Short-Term Power Load Forecasting Based On CSFLA And Wavelet Neural Network

In recent years,the rapid development of economic and social has made the power system face more and more serious challenges.The accurate power load forecast(STLF)is also directly related with the power system security scheduling and the social normal production.At present,many scholars have made load forecasting by Neural Networks,and achieved good results.Therefore,this paper studys the power load forecasting through using the method of combining wavelet neural network and shuffled frog leaping algorithms.Firstly,the phase space of the actual load has been reconstructed,which is based on the chaotic characteristic of the power load.Meantime,two crucial parameter,embedding dimension and the time delay have been calculated through C-C algorithm.And through this method more implicit information of the time series can be digged out.At the same time,this paper makes the data normalized to speed up the convergence space of the algorithm.After that,this paper establishes a short-term power load forecasting model based on wavelet neural network.And performing simulation analysis with the actual power load data,the experimental proofs that the wavelet neural network prediction model gets a better result,but there is still room for improvement.Then,aiming at the problem that the wavelet neural network is sensitive to the initial value,and easy to fall into the local optimal.Two improved algorithms are proposed respectively.The one adds adaptive learning rate and penalty factor to the gradient descent algorithm,which is IWNN.The other is achieved by introducing Shuffled Frog Leaping Algorithms instead of the gradient descent algorithm to train the wavelet neural network,which is SFLA-WNN.Then the short-term power load forecasting models based on SFLA-WNN and IWNN are established respectively.Through the comparison and analysis of the simulation results,it is proved that the improved models both have higher accuracy compared with WNN model.At last,aimed at the defects of classical shuffled frog leaping algorithms,like the initialization of the population and the update manner of frog leap step in the local search are all at a random way,it is not conducive to the convergence of the algorithm.So a Comprehensive Shuffled Frog Leaping Algorithms is proposed in this paper.The algorithm makes full use of the better solution generated by IWNN to optimize the initial population of the leapfrog algorithm,and a new adaptive moving factor is proposed when the frogs update steps at the local search.Then propose a short-term power load forecasting model of CSFLA-WNN,which can be used for simulation experiments,the results proved that the improved algorithm can enture real time prediction and achieve higher accuracy.