Currently, customer demand value evaluation work is still carried out through two methods: customer active reporting and customer manager on-site mining, with a single source of demand and a lack of intelligent methods for mining customer demand. In this paper, the prediction models in Power Grid customer data information search process have been analyzed. Through comparing with the practicability of regression forecasting model, time series forecasting model, trend extrapolation forecasting model, gray prediction model and Bayesian forecasting model, the comprehensive evaluation method and the variation coefficient method are employed to establish the mathematical model. It is finally pointed out that the practicability of Bayesian forecasting model is maximum in the process of data information search. The results show that in the data mixed sampling, the accuracy and comprehensiveness of the Power Grid customer data model based on the Bayesian forecasting are concentrated between 94% and 97%, while the accuracy and comprehensiveness of other models are unsatisfactory.