In order to rationally arrange and give priority to ensuring the power dispatching problem of a military base, a method of shot-term load forecasting based on chaotic time series and artificial neural networks is presented. According to chaos theory and neural networks method, it is based on the delay coordinates phase space reconstruction first, choose time delay ‘t ’ and embedding dimension ‘m’ by using the method of mutual information and saturation correlation dimension after, then use the BP neural networks prediction, carry out an experimental simulation about grid in the period of load sequence on a base finally. The simulation results show that the relative errors is within 5%, and 33.3% of the error is less than 1%, proved that the prediction method has higher forecasting precision and application value.