The Research Of Power System Short Term Load Forecasting Based On Principal Component Analysis’s Genetic Neural Network

Short term load forecasting is one of the important works of electricityproduction sector, which is mainly applied in forecasting the power load of the nextfew hours or one day to several days. Accurate results of the short-term loadforecasting are the key to guarantee the security of power system operation, economyand quality of supply. It can be used to arrange unit start or stop economically forreducing the spinning reserve capacity, and it also can be used to arrange themaintenance program reasonably for reducing the cost of power generation andimproving economic efficiency.There are many kinds of methods about short-term load forecasting, includingthe BP neural network technology, which is an advanced prediction method. It canmassively deal with complex systems which contain various kinds of nonlinearinformation through its self-organizing and adaptive function. However, thetraditional BP neural network has two main defects: slow training speed and sensitiveto initial weights and thresholds which easily lead the training process to fall into thelocal minimum points. In order to overcome these defects, this paper presents aprincipal component analysis’s genetic neural network (PCA-GABP neural network)model. Using PCA to reduce the dimension of the original load data can remove thecorrelation among the data and delete some redundant information. It also can addressthe defect of slow training speed. Meanwhile, combining improved genetic algorithmwith BP neural network and using the global search performance of genetic algorithmto determine the weight threshold of the BP neural network, can effectively overcomethe problem of local convergence of the BP algorithm. Because meteorological factorshave a great impact on the accuracy of the short-term load forecasting, this paperresearch and do analysis on the influences what the daily maximum temperature andthe daily minimum temperature did on the accuracy of forecasting. The dailymaximum temperature and the daily minimum temperature are included inmeteorological factors.In the end of this paper, there do some simulation experiments using theMATLAB as a platform. These experiments are divided into two large groups, onegroup does not consider meteorological factors, and another group considersmeteorological factors. Each of the large group is divided into four small groups which include the BP neural network model, the PCA-BP neural network model,GABP neural network model and the PCA-GABP neural network model. Thesimulation results show that the PCA-GABP neural network model which consideredmeteorological factors improve network performance and forecasting accuracy, andprove that the model has feasibility.