Augmented Random Search with Artificial Neural Networks for energy cost optimization with battery control

Abstract

Intermittent renewable energy production and dynamic load must be balanced through appropriate control of integrated energy storage to account for the temporal discrepancy among power supply and demand. Intelligent control systems are required to anticipate and optimize the charging and discharging of energy storage. In recent years, reinforcement learning based techniques have been applied to a multitude of problems, including building integrated energy storage solutions. In this work, the focus is on the application of reinforcement learning based techniques to the specific energy optimization problem of controlling a battery energy storage system in a smart warehouse. This paper adopts data from a real operational battery energy storage system installed in a smart warehouse, integrated with photovoltaic, for food distribution on the west coast of Norway. In the smart warehouse, an intelligent energy management system controls the on-site battery energy storage using machine learning predictions of load and photovoltaic production, and an optimization algorithm is presented to generate a schedule for effective utilization of battery energy storage in coordination with a thermal storage system. This paper presents the combination of the augmented random search reinforcement algorithm with artificial neural networks as a basis to design an intelligent energy management system for controlling energy flows of battery energy storage systems to minimize the energy cost. The developed algorithm finds very promising solutions in the considered case-study of a smart house for energy cost minimization through a battery energy storage system, achieving an average of 99.2% accuracy across 10 seeded trials.