基于改进非等时距灰色模型与PSO优化的轨道不平顺预测

Prediction of track irregularity based on improved non-equal interval grey model and PSO optimization

基于轨道动检数据开展的轨道不平顺预测研究,可用于指导以预防为主的养护维修作业。将改进非等时距灰色模型与粒子群优化(Particle Swarm Optimization,PSO)相结合来实现轨道质量指数(Track Quality Index,TQI)的高精度预测。在考虑TQI原始动检数据特征的基础上增加原始数据平滑优化、累加初始值优化和背景值优化等环节提出改进非等时距灰色模型。利用PSO算法的启发式搜索优势,以平滑优化参数、初始值优化参数、背景值优化参数为搜索目标,以预测平均相对误差为适应度函数,实现预测模型参数的自适应优化。在此基础上,基于优化参数计算得到拟合区间和外推区间上的TQI预测结果。选取沪昆线上行区段实测TQI数据对本文方法进行验证,并与既有TQI组合预测模型的预测结果进行了对比。研究结果表明,模型可有效捕捉TQI序列中的随机波动与实时演变趋势,在外推区间上的平均相对误差分别为2.04%和2.54%,预测性能优良;当TQI序列振荡特性显著时,本模型仍能保证预测结果的可靠性;与组合预测模型相比,该模型规避了残差修正、多算法融合等繁琐步骤,可通过有限优化环节提升预测精度,为轨道不平顺预测提供一种新方法。

The research of track irregularity prediction using track dynamic detection data can be used to guide the railway track preventive maintenance work.This paper combined the improved non-equal interval grey model and particle swarm optimization(PSO)to achieve high-precision prediction of track quality index(TQI).Considering the characteristics of TQI dynamic inspection data,the improved non-equal interval grey model was proposed with the raw data smoothing optimization,cumulative initial value optimization and background value optimization.The smooth optimization parameters,initial value optimization parameter and background value optimization parameter were set as the search objectives.The prediction average relative error as the fitness function,the prediction model parameters adaptive optimization was performed using the heuristic search advantage of PSO algorithm.Based on the optimization parameters,the TQI prediction results of the fitting interval and extrapolation interval were calculated.The proposed method was verified with the measured TQI data of the up line of Hukun railway,and the prediction results were also compared with the existing TQI combination prediction models.The results show that the model can capture the random fluctuation and real-time evolution trend in TQI series.The average relative errors of the extrapolation interval are 2.04%and 2.54%respectively.The prediction performance is excellent.When there is significant oscillation for TQI sequences,the reliability of prediction results can be guaranteed by the model.Compared with the combined prediction models,the model avoids some unnecessary steps such as residual correction and multi algorithm fusion,and improves the prediction accuracy through limited optimization steps,which provides a new way for track irregularity prediction.