Economic Load Forecasting of Power System Based on Hybrid Neural Network and Multi-objective Evolutionary Algorithm

Abstract

The accuracy and efficiency of power load forecasting are essential to the power grid’s stability and economic operation due to the ongoing growth of the power system scale and the widespread use of renewable energy. Traditional load forecasting methods are often difficult to balance the forecasting accuracy, economy, and environmental protection. Therefore, the study first combines wide deep neural network with temporal convolutional network to better extract the spatial features and temporal dependencies of load power data. Subsequently, the economic dispatch function under multi-objective is constructed and the function model is solved taking into account the generation cost, pollution emission and voltage deviation. The results indicate that the hybrid neural network has good convergence and loss results, with its loss value approaching 0 in the later stages of iteration. K-means clustering-support vector machines (K-means-SVM) and Improved Whale Optimization Algorithm-Long Short-Term Memory (IWOA-LSTM) are significantly affected by the number of sample points, and their minimum load power prediction error is greater than 0.5%. The training time of the hybrid model is always less than 0.2 seconds, and the maximum prediction accuracy exceeds 95%. The multi-objective evolutionary algorithm proposed in the study exhibits good solution set diversity and convergence. The average running time of the improved Multi-objective Evolutionary Algorithm based on Decomposition (MOEA/D) algorithm does not exceed 105 seconds, and the maximum electricity cost does not exceed 2.5 × 104 yuan. The research method provides a new solution for economic load forecasting of the power system, which provides a guarantee for its safe and stable operation.