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

Short-term load forecasting is the precondition and important information for a grid to dispatch,control,and make an operation plan or market orientation reasonably.Scientific forecasting is the premise and guarantee of correct decision.Take a medium city whose annual electricity consumption is 29 billion kilowatt-hour as an example,the operation costs can be saved by 145 million yuan when the forecast error is reduced by 1%.Up to present,scholars have proposed various intelligent algorithms to improve the accuracy and the efficiency in prediction.Artificial neural network can deal with the nonlinear relationship of load data well without relying on human experience,just by training network and establishing the predictive model.This dissertation regards the short-term load forecasting as the research object,and the neural network as forecasting model to improve the accuracy and efficiency.The major achievements are as follows:Firstly,this dissertation introduces historical load data sources,and some abnormal data samples are corrected by the way of horizontal and vertical processing.The load characteristics are analyzed which contains the change rules of the daily load,the weekly load and the holiday load.Then the influencing factors of load forecasting are analyzed.Among of them,the meteorological factors as the main cause that impact the power load are done a detailed analysis.At the same time,the economic,price and date factors are also discussed in the chapter.It is helpful to improve the accuracy of load forecasting by understanding the load characteristics and changing rules correctly,and grasping the relationships between various factors and power load scientifically.Secondly,four kinds of neural network models of BP,RBF,ELM and Elman,and their characteristics are introduced.Then these four networks are used to establish the prediction model,and the predictive performances are verified.Based on the actual load data of a certain area in Jiangsu Province,the short-term load forecast is conducted,and the results show that the Elman neural network have a higher prediction accuracy.The mean average percentage error of Elman is 1.98%,while the other three models are above 2%,but the efficiency and accuracy of Elman model still needs to be improved.Thirdly,on the basis of the above two parts,this dissertation puts forward a method of short-term load forecasting based on PCA and MEA-Elman neural network.This method uses principal component analysis algorithm to analyze the meteorological factors and get the comprehensive meteorological indexes;then improves the excitation function of the Elman network,optimizes its weights and thresholds using the Mind Evolutionary Algorithm,and formulates a MEA-Elman network model to forecast the power load.The simulation results show that the proposed method can further improve the accuracy and efficiency.Finally,with the development of smart grids,communication network and sensor technology,the amount of electric power data grows exponentially,which has come into massive data.The traditional forecasting models which are mainly based on small scale data samples can't meet the need of faster training speed and higher forecasting accuracy when coping with massive data,thus this dissertation proposes a parallel load forecasting model based on MR-MEA-Elman to deal with it.At last,taking EUNITE competition load data samples for analysis,the results indicate that the proposed method can improve the computational efficiency and guarantee the load forecasting accuracy at the same time.