Robust PV-BESS Scheduling for a Grid with Incentive for Forecast Accuracy

This paper proposes a robust cost-optimal scheduling of a battery energy storage system (BESS) integrated with a photovoltaic power plant (PV). A power grid with an incentive policy is considered. Power transactions between the grid and its energy resources are normally settled according to the hourly price. Additional hourly incentive is provided if the day-ahead submitted schedule is maintained. Accurate forecasting and robust scheduling are essential for PV-BESS owners to maximize both revenues. The PV power forecast model, which is based on an RNN, uses a CNN discriminator to decrease the gap between its open-loop training and closed-loop test dynamics. The application of a GAN concept to the model training process ensures a stable day-ahead hourly forecast performance. The robust BESS scheduling model handles the remaining forecast error as a box uncertainty set to consider the cost-optimality and cost-robustness of the resulting control schedule. The scheduling model is formulated as a concise mixed-integer linear programming form to enable fast online optimization with the consideration for both transaction and incentive revenues. The introduction of adversarial learning to the forecast model increased the incentive revenue by 7.33%. Moreover, the online BESS scheduling with the incentive consideration enhanced the overall revenue by 3.73%.