Agent-Based Micro-Storage Management for the Smart Grid Perukrishnen Vytelingum, Thomas D. Voice, Sarvapali D. Ramchurn, Alex Rogers and Nicholas R. Jennings University of Southampton
Outline Energy Domain Smart Grid Agent-Based Micro-Storage Management Game theoretic analysis Adaptive storage mechanism Empirical evaluation 2
Development and growth since the start of the industrial revolution due to fossil fuels Fossil Fuels Coal (1800+) Oil (1900+) Gas (1960+) Concentrated solar energy collected over millions of years 3
Continued use of fossil fuels is challenged by three impending factor Finite resources Demand outstrips production capacity Energy security Resources are not evenly distributed Climate Change Increasing atmospheric CO2 concentration 4
Addressing these issues require challenging changes in the way we use energy 80% reduction in CO2 by 2050 Increased energy efficiency through electrification Transport Heating Low carbon Wind Solar Hydro Nuclear 2008 Climate Change Act 5
Current electricity networks are challenged by these proposed changes Ageing infrastructure designed for a small number of large generators Supply follows demand Fuel inefficiency Peak demand Fixed prices Fred Schweppe proposed the need for a more dynamic grid (1978) Spot pricing and homeostatic control 6
The Smart Grid represents a modern vision of a dynamic electricity grid Imagine the possibilities: electricity and information flowing together in real time, near-zero economic losses from outages and power quality disturbances, a wider array of customized energy choices, suppliers competing in open markets to provide the world s best electric services, and all of this supported by a new energy infrastructure built on superconductivity, distributed intelligence and resources, clean power, and the hydrogen economy. US Department of Energy (2009) 7
The Smart Grid represents a modern vision of a dynamic electricity grid Imagine the possibilities: electricity and information flowing together in real time, near-zero economic losses from outages and power quality disturbances, a wider array of customized energy choices, suppliers competing in open markets to provide the world s best electric services, and all of this supported by a new energy infrastructure built on superconductivity, distributed intelligence and resources, clean power, and the hydrogen economy. US Department of Energy (2009) 8
We investigate how micro-storage can address the three challenges posed above How can consumer owned, small scale, storage devices be deployed within a Smart Grid? How much storage is required? How should storage be managed? Does the electricity network benefit? Load and Diversity Factor Carbon Emission Reduction 9
Smart Home Demand (kw) 2 1 0 ½ hour periods Macroscopic Market Model 10
We perform two types of analysis on a system composed of multiple such homes Game theoretic analysis What does equilibrium look like? Adaptive storage strategy A best-response day-ahead storage computation A learning mechanism to adapt to changing market prices (as total demand is changing now that agents are changing their individual demand) 11
We exploit the price signal to perform an aggregate game theoretic analysis Individual payoff of each agent: Seek to find the storage profile two conditions: subject to No net charge Battery capacity 12
We exploit the price signal to perform an aggregate game theoretic analysis Calculating the Nash equilibrium simplifies to minimising: Characterised by two price points: Charging price point Discharging price point We can solve for aggregate and individual storage profiles. 13
We use linear programming to optimise the storage profile within individual homes Best-response storage computation Solve using CPLEX subject to same constraints as before (with one change): 14
We need to apply a two rate learning approach to optimise storage over time Use moving average prediction of prices Find storage capacity that will minimise cost of electricity Adapt capacity toward this value: Again, find the best storage profile Adapt existing storage toward this profile: 15
We perform an empirical evaluation using price and carbon data from the UK grid Simulate 1000 homes with randomly allocated storage capacity and load profiles (generated from UK averages). Use UK grid data for the macroscopic market model 16
Empirical Evaluation Nash Equilibrium Storage Profile Storage Profile over 100 Trading Days Mean-Squared Deviation from Nash Equilibrium Electricity Prices over 100 Trading Days 17
We see improvements in system-wide metrics with varying storage uptake System-wide Grid Performance Diversity Factor (DF) Ratio of the sum of individual maximum demand to the maximum total demand. 4 kwh Load Factor (LF) Average power divided by peak power. Factor Carbon Emissions Reduction Gird carbon intensity correlated to total demand. Proportion of Population with Storage 18
Conclusions and Future Work Conclusions Shown how an adaptive agent-based storage strategy is able to reach equilibrium solution Demonstrated desirable system-wide properties at this equilibrium Future Work Better understand convergence critieria Improved market modelling Interaction between domestic, industrial and commercial use Drive demand with real smart meter data Predict future demand and price (Gaussian processes) 19