Application of temporal convolutional neural network combined with autoencoder in short-term bus load forecasting

Abstract

Improving the accuracy of bus load forecasting is a crucial step to achieve the goal of fine intelligent power grid dispatching. Aiming at the problem of small base and large fluctuation of bus load data, a short-term bus load forecasting model (AE-TCN) based on the combination of Auto-Encoder (AE) and Temporal Convolution Network (TCN) is proposed. Firstly, the Wavelet Threshold Denoising (WTD) is used to process the original bus load data to remove the burr, then, the data with high similarity is categorized into one cluster by using the Bisecting K-means clustering algorithm. AE-TCN model is formed to fit the processed data and obtain the predicted value of the load. Finally, to verify the effectiveness of the proposed method, two bus load data of 220kV and 110kV in a city of china are employed. The simulation results show that the proposed method has higher prediction accuracy than traditional prediction models.