Combining Predictive and Heuristic Control Strategies for Optimal Deployment of Distributed Energy Storage Systems

Abstract

The continued integration of variable renewable energy (VRE) sources in the energy system is essential for the mitigation of climate change. However, the intermittent, unpredictable, and distributed nature of mainly solar and wind energy will substantially increase the need for grid flexibility. Currently, flexibility in the power system is mainly provided by (thermal) generation capacity on the supply side. Consequently, continued and economically feasible decarbonisation of the energy system will require more flexibility on the demand side. Demand side management (DSM) in short timeframes can consist of demand response (DR) and distributed energy storage systems, and relies on smart grids in which energy management systems may shift loads to dynamically match supply and demand. This thesis proposes a two-step DSM program which manages distributed battery energy storage systems (BESS) within a portfolio of solar energy producers and prosumers. The first step uses model predictive control (MPC) to continuously readjust the aggregate (dis)charging schedule of the BESSs, based on new solar irradiance and temperature forecasts, in order to minimize the imbalances between expected and actual energy generation. The second step comprises a heuristic control program, which is executed within each time step of the MPC program and allows for additional (dis)charging in order to (passively) contribute to grid balance when price peaks are expected in the imbalance market. Both optimization steps also take into account battery degradation. In contrast with what is common in the literature, the DSM program is based on actual market and weather forecasts, as opposed to (model-adjusted) historical data. Perfect foresight is only assumed with respect to portfolio demand. Solar irradiance forecasts are recursively improved using a Kalman Filter. An auxiliary optimization program is used to determine the optimal size of the BESSs. To accurately gauge the potential of the DSM program, a number of representative days are selected from two sets of weather forecasts using hierarchical clustering. The results suggest that MPC is a feasible framework for DSM in an real world environment were new weather forecasts come available on a rolling basis. However, while the Kalman Filter successfully improves the accuracy of the forecasts, the remaining mismatch between actual and forecasted VRE supply more often than not results in an increase of total imbalance. This is the case for both forecast data sets, and is also a result of the 24-hour optimization horizon combined with a constraint on the final State of Charge of the BESS. Moreover, the net potential economic gain from the Lithium Iron Phosphate batteries examined is minimal, indicating the necessity of DSM deploying both BESSs and DR.