New MPC Approach Turns Distributed Energy Resources into Virtual Power Plants

A new paper from Obai Bahwal, Oliver Kosut, and Lalitha Sankar tackles the challenge of balancing net demand on the grid using distributed energy resource aggregators (DERAs). As renewable energy adds supply uncertainty, DERAs—which pool decentralized sources like rooftop solar and batteries—can act as virtual power plants. The authors model each DERA as a flexible 'virtual battery' with constrained state-of-charge and power limits. They then formulate the dispatch problem as a long-term model predictive control (MPC) task that minimizes deviations from desired charge levels, ramping, and net-load tracking errors. To keep operations real-time, they implement a rolling-horizon MPC that updates decisions using the latest marginal-demand forecasts. For forecasting, two models are tested: linear regression and a long short-term memory (LSTM) neural network.

Using high-resolution CAISO net-demand data and five representative DERA types, the simulations demonstrate how well the approach tracks marginal demand. Key trade-offs emerge between forecasting horizon length and MPC update rate, as well as dependence on the forecasting model choice. Results indicate a slight edge for LSTM models over linear regression for certain time shifts and horizon choices. The work provides a practical pathway for grid operators to integrate large numbers of small-scale energy resources into load balancing without centralized control, potentially increasing renewable penetration and grid stability.