Apply Wavelet And Neural Network To Power System Mid-Long Term Load Forecasting

Medium and long term load forecasting is the basis of power system planning and construction, and its precision will directly affect the rationality of investment, meshwork layout and running of power system. Especially today in our country which power system is walking up to market, the forecasting load data is very important to adjustment, running and building of power system.Firstly, the paper begins with the concept of load forecasting and expatiate the function, trait, principle, development of forecasting technology, analyzing and comparing the merits and shortcomings of some forecasting methods at home and abroad. Based on this, the paper:(1) present a new long term forecasting method, the model of GRNN (general neural network) .With an eye to the absence of long term load history data, the learning speed of GRNN is rapid, the performance of GRNN is qualified even with sparse sample data, the GRNN can control unstable data, and the artificial adjusted parameter is only one, this paper adopt GRNN to forecast long term load.Applying the model of GRNN to predict the long term load of some area of our country, the performance of GRNN is qualified even with sparse sample data.(2) present a new model of monthly load forecasting. The model combined the characterstic of wavelet's time-frequent localized and the merits of netural network, overcome the Windless of neural network structure design. The model possesses strongly approach abililty, the convergence speed is rapaid, and it can avoid efficiently local mini-value.Monthly load is possessed of the property of increase trend and seasonal fluctuation simultaneously, so its behavior appears as the characteristic of complex non-linearity. This paper present using the horizontal and vertical history data as input of WNN for the first time, cause the determent of WNN input have more meaning.Applying the model of WNN to predict the medium term load of some area of our country, the forecasting result shows the precision of WNN is high, its self-adaptability is well and theconvergence speed is faster than the model of BP (back propagation).