Robust model predictive control of battery energy storage with neural network forecasting for peak shaving in university campus

Abstract

This study addresses the challenge of optimizing energy consumption and managing peak demand charges in large university campuses using battery energy storage system (BESS) by demonstrating the effectiveness of a two-stage neural network-based Model Predictive Control (MPC) algorithm enhanced with robust optimization. To achieve this, we first delineate the architecture of neural networks and the Robust MPC model. Subsequent testing in simulation environments leads to the practical validation of the algorithm on a small-scale test bench configured to emulate a microgrid system. Results show that the integration of neural networks and robust optimization in an MPC framework significantly outperforms traditional control methods, achieving more effective peak shaving, reducing energy costs, and enhancing system resilience. The added robustness effectively addresses forecasting errors, making the control strategy more resilient and reliable. The successful deployment of this algorithm on a test bench underscores its practical applicability, highlighting its potential to optimize energy consumption and reduce peak demand charges in buildings. This research contributes a novel, scalable, and adaptive control strategy that bridges advanced forecasting techniques with robust MPC, providing a valuable solution to address peak demand challenges in commercial and institutional buildings.