Optimizing electricity peak shaving through stochastic programming and probabilistic time series forecasting

This paper proposes a predict-then-optimize framework to optimally schedule the charging and discharging activities of battery energy storage systems (BESS). BESS are used to eliminate peak electricity consumption loads, which pose a significant risk of microgrid system failure and increase electricity costs to end users. A major challenge in BESS lies in determining the battery usage schedule, which must be determined before actual energy consumption materializes. To address this challenge, we propose a two-phase approach. In the first phase, we develop a deep-learning-based probabilistic time-series forecasting model to predict future electricity consumption. In the second phase, the output of the prediction model is used to generate future load scenarios, which are incorporated into a two-stage stochastic programming model that determines the optimal battery usage schedule. Extensive computational experiments on real-world datasets demonstrate the effectiveness of the proposed framework in shaving peak loads and minimizing energy costs. Specifically, our proposed framework reduces daily energy peaks by up to 26%, with the potential for greater improvement as the forecast of future energy loads improves. To the best of our knowledge, this work is the first to investigate the integration of probabilistic forecasting and stochastic optimization to enhance the effectiveness of BESS in managing peak energy loads, leading to more energy-efficient buildings. •Introduced a predict-then-optimize framework for BESS peak shaving.•Framework combines machine-learning with two-stage stochastic programming (2SP).•Achieved up to 26% reduction in daily energy peaks using real-world datasets.•Accurate forecasts crucial for enhanced performance of the 2SP model.