Abstract: Data pre-processing in short-term load forecasting and the limitations of a single forecasting model are pressing issues that need to be addressed. Thus, we propose a short-term load forecasting method based on visualization dimension reduction of meteorological data and a multi-mode weighed combination. The method combines visualization dimension reduction, modal decomposition in noise reduction, single forecasting model, and weight determination theory to design a short-term load forecasting model with dimensionality reduction of meteorological data, historical load decomposition, modal component in noise reduction, and multi-mode weighed combination. Therefore, by setting up three comparative experimental environments and analyzing the electric load and meteorological data provided by a regional power supply company, the prediction results and error analysis reveal that the proposed short-term load prediction method effectively combines the characteristics of data pre-processing methods and single prediction models while retaining the essential characteristic structure of high-dimensional meteorological factors. Furthermore, it effectively improves the prediction accuracy of the grid load in the region.