Abstract: Conventionally the electric load forecasting can hardly attain a result whose accuracy meets what's required. A short-term load forecasting model is therefore developed to solve the problem, based on the process neural network of which the input is the function of time and the high forecasting accuracy is available. Describes the structure of the model, discrete data fitting method by the expansion of function orthogonal basis and learning algorithm. According to the daily load data of a certain power network in Northeast China, the model training and the accuracy of load forecasting were investigated. The simulation results showed that the load forecasting model based on process neural network is better than on BP neural network.