series forecasting

traffic prediction

load forecasting

Electric Power Systems Research

• The research paper proposes a novel short-term load forecasting model based on Temporal convolutional network (TCN) and Attention mechanism (AM). • Maximum Information Coefficient (MIC) is adopted to select high-quality input variables can reduce the parameters that the model needs to train. • Fuzzy c-means (FCM) combined with Dynamic Time Warping (DTW) is developed to classify load data with high similarity into one category for modeling. • Two datasets demonstrate that the proposed approach can effectively improve the forecasting accuracy. Load forecasting is the foundation of power system operation and planning. Accurate load forecasting can secure the safe and reliable operation of the power system, cut power generation costs, and increase economic benefits. However, it is challenging to effectively forecast load in view of the complicated effects due to a variety of factors involved. In order to enhance the forecast performance, this paper proposes a short-term load forecasting model based on Temporal Convolutional Network (TCN) with channel and temporal attention mechanism (AM), which fully exploits the non-linear relationship between meteorological factors and load. In addition, the Maximum Information Coefficient (MIC) is adopted to select high-quality input variables and eliminate irrelevant variables to reduce the parameters that the model needs to train. Meanwhile, Fuzzy c-means (FCM) combined with Dynamic Time Warping (DTW) is developed to conduct cluster analysis on load data to classify loads with high similar characteristics into the same cluster. Experiments were performed on two different datasets come from public and power grid company to verify the capabilities of the proposed forecasting approach. The experimental results show that the put forward method can effectively enhance the accuracy and generalization capability of load forecasting compared with other forecasting models.

Short-Term Load Forecasting Using Channel and Temporal Attention Based Temporal Convolutional Network