基于XGboost的省级路面技术状况指标衰变预测

Prediction of Decay of Provincial Level Road Technical Condition Index Based on XGboost

针对国省道路面技术状况指标的衰变预测问题,提出了一种基于XGboost(eXtreme Gradient boost)的路面技术状况指数衰变预测模型。以某省为例,经过数据融合将普通国省干线公路的各类数据整理,通过研究临近地区部分路段路面技术状况指数历年数值变化分析其变化规律,并结合养护历史进行数据异常值修复。构建XGboost预测模型并对比随机森林和LightGBM模型对路面技术状况指数PQI进行预测,得出XGboost预测模型的预测精度更高。结合某省特点,对该省不同公路分区域构建差异化衰变预测模型。结果表明:省级预测模型用于预估全省路面性能平均发展水平及趋势,XGBoost模型和LightGBM模型对全省级的路面性能指标预测精度较高,能够较好地还原路面性能指标值随时间变化的趋势,且XGBoost模型的预测效果最佳;片区级预测模型在考虑片区差异化因素的前提下建立;受样本量影响,分片区预测的相关系数较全省预测约低0.2。

In view of pavement technology status indicators prediction problem for national and provincial highway,the paper proposed a based on XGboost(eXtreme Gradient boost)pavement technology condition of exponential decay model with province as an example,through data fusion to ordinary of provincial trunk highways all kinds of data,through the research neighborhood sections of the road pavement condition index numerical analysis its change rule change throughout the years,The XGboost prediction model was built by repairing outliers combined with Maintenance history,and Pavement Maintenance Quality Index was predicted by comparing random forest and Light Gradient Boost Machine model,and the prediction accuracy of the XGboost prediction model was higher.Combining with the characteristics of different economic zone in one province,decay prediction model for highway subregion was built.The results showed that provincial prediction model is used to predict the average development level and trend of road performance in the whole province.XGBoost model and LightgBM model have higher prediction accuracy for road performance indexes in the whole province,which can well restore the trend of pavement performance index values over time,and the prediction effect of XGBoost model is the best;The region-level prediction model was established on the premise of considering the region-level differentiation factors.Due to the influence of the sample size,the correlation coefficient of the segmented prediction is about 0.2 lower than that of the whole province.