面向城市轨道交通智能运维的数据耦合性与独立一致性研究

Research on Data Coupling and Independent Consistency for Urban Rail Transit Intelligent Operation-maintenance

[目的]智能运维背景下,现有算法准确度低,导致虚警率高,因此有必要开展列车运营数据的耦合性分析与独立一致性研究。[方法]从统计和数据驱动的角度对耦合性与独立一致性进行定义;根据加速度绝对值变化率将列车运行状态分为4个阶段:静止、平稳运行、起动加速及制动减速,并分别生成对应数据切片综合分位图、相关系数等方法;对牵引系统、制动系统累计正线运营数据进行分析,量化系统间的耦合关系;通过构建线性回归模型、支持向量机模型、LightGBM模型和K-近邻模型对于数据进行解耦处理,使牵引制动系统数据呈现正态性,相关变量服从独立性与一致性,以满足联合条件概率分布的前置条件。[结果及结论]数据解耦操作能够提升系统间原始数据的独立一致性;从工程实用角度出发,LightGBM模型在实时与离线状态下表现出最优的性能,在所有量化分析中均取得了50%及以上的优化率;采用解耦后的数据,能够在故障样本较少或者缺失的情况下,实现对潜在故障的预警功能,能有效降低智能运维的虚警率,同时提升故障预测的准确性。

[Objective]In the context of intelligent operation-maintenance,existing algorithm exhibits low accuracy,resulting in high false alarm rates,hence it is necessary to conduct analysis of data coupling and independent consistency in train operation data.[Method]The definition of coupling and independent consistency are established from both statistical and data-driven perspectives.Train operation states are divided into four stages based on the absolute value change rate of acceleration:stationary,stable operation,start-up acceleration,and brake deceleration.Corresponding data slices are generated,and comprehensive quantile plots and correlation coefficients are utilized to analyze the cumulative mainline operation data of traction and braking systems,quantifying the coupling relationship between systems.Linear regression,support vector machine,LightGBM,and K nearest neighbors models are constructed to decouple the data,rendering the traction and braking system data normal,with related variables conforming to independence and consistency,so as to meet the prerequisites of a joint conditional probability distribution.[Result&Conclusion]The research findings indicate that data decoupling enhances the independent consistency of raw data between systems.From an engineering perspective,the LightGBM model exhibits optimal performance in both real time and offline states,achieving no less than 50%optimization rates across all quantitative analyses.By utilizing decoupled data,it becomes feasible to issue early warnings for potential faults even in cases of limited or missing fault samples,effectively reducing false alarm rates in intelligent operation-maintenance while enhancing fault prediction accuracy.