基于随机森林算法的土石坝压实质量动态评价研究

坝体压实质量评价是土石坝施工安全控制的关键,其中干密度是评价压实质量的重要指标,但是施工现场通过随机试坑试验获取干密度的方法难以全面反映仓面压实质量,同时压实质量评价模型存在缺乏深入分析和量化料源参数不确定性的问题。针对上述不足,本文基于碾压实时监控系统提出了考虑料源参数和评价过程随机不确定性的压实质量动态评价模型,其功能主要包括以下3个方面:(1)采用信息熵理论量化土石坝料源参数的不确定性;(2)在影响干密度的指标中增加P5含量和湿度两个指标,反映了级配与气象要素对压实质量的影响,并在考虑料源参数不确定性影响的条件下基于随机森林算法对压实质量评价模型进行了动态求解;(3)采用插值结果可信度高的Kriging方法进行全仓面压实质量动态评价,弥补了试坑试验有限检测点难以全面反映仓面压实质量的不足。将该模型应用于某工程心墙区的压实质量评价,采用五折交叉验证和F检验等方法验证了该模型的可行性,并与BP神经网络和线性回归评价方法进行对比分析,验证了本评价方法的一致性、代表性和优越性。

Study on dynamic evaluation of compaction quality of earth rock dam based on Random Forest

Evaluation of compaction quality of dam body is the key to control the safety of earth-rock dam construction, and the dry density is an important index to evaluate the compaction quality. However, the method of obtaining dry density through random pit test is difficult to fully reflect the compaction quality of the work area. Meanwhile, there is a lack of in-depth analysis and quantification of the uncertainty of parameters of material sources in the compacted quality evaluation model. In view of the above shortcomings, based on the real-time compaction monitoring system, a dynamic compaction quality evaluation model considering the random uncertainty in material sources and evaluation process is proposed. Its functions mainly include the following three aspects:(1)the information entropy theory is used to quantify the uncertainty of the parameters of material source;(2)P5 content and humidity are added to the indexes affecting dry density, which reflect the influence of gradation and meteorological factors on the compaction quality. Meanwhile, random forest algorithm is used to dynamically solve the compaction quality evaluation model under the condition of considering the uncertainty of the material parameters;(3)The Kriging method with high interpolation results is used to realize the dynamic evaluation of the whole work area compaction quality, which makes up for the problem that the limited detection points cannot fully reflect the compaction quality of the work area. The model is applied to evaluate the compaction quality of an engineering core wall area,and the feasibility of the model has been verified by the five-fold cross validation and F test and compared with the BP neural network and linear regression method. The analysis results show the consistency, representative and superiority of this evaluation method.