Sequence signal prediction and reconstruction for multi-energy load forecasting in integrated energy systems: A bi-level multi-task learning method

Abstract

Multi-energy load forecasting is the basis for the optimization and scheduling of integrated energy systems (IES). An IES contains much heterogeneous energy with volatility and temporality, and complex coupling relationships between the energy sources, which makes it difficult to accurately predict multiple loads simultaneously. Considering these issues, this paper proposes a bi-level multi-task learning (BiMTL) method for simultaneous sequence signal prediction and reconstruction via deep neural networks. In the feature construction process, considering the volatility and randomness of the load, the improved wavelet packet decomposition model is constructed to decompose the multiple load sequences into sub-signal with more obvious signal characteristics, and improves the ability of the forecasting model to capture and respond to the fluctuations of load sequences within a local range. In the modeling process, considering the excessive dependence of traditional signal reconstruction methods on the prediction results of sub-bands, this paper innovatively proposes a BiMTL method. This method employs a second-layer shared learning to perform signal reconstruction and error correction on the output of the first layer’s sub-band predictions. Finally, a BiMTL-LSTM model is constructed to extract the coupling and temporality of loads based on the LSTM hard-sharing mechanism. Experimental results show that the proposed method outperforms the current state-of-the-art methods on the widely used Tempe dataset, achieving an accuracy of up to 98.70%.