Short-Term Load Forecasting Based on Ant Colony Clustering and Improved BP Neural Networks

Short-term load forecasting is important for the economic and secure operation of power system. Taking the random disturbances, especially the meteorological factors into account are the key to improve forecasting precision. This paper presents ant colony clustering model to process the historical load days. Ants pick or drop samples decided by the similarity of it to its surroundings. After iterative processing, the historical load days with their meteorological characters are classified. Before load foresting, the weather conditions of forecasting day are got by weather forecast and a group of historical load data with similar meteorological characters are selected. Furthermore, in order to avoid local optimum and improve training speed, this paper presents improved BP neural network from adding dynamic parameters. By setting dynamic parameters related to input and output range, the error adjusting of output and hidden layer realizes intelligent control. As a result, at the same time to reduce the processing time, the precision of load forecasting is improved