Ultra-short-term dynamic prediction of multiple types of energy sources based on two-stage transfer learning

Abstract

Abstract In this thesis, starting from the source load of the integrated energy system, the source load in the park is analyzed for the energy side of wind turbine power generation, hot and cold energy, photovoltaics, as well as for the related electric and hot and cold load side of enterprises and buildings. The ultra-short-term time-scale load prediction is performed based on artificial intelligence algorithms, respectively. The time series prediction model is trained by three factors, namely temperature, season, and holidays, using the data of electric load, heat load, and cold load to obtain the primary model. Based on the primary model, the final prediction model is obtained by adopting the migration learning strategy.