SWAFRET: A Temporal Convolutional Network Model Integrating Dynamic Sliding Window Technology

Accurate and reliable Electricity Load Forecasting (ELF) is of paramount importance for government decision making, market operation adjustment and optimization of consumers' electricity consumption behavior. In this study, Temporal Convolutional Network (TCN) model incorporating dynamic sliding window technique is designed. Firstly, the feature importance analysis of raw electricity load data is carried out by the Random Forest algorithm, and the feature set is constructed accordingly. Subsequently, the time series data are segmented using the sliding window technology, and the TCN captures the intrinsic connection between the sequences and extracts the similar patterns from the historical load data, thus realizing the prediction. To verify the effectiveness of our proposed method, the study compared it with existing research that utilized three open-source datasets. The experimental results show that the SWAFRET model consistently outperforms other methods in terms of Mean Absolute Percentage Error (MAPE).