铁道学报
鐵道學報
철도학보
2014年
1期
81-87
,共7页
韩晋%杨岳%陈峰%吴湘华
韓晉%楊嶽%陳峰%吳湘華
한진%양악%진봉%오상화
轨道不平顺%神经网络%非等时距%灰色模型%加权%残差修正
軌道不平順%神經網絡%非等時距%灰色模型%加權%殘差脩正
궤도불평순%신경망락%비등시거%회색모형%가권%잔차수정
track irregularity%neural network%non-equal interval%grey model%weight%residual modification
对轨道不平顺的发展趋势进行有效预测,可以提高铁路线路养护的维修效率,保障行车安全。根据轨道不平顺的发展特性,提出一种基于非等时距加权灰色理论和神经网络法的组合预测方法。该方法通过构建非等时距加权灰色预测模型,将原始TQI序列的平均值作为累加序列初值,将连续累积函数的积分面积作为背景值,对累加序列进行加权处理,较好地反映了时间序列对轨道不平顺预测结果的贡献。在此基础上,引入 BP 神经网络模型对TQI预测的残差序列进行修正,较好地克服了单一模型预测精度偏低的不足。分别对沪昆线上行两段线路的轨道不平顺进行预测,结果表明该预测方法相对误差平均值分别为2.76%和2.08%,预测结果的后验差比值分别为0.121和0.151,精度等级达到1级。
對軌道不平順的髮展趨勢進行有效預測,可以提高鐵路線路養護的維脩效率,保障行車安全。根據軌道不平順的髮展特性,提齣一種基于非等時距加權灰色理論和神經網絡法的組閤預測方法。該方法通過構建非等時距加權灰色預測模型,將原始TQI序列的平均值作為纍加序列初值,將連續纍積函數的積分麵積作為揹景值,對纍加序列進行加權處理,較好地反映瞭時間序列對軌道不平順預測結果的貢獻。在此基礎上,引入 BP 神經網絡模型對TQI預測的殘差序列進行脩正,較好地剋服瞭單一模型預測精度偏低的不足。分彆對滬昆線上行兩段線路的軌道不平順進行預測,結果錶明該預測方法相對誤差平均值分彆為2.76%和2.08%,預測結果的後驗差比值分彆為0.121和0.151,精度等級達到1級。
대궤도불평순적발전추세진행유효예측,가이제고철로선로양호적유수효솔,보장행차안전。근거궤도불평순적발전특성,제출일충기우비등시거가권회색이론화신경망락법적조합예측방법。해방법통과구건비등시거가권회색예측모형,장원시TQI서렬적평균치작위루가서렬초치,장련속루적함수적적분면적작위배경치,대루가서렬진행가권처리,교호지반영료시간서렬대궤도불평순예측결과적공헌。재차기출상,인입 BP 신경망락모형대TQI예측적잔차서렬진행수정,교호지극복료단일모형예측정도편저적불족。분별대호곤선상행량단선로적궤도불평순진행예측,결과표명해예측방법상대오차평균치분별위2.76%화2.08%,예측결과적후험차비치분별위0.121화0.151,정도등급체도1급。
Effective prediction of the track irregularity development trend can improve the efficiency of railway line maintenance & repairs and so ensure traffic safety.According to track irregularity development character-istics,the combination prediction method based on the non-equal interval weighted grey theory and neural net-work method was proposed.With this method,by constructing the non-equal interval weighted grey prediction model,the average of the original TQI sequence was regarded as the cumulative sequence initial value,the inte-gral area of the continuous accumulation function was used as the background value,the cumulative sequence was processed by weighting.Therefore,the contribution of the time sequence to the track irregularity predic-tion results was better reflected.On this basis,the BP neural network model was introduced to amend the TQI prediction residuals sequence and to overcome the drawback of low prediction accuracy by a single model.The track irregularities of two up-direction sections of the Shanghai-Kunming Line were predicted respectively.The prediction results indicate that the means of the relative errors are 2.76% and 2.08% respectively,the posteri-or error ratios are 0.121 and 0.151 respectively,and the accuracy level reaches A.