Research On Rough Neural Network Based Power System Short-term Load Forecasting

Short-term load forecasting plays an important role in power system operation and planning.The changes of the load are related to load structure,time,weather,people's social activities and so on.The changes of the load have the characteristics of multiple nonlinear and time-varying,so it is difficult to express its development rule with mathematical formula.In order to meet the requirements that the power system short-term load forecasting needs higher accuracy,this paper combines rough set theory with neural network and establishes a short-term load forecasting model based on rough neural network.Specific studies are as follows:Since the short-term load has the characteristics of multiple nonlinear and time-varying,this paper combines rough set theory with RBF(Radial basis function)neural network and gives a prediction model based on rough neuron neural network.Rough set theory has obvious advantages of analyzing and dealing with imprecise,inconsistent and incomplete information.RBF neural network with nonlinear neurons has a strong ability in massively and parallel processing data and self-learning.Introducing rough neurons into network can combine the advantages of the rough set and neural network.The rough neuron network not only has a strong learning ability,but also can deal with imprecise information because the inputs of the network are double values instead of single values.As a result,the network with rough neurons can reflect the fluctuation of the load data for a period of time.Since the initial parameters of RBF neural network are usually set as random numbers,the prediction results are uncertain and the prediction accuracy is random.In this paper,the genetic algorithm is proposed to optimize the rough neural network to search the optimal initial weights of the network,and then the prediction accuracy of the network will be improved.Since wavelet neural network has advantages in signal analysis and has a strong learning ability,the training speed of the network is fast and the network is widely used in the field of prediction.This paper will introduce rough neurons into wavelet neural network and give a short-term load forecasting model based on rough wavelet neural network.This kind of network model has faster training speed and stronger generalization ability than rough RBF network.Since rough wavelet neural network has multiple parameters,the network calculation will become complicated in the process of training.In the other hand,the firefly algorithm has high computational efficiency and the parameters needed to be set are less.What's more,the algorithm's convergence speed is faster and optimization precision is higher.So the improved firefly algorithmis used to optimize related parameters of the network to find the best parameters in the global.It is helpful to accelerate the training speed and improve the generalization ability of the network to use the optimized rough neural network for short-term load forecasting.The simulation results show that the proposed method can effectively reflect the fluctuation information of load and the improved predictive model has strong nonlinear fitting ability and high prediction accuracy,so it is of great use in practical application.