Short-term electricity load forecasting based on improved sparrow search algorithm with optimized BiLSTM

Abstract

Short-term electricity load forecasts (STELF) is an essential part of power system and operation, capable of balancing electricity demand and is vital to the safety and efficient operation of the power system. The research improves the Long short-term memory (LSTM), combines it with Bidirectional recurrent neural network (BIRNN), and obtains the improved Bidirectional Long Short-Term Memory Network (BiLSTM) forecasting model. The Sparse Search Algorithm (SSA) can provide a new solution to more difficult global optimization problems and has been improved due to the shortcomings of the search and detection mechanisms. and a simplex mechanism is introduced to obtain an improved Search Mechanism Sparse Search Algorithm (SMSSA) optimized pathfinding algorithm. And constructs the SMSSA-based optimized BiLSTM for STELF model. By choosing actual data, the model's prediction behavior is confirmed. The results showed that, in descending order, BiLSTM, LSTM, and Recurrent Neural Network (RNN) had the best fitting effects between the predicted and actual values. BiLSTM also had the highest prediction accuracy, with error values of 95.7059 for Root Mean Square Error (RMSE), 79.1575 for Mean Absolute Error (MAE), and 2.1260% for Mean Absolute Percent Error (MAPE). After SMSSA optimized the parameters, SMSSA-BiLSTM had the best fit and had errors that were much lower than those of the other two models. According to the three error judgment metrics of RMSE, MAE, and MAPE, the errors were 82.6298, 71.9029, and 2.0952%, respectively. This showed that SMSSA-BiLSTM performed well in short-term power load forecasting, offering security for the power system's safe operation.