Improving the Method of Short-term Forecasting of Electric Load in Distribution Networks using Wavelet transform combined with Ridgelet Neural Network Optimized by Self-adapted Kho-Kho Optimization Algorithm

• Published in: Heliyon Vol. 10; no. 7; p. e28381
• Main Authors: Wang, Yaoying, Sun, Shudong, Fathi, Gholamreza, Eslami, Mahdiyeh
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 15.04.2024; Elsevier
• Subjects: Electric load forecasting; Optimization; Ridgelet neural network; Self-adapted kho-kho algorithm; Wavelet transform; Electric load forecasting; Wavelet transform; Ridgelet neural network; Optimization; Self-adapted kho-kho algorithm
• ISSN: 2405-8440; 2405-8440
• DOI: 10.1016/j.heliyon.2024.e28381

This paper proposes a new method for short-term electric load forecasting using a Ridgelet Neural Network (RNN) combined with a wavelet transform and optimized by a Self-Adapted (SA) Kho-Kho algorithm (SAKhoKho). The aim of this method is to improve the accuracy and reliability of electric load forecasting, which is essential for the planning and operation of competitive electrical networks. The proposed method uses the Wavelet Transform (WT) to decompose the load data into different frequency components and applies the RNN to each component separately. The RNN is, then, optimized by the SAKhoKho algorithm, which is an improved version of the KhoKho algorithm that can adapt the search parameters dynamically. The proposed method is trained and tested on the Zone Preliminary Billing Data from the PJM regulatory area, which is updated every two weeks based on the Intercontinental Exchange (ICE) figures. The proposed method is compared with six other cutting-edge methods from the literature, including SVM/SA, hybrid, ARIMA, MLP/PSO, CNN, and RNN/KhoKho/WT. The results show that the proposed method achieves the lowest Mean Absolute Error (MAE) of 7.7704 and Root Mean Square Error (RMSE) of 17.4132 among all the methods, indicating its superior performance. The proposed method can capture the temporal dependencies in the load data and optimize the RNN's weights to minimize the error function. The proposed method is a promising technique for electric load forecasting, as it can provide accurate and reliable predictions for the next hour based on the previous 24 h of data.