An Integrated Energy System Load Forecasting Method Based on CEEMDAN and MTL

Accurate load forecasting is a prerequisite for achieving optimal system energy dispatch and improving energy utilisation, but the physical characteristics and strong coupling of loads lead to a significant increase in load volatility and stochasticity, which affects forecasting accuracy. In this regard, this paper constructs a new integrated energy load forecasting model: a combined MTL-CNN-GRU and MLR forecasting model. The method firstly uses CEEMDAN algorithm to decompose the load data of the integrated energy system, splitting the time series into sub-components with different complexity; then further adopts the multi-task learning strategy to learn the complex coupling relationship between different load types. The signal can be divided into high-frequency and low-frequency components by calculating the over-zero rate, and MTL-CNN-GRU is used for the high-frequency part of the prediction and MLR for the low-frequency part of the prediction; at last, the two parts of the prediction results are superimposed to obtain the final results. The experimental results show that the constructed model can effectively improve the accuracy of load prediction.