Neural Network based Predictive Algorithm for Peak Shaving Application using Behind the Meter Battery Energy Storage System

This article aims at demonstrating that a neural network based predictive algorithm can be used to perlbrm peak shaving on a large size building using a behind the meter battery energy storage system. First, a complete model predictive control architecture is defined with a real-time synchronous layer and an asynchronous one, which enables increased accuracy and robustness to forecast errors. Then, by using neural networks for predictions and linear programming to solve the problem, it is shown that peak shaving is achieved successfully. Simulations results are computed using perfect predictions and the proposed algorithm to compare both. Finally, once the control system accuracy is demonstrated, it is used on a small-scale test-bench representing a large university campus. Doing so, validates that the proposed control algorithm can he used in real-life conditions.