Multivariate Load Forecasting Model Based on MTL-LSTM-TCN

This study introduces a multi-task learning (MTL) model that integrates a temporal convolutional network (TCN) with a long-short-term memory network (LSTM) for short-term multivariate load forecasting. Initially, the input features undergo preprocessing to ensure data quality and consistency. The MTL framework and LSTM are then employed to capture the interdependencies among multivariate loads and the temporal characteristics of load sequences, while TCN is leveraged to extract critical features from local patterns. This approach culminates in the development of the MTL-LSTM-TCN multivariate load forecasting model, aimed at enhancing the precision of multivariate load predictions. A simulation study was performed using a multivariate load dataset from Arizona State University's Tempe campus. The findings confirm that the proposed MTL-LSTM-TCN model delivers outstanding performance across various load forecasting scenarios, significantly boosting both the accuracy and robustness of short-term load predictions.