Combined electricity-heat-cooling-gas load forecasting model for integrated energy system based on multi-task learning and least square support vector machine

Abstract

Accurate forecasting of the combined loads of electricity, heat, cooling and gas in the integrated energy system is the key to improve the comprehensive efficiency and gain more economic benefits of various types of energy. As an important part of the new generation of energy systems, the integrated energy system contains energy subsystems such as electricity, heat, cooling and gas, and each subsystem employs energy supply, conversion and storage equipment. This form of energy system achieves the coupling of different types of energy in different links. Based on this, this paper firstly combs the coupling relations among different integrated energy subsystems. Secondly, with the help of the weight sharing mechanism in the multi-task learning and the idea of least square support vector machine, a combined forecasting model of electricity, heat, cooling and gas loads based on the multi-task learning and least square support vector machine is constructed. Finally, in order to verify the effectiveness of the forecasting model proposed in this paper, the actual data from the integrated energy system in Suzhou Industrial Park are selected for a case study. The results show that: (1) the combined forecasting model based on multi-task learning and least square support vector machine can accurately predict the electricity, heat, cooling and gas loads of the park integrated energy system. (2) Compared with extreme learning machine and least square support vector machine, the combined forecasting model based on the multi-task learning and least square support vector machine increased the forecasting accuracy of a workday and a weekend by 18.00% and 19.19%, and the average forecasting accuracy increased by 18.60%. (3) Compared with extreme learning machine and least square support vector machine, the combined forecasting model can effectively shorten the training time which is reduced by 58.1% and 35.22%. The results further reflect the application effect of the multi-task learning in energy demand forecasting of the integrated energy system and have a very broad reference value.