Optimal power allocation of battery energy storage system (BESS) using dense LSTM in active distribution network

Abstract

AbstractForecasting of renewable energy plays a major role in deregulated power systems. The rapid change in climatic conditions poses many challenges in recent years throughout the globe to policymakers and due to this, the forecasting of solar energy has become quite difficult to forecast with conventional forecasting methods. As compared to conventional methods, machine learning algorithms have shown better accuracy due to their learning methods with uncertain parameters. The market participants such as solar photovoltaic (SPV), battery energy storage system (BESS), and thermal units undergo challenges with the optimal dispatch strategy under such uncertainties of renewable energy. In addition to the concerned integrated system, other uncertainties affect the optimal operation of the integrated system and these are line contingencies and SPV. In this paper, we have used supervised learning methods such as multilayer perceptron (MLP), recurrent neural network (RNN), and long short‐term memory (LSTM) to forecast the hourly SPV in the day‐ahead market. Among the three methods of machine learning, results show LSTM with dense has been a better forecasting method with high accuracy obtained. The role of BESS in the optimal operation of day‐ahead hourly forecasted SPV along with the hourly thermal unit dispatch in DC optimal power flow (OPF) has been investigated in the paper. The main objective of the paper is to optimally allocate the BESS in the SPV‐BESS‐Thermal unit integrated system forming an active distribution network (ADN) to minimize the operating cost under different uncertainties such as line contingencies and SPV. The hourly dispatch of thermal units, BESS, and forecasted SPV is obtained for the short‐term market. The proposed approach is validated by a modified IEEE 33 bus system and solved by mixed integer linear programming (MILP).