The Research Of Short-Term Load Forecasting Based On Rough Sets And Support Vector Machines

Power load forecast is very important management that can guarantee the safety of the system, save energy and raise production efficiency to the maximal degree. Short-term load forecast is the most important load forecast in power system.Artificial neural network (ANN) is a common method. Based on the experiential risk minimization principle, ANN has the advantage of higher forecast accuracy, and but it has some disadvantages, such as difficult net-structure-parameter- confirmation, and weak generalization ability. A method combined support vector machine (SVM) with rough set theory (RS) is presented. It can solve the small sample, simplify the dimensions of the sample data, and has the advantages of the global optimization and the better generalization ability. The method makes the abnormal data completed, smooth and uniform, and combines the improved attribute reduction algorithm based on binary discernibility matrix and training model of support vector regression(SVR) to forecast the short-term load. Using the data of Nanchang between 2005 and 2007, simulation experiments show:1. The historical data pre-processed by filter method can guarantee the data completeness and smoothness;2. The load forecast of the improved attribute reduction algorithm, based on binary discernibility matrix and training model of SVR, can reduce the test samples and the dimensions of the sample data, shorten the training time, and further improve the forecast accuracy.The method mentioned in the paper provides a theoretical guide for power production and marketing, and has a good social and economic benefit and a nice prospect.