We’ve just posted a new preprint on arXiv: "Joint Scheduling of DER under Demand Charges: Structure and Approximation".
This paper addresses the joint scheduling of behind-the-meter distributed energy resources (DERs)—flexible loads, renewable generation, and batteries—under net metering with demand charges. The problem is formulated as a stochastic dynamic program to maximize expected operational surplus while accounting for uncertainty in renewable output. The optimal control policy exhibits a threshold-based structure, but strong temporal coupling from storage dynamics and demand charge constraints leads to combinatorial policy complexity. To mitigate such complexity, we propose a scalable approximation algorithm that relaxes the problem and searches for peak demand efficiently, with linear complexity in the time horizon. Simulation results on real-world datasets demonstrate the algorithm's superior performance and near-optimality relative to benchmark strategies, including reinforcement learning.
