The invention discloses a prediction method for a heat supply load of a combined heat and power system. The method comprises the steps that firstly, main influence factors of a thermal load are selected as variables, the outdoor temperature, the water supply temperature, the water return temperature and the water supply flow are used as input variables, the thermal load is used as an output variable, and historical data of all variables in 24 hours of the past day are collected and subjected to normalization preprocessing; According to historical data, a fuzzy C-means clustering algorithm combined with subtraction clustering is used for structure identification, an input space is divided, a fuzzy set and a membership function corresponding to variables are defined, the number of rules andthe like are determined, a least square algorithm is used for parameter identification, and a linear regression equation with the thermal load as output is obtained; Finally, input variables such as the outdoor temperature of the prediction day are substituted into the linear regression equation, then the corresponding thermal load size and change rule are estimated and predicted, the problem thata prediction model is difficult to build by thermal load multiple variables is solved, and the thermal load of the combined heat and power system in the future 24 hours can be predicted accurately.