Short-Term Load Forecasting for an Industrial Park Using LSTM-RNN Considering Energy Storage
Feng Li, Xiaowei Yu, Xin Tian, Zilong Zhao
Abstract
Short-term forecasting approaches for conventional load can be generally divided into the model-based methods and the data-driven methods. With the increasing trend of the incorporation of energy storage systems (ESSs) into modern industrial parks, the conventional short-term load forecasting techniques become less effective. In this paper, a short-term load prediction method that takes into account the effect of energy storage is proposed. In this initial study, we specifically investigate the prevailing “two-charging-two-discharging” operation strategy for the ESS designed according to common 24-hour load characteristics in an industrial park, which can effectively reduce the peak load. A long short-term memory recurrent neutron network (LSTM-RNN) is used as the forecasting algorithm due to its excellent capability to correlate the historical data. Furthermore, we compare the results of two strategies to incorporate the ESS power flow into the LSTM-RNN algorithms. Simulation results show that by preprocessing the historical data without virtually manipulating control strategies has a smaller error than the post-processing of ESS effect.