基于机器学习的中小跨径桥梁技术状况评估

Evaluation of technical condition of medium and small span bridge based on machine learning

针对基于规范方法的公路桥梁技术状况评估需要人为确定各因素影响权重的问题,提出了一种基于机器学习的桥梁技术状况智能评估方法。参考桥梁检测评定规范建立了技术状况评估指标体系,基于山西省某地区388座中小跨径桥梁的检测数据建立桥梁状态数据库,利用机器学习算法构建桥梁各部件状态评估值和桥梁状态评估值非线性映射模型,并通过画学习曲线、网格搜索寻找算法最优参数。同时,比较了决策树(CART)、支持向量机(SVM)、随机森林(RF)、改进的梯度提升树(XGBoost)、自适应提升(AdaBoost)五种机器学习算法的评估性能。结果表明:采用集成算法比单一算法具有更好的效果,在训练集数量较少的情况下,各算法的预测准确率都在85%以上,尤其是集成算法AdaBoost对桥梁技术状况进行分类预测准确率达93%,说明AdaBoost算法可以较好地用于评估该地区桥梁的技术状况;该方法不仅可以利用检测报告中积淀的数据价值,还可以避免基于检测规范中繁琐的计算公式和固定权重计算桥梁总体技术状况等级,可为改进桥梁技术状况评估方法提供参考。利用该方法还可以建立桥梁构件状况数据库,实现对主梁、桥墩等构件的状态分类预测,进一步减小公路桥梁评估规范的分级扣分缺陷和人为确定因素权重对整体桥梁技术状况等级评估的影响。

Aiming at the problem that the evaluation of highway bridge technology condition based on specification method needs to determine the weight of each factor artificially,an intelligent evaluation method of bridge technology condition based on machine learning was proposed.The technical condition evaluation index system was established with reference to the bridge inspection and evaluation specification,and the bridge condition database was established based on the inspection data of 388 medium-and-small-span bridges,in Shanxi Province.The machine learning algorithm was used to construct the non-linear mapping model of the evaluation value of each component of the bridge and the evaluation value of the bridge condition,and the optimal parameters of the algorithm were searched by drawing the learning curve and grid search.At the same time,the evaluation performance of five widely used machine learning algorithms,classification and regression tree(CART),support vector machine(SVM),random forest(RF),improved gradient boosting tree(XGBoost)and adaptive boosting(AdaBoost)were compared.The results show that the ensemble algorithm has better effect than the single algorithm,and in the case of the few training sets the prediction accuracy of each algorithm is more than 85%,especially the prediction accuracy of AdaBoost algorithm for bridge technical condition reaches 93%,indicating that AdaBoost algorithm can be better used to evaluate the bridge technical condition in the region.The method in this paper can not only make use of the data value accumulated in the inspection report,but also avoid the use of tedious calculation formulas and fixed weights to calculate the overall technical condition of the bridge based on the inspection specifications.It can be used as an auxiliary evaluation method and provide reference for improving the evaluation method of bridge technical conditions.Using this method,we can also establish bridge component condition database to realize the condition prediction of beams,piers and other components,and further reduce the impact of grading deduction defects in highway bridge evaluation specifications and artificially determined factor weights on the grade evaluation of overall bridge technical conditions.7 tabs,8 figs,28 refs.