A new hybrid optimization prediction strategy based on SH-Informer for district heating system

Abstract

The district heating system (DHS) is an essential social service. Heat load forecasting is one of the critical steps in DHS. However, the manual and empirical operation in the dispatch process may exhibit some deviations which will cause the irrationality of historical data. This paper proposes a new hybrid optimization prediction strategy, which consists of the similar hour (SH) method and a prediction model namely Informer. In the SH approach module, light gradient boost machine (LightGBM) and Euclidean norm (EN) is used to select the SH dataset. In the prediction module, Informer and other four popular models are constructed. The evaluation module contain four frequently-used evaluation criteria and energy-saving rate. Especially, the energy-saving rate is defined for the new strategy. The historical operation data of a DHS in Tianjin is studied as the case for model training. Experimental results indicate that: (a) Informer can effectively sense the change of heat load and performs excellent in the prediction task; (b) The hybrid models based on SH can improve the prediction performance; (c) SH_Informer achieves the highest energy-saving rate, reaching 11.09%, 10.05% and 10.36% in the prediction length of 24, 48 and 168 h, which demonstrate the feasibility of the new prediction strategy.