Meteorological factors are important in short － term load forecasting． Considering the cumulative effect of meteorology，similar days are selected as training samples，a BP neural network load forecasting method based on improved particle swarm optimization algorithm ( IPSO － BP) is proposed ． Firstly，the meteorological factors more relevant to daily loads are determined by means of correlation analysis． On this basis，the weighted geometric distance is used to select the historical days which have a greater correlation with the predicted day as the similar day，and the IPSO － BP neural network model is trained and used in short － term load forecasting． The practical application results show that the proposed prediction model and data processing method can achieve more accurate prediction results．