Power load data has time-sequence and non-linear characteristics, and Long short-term memory (LSTM) neural network can handle the above data characteristics. However, the performance of the LSTM algorithm has a great dependence on the preset parameters, and the parameters set by experience will make the model have low generalization performance and reduce the prediction effect. In order to solve the above problems, this paper proposes a prediction model NIWPSO-LSTM combining the nonlinear dynamic inertia weight particle swarm optimization (NIWPSO) and long-short-time memory(LSTM) neural network . The nonlinear dynamic inertial weights are used to improve the global optimization ability of PSO, and then, the key parameters of LSTM are optimized through NIWPSO. The results show that the prediction accuracy of NIWPSO-LSTM is much higher than other models, which verifies the feasibility of the proposed scheme.