Power grid load forecasting is of great significance to ensure power quality. A method of using time series analysis to predict the load change of power grid in the next 72 hours is proposed in this paper, which enables the power grid dispatching department to understand the change trend of user load in the short term, so as to adopt the corresponding dispatching strategy to ensure the power quality. In addition, a statistical study on the load category of users is also made. Thus, the load type that accounts for a large proportion in the peak load period is found out, and users are encouraged to transfer the use of this type of load to non peak hours, which is beneficial to realize the peak shaving of load curve and ensure the safe operation of power grid. Finally, the data collected by the power grid company and the intelligent internet of things system are tested and verified, and the results also show the effectiveness of this method.