基于改进多重BRB的网络货运单据异常识别方法

An Improved Multiple BRB Based Network Freight Bill Anomaly Identification Method

为解决网络货运平台单据异常识别准确率不高和网络货运中平台数据混乱的问题,提出了一种基于置信规则库(BRB)和证据推理(ER)算法的网络货运单据异常识别方法。结合网络货运平台运输业务异常逻辑,考虑相关性对特征量进行选择,将特征量作为BRB模型输入,建立了多异常模式并行检测BRB模型。运用ER推理算法对置信规则进行推理,形成了改进多重BRB异常识别模型,并运用差分进化算法对改进BRB模型的子系统进行了参数训练,将所有子系统异常识别数据作为BRB模型识别的结果。通过分析运输业务逻辑中的运输逻辑异常及成本比异常,提取BRB模型的先验属性,将2个子系统进行了并行检查并作为最终的异常识别结果。运用交通运输部网络货运信息交互系统中2020年某省部分特定商品的长途运输数据进行了研究,经过人工清洗验证及标记,可以准确得知的数据集22793条样本中有10986条数据被标为运输业务异常。根据人工标记数据集,可以得知BRB模型的异常识别正确率。通过对模型所得结果与人工数据进行十折交叉验证确定了模型的可用性,并与非BRB方法所得结果进行了对比。结果表明:改进后的BRB模型可以用来高效识别网络货运单据中难以进行识别的数据异常,所研究的部分规则已应用于交通运输部网络货运信息交互系统。

In order to solve the problems of low accuracy rate of freight bill anomaly identification in network freight platform and the platform data confusion in network freight,a method of network freight bill anomaly identification based on BRB and Evidence Reasoning(ER)algorithm is proposed.Combining the anomaly logic of transport business in the network freight platform and considering correlation,the feature quantity is selected as the input of the BRB model,and a multiple anomaly pattern parallel detection BRB model is established.The reasoning of the evidence rule is conducted by applying ER reasoning algorithm to form an improved multiple BRB anomaly identification model,and the parameters of the subsystems of the improved BRB model are trained by using differential evolution algorithm,and the data of all subsystem anomaly identification are used as the result of BRB model identification.The anomalies in the transport business logic and the cost ratio anomalies are analysed and prior attributes of the BRB model are extracted,and the 2 subsystems are checked in parallel and used as the final anomaly identification result.The long-distance transport data for selected specific commodities in a province in 2020 from MOT network freight information exchange system are used for study.After manual cleaning,verification and labeling,it can be accurately determined that 10986 data are marked as transport business anomaly out of the 22793 samples in the dataset.According to the manually labeled dataset,the accuracy of anomaly identification in the BRB model can be determined.The usability of the model is determined by cross verification on the result obtained from the model and the artificial data,and compared with the result obtained from non BRB methods.The result shows that the improved BRB model can be used to efficiently identify data anomalies that are difficult to be identified in network freight bills.Some of the studied rules are applied to the MOT network freight information exchange system.