Abstract This thesis presents a Moving Average multi-dimensional regression technique with back-propagation neural network method to do load forecast based on data about temperature, passengers coming in and out. In practical application, the short-term, mid-term and long-term load variation of Taiwan High Speed Railway (THSR) Taichung station is predicted by the proposed method using relative data about load per half hour, temperature, passengers coming in and out, and holiday load from 97 to 99 year. The effectiveness of the proposed method is identified, and it can be applied to load forecast of new building THSR stations and factories in future. Keywords:Regression; Moving Average; BPN (Back-Propagation neural Network );Load forecast