Enhancing Load Forecasting for Large Industrial Users Through Feature Preference and Error Correction

Accurate power load forecasting for large industrial users plays a crucial role in developing effective power consumption plans and energy-saving strategies. It also contributes to optimizing the operational efficiency of the power grid. To enhance the forecasting accuracy, a load forecasting method for large industrial users that combines signal decomposition, feature preference, and error correction is proposed. Firstly, the variational modal decomposition (VMD) is employed to decompose the power load series into multiple intrinsic mode functions (IMFs). For each IMF, the most influential factors are identified by utilizing the maximum information coefficient (MIC) based on their highest correlation. Subsequently, separate Informer models are constructed for each IMF, and both historical data and impact factor data are employed as inputs for forecasting. Furthermore, a gated recurrent unit (GRU) network is used to predict the error of Informer model, thereby the forecasting accuracy is refined. Then the IMF-based forecasting series and prediction error series are combined to obtain the final power load forecasting series. To validate the effectiveness of our proposed method, a real power load dataset is utilized from the large industrial users in China. The experimental results show that our proposed model surpasses other baseline models in terms of accuracy and stability. Thus, it proves to be a valuable tool for accurate power load forecasting of large industrial users.