WO2016004774A1 - 一种基于时间序列分析的轨道交通故障诊断方法和系统 - Google Patents

一种基于时间序列分析的轨道交通故障诊断方法和系统 Download PDF

Info

Publication number
WO2016004774A1
WO2016004774A1 PCT/CN2015/075006 CN2015075006W WO2016004774A1 WO 2016004774 A1 WO2016004774 A1 WO 2016004774A1 CN 2015075006 W CN2015075006 W CN 2015075006W WO 2016004774 A1 WO2016004774 A1 WO 2016004774A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
curve
fault
model
time
Prior art date
Application number
PCT/CN2015/075006
Other languages
English (en)
French (fr)
Inventor
鲍侠
Original Assignee
北京泰乐德信息技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京泰乐德信息技术有限公司 filed Critical 北京泰乐德信息技术有限公司
Publication of WO2016004774A1 publication Critical patent/WO2016004774A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
    • G16Z99/00—Subject matter not provided for in other main groups of this subclass

Definitions

  • the invention belongs to the field of rail transit information technology, and particularly relates to a rail transit fault diagnosis method and system based on time series analysis.
  • CSM signal centralized monitoring system
  • equipment maintenance machines equipment maintenance machines
  • communication network management systems communication network management systems.
  • TJWX-I and TJWX-2000 models which are continuously upgraded to monitor the CSM system.
  • most stations use a computer monitoring system to realize real-time monitoring of the status of the station signal equipment, and provide basic information for the electrical department to grasp the current state of the equipment and conduct accident analysis by monitoring and recording the main operational status of the signal equipment. Based on, played an important role.
  • centralized monitoring CSM system is also widely deployed in urban rail centralized stations/vehicle sections, etc., for urban rail operation and maintenance.
  • the high-speed railway's unique RBC system, TSRS system, ATP system also faces the need to incorporate the signal centralized monitoring system, and also face to improve its monitoring capabilities, operation and maintenance capabilities, and equipment self-diagnosis capabilities. Demand.
  • Data mining analysis is the use of mathematical knowledge of statistical analysis, analysis of text, images, numerical data and other data, the hidden rules and relationships of data, the establishment of a data model for classification, clustering, statistics and other operations.
  • the mining and analysis of rail transit monitoring data is of great significance for judging and analyzing the technical faults of rail transit.
  • relying on manual experience analysis and judgment relying on manual analysis and analysis of faults in massive monitoring data requires a large amount of labor cost and time for analysis of failure causes. In many cases, faults are only found in the event of an accident, resulting in failures.
  • the technical problems of large workload, fault monitoring and low diagnostic efficiency when manually diagnosing railway signal system failures increase the risk of driving, and it is difficult to provide time guarantee for subsequent maintenance and rescue work. Therefore, it is the rail transit field to study more efficient rail traffic monitoring data analysis and fault analysis methods, improve the ability of rail transit fault analysis, check hidden dangers, treat hidden dangers, and promote fault repair status maintenance, thereby ensuring driving safety and improving transportation capacity. Urgent needs.
  • the invention provides a method and system for analyzing and troubleshooting faults of rail transit monitoring data based on time series analysis.
  • a method for fault diagnosis of rail transit based on time series analysis the steps of which include:
  • the fault identification parameter obtained by the real-time data in step 3) is used for early warning of the fault, and the early warning result is output;
  • step 5 Using the fault discriminating parameter obtained from the historical data of step 2) to train the fault classifier, and for the fault curve in the warning result of step 4), input the extracted curve model parameter into the fault classifier to obtain the fault diagnosis result.
  • a time-sequence analysis-based rail transit fault diagnosis system using the above method comprising:
  • Historical database for storing historical monitoring data, including normal data and fault data
  • Real-time database used to store real-time monitoring data
  • a data acquisition interface for receiving real-time data of the data acquisition system
  • the data early warning module connects the real-time database and the knowledge base, and is used for reading the real-time data by using the data early warning algorithm, and outputting the interpretation conclusion about the abnormal data;
  • the fault diagnosis module connects the historical database and the knowledge base, and uses the fault diagnosis algorithm to classify the abnormal data and output the fault diagnosis result.
  • the invention provides a rail transit monitoring data analysis and fault diagnosis scheme based on time series analysis, which can solve the influence of small curve shape changes on fault interpretation results, reduce the risk of misjudgment, and can change with time and environment.
  • the adaptive adjustment curve classification model enables the model to adapt to the dynamic changes, which can effectively solve the problems of large workload, low efficiency and high risk in the manual diagnosis of railway signal system faults in the prior art.
  • FIG. 1 is a schematic structural diagram of a rail transit signal device fault diagnosis system based on time series analysis.
  • FIG. 2 is a flow chart showing the steps of a method for fault diagnosis of a rail transit signal device based on time series analysis.
  • 3A, 3B, and 3C are graphs of ballast current data in the embodiment.
  • FIG. 1 is a schematic structural view of a rail transit monitoring data analysis and fault diagnosis system based on time series analysis of the present invention.
  • the system consists of a historical database, a real-time database, a knowledge base, a data collection interface, a data early warning module, and a fault diagnosis module, wherein:
  • Knowledge base a parametric model for establishing and storing time series current curves
  • Historical database used to store historical normal data and fault data
  • Real-time database used to store real-time monitoring data collected by the current data acquisition system
  • Data acquisition interface used to receive real-time data of the data acquisition system
  • Data early warning module Connect real-time database and knowledge base, use data alerting algorithm to interpret real-time data, and output judgment conclusions about abnormal data;
  • Fault diagnosis module connects the historical database and the knowledge base, and uses the fault diagnosis algorithm to classify the abnormal data and output the fault diagnosis result.
  • FIG. 2 is a flow chart showing the steps of a method for fault diagnosis of a rail transit signal device based on time series analysis using the above system, which is specifically described as follows:
  • the existing CSM system of the railway equipment is used to collect data for the rail transit signal equipment, and the rail transit signal equipment includes a power screen, a switch, a switch machine and the like.
  • the collected monitoring data includes historical data and real-time data.
  • Historical data refers to previously collected monitoring data stored in a database that is used to record various states of the device's past work.
  • Real-time data refers to the monitoring data collected by the current data acquisition system, which is used to judge the current working state of the device.
  • the purpose of the pre-processing is to process the data to be analyzed and generate data suitable for the analysis.
  • the pre-processing includes:
  • the invention is applicable to all time series curves, such as a switch current curve, an analog change trend curve, and the like.
  • the turnout current curve is a curve drawn by point-by-point connection of current values with a current as a vertical axis and a horizontal time at a fixed measurement interval, and implies electrical and mechanical characteristics during the turnout conversion process.
  • the ballast current curve has a certain relationship with the ambient temperature, the parameters of the ballast current curve constitute a seasonal time series.
  • ARIMA seasonal model Differential autoregressive moving average model
  • the autocorrelation coefficient and partial autocorrelation coefficient of the stationary process will attenuate to 0 in some way.
  • the former measures the simple and conventional correlation between the current sequence and the previous sequence, and the latter measures the influence of other previous sequences.
  • the degree of correlation between the current sequence and a previous sequence If the autocorrelation function of a time series quickly drops to 0 as the lag k increases, then we consider the sequence to be a stationary sequence; if the autocorrelation function does not rapidly decrease to 0 as k increases, it indicates This sequence is not stable. If the autocorrelation and partial correlation graphs of a time series do not have any patterns and the values are small, then the sequence may be independent variables that are independent of each other.
  • Difference is a method of eliminating the correlation between data before and after by item-by-item subtraction. It can eliminate the trend in the sequence and is the pre-processing of the mean smoothing of non-stationary sequences.
  • the parameters of the model are determined, and the ARIMA prediction model is established.
  • Is the maximum likelihood estimate of the time series model parameter ⁇ [ ⁇ 1 , ⁇ 2 , I, ⁇ N ] T ,
  • n is the number of independent variables. Make or Minimal A relatively reasonable order for the model.
  • the typical similarity measure is mostly the application of Euclidean distance, or some improvement techniques based on this.
  • the Euclidean distance measure has certain limitations, the main reason is the Euclidean distance as a similar measure, the time series data.
  • the shape distortion of the data on the time axis does not have a certain ability to recognize, and the robustness to data noise is poor. Some slight changes may cause the Euclidean distance between sequences to vary greatly.
  • Dynamic time warping is a pattern matching algorithm based on nonlinear dynamic programming. It obtains a set of dynamic time warping paths by measuring the similarity coefficients of two sets of time series. Generally, dynamic time warping path sets have different characteristics in different working states. If the total length of the curved path is the smallest, the data is most similar. It allows the sequence to be offset on the time axis. The points of the sequence do not require a one-to-one correspondence, and the distance between sequences of different lengths can be calculated, so that it has better robustness.
  • the data early warning algorithm is used to judge the abnormal data. The specific steps of the early warning algorithm are described later.
  • the classification algorithm is used to train the classifier, and then the fault diagnosis algorithm is used to obtain the fault diagnosis result.
  • the specific steps of the fault diagnosis algorithm are described later.
  • the classifier is trained by SVM (Support Vector Machine), Bayesian and other classification algorithms to obtain the fault classification model. Then, for the abnormal data judged by the early warning algorithm, the classifier is used to classify and obtain the corresponding fault.
  • SVM Serial Vector Machine
  • Bayesian Bayesian
  • other classification algorithms to obtain the fault classification model. Then, for the abnormal data judged by the early warning algorithm, the classifier is used to classify and obtain the corresponding fault.
  • the flow of the above algorithm is divided into two parts. First, data warning is performed on the real-time monitoring data, that is, step 4), and the data warning can determine the current data normal data and abnormal data, thereby obtaining the running state of the device. An alarm prompt is given for the abnormal data, and the fault diagnosis algorithm is used for fault diagnosis, that is, step 5).
  • the data early warning algorithm and fault diagnosis algorithm are specifically described below.
  • the dynamic path bending method is used to calculate the bending path length of the monitoring data and the normal curve model.
  • This example performs data processing on the upstream channel current curve data and the downlink channel current curve data monitored by a CSM.
  • the data is decoded to obtain current data at each moment.
  • the curve of the obtained ballast current data is extracted as follows:
  • the curve is basically horizontal, slightly downward.
  • Locking current segment a slightly upward smooth curve.
  • data analysis starts by judging the similarity between each set of data.
  • the similarity between the curve and the normal curve and the abnormal curve is calculated by calculating the dynamic time bending distance between each set of curves, and then the abnormal data is determined and based on the abnormal data. Diagnose the fault.
  • the curve of the data is shown in Figure 3B. It can be seen that the points of the two curves of FIG. 3A and FIG. 3B are not the same, so the distance calculated by the general Euclidean distance measurement method is 2.4951, and the distance is large.
  • the dynamic bending distance calculated by the method is 0.20915, and it can be determined that the curves are basically of the same type.
  • the curve of the data is shown in Figure 3C. It can be seen that this curve is significantly different from the first curve, that is, FIG. 3A.
  • the dynamic bending distance calculated by the method is 3.44175, and it can be determined that the curve does not belong to the same type, and should be a fault state curve.
  • By inputting the characteristic parameters into the classifier it can be obtained that the fault belongs to a start delay fault. It can also be seen from the graph that there is a period of time (about a few tenths of a second) before the start of the turnout action current is zero.

Landscapes

  • Testing And Monitoring For Control Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于时间序列分析的轨道交通故障诊断方法和系统。该方法包括：1)采集轨道交通信号设备的历史和实时监测数据；2)根据历史数据生成初始的时间序列曲线的参数模型，得到故障的判别参数；3)根据实时数据重新调整时间序列曲线的参数模型的参数，得到当前环境条件下的故障判别参数；4)对于当前环境条件下的时间序列曲线数据，通过步骤3)得到的故障判别参数进行故障预警，输出预警结果；5)利用步骤2)得到的故障判别参数训练故障分类器，对于步骤4)所述预警结果中的故障曲线，将提取的曲线模型参数输入到故障分类器，得到故障诊断结果。该方法能够有效解决人工诊断铁路信号系统故障时工作量大、效率低、风险高等问题。

Description

一种基于时间序列分析的轨道交通故障诊断方法和系统 技术领域
本发明属于轨道交通信息技术领域，具体涉及一种基于时间序列分析的轨道交通故障诊断方法和系统。
背景技术
目前,轨道交通(国有铁路、企业铁路和城市轨道交通)领域的监测维护产品主要有三类：CSM(信号集中监测系统)、各设备维护机、通信网管系统。为了提高我国铁路信号系统设备的现代化维修水平，从90年代开始，我国先后自主研制了TJWX-I型和TJWX-2000型等不断升级中的信号集中监测CSM系统。目前大部分车站都采用了计算机监测系统，实现了对车站信号设备状态的实时监测，并通过监测与记录信号设备的主要运行状态，为电务部门掌握设备的当前状态和进行事故分析提供了基本依据，发挥了重要作用。并且，对城市轨道交通信号设备，集中监测CSM系统也被广泛部署在城轨集中站/车辆段等处，供城轨运维使用。此外，伴随我国高速铁路的建设发展，高铁特有的RBC系统、TSRS系统、ATP系统，也面临着纳入信号集中监测系统的需求，也面临着提高其监测能力、运维能力，以及设备自诊断能力的需求。
数据挖掘分析是利用统计分析的数学知识，分析文本、图像、数值等数据，发现数据的隐含规则、关系，建立数据模型，用于对数据进行分类、聚类、统计等操作。轨道交通监测数据的挖掘分析，对于判断和分析轨道交通的技术故障具有重要的意义。但目前多是依靠人工经验分析判断，靠人工在海量的监测数据中进行故障的判断和分析，需要大量的人力成本以及故障原因分析的时间，很多情况下只有在出现事故时才发现故障，导致了人工诊断铁路信号系统故障时工作量大、故障监测与诊断效率低下等技术问题，增加了行车的危险，难以为后续的维修、救援等工作提供时间保障。因此，研究更高效的轨道交通监测数据分析和故障分析方法，提高轨道交通故障分析能力，查隐患，治隐患，推动故障修向状态修发展，从而保障行车安全、提高运力，是轨道交通领域的迫切需求。
发明内容
为了解决现有技术中人工诊断铁路信号系统故障时工作量大、效率低下、风险性高等技 术问题，本发明提供一种基于时间序列分析的轨道交通监测数据分析和故障诊断方法和系统。
本发明采用的技术方案如下：
一种基于时间序列分析的轨道交通故障诊断方法，其步骤包括：
1)采集轨道交通信号设备的历史和实时监测数据；
2)根据历史数据生成初始的时间序列曲线的参数模型，得到故障的判别参数；
3)根据实时数据重新调整时间序列曲线的参数模型的参数，得到当前环境条件下的故障判别参数；
4)对于当前环境条件下的时间序列曲线数据，通过步骤3)的由实时数据得到的故障判别参数进行故障预警，输出预警结果；
5)利用步骤2)的由历史数据得到的故障判别参数训练故障分类器，对于步骤4)所述预警结果中的故障曲线，将提取的曲线模型参数输入到故障分类器，得到故障诊断结果。
一种采用上述方法的基于时间序列分析的轨道交通故障诊断系统，其包括：
知识库，用于建立并存储时间序列曲线的参数模型；
历史数据库，用于存储历史监测数据，包括正常数据和故障数据；
实时数据库：用于存储实时监测数据；
数据采集接口，用于接收数据采集系统的实时数据；
数据预警模块，连接实时数据库和知识库，用于采用数据预警算法对实时数据进行判读，输出关于异常数据的判读结论；
故障诊断模块，连接历史数据库和知识库，用于采用故障诊断算法对异常数据进行分类，输出故障诊断结果。
本发明提供了一种基于时间序列分析的轨道交通监测数据分析和故障诊断方案，可以解决曲线形状微小变化对故障判读结果的影响，减小误判的风险，并且可以随时间和环境的变化，自适应的调整曲线的分类模型，使模型能够适应动态变化的情况，能够有效解决现有技术中人工诊断铁路信号系统故障时工作量大、效率低下、风险性高等问题。
附图说明
图1是基于时间序列分析的轨道交通信号设备故障诊断系统的结构示意图。
图2是基于时间序列分析的轨道交通信号设备故障诊断方法的步骤流程图。
图3A、图3B和图3C是实施例中的道岔电流数据曲线图。
具体实施方式
下面通过具体实施例和附图，对本发明做进一步说明。
图1是本发明的基于时间序列分析的轨道交通监测数据分析及故障诊断系统的结构示意图。该系统由历史数据库、实时数据库、知识库、数据采集接口、数据预警模块和故障诊断模块组成，其中：
知识库：用于建立并存储时间序列电流曲线的参数模型；
历史数据库：用于存储历史正常数据和故障数据；
实时数据库：用于存储当前数据采集系统所采集到的实时监测数据；
数据采集接口：用于接收数据采集系统的实时数据；
数据预警模块：连接实时数据库和知识库，用于采用数据预警算法对实时数据进行判读，输出关于异常数据的判读结论；
故障诊断模块：连接历史数据库和知识库，用于采用故障诊断算法对异常数据进行分类，输出故障诊断结果。
图2是采用上述系统的基于时间序列分析的轨道交通信号设备故障诊断方法的步骤流程图，对其具体说明如下：
1)采集轨道交通信号设备的监测数据
该步骤采用铁路设备既有的CSM系统对轨道交通信号设备进行数据采集，轨道交通信号设备包括电源屏、道岔、转辙机等设备。采集的监测数据包括历史数据和实时数据。历史数据是指存储在数据库中的以前采集到的监测数据，这些数据用来记录设备过去工作的各种状态。实时数据是指当前数据采集系统所采集到的监测数据，这些数据用来对设备当前的工作状态进行判断。
2)对采集的数据进行预处理
进行预处理的目的是为了对待分析的数据进行处理，生成适合于分析的数据，预处理包括：
(1)数据选择，选择合适的数据源，从数据中提取与分析任务相关的数据
(2)数据清理和集成，清除噪声数据、非可用数据，将原始数据规范化、标准化并将多个数据源组合在一起；
(3)数据转换，以合适的方式组织数据，将数据类型转换为可应用的类型，定义新的数据属性，减小数据维数和尺寸。
3)利用预处理后的数据建立曲线参数时间序列模型
本发明适用于所有的时间序列曲线，如道岔电流曲线、模拟量变化趋势曲线等。道岔电流曲线是一条以电流为纵轴、时间为横，以固定测量间隔的各电流值逐点连接绘制而成的曲线，蕴涵了道岔转换过程中的电气特性和机械特性。因为道岔电流曲线与环境温度有一定的关系，因此，道岔电流曲线的参数又构成了一种季节性变化的时间序列。这里采用ARIMA季节模型(差分自回归移动平均模型)建立道岔电流曲线特征参数的模型：
①利用自相关分析和偏自相关分析等方法，对时间序列的随机性、平稳性及季节性进行分析，并采用差分的方法对数据进行平稳化处理。然后根据自相关和偏自相关图，确定备选模型。
平稳过程的自相关系数和偏自相关系数都会以某种方式衰减趋于0,前者测度当前序列与先前序列之间简单和常规的相关程度,后者是在控制其它先前序列的影响后,测度当前序列与某一先前序列之间的相关程度。如果某一时间序列的自相关函数随着滞后k的增加而很快地下降为0,那么我们就认为该序列为平稳序列；如果自相关函数不随着k的增加而迅速下降为0,就表明该序列不平稳。如果一个时间序列的自相关和偏相关图没有任何模式,而且数值很小,那么该序列可能就是一些互相独立的无关的随机变量。
差分是通过逐项相减消除前后期数据相关性的方法，可剔除序列中的趋势性，是非平稳序列的均值平稳化的预处理。
②依据赤池信息准则或者Schwarz贝叶斯准则确定模型的参数，建立ARIMA预测模型。
具体的，采用的赤池信息准则如下：
Figure PCTCN2015075006-appb-000001
或者也可以采用Schwarz贝叶斯准则，如下：
Figure PCTCN2015075006-appb-000002
其中，
Figure PCTCN2015075006-appb-000003
为时间序列模型参数θ＝[θ1，θ2，I，θN]T的极大似然估计值，
Figure PCTCN2015075006-appb-000004
为在
Figure PCTCN2015075006-appb-000005
条件下的似然函数，
Figure PCTCN2015075006-appb-000006
为模型阶次或独立参数个数的估计,n为自变量的个数。使
Figure PCTCN2015075006-appb-000007
或者
Figure PCTCN2015075006-appb-000008
为最小的
Figure PCTCN2015075006-appb-000009
为模型相对合理的阶次。
③用选定的模型对将来某个时期的数值及可信区间做出预测。
4)计算动态时间弯曲路径，进而采用数据预警算法判断异常数据
典型的相似性测度绝大多数是应用欧几里得距离，或者是在此基础上的一些改进技术。但欧式距离测度存在一定的局限性，其主要原因是欧式距离作为相似测度，对时间序列数据 在时间轴上的数据形状扭曲变形没有一定的辨识能力，对数据噪声的鲁棒性较差，一些轻微的变化可能会使得序列之间的欧氏距离变化很大。
动态时间弯曲技术是基于非线性动态规划的一种模式匹配算法。它通过测度两组时间序列的相似性系数，得到一组动态时间弯曲路径集。通常，在不同的工作状态，动态时间弯曲路径集具有不同的特征。若弯曲路径总长度最小，则数据相似程度最大。它允许序列在时间轴上偏移，序列各点不要求一一对应，并且能够计算不同长度的序列之间的距离，因此具有更好的鲁棒性。
计算得到动态时间弯曲路径后，进而采用数据预警算法判断异常数据，预警算法的具体步骤见后文。
5)采用分类算法训练分类器，进而采用故障诊断算法得到故障诊断结果，故障诊断算法的具体步骤见后文。
采用SVM(支持向量机)、贝叶斯等分类算法训练分类器，得到故障分类模型。然后对于预警算法判断的异常数据，采用分类器进行分类，得到相应的故障。
上述算法的流程分为两个部分，首先对实时监测数据进行数据预警，即步骤4)，数据预警能够判断出当前的数据正常数据和异常数据，以此得到设备的运行状态。对于异常数据给出报警提示，并采用故障诊断算法进行故障诊断，即步骤5)。下面具体说明数据预警算法与故障诊断算法。
(1)数据预警算法：
①根据历史数据，得到曲线各特征点的季节变化的初始模型。
②根据曲线各特征点的季节变化模型，生成标准的正常曲线模型。
③对于实时的监测数据，采用动态时间弯曲方法计算监测数据与正常曲线模型的弯曲路径长度。
④将弯曲路径长度与事先设定的阈值进行比较，如果超出阈值，则进行报警。
⑤根据当前一段时间内的正常曲线数据，对参数的时间序列模型进行更新。
(2)故障诊断算法：
①根据历史数据，得到曲线各特征点的季节变化的初始模型。
②计算故障曲线与各种故障模型下的曲线参数的弯曲路径，进而对异常数据进行分类，输出故障诊断结果。
下面提供一个具体应用实例。本实例对某CSM监测的上行道岔电流曲线数据和下行道岔电流曲线数据进行数据处理。
根据数据格式，对数据进行解码，得到每一时刻的电流数据。对得到的道岔电流数据的曲线提取如下特征点：
启动段：电机启动时曲线骤升，形成一个尖峰，峰顶值通常为2至4A。
回落段：电流至峰点后迅速回落，弧线应平顺。
工作电流段：曲线基本呈水平状，略微向下。
锁闭电流段：为一略微向上的平顺曲线。
缓放段：电流缓慢下降到0。
为了对正常和异常的数据进行判断，数据分析从判断每组数据之间的相似性入手。在给出正常数据和异常数据的前提下，通过计算每组曲线之间的动态时间弯曲距离方法，计算出曲线和正常曲线和异常曲线之间的相似性，进而判定异常数据，并根据异常数据对故障进行诊断。
以上行道岔电流曲线数据为例：
2014年1月27日0点30分时所采集的数据如下：
0.000000,0.000000,0.000000,1.843137,3.647059,2.627451,2.000000,1.647059,1.450980,1.333333,1.215686,1.137255,1.098039,1.019608,0.980392,0.941176,0.901961,0.862745,0.823529,0.823529,0.784314,0.784314,0.745098,0.745098,0.745098,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.666667,0.666667,0.705882,0.666667,0.705882,0.666667,0.666667,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.745098,0.745098,0.745098,0.745098,0.784314,0.745098,0.784314,0.745098,0.784314,0.745098,0.745098,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.823529,0.823529,0.823529,0.823529,0.823529,0.823529,0.823529,0.823529,0.784314,0.784314,0.823529,0.823529,0.784314,0.784314,0.784314,0.745098,0.745098,0.705882,0.705882,0.666667,0.666667,0.666667,0.666667,0.627451,0.627451,0.627451,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.627451,0.627451,0.627451,0.627451,0.627451,0.627451,0.588235,0.588235,0.588235,0.588235,0.588235,0.509804,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000
数据的曲线如图3A所示。这些等时间间隔的数据可以看作是一个时间序列。
2014年1月28日4点55分时所采集到数据如下：
0.000000,0.000000,0.000000,0.000000,2.431373,2.588235,2.000000,1.686275,1.450980,1.333333,1.215686,1.137255,1.098039,1.019608,0.980392,0.941176,0.941176,0.901961,0.862745,0.823529,0.823529,0.784314,0.784314,0.745098,0.745098,0.745098,0.705882,0.705882,0.705882,0.705882,0.666667,0.705882,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.745098,0.745098,0.745098,0.745098,0.745098,0.745098,0.745098,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.823529,0.823529,0.823529,0.823529,0.823529,0.823529,0.823529,0.862745,0.823529,0.823529,0.823529,0.823529,0.823529,0.784314,0.784314,0.784314,0.784314,0.745098,0.745098,0.745098,0.705882,0.705882,0.705882,0.666667,0.666667,0.627451,0.627451,0.627451,0.627451,0.627451,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.627451,0.627451,0.627451,0.588235,0.627451,0.588235,0.588235,0.588235,0.588235,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000
数据的曲线如图3B所示。可以看出图3A和图3B两条曲线的点数并不相同，因此采用一般的欧式距离度量方法计算距离为2.4951，距离较大。而采用本方法计算得到的动态弯曲距离为0.20915，可以判定基本上属于同一类型的曲线。
2014年1月31日2点17分时所采集到数据如下：
0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,2.431373,2.588235,2.000000,1.686275,1.450980,1.333333,1.215686,1.137255,1.098039,1.019608,0.980392,0.941176,0.941176,0.901961,0.862745,0.823529,0.823529,0.784314,0.784314,0.745098,0.745098,0.745098,0.705882,0.705882,0.705882,0.705882,0.666667,0.705882,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.666667,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.705882,0.745098,0.745098,0.745098,0.745098,0.745098,0.745098,0.745098,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.784314,0.823529,0.823529,0.823529,0.823529,0.823529,0.823529,0.823529,0.862745,0.823529,0.823529,0.823529,0.823529,0.823529,0.784314,0.784314,0.784314,0.784314,0.745098,0.745098,0.745098,0.705882,0.705882,0.705882,0.666667,0.666667,0.627451,0.627451,0.627451,0.627451, 0.627451,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.588235,0.627451,0.627451,0.627451,0.588235,0.627451,0.588235,0.588235,0.588235,0.588235,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000,0.000000
数据的曲线如图3C所示。可以看出这条曲线和第一条曲线即图3A有明显差异，采用本方法计算得到的动态弯曲距离为3.44175，可以判定不属于同一类型的曲线，应为故障状态曲线。通过将特征参数输入分类器，可以得到该故障属于启动延迟故障。从曲线图中也可以看到启动前有一段时间(大约是零点几秒)道岔动作电流为零。
以上实施例仅用以说明本发明的技术方案而非对其进行限制，本领域的普通技术人员可以对本发明的技术方案进行修改或者等同替换，而不脱离本发明的精神和范围，本发明的保护范围应以权利要求所述为准。

Claims (8)

  1. 一种基于时间序列分析的轨道交通故障诊断方法，其步骤包括：
    1)采集轨道交通信号设备的历史和实时监测数据；
    2)根据历史数据生成初始的时间序列曲线的参数模型，得到故障的判别参数；
    3)根据实时数据重新调整时间序列曲线的参数模型的参数，得到当前环境条件下的故障判别参数；
    4)对于当前环境条件下的时间序列曲线数据，通过步骤3)的由实时数据得到的故障判别参数进行故障预警，输出预警结果；
    5)利用步骤2)的由历史数据得到的故障判别参数训练故障分类器，对于步骤4)所述预警结果中的故障曲线，将提取的曲线模型参数输入到故障分类器，得到故障诊断结果。
  2. 如权利要求1所述的方法，其特征在于，步骤1)对采集的数据进行预处理，包括：
    数据选择，选择合适的数据源，从中提取与分析任务相关的数据；
    数据清理和集成，清除噪声数据和非可用数据，将原始数据规范化、标准化并将多个数据源组合在一起；
    数据转换，以合适的方式组织数据，将数据类型转换为可应用的类型，并定义新的数据属性，减小数据维数和尺寸。
  3. 如权利要求1所述的方法，其特征在于，所述时间序列曲线为下列中的一种：道岔电流曲线、模拟量变化趋势曲线。
  4. 如权利要求3所述的方法，其特征在于：所述时间序列曲线为道岔电流曲线，采用ARIMA季节模型建立所述道岔电流曲线的特征参数的模型，首先利用自相关分析和偏自相关分析方法对时间序列的随机性、平稳性及季节性进行分析，并采用差分的方法对数据进行平稳化处理，然后根据自相关和偏自相关图，确定备选模型；再依据赤池信息准则或者Schwarz贝叶斯准则确定模型的参数，建立ARIMA预测模型。
  5. 如权利要求4所述的方法，其特征在于，采用的所述赤池信息准则为：
    Figure PCTCN2015075006-appb-100001
    采用的所述Schwarz贝叶斯准则为：
    Figure PCTCN2015075006-appb-100002
    其中，
    Figure PCTCN2015075006-appb-100003
    为时间序列模型参数θ＝[θ1，θ2，|，θN]T的极大似然估计值，
    Figure PCTCN2015075006-appb-100004
    为在
    Figure PCTCN2015075006-appb-100005
    条件下的似然函数，
    Figure PCTCN2015075006-appb-100006
    为模型阶次或独立参数个数的估计,n为自变量的个数。
  6. 如权利要求1所述的方法，其特征在于：步骤4)进行故障预警的具体步骤是：
    ①根据历史数据得到曲线各特征点的季节变化的初始模型；
    ②根据曲线各特征点的季节变化模型，生成标准的正常曲线模型；
    ③对于实时的监测数据，采用动态时间弯曲方法计算监测数据与正常曲线模型的弯曲路径长度；
    ④将弯曲路径长度与事先设定的阈值进行比较，如果超出阈值，则进行报警；
    ⑤根据当前一段时间内的正常曲线数据，对参数的时间序列模型进行更新。
  7. 如权利要求6所述的方法，其特征在于：步骤5)采用SVM或者贝叶斯分类算法训练故障分类器。
  8. 一种采用权利要求1所述方法的基于时间序列分析的轨道交通故障诊断系统，其特征在于，包括：
    知识库，用于建立并存储时间序列曲线的参数模型；
    历史数据库，用于存储历史监测数据；
    实时数据库：用于存储实时监测数据；
    数据采集接口，用于接收数据采集系统的实时数据；
    数据预警模块，连接实时数据库和知识库，用于对实时数据进行判读，输出关于异常数据的判读结论；
    故障诊断模块，连接历史数据库和知识库，用于采用故障分类器对异常数据进行分类，输出故障诊断结果。
PCT/CN2015/075006 2014-07-07 2015-03-25 一种基于时间序列分析的轨道交通故障诊断方法和系统 WO2016004774A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201410321920.3A CN104091070B (zh) 2014-07-07 2014-07-07 一种基于时间序列分析的轨道交通故障诊断方法和系统
CN201410321920.3 2014-07-07

Publications (1)

Publication Number Publication Date
WO2016004774A1 true WO2016004774A1 (zh) 2016-01-14

Family

ID=51638786

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/075006 WO2016004774A1 (zh) 2014-07-07 2015-03-25 一种基于时间序列分析的轨道交通故障诊断方法和系统

Country Status (2)

Country Link
CN (1) CN104091070B (zh)
WO (1) WO2016004774A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106503439A (zh) * 2016-10-21 2017-03-15 国网福建省电力有限公司 一种基于数据挖掘的采集故障预警系统的方法
CN110097134A (zh) * 2019-05-08 2019-08-06 合肥工业大学 一种基于时间序列的机械故障早期诊断方法
US11282240B2 (en) 2020-04-23 2022-03-22 Saudi Arabian Oil Company Generation and implementation of multi-layer encoded data images

Families Citing this family (40)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104091070B (zh) * 2014-07-07 2017-05-17 北京泰乐德信息技术有限公司 一种基于时间序列分析的轨道交通故障诊断方法和系统
CN105045256B (zh) * 2015-07-08 2018-11-20 北京泰乐德信息技术有限公司 基于数据对比分析的轨道交通实时故障诊断方法和系统
CN106528565B (zh) * 2015-09-11 2019-08-06 北京邮电大学 监测系统的数据处理方法和装置
CN105787511A (zh) * 2016-02-26 2016-07-20 清华大学 基于支持向量机的道岔故障诊断方法及系统
CN106326933B (zh) * 2016-08-25 2019-05-24 中科君胜(深圳)智能数据科技发展有限公司 基于人在回路的智能自学习故障诊断方法
CN106446538B (zh) * 2016-09-19 2019-06-25 中山大学 基于动态时间规整的车辆终点及行程时间计算方法
CN106406295B (zh) * 2016-12-02 2019-02-26 南京康尼机电股份有限公司 基于多工况的轨道交通车辆门系统故障诊断及预警方法
CN106649727B (zh) * 2016-12-23 2019-12-24 南京航空航天大学 一种用于无人机飞行控制系统故障检测的数据库构建方法
CN106953766B (zh) * 2017-03-31 2020-06-26 北京奇艺世纪科技有限公司 一种报警方法及装置
CN107144721A (zh) * 2017-05-26 2017-09-08 北京戴纳实验科技有限公司 用于实验室设备的电流和电压的大数据分析构架
CN107451004B (zh) * 2017-07-01 2020-07-31 南京理工大学 一种基于定性趋势分析的道岔故障诊断方法
CN107832173B (zh) * 2017-11-02 2020-11-10 河海大学 一种基于工况检测的城市轨道交通车辆实时故障诊断方法
CN108039971A (zh) * 2017-12-18 2018-05-15 北京搜狐新媒体信息技术有限公司 一种告警方法及装置
CN108052092B (zh) * 2017-12-19 2020-02-21 南京轨道交通系统工程有限公司 一种基于大数据分析的地铁机电设备状态异常检测方法
CN108268901B (zh) * 2018-01-25 2021-05-18 中国环境监测总站 一种基于动态时间弯曲距离发现环境监测异常数据的方法
CN108459579B (zh) * 2018-02-02 2019-08-27 郑州轻工业学院 基于时间序列模型系数的半导体批次过程故障诊断方法
CN108416022B (zh) * 2018-03-07 2020-06-09 电信科学技术第五研究所有限公司 一种流式数据实时保真曲线绘制模型实现系统及方法
CN108985380B (zh) * 2018-07-25 2021-08-03 西南交通大学 一种基于聚类集成的转辙机故障识别方法
CN109163683B (zh) * 2018-08-27 2021-03-02 成都云天智轨科技有限公司 轨道波磨病害诊断方法与装置
CN109446484A (zh) * 2018-10-12 2019-03-08 湖南磁浮技术研究中心有限公司 一种基于曲线比对的数据判别方法及系统
CN109656228A (zh) * 2018-12-04 2019-04-19 江苏大学 一种地铁信号系统车载设备故障自动诊断方法
CN109783903B (zh) * 2018-12-28 2023-01-24 佛山科学技术学院 一种基于时间序列的工业用水管道故障诊断方法及系统
CN109948812A (zh) * 2019-02-27 2019-06-28 东软集团股份有限公司 确定故障原因的方法、装置、存储介质及电子设备
CN110148132B (zh) * 2019-05-28 2022-04-19 中南大学 一种基于尺寸特征相似性度量的模糊故障诊断预报表示方法
CN110515781B (zh) * 2019-07-03 2021-06-22 北京交通大学 一种复杂系统状态监测及故障诊断方法
CN110809280B (zh) * 2019-10-21 2022-10-18 北京锦鸿希电信息技术股份有限公司 一种铁路无线网络质量的检测预警方法及装置
CN110912775B (zh) * 2019-11-26 2021-03-16 中盈优创资讯科技有限公司 物联网企业网络故障的监控方法及装置
TWI737073B (zh) * 2019-12-10 2021-08-21 中華電信股份有限公司 陳情案件之時序分析系統與方法
CN110968076A (zh) * 2019-12-14 2020-04-07 中车大连电力牵引研发中心有限公司 一种基于以太网技术的列车智能网络监视系统
CN113541599B (zh) * 2020-04-15 2023-02-07 阳光新能源开发股份有限公司 一种逆变器温升降额的诊断方法及其应用系统
CN111537841B (zh) * 2020-06-30 2021-08-06 上海交通大学 一种适用于接地故障类型识别的优化方法及系统
CN112565422B (zh) * 2020-12-04 2022-07-22 杭州佳速度产业互联网有限公司 一种对电力物联网故障数据的识别方法、系统和存储介质
CN113033838A (zh) * 2021-03-10 2021-06-25 哈尔滨市科佳通用机电股份有限公司 一种机车信号全生命周期监测管理系统及管理方法
CN113759785A (zh) * 2021-04-28 2021-12-07 龙坤(无锡)智慧科技有限公司 一种基于大数据分析技术实现设备故障预警的方法
CN112883340B (zh) * 2021-04-30 2021-07-23 西南交通大学 基于分位数回归的轨道质量指数阈值合理性分析方法
CN113339699A (zh) * 2021-05-10 2021-09-03 上海氢枫能源技术有限公司 一种加氢站数字化诊断系统及方法
CN113521484B (zh) * 2021-08-20 2023-03-28 华东师范大学 神经反馈训练系统
CN114066274A (zh) * 2021-11-22 2022-02-18 通用电气（武汉）自动化有限公司 一种设备性能预警的方法及装置
WO2023116246A1 (zh) * 2021-12-22 2023-06-29 卡斯柯信号有限公司 一种基于铁路信号系统监测数据的智能浏览装置
CN114661802B (zh) * 2022-01-25 2024-04-05 桂林电子科技大学 一种工厂设备数据高效采集分析系统及方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050149570A1 (en) * 2003-12-19 2005-07-07 Kabushiki Kaisha Toshiba Maintenance support method, storage medium, and maintenance support apparatus
CN102829967A (zh) * 2012-08-27 2012-12-19 中国舰船研究设计中心 一种基于回归模型系数变化的时域故障识别方法
CN104091070A (zh) * 2014-07-07 2014-10-08 北京泰乐德信息技术有限公司 一种基于时间序列分析的轨道交通故障诊断方法和系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103345207B (zh) * 2013-05-31 2015-06-24 北京泰乐德信息技术有限公司 一种轨道交通监控数据的挖掘分析与故障诊断系统
CN112949715A (zh) * 2013-12-31 2021-06-11 北京泰乐德信息技术有限公司 一种基于svm的轨道交通故障诊断方法
CN103714383B (zh) * 2014-01-09 2017-03-22 北京泰乐德信息技术有限公司 一种基于粗糙集的轨道交通故障诊断方法和系统

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050149570A1 (en) * 2003-12-19 2005-07-07 Kabushiki Kaisha Toshiba Maintenance support method, storage medium, and maintenance support apparatus
CN102829967A (zh) * 2012-08-27 2012-12-19 中国舰船研究设计中心 一种基于回归模型系数变化的时域故障识别方法
CN104091070A (zh) * 2014-07-07 2014-10-08 北京泰乐德信息技术有限公司 一种基于时间序列分析的轨道交通故障诊断方法和系统

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
BAI, LIANG ET AL.: "Vibration Fault Diagnosis Based on Time-Series Similarity Mining for Hydropower Units", JOURNAL OF HYDROELECTRIC ENGINEERING, vol. 29, no. 6, 31 December 2010 (2010-12-31), ISSN: 1003-1243 *
HU , XIAOPING ET AL.: "Fault Diagnosis of Liquid Rocket Engine by Dynamic Time Warping Combined with Decision Tree Method", JOURNAL OF NATIONAL UNIVERSITY OF DEFENSE, vol. 29, no. 4, 31 December 2007 (2007-12-31), ISSN: 1001-2486 *
LI, YINGSHUN ET AL.: "Fault Diagnosis of the System for Conveying the Ammunition Based on Time Series Analysis", CHINA SCIENCE AND TECHNOLOGY INFORMATION, 30 November 2012 (2012-11-30) *
ZHANG, RONGBIAO ET AL.: "Study of PH Sensor Fault Diagnosis Based on Time Series and Expert System", JOURNAL OF ELECTRONIC MEASUREMENT AND INSTRUMENT, vol. 15, no. 3, 30 September 2001 (2001-09-30), ISSN: 1000-7105 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106503439A (zh) * 2016-10-21 2017-03-15 国网福建省电力有限公司 一种基于数据挖掘的采集故障预警系统的方法
CN110097134A (zh) * 2019-05-08 2019-08-06 合肥工业大学 一种基于时间序列的机械故障早期诊断方法
US11282240B2 (en) 2020-04-23 2022-03-22 Saudi Arabian Oil Company Generation and implementation of multi-layer encoded data images

Also Published As

Publication number Publication date
CN104091070A (zh) 2014-10-08
CN104091070B (zh) 2017-05-17

Similar Documents

Publication Publication Date Title
WO2016004774A1 (zh) 一种基于时间序列分析的轨道交通故障诊断方法和系统
CN110764493B (zh) 一种适用于高速铁路的phm应用系统、方法及存储介质
CN107941537B (zh) 一种机械设备健康状态评估方法
US20210374506A1 (en) Method for predicting bearing life based on hidden markov model and transfer learning
CN105045256B (zh) 基于数据对比分析的轨道交通实时故障诊断方法和系统
CN107451004B (zh) 一种基于定性趋势分析的道岔故障诊断方法
WO2011145496A1 (ja) 監視診断装置および監視診断方法
CN112660745B (zh) 托辊故障智能诊断方法、系统及可读存储介质
CN107703920A (zh) 基于多元时间序列的列车制动系统的故障检测方法
KR102176638B1 (ko) 선로전환기의 고장진단 및 장애예측이 가능한 철도신호시스템 및 이를 이용한 선로전환기 고장진단 및 장애예측방법
CN113610188B (zh) 弓网接触力非区段异常识别方法及装置
CN108776452B (zh) 一种特种设备现场维保监控方法及系统
CN117131110B (zh) 一种基于关联分析的容性设备介质损耗监测方法及系统
CN111931849A (zh) 一种水电机组运行数据趋势预警方法
CN107688820B (zh) 一种基于bcsa优化支持向量机的电梯故障诊断方法
CN103487250A (zh) 基于二维投射的煤矿设备预知维护方法
CN116610105B (zh) 一种基于数据融合的轧机机械运行故障监测方法及系统
CN110633314A (zh) 车联网数据处理方法及装置
CN107562041A (zh) 转辙机故障预警方法、装置、设备及计算机可读存储介质
CN113987905A (zh) 一种基于深度信念网络的自动扶梯制动力智能诊断系统
Davari et al. A fault detection framework based on lstm autoencoder: A case study for volvo bus data set
Giannoulidis et al. Investigating thresholding techniques in a real predictive maintenance scenario
CN112434979A (zh) 道岔系统健康评估方法
Doboszewski et al. Data driven detection of railway point machines failures
CN116757336B (zh) 基于数据驱动的轨道交通风险的预测方法及系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 15819078

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 15819078

Country of ref document: EP

Kind code of ref document: A1