracker AC output AC nominal power at 230 V, 50 Hz Maximum AC apparent power Nominal AC frequency Nominal AC current at 220 V / 230 V / 240 V Maximum AC current Total harmonic distortion of output current at AC THD voltage < 2 %, AC power > 0.5 AC nominal power AC voltage range* Nominal AC frequency* Operating range at nominal AC frequency 50 Hz Operating range at nominal AC frequency 60 Hz cos φ at nominal AC power 3,200 W 550 V 188 V … 440 V 400 V 125 V 150 V 17 A 17 A 1 2 3,000 W 3,000 VA 220 V / 230 V / 240 V 13.6 A / 13 A / 12.5 A 16 A ≤3% 180 V … 280 V 50 Hz / 60 Hz 45 Hz … 55 Hz 55 Hz … 65 Hz 1 Supply phases 1 Connection phases 1 Overvoltage category** III * Depends on country configuration ** Voltage surge resistance in accordance with IEC 60664-1 Power losses in the inverter Ploss and conversion efficiency of the inverter ηinv are calculated from two fundamental measurements namely the actual power delivered to the output, Pac, and the input power from the generator Pdc. There are various mays to express the relationship between these variables. 𝜂𝑖𝑛𝑣 = 𝑓(𝑃𝑎𝑐 , 𝑉𝑑𝑐 ) 𝜂𝑖𝑛𝑣 = 𝑓(𝑃𝑑𝑐 , 𝑉𝑑𝑐 ) 𝑃𝑙𝑜𝑠𝑠 = 𝑓(𝑃𝑎𝑐 , 𝑉𝑑𝑐 ) 𝑃𝑙𝑜𝑠𝑠 = 𝑓(𝑃𝑑𝑐 , 𝑉𝑑𝑐 ) 𝑃𝑎𝑐 = 𝑓(𝑃𝑑𝑐 , 𝑉𝑑𝑐 ) (1) Where 𝑉𝑑𝑐 in the output voltage from the generator. Fig. 3. Example of efficiency data available from the CEC. These curves are for SMA/ Sunny Boy 5000TL A. Euro model: Efficiency varies for an inverter depending on the input power from the generator Pdc; so instead of looking at the efficiency when the inverter is operating at its rated capacity; calculating weighted efficiency requires selecting a few DC input levels relative to the inverter’s rated capacity. European (EU) efficiency and California Energy Commission (CEC) inverter efficiency, these types of efficiency are ‘weighted’ efficiencies. 3 European efficiency is widely used all over the world for grid-connected inverter comparison; it is weighted conversion efficiency. The European conversion efficiency combines the weighed inverter efficiency at six points of operation, at nominal DC voltages. 𝜂𝐸𝑈𝑅𝑂 = 0.03 ∙ 𝜂5% + 0.06 ∙ 𝜂10% + 0.13 ∙ 𝜂20% + 0.10 ∙ 𝜂30% + 0.48 ∙ 𝜂50% + 0.20 ∙ 𝜂100% (2) Where 𝜂𝑖% is the conversion efficiency at i% of the inverter output rated power [5]. The European efficiency is mainly used as a reference for comparison between different inverters; it is not used for continuous monitoring of inverter efficiency. B. SANDIA model: Inverter efficiency, ηinv, derives from the ratio between the actual power delivered to the output, Pac, and the input power from the generator Pdc. The difference between Pdc and Pac, Ploss is converted to heat inside the inverter. The quadratic function 2 𝑃𝑙𝑜𝑠𝑠 = 𝑎0 + 𝑎1 . 𝑃𝑎𝑐 + 𝑎2 . 𝑃𝑎𝑐 (3) Fig. 4: Field test results for a 3-kWp SMA/ Sunny Boy 3000TL inverter recorded during system operation at CDER C. DRIESSE model: Sandia model uses true voltages and power values which will lead to parameters values that span several orders of magnitude. To overcome this problem Driesse& al proposed normalization of the measured data. Power can be normalized to the nominal maximum rating, 𝑃𝑛𝑜𝑟𝑚 (usually output power) [6]; and input voltage can be normalized to a nominal input voltage, 𝑉𝑛𝑜𝑟𝑚 giving the form: 𝑃𝑙𝑜𝑠𝑠 𝑃𝑛𝑜𝑟𝑚 Provide a good fit for empirical data [6] Efficiency of the inverter is calculated as follows 𝜂𝑖𝑛𝑣 = 𝑃𝑎𝑐 /(𝑃𝑎𝑐 + 𝑃𝑙𝑜𝑠𝑠 ) (4) Sandia model expresses 𝑃𝑎𝑐 as a function of 𝑃𝑑𝑐 𝑃𝑎𝑐 = {( 𝑃𝑎𝑐𝑜 𝐴−𝐵 ) – 𝐶 ⋅ (𝐴 − 𝐵)} ⋅ (𝑃𝑑𝑐 − 𝐵) + 𝐶 ⋅ (𝑃𝑑𝑐 – 𝐵)2 (5) Where 𝐴 = 𝑃𝑑𝑐𝑜 ⋅ {1 + 𝐶1 ⋅ (𝑉𝑑𝑐 − 𝑉𝑑𝑐𝑜 )} 𝐵 = 𝑃𝑠𝑜 ⋅ {1 + 𝐶2 ⋅ (𝑉𝑑𝑐 − 𝑉𝑑𝑐𝑜 )} 𝐶 = 𝐶𝑜 ⋅ {1 + 𝐶3 ⋅ (𝑉𝑑𝑐 − 𝑉𝑑𝑐𝑜 )} We can see from the equation (3) above that it is a quadratic function just like equation (1). 𝑃𝑑𝑐𝑜 is the rated DC power, 𝑃𝑎𝑐𝑜 the rated AC power and 𝑃𝑠𝑜 the self-consumption are directly for the inverter specification data sheet [6]. Co is a parameter defining the curvature (parabolic) of the relationship between ac-power and dc-power at the reference operating condition, default value of zero gives a linear [6]. The performance model was very effective in fitting measured data for inverters Xantrex GT3.8 and PVP3200 with a standard error in measured versus modeled efficiency of about 0.1% [7]. And the model was also very effective with SMA/ Sunny Boy 3000TL [1]. Relations between 𝑃𝑑𝑐 vs 𝑃𝑎𝑐 and 𝑃𝑑𝑐 vs 𝜂𝑖𝑛𝑣 are illustrated in fig 4 using field data of the CDER PV system with hundreds of measurements recorded over a 5-days period from September 1st 2018 to September 5th 2018. 𝑃𝑑𝑐 = 𝑓(𝑃 𝑛𝑜𝑟𝑚 𝑉𝑑𝑐 ,𝑉 𝑛𝑜𝑟𝑚 ) (6) This model is essentially the double quadratic model expressing this in normalized form, and using voltage deviation from nominal rather than absolute voltage, gives the equation: 𝑃𝑙𝑜𝑠𝑠 = (𝑏(0,0) + 𝑏(0,1) (1 − 𝑣𝑑𝑐 )) + (𝑏(1,0) + 𝑏(1,1) (1 − 𝑣𝑑𝑐 )) 𝑝𝑑𝑐 2 + (𝑏(2,0) + 𝑏(2,1) (1 − 𝑣𝑑𝑐 )) 𝑝𝑑𝑐 (7) This model was successfully tested with the California Energy Commission (CEC) data for 12 different inverters and achieved a very good accuracy[6]. IV. PROPOSED MODEL In field testing the inverter performance data recorded are completely representative of the actual system operating conditions. The two fundamental measurements, to calculate the inverter performance are 𝑃𝑑𝑐 the input power from generator and 𝑃𝑎𝑐 the output power from inverter that are available in field installed PV system. Most of the existing models departs from the quadratic function equation (3) to estimate PV inverter efficiency; our proposed model will use the double exponential as it is clear from 𝜂𝑖𝑛𝑣 versus 𝑃𝑎𝑐 curve (fig 5) that 𝜂𝑖𝑛𝑣 (𝑃𝑎𝑐 ) is of the following form. 𝜂𝑖𝑛𝑣 (𝑃𝑎𝑐 ) = 𝛼1 . 𝑒 𝛽1.𝑃𝑎𝑐 + 𝛼2 . 𝑒 𝛽2.𝑃𝑎𝑐 (7) Where 𝛼1 , 𝛽1 , 𝛼2 and 𝛽2 are constants to be determined. 4 Figure 5 illustrates a field test carried out at the CDER facility with hundreds of measurements recorded over a 6days period. The coefficients 𝛼1 ,𝛽1 , 𝛼2 and 𝛽2 must be found by fitting the function to the measured field data. The fit is done using the least square method, MATLAB software is used. Results are shown in table II TABLE II PARAMETERS OF THE MODEL Parameters Values 𝛼1 𝛽1 𝛼2 𝛽2 0.9885 -1.56E-05 -0.6678 -0.02325 Goodness of fit: SSE: 0.3249 R-square: 0.9426 Adjusted R-square: 0.9423 RMSE: 0.02428 Where : Sum of Squares Due to Error (SSE): It measures the total deviation of the response values from the fit to the response values. The closer its value to 0 the better is the fit. R-Square.: This statistic measures how successful the fit is in explaining the variation of the data, a value closer to 1 indicating a better fit. Root Mean Squared Error: This statistic is also known as the fit standard error. RMSE value closer to 0 indicates a better fit [8]. Fig. 7: Model residual plot V. MODEL VALIDATION This model is tested using the CDER facility PV system using the SMA/ Sunny Boy 3000TL single phase inverter. Model parameters are shown in table II they are calculated using 5-days period data from September 1st 2018 to September 5th 2018. Validation data used are from 5-days period of November and December. Results for November are shown Figure 8, residuals are shown in figure 9 and figure 10 shows a linear regression of estimated efficiency 𝜂𝑖𝑛𝑣 versus measured efficiency. Fig. 8: 𝜂𝑖𝑛𝑣 Versus 𝑃𝑎𝑐 curve estimated and measured 5-days period November Fig. 6: 𝜂𝑖𝑛𝑣 Versus 𝑃𝑎𝑐 curve Model residual plot is shown in Figure. 7 the difference between the measured efficiencies and the predicted efficiencies using the prediction equation (7). As we can see the data points for Pac greater than 100 W the residual is around zero. Model residual plot shown in Figure. 9 describes the difference between the measured efficiencies and the predicted efficiencies using the prediction equation (7) with the calculated parameters in table II. As we can see the data points for Pac greater than 100 W the residual is around zero almost the same as for the September data used for parameters calculation. 5 Fig. 9: 𝜂𝑖𝑛𝑣 measured residual plot (Nov) Fig. 13: 𝜂𝑖𝑛𝑣 estimated versus 𝜂𝑖𝑛𝑣 measured linear regression (December) VI. CONCLUSION Fig. 10: 𝜂𝑖𝑛𝑣 estimated versus 𝜂𝑖𝑛𝑣 measured linear regression November Results for December are shown Figure 11 residuals are shown in figure 12 and figure 13 shows a linear regression of estimated efficiency 𝜂𝑖𝑛𝑣 versus measured efficiency High-resolution (five minutely) performance data, recorded for the CDER PV system over a 1- year period, model has been used to construct simple empirical model of the performance of the SMA/ Sunny Boy 3000TL Inverter during normal operation. This paper describe a new model for inverter electrical conversion efficiency whose parameters are calculated using only measured actual power delivered to the output, Pac, and the input power from the generator Pdc ; measurements that are available in field. Once the inverter model 𝜂𝑖𝑛𝑣 = 𝑓(𝑃𝑎𝑐 ) parameters calculated, they are used to predict efficiency for other periods of the year namely November 2018 and December2018. The model very effective with SMA/ Sunny Boy 3000TL and as future work it will be tested on other types of inverters. . By incorporating inverter performance model in the process of PV monitoring system it will allow users to continuously monitor the inverter efficiency and it’s deviation from the estimated efficiency. This will help in diagnose causes of inverter performance degradation, and facilitate expedient and cost-effective field maintenance as well as need for repair and maintenance [7]. REFERENCES Fig. 11: 𝜂𝑖𝑛𝑣 Versus 𝑃𝑎𝑐 curve estimated and measured 5-days period December [1] [2] [3] [4] [5] Fig. 12: 𝜂𝑖𝑛𝑣 measured residual plot (December) A. Hadj Arab & al“ Analyse des performances des onduleurs du système photovoltaïque connecté au réseau du ”. Revue des Energies Renouvelables Vol. 22 N°1 (2019) 123 – 134 Elena Koumpli“Impact of data quality on photovoltaic (PV) performance assessment” A Doctoral thesis Loughborough University. November 2017. S.K. Firth & al “A simple model of PV system performance and its use in fault detection” Solar Energy 84 (2010) 624– 635. F.P. Baumgartner, “euro realo inverter efficiency: dcvoltage dependency” 20th European photovoltaic solar energy conference, 6-10 June 2005; Barcelona, Spain assimo Valentini “PV Inverter Test Setup for European efficiency, Static and Dynamic MPPT Efficiency Evaluation”. Conference Paper · June 2008 DOI: 10.1109/OPTIM.2008.4602445 · Source: IEEE Xplore 6 Anton Driesse, Praveen Jain “beyond the curves: modeling the electrical efficiency Of photovoltaic inverters” 2008 [7] David L. King & al “Performance Model for GridConnected Photovoltaic Inverters” SANDIA REPORT SAND2007-5036 Unlimited Release Printed September 2007. [8] Mathworks user guide “Curve Fitting Toolbox” p 29 [9] Adarsh Nagarajan & al “Photovoltaic Inverter Reliability Assessment”. Technical Report NREL/TP-5D00-74462 October 2019 [10] BENDIB D, “Experimental setup for testing commercial pv inverters” Conference Paper · December 2015 [6]