Research On Visual Analysis Of Multi-factor Dimensionality Reduction Of Power Load Meteorology And Improved ELM Forecasting Method

The development of social economy brings challenges to the power supply sector.Accurate power load forecasting can ensure the stable operation of power system.Based on the analysis of the classification and basic characteristics of power load,the paper analyzes the visual dimension reduction of various meteorological factors which affect the power load forecasting,and reduces the coupling between multi-dimensional data.The improved Extreme Learning Machine(ELM)prediction method is studied to improve the prediction accuracy.The main work of this paper is as follows:First,introduce the influence of regional meteorological factors on the accuracy of power load forecasting.To simplify the prediction model,An improved T-distributed Neighbor Embedding(Tsne)algorithm is proposed to solve the Stochastic Neighbor Embedding(SNE)visualization dimension reduction algorithm is used to reduce the dimension of meteorological data,which solves the problem that SNE algorithm has poor visualization effect and the data structure is easy to change.In view of the difficulty in screening the confusion parameters of Tsne algorithm,this paper proposes an automatic screening criterion method of confusion parameters,which simplifies the complex interactive selection of graphics and text to numerical comparison,and completes the automatic screening of parameters,effectively avoiding the subjectivity of artificially setting parameters in the process of using Tsne algorithm.Through comparative experiments,it is verified that Tsne dimension reduction has a higher degree of retention of the original data,and the effectiveness of automatic screening of confusion parameters is verified.Secondly,aiming at the limitation of ELM power load forecasting model,the superiority of Moth Flame Optimization(MFO)algorithm in solving complex problems with constraints and unknown search space is proposed to improve ELM.Then,the MFOELM prediction model is obtained,which solves the problem of unstable ELM weight output and insufficient prediction accuracy.The optimization experiment of the function verifies the superiority of MFO parameters.Through the 48-point power load forecasting experiment of the power grid,the method in this paper and the Particle Swarm Optimization(PSO)algorithm are used to compare and analyze the parameter optimization performance and prediction performance of the improved ELM model.It is verified that the MFO algorithm improves the problem of unstable output weights of the ELM neural network,and better improves the prediction accuracy of the ELM model.Finally,aiming at the problem of the above method that the forecast error is larger than the forecast error at other time points at the moment when the load change is extremely large,the Adaptive Grasshopper Optimization Algorithm(AGOA)is proposed to improve the ELM algorithm.Based on the Grasshopper Optimization Algorithm(GOA),an adaptive strategy is introduced to adjust the process of grasshopper exploration and development,and a new update mechanism is obtained.Through the mechanism of variable step length to guide the particles to update,it can improve the individual particles to reach the optimum,increase the reliability of the update guidance process and accelerate the particle convergence speed.The optimization experiment of multi-peak and multi-valley function verifies the superiority of AGOA optimization parameters.Through the 96-point power load forecasting experiment of the power grid,AGOAELM eliminated the points with large forecast errors,and further improved the forecast accuracy and forecast stability.