基于机器学习法的高速公路桥涵小修工程费用模型预测

Cost model prediction of minor repair project of bridge and culvert based on machine learning method

基于陕西省11条高速公路2008~2015年桥涵小修工程量清单历史数据,采用机器学习算法,以组成桥涵各构件的小修费用作为因变量,研究桥涵小修费用预测模型。通过灰色关联度模型分析桥涵小修费用的影响因素,采用皮尔逊相关系数检验法对各影响因素进行多重共线性检验,筛选出解释变量,使用岭回归和Lasso回归对桥涵各构件小修费用模型进行回归分析,得到桥涵小修总费用预测模型。结果表明:桥涵小修工程费用的影响因素主要有通车年限、桥梁(涵洞)长度、年平均当量轴次、桥涵所处地区的年均降雨量和温度及车道数;基于模型预测所依托高速公路2016~2017年桥涵小修费用,并与该年度实际费用进行Wilcoxon符号秩检验,检验结果均大于0.05,验证了机器学习法预测桥涵小修工程费用模型的有效性,预测结果能为分配养护费用、提高养护决策水平提供合理建议。

Based on the historical data of bill of quantities of bridge and culvert minor repair project on 11 expressways in Shaanxi province from 2008 to 2015,the machine learning algorithm was used to study the prediction model of bridge and culvert minor repair cost with the minor repair cost of each component of bridge and culvert as the dependent variable.Through the grey correlation degree model,the influencing factors of the minor repair cost of bridge and culvert were analyzed.The Pearson correlation coefficient test method was used to test the multicollinearity of each influencing factor,and the explanatory variables were selected.Ridge regression and Lasso regression were used to analyze the minor repair cost model of each component of bridge and culvert,and the total cost prediction model of minor repair of bridge and culvert was obtained.The results show that the main influencing factors of the cost of bridge and culvert minor repair projects are years of operation,length of bridge(culvert),annual average equivalent axles,annual average rainfall and temperature in the area where the bridges and culverts are located,and number of lanes.Based on the model prediction,the bridge and culvert minor repair cost of the expressway from 2016 to 2017 is carried out,and the Wilcoxon signed rank test is carried out with the actual cost of the year.The test results are all greater than 0.05,which verifies the effectiveness of the machine learning method to predict the bridge and culvert minor repair project cost model.The prediction results can provide reasonable suggestions for allocating maintenance costs and improving maintenance decision-making level.