In order to improve the operation efficiency and resource utilization of the power grid, improve the accuracy of budget and cost control, at the same time, help users to develop reasonable power supply plan and electricity consumption strategy, promote the continuous optimization of electricity service and electricity business environment, and build a power connection cost prediction model based on big data analysis of power. Firstly, analysis the data types in the power data, and using the MapReduce parallelization processing cluster mining algorithm, mining from the power system and the power cost related power data, obtain the clustering results. Then, through the time series analysis method build the total cost prediction model, and through the multivariate regression method to build the cost of factors prediction model, after the model prediction, get the best cost prediction results. Finally, by experiment, the model can accurately predict the power connection cost generated when enterprise users access, and can also effectively predict the change of power connection cost under different equipment prices and different voltage levels.