Research On Improved Neural Network Load Forecasting Based On Large Data Analysis

With the rapid development of the economy,the power industry is facing more and more challenges,of which the power system load forecast for the entire power industry is significant.At present,the power system load forecast has become a hot topic of scientific research.Only by grasping accurately the future load trends,can the overall layout of the power system be mastered exactly and can the safe and stable operation environment of the grid be maintained.Human life and the environment changes greatly,making the influence of the power load gets become increasingly complex and relevant data show high-dimensional large data characteristics.This paper takes into account the influence of meteorological factors on power load forecasting,and puts forward the method based on multi-weather factor LASSO and principal component analysis feature extraction and neural network load forecasting improved by LM-BP,which is effectively improve the prediction effect and increases the prediction accuracy.First of all,the paper analyzes the load characteristics in detail,highlights the importance of the external environment on the load characteristics and describes the concrete implementation of load forecasting steps and analysis methods of prediction error.Secondly,considering the importance of weather factors on the impact of power load,it proposes to introduce the multi-weather factors into the short-term load forecasting model,and use the principal component analysis method to preprocess the multi-parameter weather factors.This paper analyzes the theory of principal component analysis algorithm and takes the economic indicators of multi-provincial autonomous regions as an example to prove the advantages of the method to deal with multi-parameter complex problems.Then,the BP neural network structure principle and its algorithm improvement are studied in detail.The LM algorithm is used to optimize the BP neural network training process,the advantages of both the Newton method and the gradient descent method are synthesized to significantly improve the learning rate and effectively avoid falling into the local minimum and through the simulation to fully verify the improved network model of good performance.Finally,the principal component analysis method is used to preprocess the multi-parameter weather factors,simplify the data structure,eliminate the redundant information,get the weather feature quantity and the historical load as the modeling object and establish the LM-BP improved neural network short-term load forecasting model.Taking the actual power load in the southern part of the United States as an example,the feasibility and superiority of the proposed method are proved by load forecasting and error analysis.