Wavelet Neural Network For Short-term Power Load Forcecasting

Power load forecasting is an important part of power systemscheduling,operation,planning and is the basis of power system economic operationand the safe operation.It has important significance to power system stability anddevelopment.Electric load data has the characteristics of uncertainty andcomplexity,and currently used by most prediction methods are derived from thetraditional linear statistical theory,so it is difficult to accurately predict load valuesand is necessary to find new ways to prediction.Then wavelet neural network isproduced in this context.Firstly, this paper introduces the overview of the load forecasting,the status andthe development of forecasting techniques at home and abroad,and then introduces thebasic background,concepts and principles of artificial neural network and waveletneural network.Wavelet neural network is a combination of wavelet analysis andneural network theory and is a new type of feedforward neural networks.self-adaptability of the model,convergence speed and prediction accuracy than simplyusing the neural network is better.Then,the BP neural network model and the WNNmodel are established on a load of data of changsha in Hunan Province respectively,the three sets of experimental simulation of the two models are made.By comparingthe experimental results and error analysis,wavelet neural network is more responsivethan the BP neural network prediction of the volatility signal.And when the samehidden layer nodes of those models, wavelet neural network prediction accuracy ishigher than the BP neural network.So wavelet neural network could help to rudece thenumber of the hidden layer node and improve the network performance.Because the factors affecting the change of the power load is also uncertainly, soit will be a very difficult task to make the future load forecast and the high accuracyof prediction.The new approach to power system load forecasing presented in thispaper will combine the traditional neural network and wavelet function to predict, ithas higher accuracy than simply using the neural network.