EE5900 Advanced Embedded System For Smart Infrastructure Electricity Market With Smart Home Integration Smart Home Scheduling Basic Idea Prefer using cheap time interval Pricing for one-day ahead time period 2 Smart Home Scheduling System Power flow Internet Control flow 3 Home Appliances in Smart Home Not schedulable Restrictively schedulable Fully schedulable 4 Variable Frequency Drive Power level 350 W 500 W 820 W 1350 W Powerr Power 10 cents/kwh 5 cents / kwh 10 cents/kwh 5 cents / kwh 10 kwh 5 kwh 1 2 Time (a) cost = 10 kwh * 10 cents/kwh = 100 cents 1 2 (b) 3 Time cost = 5 kwh * 10 cents/kwh + 5 kwh * 5 cents/kwh = 75 cents 5 Smart Home Scheduling ο§ Given the pricing curve, to decide – when to launch a home appliance – – We will describe the at what frequency algorithm later. Assume for how long that we have it, then – subject to scheduling constraints such as start time and end time ο§ Targets – Reduce monetary cost of each user – Reduce peak to average ratio of grid energy usage ο§ From the utility point of view – Change the pricing curve to guide the usage of grid energy – Result in balanced usage of energy, and thus balanced generation from power plant 6 Key Contribution: What Are We Modeling? Given a pricing curve at the aggregator (community) side, customers will schedule home appliances (i.e., distribute grid energy usage over time intervals) using the smart home scheduling algorithm For each community, there is load distribution over time intervals All the communities will send energy demands per time interval to utilities Utility companies will set their own pricing per time interval. The aggregators can choose to buy from whom, when and how much Utility companies will compete with each other in deciding pricing Repeat until converging to equilibrium Aggregators will compete with each other in deciding where to buy When all utilities and aggregators make up the decisions, each aggregator modifies the pricing curve at the aggregator side 7 Electricity Market Generators We will model them in a bottom up fashion Utilities Electricity Market Aggregators Renewable Energy Customers 8 Single User Smart Home Scheduling Generators Utility Companies Aggregators Customers Home Appliances 9 Dynamic Programming For Scheduling Single Appliance Energy 0 t1 t2 t3 t4 Time Schedule the home appliance from the first time interval 10 Solution Characterization ο§ For a solution in time slot i, energy consumption e and cost c uniquely characterize its state Time slot i Time slot i+1 (ei, ci) (ei+1, ci+1) π0 = 0, π0 = 0 ππ+1 = ππ + ππ+1 ππ+1 = ππ + ππ+1 β πΌπ+1 for π = 1,2, … , π 11 Dynamic Programming ο§ For a solution in time slot i, energy consumption e and cost c uniquely characterize its state Time slot i Time slot i+1 (ei, ci) (ei+1, ci+1) π0 = 0, π0 = 0 ππ+1 = ππ + ππ+1 ππ+1 = ππ + ππ+1 β πΌπ+1 for π = 1,2, … , π 12 Solution Pruning ο§ For an time interval, solution (e1, c1) will dominate solution (e2, c2), if and only if e1≥e2 and c1≤c2. ο§ Dominated solutions will be pruned. Time interval (15, 20) (15, 25) (11, 22) 13 Dynamic Programming Based Appliance Optimization Power levels {1, 2, 3} πΌ1 = 2 Price 0 πΌ2 = 1 (3,6) (3,3) (2,4) (2,2) (1,2) (1,1) (0,0) Dynamic Programming returns the optimal solution t1 (0,0) Runtime : (6, 9) (5, 7) (4, 5) (3, 3) (5, 8) (4, 6) (3, 4) (2, 2) (4, 7) (3, 5) (2, 3) (1, 1) Time t2 2 O(m k ) – # of distinct power levels = k – # time slots = m 14 Scheduling Multiple Appliances for One User Appliances Determine Scheduling Appliances Order An appliance Schedule Current Appliance Not all the appliances processed Update Energy Upper Bound of Each Time Interval All appliances processed Schedule 15 Multiple User Smart Home Scheduling Generators Utility Companies Aggregators Customers Home Appliances 16 Motivation Price Curve 0 1.5 3 4.5 6 7.5 9 10.5 12 13.5 15 16.5 18 19.5 21 22.5 24 14 12 10 8 6 4 2 0 Customer 1 Customer 2 ............. Customer 3 Game theory is used to handle the interactions among customers. 17 Game Theory ο§ For every player in a game, there is a set of strategies and a payoff function, which is the profit of the player. ο§ Each player chooses from a set of strategies in order to maximize its payoff. ο§ When no player can increase its payoff without decreasing the payoffs of others, Nash Equilibrium is reached. ο§ In our problem, a customer is a player, the strategy is the dynamic programming based scheduling, payoff function is the negative of the payment of each customer. 18 An Example of Equilibrium 19 Game Formulation at Community Level Players: All the customers in the community Payoff: − πβ,π πΆβ πΏβ Strategy: Scheduling the appliances of all customers to maximize payoff while the scheduling constraints are satisfied 20 Game Theory Based Multiple User Scheduling Each user schedules their own appliances separately All users share information with each other Each user reschedules their own appliances separately Converge No Yes Schedule 21 Dynamic Pricing From ComED Illinois Corporation How do we set unit price? 22 Load Based Pricing Per Time Interval ο§ In a local community, when the total load over all How to decide the customers is Lh, theconstant total factor ah? cost (price) is approximately Ch=ahLh2 at time interval h ο§ If lh,j denotes the load of customer j, the customer j pays lh,j·Ch/Lh 23 Top Level Generators Utility Companies Aggregators Customers Home Appliances 24 Electricity Market Forward Market Fixed Amount Utility Generator Cheap Price Aggregators Fixed Cost Whole Sale Market Generator LMP Utility Bid Aggregators 25 An Example ο§ Suppose that a utility has a contract with a generator that the generator can provide the utility 100MWh every hour with the price 1οΏ /kWh. ο§ The utility company can sell this 100MWh at a price 1.2οΏ /kWh. ο§ When the demand received by the utility exceeds 100MWh, the utility needs extra amount of energy with a much higher cost and sell it with a much higher price. ο§ The 100MWh in the contract is called forward limit. 26 Generator Model Total Price ($) Forward limit (kWh) ο§ Within forward limit, electricity can be generated with low price in power plant ο§ Beyond forward limit, it costs much higher to generate electricity 27 Utility Buying Local generator ($) Forward limit Remote generator 2 ($) ($) Forward limit (kWh) Over buying Waste Remote generator 1 sign in a forward contract with local generator Forward limit (kWh) (kWh) Not enough LMP Utility 28 Illustration of LMP Local generator Remote generator 1 10οΏ /kWh, 100kWh 5οΏ /kWh, 50kWh Utility, 100kWh Remote generator 2 8οΏ /kWh, 100kWh 29 Market Level Modeling and Optimization Generators Utility Companies Aggregators Customers Home Appliances 30 Strategy of Utilities and Aggregators Aggregators Utilities Price Adjust price to attract more demand Demand Decide purchase from each utility to minimize payment Conducted in each time slot 31 Illustration of Our Model ($) Utility Generator Forward limit (kWh) Utility designs the pricing model according to the generator model. Utility will never get negative profit as long as there is some electricity sold. 32 Iterative Procedure of Game Initialize the demand of aggregators to utilities If the total demand from utility j decreases, utility j Each utilitywill designs their own pricingstrategy strategyf;f decreases the pricing according to the generator’s If the total demand from utility jprice increases, utility j will increase the pricing strategy f; Each aggregator solves the minimization problem to obtain the solution of dij according to the pricing strategy f of utilities Each utility combines all the demands from aggregators and adjusts the pricing strategy f Not converge Each aggregator solves the minimization problem to obtain the solution of dij according to the updated pricing strategy f of utilities Converge Finish 33 Utility Adjusts Pricing in Game At different iterations of game, utility has different pricing curve for different profits ($) Generator Forward limit (kWh) 34 Minimizing Total Payment of Aggregators f1($) π πππ ππ (πππ ) , π = 1, 2, … , π π=1 π Forward limit (kWh) π . π‘. πππ = π·π , π = 1, 2, … , π π=1 f2($) fj is the pricing function of utility j dij is the demand that aggregator i buy from utility j Di is the total demand of aggregator i m is the total number of utility n is the total number of aggregator Forward limit (kWh) 35 How Do Aggregators Distribute Demands? f1($) Utility 1 Forward limit (kWh) f1($) Utility 2 Forward limit (kWh) f1($) Utility 3 Forward limit (kWh) Aggregator Given the pricing strategy of utilities, aggregators choose the utility with lowest price and we propose two models 36 Model 1 First In First Schedule ($) Request sequence: d2j, d3j, d1j d2j d3j Forward limit d1j (kWh) Utility schedules the demands of aggregators according to the sequence of request of aggregators 37 Model 2 Most Balanced Aggregator First Load 1 ($) Time Load 2 Time d1j d2j Forward limit (kWh) Utility schedules the demands of aggregators according to the rate of balance of aggregators, where rate of balance is current demand over total demand of aggregators. 38 Game Formulation at Market Level Player: Utility and Aggregator Strategy: 1. Utility: Decide price both within and beyond forward limit 2. Aggregator: Decide the amount of energy to buy from each utility Payoff: ο§ 1. Utility: π π = ππ ( ππ=1 πππ ), where dim is the demand of aggregator i to utility m ο§ 2. Aggregator: −πΆπ where πΆπ is the total payment of aggregator π Since utilities play the leading role and aggregators play the following role, this is a Stackelberg game. 39 Game Formulation at Community Level Players: All the customers in the community Payoff: − πβ,π πΆβ πΏβ Strategy: Scheduling the appliances of all customers to maximize payoff while the scheduling constraints are satisfied 40 Interaction Between Market Level and Community Level Market Level Price Schedule Community Level • Impact of market pricing to customers • Customers feedback to market 41 Between Aggregator and Customer ο§ Assume that the aggregator makes no or fixed profit from customers ο§ Since the payment from the aggregator n to all utilities is Cn,h for time interval h, one can set ah=Cn,h/Lh2 – Cn,h is the new cost due to the game between utilities and aggregators – Lh is the old load from all the customers inside community n ο§ Due to the changes in the pricing curve, the customers will redistribute the load during time intervals 42 Initialization Community Level Each aggregator changes community level pricing Aggregators distribute demands to utilities for payment minimization Customers reschedule smart home appliances with the new pricing curve Utilities adjust pricing to attract more demands Change of payment remains big Change of payment is small enough Market Level Change of system cost is small enough Change of system cost is big End 43 Game Architecture Generators Adopted LMP based game Utility Companies Stackelberg Game Aggregators Customers Standard Nash Game Home Appliances 44 Case Study ο§ The testcase in our study – One day time horizon with the time interval of 15 minutes time – 2 utilities – 5 communities where each one has an aggregator and 400 smart home customers – Each customer has 5 flexible appliances including dish washer, washing machine, dryer and charger – The daily consumption of these appliances varies from 1kWh to 5kWh, and the run time under the normal power level varies from 45 minutes to 2 hours ο§ The market assumption – In the forward market, the unit price of the 2 utilities are in the range 1 οΏ /kWh to 3 οΏ /kWh and 1.2 οΏ /kWh to 2.6 οΏ /kWh, respectively. – In the whole sale market, the slopes of unit price are in the range 0.012 οΏ /kWh2 to 0.03 οΏ /kWh2 and 0.01 οΏ /kWh2 to 0.025 οΏ /kWh2 respectively. The upper bound of the utilities are regulated by the government. 45 Background Energy Usage 46 Average Energy Reduction Per Customer Peak to average ratio is 2.23 Peak to average ratio is 1.43 47 Average Monetary Cost Reduction Per Customer 48 Pricing and Load Change During Iterations at A Community 49 Profit (USD) Profit (USD) Utility Profit • With andNowithout smart home scheduling, utility Smart Home Scheduling, 1573 in total 150 profit is similar • The essential reason for total customer payment 100 reduction is due to the reduction in the expense of generation 50 • With smart home scheduling, the peak generation is 0 avoided and generation is much more balanced 20 40 60 80 100 • 0With smart home scheduling, generation is mostly Time Horizon (15 min) Smart Home limit Scheduling, 1397 inquadratic total within the forward where charge is 150 not applied 100 50 0 0 20 40 60 Time Horizon (15 min) 80 100 50 Balanced Generation Power Plant 1 Power Plant 2 No Smart Home Scheduling No Smart Home Scheduling 300 Generation (kWh) Generation (kWh) 300 200 100 0 0 20 40 60 Time Horizon (15 min) Smart Home Scheduling 80 0 20 40 60 Time Horizon (15 min) Smart Home Scheduling 80 100 0 20 40 60 Time Horizon (15 min) 80 100 300 Generation (kWh) Generation (kWh) 100 0 100 300 200 100 0 200 0 20 40 60 Time Horizon (15 min) 80 100 200 100 0 51 Conclusion ο§ This work is to evaluate the impact of the smart home scheduling to the whole power system. – The complete power system model is provided with integration of smart home – Provide two level game structure for electricity market – Stackelberg game to model market level – Standard game to model community level – Design the dynamic programming and distributed algorithm for the game ο§ Smart home scheduling can reduce the payment of customers payment by about 30%, and reduce peak to average ratio by about 35% in the power system. The energy generation is also much more balanced. 52