In order to ensure that the new generation of smart grid can dynamically adjust the regional power distribution and scheduling according to the real-time power consumption, it is necessary to achieve efficient and accurate power consumption prediction. The traditional power consumption prediction method is to calculate the possible power consumption roughly through manual statistics or analysis of the power consumption in the same period of history, which not only consumes a lot of manpower and material resources, but also can not meet the accurate power consumption prediction under the background of smart grid. In order to replace the traditional power consumption forecasting methods, the differential integrated moving average autoregressive forecasting model, long-term memory network prediction model and generative confrontation network prediction model are used to study the power consumption prediction. The results show that the intelligent algorithm can greatly improve the accuracy of power consumption prediction, but to achieve short-term and efficient prediction, it is necessary to use the intelligent algorithm reasonably in the smart grid system.