Benchmarking Genetic Algorithms for Short-Term Battery Energy Storage Systems Optimization

Battery Energy Storage System (BESS) arbitrage is a topic of growing interest given the widespread use of storage systems by end users and the recent developments in European electricity market policies. The optimal operation and management of a storage unit depend on multiple inputs, i.e., load demand (connected to the BESS), the wholesale electricity price, and the production from power plants. Genetic Algorithms (GAs), which belong to the family of meta-heuristic algorithms, can be used for the optimal dispatch of the BESS over a short time window, i.e., a timespan over which the forecasted data of demand, production, and energy prices can be considered reliable. The value of GAs for BESS optimization can be ascribed to three key factors: their simple implementation, their effectiveness, and the variety of parameters that can be customized to better suit the optimization problem. Additionally, GAs do not require any training data (which can be often difficult to obtain, especially when considering real data), in contrast to Deep Reinforcement Learning algorithms. In this article, a comparison of different GAs is made to provide more insight when evaluating such algorithms for BESS short-term optimization when connecting the storage systems with a commercial end-user type and a photovoltaic (PV) power plant, both located in northern Italy, for four different typical days. GAs performance is evaluated in terms of the quality of the optimal solution obtained and the execution time. Additionally, the effects of installing a BESS are evaluated by comparing the obtained economic results with the benefits obtained from a single PV plant.