Accurate forecasting of short-term electricity load has been one of the most important issues in the electricity industry. And the forecasting accuracy is influenced by many unpredicted factors. Artificial neural network is a novel type of learning method, which has been successfully employed to solve nonlinear regression and time series problems. In this paper, it is proposed a new optimal model to train ANN. The model that calls simulated annealing particle swarm optimization algorithm (SAPSO) combines the advantages of PSO algorithm and SA algorithm. The new model is proved to be able to enhance the accuracy and improve the convergence ability and reduce operation time by numerical experiment. Subsequently, examples of electricity load data from a city in China are used to illustrate the proposed SAPSO-ANN. Results show that forecasters trained by this method consistently produce lower prediction error than other methods.