Multi-energy Load Short-term Forecasting for Integrated Energy System based on Time Convolutional Networks and Multi-task Learning

Accurate multi-energy load forecasting is the premise of ensuring economic scheduling and optimal operation of integrated energy systems (IES). Aiming at the characteristics of large load fluctuation and complex energy coupling, a short-term multi-energy load forecasting method based on time convolutional networks (TCN) and multi-task learning is proposed for IES. Firstly, the maximum information coefficient (MIC) is used to consider the multi-energy interaction mechanism, multi-load coupling characteristics and the correlation of meteorological factors. Secondly, A multi-task hybrid A-TCN model based on Attention mechanism combined with TCN is proposed. Then, the A-TCN method was validated using the actual data of IES of Arizona State University. Experimental results show that compared with traditional model and single-task learning, the proposed model can accurately predict IES short-term load and has excellent performance.