摘要
为实现高速铁路路基压实质量快速可靠的评价,基于不确定性理论提出路基全断面压实度的评估方法。首先,建立预测最大干密度ρ_(dmax)的PSO-BPNN-AdaBoost(PBA)模型,通过填料的料源参数(d_(max)、b、m、EI、LAA、W_(ac)和W_(af))快速准确地获得ρ_(dmax);其次,引入Bootstrap算法修正PBA模型,通过区间形式量化ρ_(dmax)预测过程中误差引起的不确定性;最后,开展现场试坑试验,获取现场填料的实测干密度ρ_d和料源参数,并基于克里金插值(Kriging)算法获得路基试验段全断面的ρ_d和料源参数分布,进一步通过计算得到路基全断面压实度区间评估结果。结合现场试验,将全断面压实度区间评估方法应用于西南地区某站场路基施工最优摊铺厚度的确定,克服传统填料填筑碾压质量评价中仅依赖随机点干密度测试结果作为评价标准的局限性。结果表明,ρ_(dmax)预测中的不确定性包括认知不确定性和随机不确定性,计算认知误差和随机误差的方差可以获得预测总误差的方差,从而实现ρ_(dmax)预测过程中不确定性的量化。选取置信度95%对应的参数构建填料ρ_(dmax)预测区间,此时预测区间覆盖率(Prediction Interval Coverage Probability,P_i)、平均预测区间宽度(Mean Prediction Interval Width,M_p)和覆盖宽度综合指标(Coverage Width-based Criterion,C_w)分别为100%、0.469 0 g/cm和0.469 0 g/cm~3,且预测区间可较好地覆盖填料ρ_(dmax)实测曲线。在现场碾压过程中,选取填料摊铺厚度为40~50 cm,可使路基结构压实质量达到较好的状态。研究成果提高了基于机器学习评估路基压实度结果的可靠性,并对高铁路基的压实施工提供理论指导。
To achieve a rapid and reliable evaluation of compaction quality for high-speed railway subgrades,an innovative assessment method based on uncertainty theory for the compaction degree of the subgrade full-section was proposed.First,the PSO-BPNN-AdaBoost(PBA)model was established to predict the maximum dry densityρdmax,which can efficiently and accurately obtainρdmax by using various physical index properties(dmax,b,m,EI,LAA,Wac,and Waf).Second,the Bootstrap algorithm was introduced to enhance the PBA model and quantify the uncertainty caused by errors in predictingρdmax within intervals.Finally,on-site pit tests provided actual dry densityρd and physical index property data,which were then used alongside the Kriging interpolation algorithm to map the distribution ofρd and physical index properties across the subgrade test sections.This enabled a comprehensive assessment of the compaction degree for the full-section subgrade.By integrating field experiments,the interval assessment method for compaction degree of the subgrade full-section was applied to determine the optimal paving thickness for subgrade construction at a station in the southwest region of China.This method addressed the limitations of conventional fill material compaction quality assessment,which typically depends solely on dry density measurements from randomly selected points.The results indicate that the uncertainty in predictingρdmax includes cognitive and stochastic uncertainties.By calculating the variances of cognitive and stochastic errors,the variance of total prediction error can be obtained,thereby quantifying the overall uncertainty in predictingρdmax.Parameters corresponding to a 95%confidence level were selected to construct the prediction interval forρdmax of the fill material.At this confidence level,the prediction interval coverage probability(Pi),mean prediction interval width(Mp),and coverage width-based criterion(Cw)are 100%,0.4690 g/cm3,and 0.4690 g/cm3,respectively.Additionally,the prediction interval at this confidence level adequately covers the measured curve of fillρdmax.On-site compaction operations indicate that selecting a paving thickness of 40~50 cm can achieve a desired state of compaction quality for the subgrade structure.The research findings enhance the reliability of assessing subgrade compaction degree using machine learning techniques and offer theoretical guidance for the construction of high-speed railway subgrades.
作者
郝哲睿
陈晓斌
肖宪普
闫宏业
李泰灃
尧俊凯
谢康
HAO Zherui;CHEN Xiaobin;XIAO Xianpu;YAN Hongye;LI Taifeng;YAO Junkai;XIE Kang(School of Civil Engineering,Central South University,Changsha 410075,China;School of Civil Engineering,Shijiazhuang University of Railways,Shijiazhuang 050043,China;Railway Engineering Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)
出处
《铁道科学与工程学报》
北大核心
2025年第2期649-663,共15页
Journal of Railway Science and Engineering
基金
国家自然科学基金资助项目(51978674)
中国铁道科学研究院科技研究开发计划(2023QT002)。
关键词
高速铁路路基
级配碎石
振动压实
机器学习
不确定性理论
质量控制
high-speed railway subgrade
graded gravel
vibratory compaction
machine learning
uncertainty theory
quality control
