Based On The Similarity And The Dynamic Fuzzy Neural Network Short-term Power Load Forecasting

Power load forecasting can not only improve the safety of the power distribution,and also make the reasonable and effective operation of the power companies. It’s an important method to improve the energy utilization efficiency. Short-term load forecasting has random,time-varying, nonlinear characteristics. There are the traditional statistics and modern artificial intelligence two methods to forecast the power load at present stage. The traditional statistical methods focus on the establishment of data fitting and model,but the impact factor of the power load forecasting also underutilized,so the predicted results are clearly not ideal.In modern artificial intelligence algorithm, BP neural network is the most used for the power load forecasting. This method also have several of choices uncertainty in the numbers of hidden layer、hidden layer nodes、initial weights,and rely on the experience experts of user itself,and also have the black box handle hidden layer,over-fitting,too training and a series of shortcomings.In order to improve the short-term load forecasting accuracy,this paper contains the following aspects:(1)The fuzzy rules of traditional fuzzy neural network were determined based on the experiences of the expert,but now the amounts of power load data was too large and the data is too multifarious,so the fuzzy rules have become a problem at this time. In order to deal with this shortcomings in fuzzy neural network, the dynamic fuzzy neural network to predict the short-term power load is used in this paper. The biggest feature of dynamic fuzzy neural network is fuzzy rules aren’t determined in advance, however, are according to the input samples to dynamically adjusted. Firstly, systematic errors and accommodate boundary the two factors determine whether the system requires new fuzzy rules,and then used hierarchical learning techniques to speed up the whole network established model, and also used the algorithm of error decline rate to eliminate the fuzzy rules which do not affect the entire network very well.(2) When using the entire dynamic load data to trained the dynamic fuzzy neural network, the complexity of network modeling will increased,and also the modeling speed of network will affected. To solve this problem,this paper uses the similar day and dynamic fuzzy neural network combined to predict the short-term load. Before training the model, the training samples are processed by similar day. By automatically finding out the sampleswhich are similar in average temperature and week factor of the predict daily’s,the load consumption levels of similar day are similar with the load consumption levels of predict daily. The prediction on the similar day samples as the training data rather than all the data as the training data, so as to improve the modeling speed and simplify the complexity of the model.(3) The experiments are arranged respectively using the dynamic fuzzy neural network approach and the similar day combined dynamic fuzzy neural network approach. The electric load data come from the EUNITE competition. The experiments have achieved the ideal results, which provides a new method for power load forecasting.