Short-term Load Prediction Based on Whale Optimized WTCN-BiGRU Hybrid Neural Network

Abstract

AbstractThe accuracy of short-term power load prediction provides according to power department companies to make reasonable production scheduling plans and avoid resource waste. In this paper, a Temporal Convolutional Network (TCN) -BiGRU-Attention short period power load forecasting means on the strength of the Whale Optimization Algorithm (WOA) is proposed. TCN is improved by the Morlet mother wavelet basis function. Train network hyperparameters by WOA. The WOA-WTCN- BiGRU-Attention hybrid neural network is formed. Firstly, the high-dimensional features of power load are extracted by the WTCN network. Secondly, the BiGRU network is used to learn the dynamic change rules of high-dimensional characteristics. Ultimately, the Attention mechanism is important to endow homologous proportion to characteristics through mapping weighting and learning parameter matrices to reduce information loss. The proposed method is compared with the prediction results of WTCN-LSTM, WTCN-BiGRU, and TCN-BiGRU-Attention models by taking an open data set from an Australian region as an example. The results show that the proposed method has a more accurate prediction effect.KeywordsPower load forecastingGated recurrent networksAttention mechanismsTemporal convolutional networksWhale optimization algorithm