Study On Short Term Load Forecasting Based On Soft Computing

Short term load forecasting (STLF) of power system is not only the fundamental information of grid dispatching and scheduling departments, but also the foundation of grid dispatching, operation and development. Furthermore, with the progress of power market, demand statistic and forecasting data of grid load will be declared to all the communities and it will be one of important parts of power market real time trade.There are traditional model methods of forecasting short-term load, such as time series, regression analysis, and so on. Lots of non-linear relationships exist between load and factors that influence it. Artificial neural network (ANN) was put forward for forecasting short-term load in 1990s because of its ability to approach any non-linear functions. In the same term, the other intelligent computing technologies develop fast, such as genetic algorithm (GA), fuzzy computing. In the thesis, ANN and ANN combining with GA are adopted to forecastingshort-term load.Before forecasting short-term load, defective data are eliminated from historical records by mathematical statistical means, and historical data are pretreated lengthways and transversely to get rid of abnormal data and smooth load curve. Whereas urban resident and commercial load have considerable percentage of Taiyuan load and they are sensitive to weather factors. Therefore, when forecasting short-term load by three layers ANN and four layers ANN, it is sorted according to whether considering weather factors or not. Because ANN has local minimum value and its convergence speed is slow, ANN's weight value and threshold value are confirmed by GA.Short-term load of Taiyuan area is forecasted by the above methods in the thesis and the summary is as follows: Though weather factors influence load, as far as Taiyuan grid is concerned that the results without weather factors are superior to those with weather factors when electricity is severe shortage; On the basis of lots of calculation, four layers ANN is superior to three layers ANN at the aspect of function mapping ability, and when ANN combining with GA, numbers of chromosome gene of four layers ANN are less than those of three layers ANN. So compared to three layers ANN, four layers ANN saves computing time much. When combining ANN with GA to keep ANN from falling into local minimumvalue, it is at the cost of increasing computing time; The research of load characteristic should be done well to confirm the relationships between load and the factors that influnce it, and to select better similar historical days.In a word, STLF in Taiyuan area is discussed in the thesis and there comes to the conclusion that four layers ANN combining with GA is valid. The next job is to reduce computing time and analyze load characteristic.