Bp Neural Network Based On APSOwM And Its Application In High-speed Railway Subsidence Prediction

Its designed speed is high and it builds costly,so it involves safety of so many passenger’s life.Its joint damage is immeasurable.Thus,high-speed railway has the high requirements to driving safety factor and stability of railway subgrade.Given these factors,it’s important to monitor and forecast the settlement of subgrade for its normal,safe and efficient operation of high-speed railway.At the same time,the existing various mathematical model based on mathematical statistics,such as Regression Analysis method,Artificial Neural Network,Grey theory,time series analysis and so on.Their structure is too simple,mostly used in the monitoring of dam,foundation pit,landslides and so on,and there are few applications in the field of high-speed railway subgrade settlement monitoring.These models has the weak ability of nonlinear mapping to the settlement data and their convergence process is easy to fall into local minimum.So studying on method of combining of the multiple model and established mathematical model of high-railway subgrade settlement prediction is the top priority.Starting from the ideas above,this paper introduce BP neural network with strong ability of nonlinear mapping,and Adaptive Particle Swarm Optimization with Mutation(APSOwM)with neighborhood impact factor,Research on optimized model that combine APSOwM with BP neural network,then predict high-speed railway subgrade settlement data by several optimized BP neural network model and finally contrastive analysis.In this paper,research results are as follows:1)Studying on Particle Swarm Optimization and Artificial Neural Network theory.Research finds that BP Neural Network has highly nonlinear mapping characteristics,also has great effect on forecast of nonlinear data.But single BP Neural Network is easily trapped into local minimum,and its connection weights and threshold are randomness,which leading to a long convergence procedure,so it can’t accurately forecast the subgrade settlement data.2)The adaptive mutation strategy and neighborhood impact factor are introduced into the particle swarm optimization.Thus,the update of particle velocities is affected by the optimal particle of neighborhood.At the same time,according to the average distance of particles,we can mutate several particles that fitness is smallest so that the Particle Swarm Optimization can jump out of local minimum and reach the best value.3)The connection weights and threshold of BP Neural Network are optimized by APSOwM,and establish a compositional model combine APSOwM with BP Neural Network.The program is composed by MATLAB,and applied to the monitoring and forecast of high-railway subgrade settlement.This compositional model is compared with single BP model,PSO-BP model and IPSO-BP model,and finally the result shows that the prediction accuracy of APSOwM-BP model is better than the other model mentioned above in the field of high-railway subgrade settlement forecast.