Abstract: With the increase of electrification rate, the characteristics of "double heights and double peaks" of the new power system are becoming more and more obvious. The importance of accurately predicting the maximum load in summer and winter is also more prominent for the power balance arrangement, the peak power and supply guarantee. With load decomposition theories, this paper proposes a method of forecasting daily maximum load in summer and winter based on the spatial distribution of sensitive temperatures. Making full use of the social development factors, a correlation model between the basic load and the gross domestic product (GDP) and the consumer price index (CPI) is established in order to provide a more objective and rigorous theoretical basis for the estimation of the basic load. Key sub-regions are identified according to the load proportion of each regional power grid, the sensitive temperature of sub-regions is defined by the M-K (Mann-Kendall) test, and the range of sensitive temperature of sub-regions is then obtained by statistics. Forecasting models of temperature-related load of different grades is established referring to the cluster analysis method. In the case of forecasting maximum daily load of the State Grid business area in 2021 summer and winter, results of this paper are superior to which of the ARIMA (autoregressive integrated moving average) time series forecasting method and the SVR (support vector regression) method. The effectiveness of the algorithm and model is further verified in the case of forecasting maximum daily load in 2022 summer.