Summer short-term load forecasting in the city is not only related with the past power load data and affected by temperature, wind, precipitation and other factors, etc. Mutation structure is obvious in this data. In order to forecast summer short-term power load in the city, this paper establishes the ARIMAX model between the power load sequence and the input sequence "temperature" based on Cointegration Relation Theory by SAS, fully exploits self-relevant information of the internal sequence as well as the correlation between sequences. It can be seen from the minimum amount of information standards "AIC-SBC", the amount of information about ARIMAX model is smaller than the classic time series method ARMA model and the relative error is smaller. The fitting results are more accurate. This model has a high value under the presence of mutation structure and significant influence factors in the short-term load forecasting field.