Sizing and Scheduling Solar Photovoltaic Battery Systems for Peak Electrical Demand Management

Abstract

For Texas commercial and industrial customers, peak electrical demand charges are largely based on the highest 15-minute monthly peak. By storing energy from photovoltaic cells in batteries for use during high-demand times, electrical utility customers can shave the electrical demand peak by load-shifting. The potential for this type of load shifting strategy is demonstrated in this study using data from commercial buildings located in Texas.

Shaving the monthly peak demand is a challenging optimization problem due to the long scheduling horizon of one month. Moreover, the initial and the operation costs of both the photovoltaic system and battery storage are significant; thus, the sizes of these systems need to be optimized. To optimize a photovoltaic battery system, the sizing and the scheduling problems are solved in tandem. One year’s worth of data is usually considered for sizing energy systems, but scheduling the load over one year is computationally intensive. Consequently, sizing a photovoltaic battery system for peak shaving under Texas’s electricity tariff is a complex optimization problem.

In its most basic form, shaving peak electrical demand is achieved by judiciously scheduling battery storage, assuming knowledge of photovoltaic and load profiles. Forecasting errors of the photovoltaic and load profiles to construct the schedule might lead to higher peaks. Therefore, scheduling a photovoltaic battery system for peak shaving requires a real-time control mechanism to manage the PV power and load uncertainties.

In this work, a framework based on convex optimization is developed for sizing combined photovoltaic battery systems under different pricing policies. In addition, a control scheme is established to shave the electrical site demand peak with long control-horizon and propagated uncertainty based on stochastic dual-dynamic programming. Also, the sophisticated prediction methods are replaced by simple estimation methods based on imitating the patterns as well as the sub-optimality decision-making.

Overall, this research has led to great contributions to grid-connected energy storage sizing and scheduling. The results show that reformulating the optimization problem reduces the problem complexity significantly. Furthermore, the effectiveness of scheduling battery storage in shaving the demand peak under the uncertainty of prediction for different buildings was investigated in this study.