Based on the 15-min power load of Shijiazhuang City from 2017 to 2020 and the meteorological data of the same period, we calculate the effective temperature and the temperature-humidity index of human comfort index. Considering the periodicity and growth of the benchmark load, we adopt the grey model GM (1, 1), and at the sametime, combining filtering method and correlation analysis method, we establish a piecewise regression model of daily peak cooling load and human body comfort index. The results show that the power load in Shijiazhuang has an obvious year-on-year growth trend. The stripped-off daily cooling load curve shows a “W”-shaped distribution. The models are fitted with primary, secondary and piecewise functions respectively. The three prediction models are tested， and the fixdings suggest that the prediction accuracy of the piecewise function is relatively high, the average relative error is between 4.8% and 5.2%, and the piecewise function error between the effective temperature and the temperature-humidity index is between -10% and 10%, and the proportions are 88.1% and 90.5%, respectively. The effective temperature ratio considering temperature, humidity and wind speed has a higher forecast accuracy than the summer daily peak cooling load forecasting model which is based on the temperature-humidity index. The regression model piecewise point is 26.2℃, which has better reference value for power dispatch during the “peak summer” period.