Integrated loads forecasting with absence of crucial factors

Accurate forecasting of multi energy loads is critical for the stable operation of urban integrated energy systems. However, urban integrated energy systems operate under complex and dynamic conditions, where crucial input factors, such as meteorological conditions, are often difficult to obtain with high accuracy due to sensor malfunctions, data transmission failures, or inherent uncertainties in weather predictions. The absence of these crucial factors leads to incomplete input information, thereby compromising the reliability of load forecasting. Addressing this challenge requires method that can mitigate the impact of absent crucial factor data while maintaining high forecasting accuracy. Therefore, this study proposes an integrated loads forecasting method that enhances predictive accuracy despite absent crucial factors. First, the proposed method reconstructs the multi energy loads sequences to obtain more information about their temporal variations. Then, a multi-task learning model is developed to capture the intrinsic interdependencies among cooling, heating, and electric power loads through a parameter sharing mechanism. This enables the model to compensate for absent critical factors by leveraging the correlations among different energy loads. Finally, compared with the bidirectional gated recursive unit, the proposed method improves the accuracy of integrated forecasting cooling, heating, and electric power loads by 76.51 %, 38.52 %, and 42.42 % respectively while reducing total computation time for integrated loads forecasting by 27.23 %. These improvements facilitate more reliable energy demand predictions, supporting the optimal management and utilization of multi energy resources in urban integrated energy systems.