The invention discloses an electric load predication optimization method for N-section intervals of a combined heat and power generation set. The method comprises the following steps: reading heating historical data of the combined heat and power generation set in each hour in N time intervals when approximate linear change of the outdoor temperature occurs in 24 hours of the past day, fitting N linear regression curves of the outdoor temperature and the thermal load of a building through a least square method according to the historical data, substituting the outdoor temperature in a predicate day into the linear regression curves, estimating the thermal load of predicate day, establishing a digital model of a set heating condition diagram through the historical data of the combined heat and power generation set and the least square method, substituting the thermal load of predicate day into the digital model, and calculating the minimum generated power and the maximum generated power of set at certain thermal load according to boundary conditions of the digital model, so as to fix a generation load interval of the set at certain thermal load. The method can be used for accurately predicating thermal load and electric load intervals of a heating system of the combined heat and power generation set in the next 24 hours.