CN109754110B - Early warning method and system for traction converter faults - Google Patents
Early warning method and system for traction converter faults Download PDFInfo
- Publication number
- CN109754110B CN109754110B CN201711094900.7A CN201711094900A CN109754110B CN 109754110 B CN109754110 B CN 109754110B CN 201711094900 A CN201711094900 A CN 201711094900A CN 109754110 B CN109754110 B CN 109754110B
- Authority
- CN
- China
- Prior art keywords
- data
- fault
- early warning
- information
- train
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/60—Other road transportation technologies with climate change mitigation effect
- Y02T10/72—Electric energy management in electromobility
Abstract
The application discloses a traction converter fault early warning method, which comprises the following steps: extracting operation data of a rectification inversion module and a network from historical data, and constructing an EOVW energy index by utilizing the operation data of the rectification inversion module to obtain data characteristics and fault information; correlating the data characteristics with network operation data according to the vehicle type, the vehicle number and time, performing cluster analysis operation on the correlated data according to the vehicle type and the fault type to obtain a time range of data abnormality before the fault, and attaching a fault early warning label to the abnormal data; and introducing all data into the random forest classifier for iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, classifying the actual running data by using the trained classifier, and sending early warning information if the data does not pass. The service life of the traction converter is prolonged, and the operation and maintenance cost is reduced. The application also discloses a traction converter fault early warning system simultaneously, which has the beneficial effects.
Description
Technical Field
The application relates to the technical field of rail transit fault early warning, in particular to a traction converter fault early warning method and system.
Background
The train is one of the modern main transportation tools, plays a great role in passenger and goods transportation, and the safety problem of the train is always focused by people in all areas, wherein the traction converter in the train traction main circuit is taken as an important component of the train, and the safety problem of the train traction main circuit is more important.
In the prior art, in order to monitor the working state of the converter, various sensors (such as voltage and current sensors) are installed in the converter rectification inversion module, and when the module fails, the converter traction control unit sends a failure signal to a cab to remind a driver to take measures such as resetting, blocking the failure converter, starting a standby converter, even stopping the train, and the like, so that the safety of the train is ensured.
However, in the prior art, only the operation state of the rectifying inversion module is detected by using a sensor, and an alarm is given when a fault signal is detected, namely, the fault signal is detected and sent to a cab only after the fault is actually detected. And once a serious fault (such as thyristor or diode burst) occurs in the converter, the repairing and updating costs are too high.
Therefore, how to provide a fault early warning mechanism for a traction converter, which can analyze according to data and discover the sign of fault occurrence in advance, is a problem to be solved by those skilled in the art.
Disclosure of Invention
The utility model provides a traction converter fault early warning method and system, which has the advantages that the data-driven model early warning method is adopted, so that on one hand, the accuracy is higher, on the other hand, invisible connection between the fault which is not easy to detect and the device can be dug out from a large amount of data, the service life of the traction converter is prolonged, the operation and maintenance cost is reduced, and the stability, the safety and the reliability of a train are promoted.
In order to solve the technical problems, the application provides a method for early warning of traction converter faults, which comprises the following steps:
respectively extracting from traction converter historical data and network historical data of a train to obtain rectifying and inverting module operation data and network operation data of the train, and constructing an EOVW energy index by utilizing the rectifying and inverting module operation data to obtain data characteristics and fault information of the rectifying and inverting module; wherein the fault information includes a fault type;
The data characteristics and the network operation data are associated according to different vehicle types, vehicle numbers and time of the train, and associated data are obtained;
performing cluster analysis operation on the associated data according to different vehicle types and fault types to obtain a time range of data abnormality before the fault, and attaching a fault early warning label to the abnormal data in the time range;
introducing the abnormal data and the normal data added with the fault early warning label into a random forest classifier for iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, and obtaining a trained random forest classifier; the random forest classifier is constructed according to the fault types and the dichotomy;
and analyzing the actual running data of the train by using the trained random forest classifier, and if the train does not pass through the trained random forest classifier, sending early warning information through a preset path.
Optionally, constructing an EOVW energy index by using the operation data of the rectifying and inverting module to obtain data characteristics and fault information of the rectifying and inverting module, including:
extracting EOVW energy indexes of the sensor information by wavelet transformation from the operation data of the rectifying inversion module in a second unit;
And taking the EOVW energy index as the data characteristic of the rectification inversion module of the current second, and obtaining the fault information.
Optionally, performing cluster analysis operation on the associated data according to different vehicle types and the fault types to obtain a time range of abnormality of the data before the fault, including:
performing hierarchical clustering processing on the associated data according to different vehicle types and different fault types to obtain hierarchical clustered data; the hierarchical clustering processed data comprises cluster information and center point position information;
and carrying out K-means clustering processing by utilizing the cluster information and the central point position information, and determining the time range of data abnormality before failure.
Optionally, after the early warning information is sent through the preset path, the method further includes:
judging whether the same early warning information is received for a plurality of times within a preset time;
if yes, judging that the train is abnormal, and sending a judging result of the train abnormality to a train driver.
Optionally, the early warning method further comprises;
recording and storing all received early warning information, and generating an early warning information log so as to carry out subsequent analysis and investigation according to the early warning information log.
The application also provides an early warning system of traction converter trouble, this early warning system includes:
the operation data acquisition and analysis unit is used for respectively extracting the operation data of the rectification inversion module and the network operation data of the train from the traction converter historical data and the network historical data of the train, and constructing an EOVW energy index by utilizing the operation data of the rectification inversion module to obtain the data characteristics and the fault information of the rectification inversion module; wherein the fault information includes a fault type;
the data association unit is used for associating the data characteristics with the network operation data according to different vehicle types, vehicle numbers and time of the train to obtain associated data;
the cluster analysis operation execution unit is used for executing cluster analysis operation on the associated data according to different vehicle types and fault types to obtain a time range of data abnormality before the fault, and attaching a fault early warning label to the abnormal data in the time range;
the classifier training unit is used for introducing the abnormal data and the normal data added with the fault early warning label into the random forest classifier for iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, so as to obtain a trained random forest classifier; the random forest classifier is constructed according to the fault types and the dichotomy;
And the actual fault prediction unit is used for analyzing the actual running data of the train by using the trained random forest classifier, and if the train does not pass through the trained random forest classifier, the pre-warning information is sent through a preset path.
Optionally, the operation data acquisition and analysis unit includes:
the wavelet transformation subunit is used for extracting EOVW energy indexes of the sensor information through wavelet transformation by taking the operation data of the rectifying and inverting module as a unit of seconds;
and the processing subunit is used for taking the EOVW energy index as the data characteristic of the rectification inversion module of the current second and obtaining the fault information.
Optionally, the cluster analysis operation execution unit includes:
the hierarchical clustering processing subunit is used for performing hierarchical clustering processing on the associated data according to different vehicle types and different fault types to obtain hierarchical clustering processed data; the hierarchical clustering processed data comprises cluster information and center point position information;
and the K-means clustering processing subunit is used for carrying out K-means clustering processing by utilizing the cluster information and the central point position information and determining the time range of data abnormality before failure.
Optionally, the actual fault prediction unit further includes:
the expert rule judging subunit is used for judging whether the same early warning information is received for a plurality of times within the preset time;
and the expert rule judgment pass through the subunit and is used for judging that the train is abnormal and sending the judging result of the train abnormality to a train driver.
Optionally, the early warning system further includes:
and the record storage unit is used for recording and storing all the received early warning information and generating an early warning information log so as to carry out subsequent analysis and investigation according to the early warning information log.
According to the early warning method for the traction converter faults, the rectification inversion module operation data and the network operation data of the train are respectively extracted from the traction converter historical data and the network historical data of the train, and the rectification inversion module operation data are utilized to construct an EOVW energy index to obtain the data characteristics and the fault information of the rectification inversion module; the data characteristics and the network operation data are associated according to different vehicle types, vehicle numbers and time of the train, and associated data are obtained; performing cluster analysis operation on the associated data according to different vehicle types and fault types to obtain a time range of data abnormality before the fault, and attaching a fault early warning label to the abnormal data in the time range; introducing the abnormal data and the normal data added with the fault early warning label into a random forest classifier for iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, and obtaining a trained random forest classifier; and analyzing the actual running data of the train by using the trained random forest classifier, and if the train does not pass through the trained random forest classifier, sending early warning information through a preset path.
Obviously, the technical scheme provided by the application is that firstly, the situation that the historical data of the rectification inversion module is inconsistent with the network data acquisition frequency is processed, the processed data are associated, the associated data are divided according to the vehicle type and fault type information and sequentially pass through two hierarchical clustering algorithms, so that the time range of the abnormal data before the fault occurs is obtained, a fault early warning label is added to the abnormal data, and finally, the normal data and the abnormal data are uniformly put into random forest classification generated by the fault type to carry out binary judgment. The early warning method has higher accuracy on one hand and can dig out invisible connection between faults which are not easy to detect and devices from a large amount of data on the other hand by adopting a data-driven model early warning method, so that the service life of the traction converter is prolonged, the operation and maintenance cost is reduced, and the stability, the safety and the reliability of a train are promoted. The application also provides an early warning system for the fault of the traction converter, which has the beneficial effects and is not repeated here.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings may be obtained according to the provided drawings without inventive effort to a person skilled in the art.
Fig. 1 is a flowchart of a method for early warning of a traction converter fault provided in an embodiment of the present application;
fig. 2 is a flowchart of another method for early warning of a traction converter fault according to an embodiment of the present disclosure;
fig. 3 is a flowchart of another method for early warning of a traction converter fault according to an embodiment of the present disclosure;
fig. 4 is a block diagram of a traction converter fault early warning system according to an embodiment of the present application;
fig. 5 is a schematic actual flow chart of a traction converter fault early warning system provided in an embodiment of the present application.
Detailed Description
The core of the application is to provide a traction converter fault early warning method and system, which have higher accuracy on one hand and can dig invisible connection between a fault which is not easy to detect and a device from a large amount of data on the other hand by adopting a data-driven model early warning method, so that the service life of the traction converter is prolonged, the operation and maintenance cost is reduced, and the stability, the safety and the reliability of a train are promoted.
For the purposes of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
With reference to fig. 1, fig. 1 is a flowchart of a method for early warning of a traction converter fault according to an embodiment of the present application.
The method specifically comprises the following steps:
s101: respectively extracting from traction converter historical data and network historical data of a train to obtain rectifying and inverting module operation data and network operation data of the train, and constructing an EOVW energy index by utilizing the rectifying and inverting module operation data to obtain data characteristics and fault information of the rectifying and inverting module; wherein the fault information includes a fault type;
the method aims at acquiring the most important operation data of the rectification inversion module in the traction converter and the network operation data of the whole train, constructing an EOVW energy index by utilizing the operation data of the rectification inversion module to obtain the data characteristics of the rectification inversion module and fault information acquired according to a sensor arranged on the rectification inversion module.
In actual operation, the frequency of acquiring the related data of the rectifying and inverting module by the sensor is high, which may reach thousands of times per second, and the acquisition frequency of the network operation data may be only once per second, if the data of the rectifying and inverting module are simply correlated with the network operation data and are subjected to subsequent analysis, a great amount of processing capacity is obviously consumed, and since not all the data acquired by the sensor have representativeness and use value, the data characteristics of thousands of data can be summarized by wavelet transformation, and the data characteristics can be rapidly and effectively acquired by combining the qualitative fault information and are conveniently correlated with the network operation data.
The main idea and structure of EOVW (Energy of Variation Wavelet) energy index, i.e. the adjusted wavelet decomposition energy value, still depends on wavelet analysis, but when extracting signal characteristic indexes, the energy value (characteristic quantity) is not obtained by simply calculating the square sum of decomposition coefficients, but the adjusted variation coefficient of the wavelet decomposition coefficients is calculated as a new characteristic quantity. And the calculation of the adjustment variation coefficient is carried out by taking the normal signal calculation adjustment mean value and the standard deviation as the standard instead of relying on the mean value and the standard deviation of the overall signal (including faults).
In wavelet transformation, a "wavelet" is a small waveform, which means that it has attenuation property, and a "wave" is a fluctuation of the waveform, and the amplitude of the waveform is an oscillation form between positive and negative. Compared with the Fourier transform, the wavelet transform is a localized analysis of time (space) frequency, gradually performs multi-scale refinement on signals (functions) through telescopic translation operation, finally achieves time subdivision at high frequency and frequency subdivision at low frequency, and can automatically adapt to the requirement of time-frequency signal analysis, thereby focusing on any details of the signals and solving the difficulties encountered in the Fourier transform.
Of course, as to how to obtain the data characteristics of the rectifying and inverting module from the processing in this step, wavelet transformation which is better for wavelet analysis can be used, and methods such as maximum value, minimum value, mean value, variance, quantile and the like can be selected to extract and obtain the EOVW energy index, and the specific EOVW energy index finally extracted by different methods has slight deviation, but is still representative. In addition, the analysis time and the calculation complexity of each analysis method are different, and in actual situations, the analysis time and the calculation complexity can be comprehensively considered according to the performance intensity of equipment equipped by each manufacturer, and the analysis method is not particularly limited.
S102: the data characteristics and the network operation data are associated according to different vehicle types, vehicle numbers and time of the train, and associated data are obtained;
on the basis of S101, this step aims to correlate the analyzed data features and the extracted network operation data according to the model and the number of different trains with reference to the time axis, because the specifications of traction converters installed on the trains of different models are not necessarily the same, and other devices which may affect the traction converters are also different. Meanwhile, trains with different numbers and the same vehicle type can have different influences due to the variability in manufacturing equipment parts with the same specification.
The aim of the step is to consider the possible difference factors, and correlate the data characteristics obtained by analyzing and transforming the operation data of the rectification inversion module with the network operation data on a time axis to obtain correlated data which is convenient for subsequent processing.
Of course, the possible difference factors include not only the vehicle type and the vehicle number, but also other influencing factors in actual situations, and in order to obtain the correlation result more accurately so as to obtain the determination result with the minimum error in the final bipartite determination, the possible difference factors may be comprehensively considered and selected according to the requirements of each manufacturer and combined with the performance of the vehicle-mounted processing device, which is not limited in detail herein.
S103: performing cluster analysis operation on the correlated data according to different vehicle types and fault types to obtain a time range of data abnormality before the fault, and adding a fault early warning label to the abnormal data in the time range;
on the basis of S102, the step aims to perform cluster analysis on the associated data according to different vehicle types and fault types, so as to find out the contained abnormal data from the associated data through cluster analysis operation, locate the time range of the abnormal data on a time axis and attach a fault early warning label to the abnormal data.
The cluster analysis refers to an analysis process of grouping a collection of physical or abstract objects into a plurality of classes composed of similar objects, and is a method of classifying the objects according to their characteristics (samples or indexes), aiming at reducing the number of the objects. The purpose of cluster analysis is to classify things with similar properties into one category, and the manner of dependence is how much the indexes have correlation.
The clustering is a process of classifying data into different classes or clusters, so that objects in the same cluster have great similarity, and objects among different clusters have great dissimilarity. From a statistical point of view, cluster analysis is one way to simplify data by data modeling. The traditional statistical cluster analysis method comprises a systematic cluster method, a decomposition method, an addition method, a dynamic cluster method, ordered sample clustering, overlapped clustering, fuzzy clustering and the like. Cluster analysis tools employing k-means, etc. algorithms have been added to many well-known statistical analysis software packages, such as SPSS, SAS, etc.
From a machine learning perspective, the clusters correspond to hidden modes. Clustering is an unsupervised learning process of searching clusters. Unlike classification, unsupervised learning does not rely on pre-defined classes or training instances with class labels, which require automatic determination by a cluster learning algorithm, i.e., the clusters are observational learning, rather than exemplary learning.
Namely, cluster analysis is exploratory analysis, people do not need to give a classification standard in advance in the classification process, and the cluster analysis can automatically classify from sample data. Different methods of cluster analysis often lead to different conclusions. Different researchers perform cluster analysis on the same set of data, and the number of clusters obtained is not necessarily consistent.
From the practical point of view, cluster analysis is one of the main tasks of data mining, and the clustering can be used as an independent tool to obtain the distribution condition of data, observe the characteristics of each cluster of data, concentrate on further analysis of specific clustering cooperation, and can also be used as a preprocessing step of other algorithms (such as classification and qualitative induction algorithms).
The method aims at analyzing abnormal data from the associated data by using cluster analysis, finding out sign data which finally causes faults when the sign data appear according to the correlation among the data, and summarizing an abnormal data set, namely carrying out fault early warning through the time range of the abnormal data set on a time axis. The above-mentioned clustering analysis methods are many, and the most suitable algorithm can be selected according to the actual situation, because different clustering algorithms may draw different conclusions for the same data, and it may be necessary to perform experiments to find the most suitable clustering algorithm, which is not limited herein, and only needs to be able to obtain the time range of the abnormal data according to the clustering algorithm.
S104: introducing abnormal data and normal data with fault early warning labels into the random forest classifier for iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, and obtaining the trained random forest classifier; the random forest classifier is constructed according to the fault types and the dichotomy;
on the basis of S103, the step aims to introduce the abnormal data set with the fault early warning label obtained through cluster analysis processing into a random forest classifier together with normal data, wherein the random forest classifier is built by combining all possible fault types obtained through historical data analysis with a dichotomy, the aim is to judge the abnormal data and the normal data one by one sequentially through all fault types by using the random forest classifier, judge which fault types the data are through learning the abnormal data, judge the data as the normal data through learning the normal data, finally ensure that the classification accuracy of all the random forest classifiers reaches a threshold value, namely, the abnormal data in the actual running data of the train can be accurately judged, and the specific fault type of the abnormal data is given out.
Random forests are a relatively new model of machine learning. The classical machine learning model is a neural network, which predicts accurately, but with a large computational effort. The algorithm of the classification tree invented by the Breiman et al in the eighties of the last century can greatly reduce the calculated amount by repeating the classification or regression of binary data. In 2001 Breiman combines classification trees into a random forest, i.e., randomizes the use of variables (columns) and the use of data (rows), generates many classification trees, and aggregates the results of the classification trees. The prediction precision of the random forest is improved on the premise that the operand is not obviously improved. And the random forest is insensitive to the multi-element collinearity, so that the random forest is more robust to missing data and unbalanced data, and the effect of thousands of explanatory variables can be well predicted.
Random forests, as their name implies, are structured in a random manner, where there are many decision trees in the forest, and there is no association between each decision tree in the random forest. After a forest is obtained, when a new input sample enters, each decision tree in the forest is judged, which class the sample should belong to (for a classification algorithm) is looked at, which class is chosen most, and the sample is predicted to be which class.
S105: and analyzing the actual running data of the train by using the trained random forest classifier, and if the train does not pass through the trained random forest classifier, sending early warning information through a preset path.
On the basis of S104, the step aims to carry out classification judgment on traction converter related data and network data obtained in actual operation of a train by using trained random forest classifiers, and if judgment of all the classifiers is not passed, early warning information is sent through a preset path. Because a decision tree is established for each fault type of the random forest classifier, if one decision tree does not pass, the decision tree represents that abnormality exists certainly, and the equipment can be considered to be in a normal state only if all the decision trees pass.
The preset path has various expression modes, can be transmitted through a network with communicated train contents according to a preset network channel, can also use a display screen which is simultaneously sent to a cab by using a wireless communication module as a backup so as to prevent which line fails and alarm information cannot be received, and the like, can be comprehensively considered and selected according to actual conditions, and is not particularly limited.
Furthermore, instead of using a random forest classifier as a training model, a classifier with similar effects such as GBDT and Xgboost can be used, and similarly, a method of combining hierarchical clustering with K-means clustering is selected for determining the time range of abnormal data, and other clustering algorithms or analysis by using correlation coefficients can be used to achieve the purpose, which is not limited in detail herein.
Further, due to possible short-time faults of the sensors, collected data are abnormal such as 'burrs', 'spikes', and the like, so that related decision trees are alarmed, and aiming at the possible problems, if each alarm signal is directly sent to a cab, the possibility of misjudgment is possibly greatly increased, and whether the same alarm signals are continuously received within a preset time and are in a preset number can be adopted to judge whether the abnormality really occurs or not, instead of misjudgment. Of course, the predetermined time may be set by itself, because the probability of erroneous judgment of different fault types may be different, and the specific predetermined number may also be set by itself, so as to meet the specific requirements and standards of different manufacturers, which are not limited herein.
Furthermore, all received early warning information can be recorded and stored, and all relevant information can be recorded and a detailed early warning information log can be generated, so that subsequent analysis and searching use can be performed according to the early warning information log.
Based on the above technical scheme, the early warning method for the traction converter fault provided by the embodiment of the application processes the situation that the historical data of the rectification inversion module is inconsistent with the network data acquisition frequency, correlates the processed data, divides the correlated data according to the vehicle type and fault type information and sequentially passes through two clustering algorithms to obtain the time range of the abnormal data before the fault occurs, adds a fault early warning label to the abnormal data, and finally uniformly puts the normal data and the abnormal data into random forest classification generated by the fault type to perform binary judgment. The early warning method has higher accuracy on one hand and can dig out invisible connection between faults which are not easy to detect and devices from a large amount of data on the other hand by adopting a data-driven model early warning method, so that the service life of the traction converter is prolonged, the operation and maintenance cost is reduced, and the stability, the safety and the reliability of a train are promoted.
With reference to fig. 2, fig. 2 is a flowchart of another method for early warning of a traction converter fault according to an embodiment of the present application.
The present embodiment is a specific limitation on how to construct the EOVW energy index to obtain the data characteristics and fault information of the rectifying and inverting module in S101 and how to perform hierarchical clustering operation in S103 in the previous embodiment, and other steps are substantially the same as those in the previous embodiment, and the same parts are referred to the relevant parts of the previous embodiment and are not repeated herein.
The method specifically comprises the following steps:
s201: extracting EOVW energy indexes of the sensor information by wavelet transformation from the operation data of the rectification inversion module in a second unit;
s202: taking the EOVW energy index as the data characteristic of the rectification inversion module of the current second, and obtaining fault information;
s201 and S202 of the present embodiment aim to extract an EOVW energy index of each sensor information in units of seconds from the operation data of the rectifying and inverting module by using wavelet transformation with a better waveform analysis result on small fluctuations, and use the EOVW energy index as the data feature of the rectifying and inverting module of the current second, and obtain fault information including the qualitative fault type from each sensor.
S203: the data characteristics and the network operation data are associated according to different vehicle types, vehicle numbers and time of the train, and associated data are obtained;
the step is the same as S102, and the related content can be referred to the description content of S102, which is not repeated here.
S204: performing hierarchical clustering processing on the correlated data according to different vehicle types and different fault types to obtain hierarchical clustered data; the hierarchical clustering processed data comprises cluster information and center point position information;
s205: and carrying out K-means clustering processing by using the cluster information and the central point position information, and determining the time range of data abnormality before failure.
In S204 and S205 of this embodiment, first, hierarchical clustering is used to process the associated data according to different vehicle types and different fault types, so as to obtain processed data including cluster information and center point position information, then K-means clustering is performed by using the cluster information and the center point position information, and a time range of data abnormality before the fault is determined.
Hierarchical clustering (Hierarchical Clustering) is a method of hierarchically decomposing a data set until a certain condition is satisfied. According to the classification principle, two methods, i.e., aggregation and splitting, can be classified.
Wherein, the aggregated hierarchical clustering is a bottom-up strategy, each object is first used as a cluster, then the clusters are combined into larger clusters until all objects are in one cluster or a certain termination condition is met, most hierarchical clustering methods belong to the class, and they are only different in definition of similarity among clusters; split hierarchical clustering is contrary to condensed hierarchical clustering, and adopts a top-down strategy, wherein all objects are firstly placed in the same cluster, and then gradually subdivided into smaller and smaller clusters until each object is self-clustered or a certain termination condition is reached.
The K-means algorithm is a hard clustering algorithm, is representative of a typical prototype-based objective function clustering method, is an adjustment rule of iterative operation by taking a certain distance from a data point to a prototype as an optimized objective function and utilizing a function extremum solving method. The K-means algorithm takes Euclidean distance as similarity measure, and aims at solving the optimal classification of a certain initial clustering center vector so as to minimize an evaluation index.
The K-means algorithm is a very typical distance-based clustering algorithm, and uses distance as an evaluation index of similarity, that is, the closer the distance between two objects is, the greater the similarity is. The algorithm considers clusters to be made up of objects that are close together, thus targeting a compact and independent cluster as the final target. The selection of k initial class cluster center points has a larger influence on the clustering result, because in the first step of the algorithm, any k objects are randomly selected as the centers of the initial clusters, and initially represent one cluster. The algorithm reassigns each object remaining in the dataset in each iteration to the nearest cluster based on its distance from the center of the respective cluster. After all the data objects are inspected, one iteration operation is completed, and a new cluster center is calculated. If the values do not change before and after an iteration, it is stated that the algorithm has converged.
Firstly, carrying out preliminary clustering analysis processing through hierarchical clustering to obtain cluster information and central point position information which are convenient for the subsequent K-means algorithm to use, and finally utilizing the K-means algorithm to induce the abnormal data set wanted by the application.
Fig. 3 is a flowchart of another method for early warning of a traction converter fault according to an embodiment of the present disclosure.
The content of this embodiment is added to the above embodiment, and other steps are substantially the same as those of the previous embodiment, and the same parts can be referred to the relevant parts of the previous embodiment, which are not described herein.
The method specifically comprises the following steps:
s301: analyzing actual running data of the train by using the trained random forest classifier, and if the train does not pass through the trained random forest classifier, sending early warning information through a preset path;
s302: judging whether the same early warning information is received for a plurality of times within a preset time;
s303: judging that the train is abnormal, and sending a judging result of the train abnormality to a train driver;
s304: judging that the train is not abnormal, and executing no operation;
s305: recording and storing all received early warning information, and generating an early warning information log so as to carry out subsequent analysis and investigation according to the early warning information log.
Based on the above technical scheme, the early warning method for the traction converter fault provided by the embodiment of the application obtains the data characteristic formed by the representative EOVW energy index by utilizing wavelet transformation to solve the problem of inconsistent information acquisition frequency, and sequentially passes the associated data through two clustering algorithms of hierarchical clustering and K-means so as to obtain the time range of the abnormal data before the fault occurrence by utilizing the advantages of the two clustering algorithms more accurately, and finally uniformly placing the normal data and the abnormal data into random forest classification generated by the fault type to perform binary judgment. The early warning method has higher accuracy on one hand and can dig out invisible connection between faults which are not easy to detect and devices from a large amount of data on the other hand by adopting a data-driven model early warning method, so that the service life of the traction converter is prolonged, the operation and maintenance cost is reduced, and the stability, the safety and the reliability of a train are promoted.
Because of the complexity and cannot be illustrated by one, those skilled in the art will recognize that many more examples exist with respect to the basic method principles provided in this application, and that such examples are within the scope of this application without undue creative effort.
Referring to fig. 4, fig. 4 is a block diagram of a traction converter fault early warning system according to an embodiment of the present application.
The early warning system may include:
the operation data acquisition and analysis unit 100 is used for respectively extracting the operation data of the rectification inversion module and the network operation data of the train from the traction converter historical data and the network historical data of the train, and constructing an EOVW energy index by utilizing the operation data of the rectification inversion module to obtain the data characteristics and the fault information of the rectification inversion module; wherein the fault information includes a fault type;
the data association unit 200 is configured to associate the data characteristics with the network operation data according to different vehicle types, vehicle numbers and time of the train, so as to obtain associated data;
the cluster analysis operation execution unit 300 is used for executing cluster analysis operation on the associated data according to different vehicle types and fault types to obtain a time range of abnormality of the data before the fault, and attaching a fault early warning label to the abnormal data in the time range;
the classifier training unit 400 is configured to introduce the abnormal data and the normal data with the fault early warning label attached to the random forest classifier to perform iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, thereby obtaining a trained random forest classifier; the random forest classifier is constructed according to the fault types and the dichotomy;
The actual fault prediction unit 500 is configured to analyze actual running data of the train by using the trained random forest classifier, and if the actual running data does not pass through the trained random forest classifier, send early warning information through a preset path.
Wherein the operation data acquisition and analysis unit 100 includes:
the wavelet transformation subunit is used for extracting EOVW energy indexes of the sensor information through wavelet transformation by taking second as a unit of rectifying and inverting module operation data;
and the processing subunit is used for taking the EOVW energy index as the data characteristic of the rectification inversion module of the current second and obtaining fault information.
Wherein the cluster analysis operation execution unit 300 includes:
the hierarchical clustering processing subunit is used for performing hierarchical clustering processing on the associated data according to different vehicle types and different fault types to obtain hierarchical clustering processed data; the hierarchical clustering processed data comprises cluster information and center point position information;
and the K-means clustering processing subunit is used for carrying out K-means clustering processing by utilizing the cluster information and the central point position information and determining the time range of data abnormality before failure.
The actual fault prediction unit 500 may further include:
the expert rule judging subunit is used for judging whether the same early warning information is received for a plurality of times within the preset time;
And the expert rule judgment pass through the subunit and is used for judging that the train is abnormal and sending the judging result of the train abnormality to a train driver.
Further, the early warning system may further include:
and the record storage unit is used for recording and storing all the received early warning information and generating an early warning information log so as to carry out subsequent analysis and investigation according to the early warning information log.
The above units may be applied to the following specific practical examples, and reference may be made to fig. 5, where fig. 5 is a schematic actual flow diagram of a traction converter fault early warning system provided in the present application:
firstly, data during train operation are respectively extracted from historical data of a train traction converter module and historical data of a train network. Because sampling frequencies of the two-side data are inconsistent, in order to correlate the two-side data, characteristic extraction operation is required to be carried out on the converter module data after data extraction, and the EOVW energy index of each sensor signal in one second is extracted by wavelet transformation. And after feature extraction is completed, the train network data and the converter module feature data are related according to the vehicle type, the vehicle number and the time. And then dividing the data according to the information of the vehicle type and the fault type, and obtaining the time range of the abnormal data before the fault by using hierarchical clustering and K-means clustering algorithm for two times. And finally, marking fault early warning labels on the corresponding abnormal data according to the fault types, merging all normal and abnormal data according to the vehicle types, putting the merged abnormal data into a random forest classifier to conduct classification judgment on the abnormal data and the normal data, and conducting iterative training through the labels until the accuracy rate meets the requirement.
Because the sampling frequency of the related sensor of the traction converter of the train is far higher than the sampling frequency of the network data of the train, and the driver can decide in consideration of enough time required for early warning, the invention extracts EOVW energy indexes of each sensor signal by wavelet transformation in the unit of 'second' of the traction converter data, and then sends the EOVW energy indexes to the network of the train, and the EOVW energy indexes are input into a trained vehicle-mounted random forest two-class classifier for data classification judgment after being correlated with the network data. Since there is a random forest two-classification model for each fault, each piece of prediction data needs to be predicted and judged by all the two-classification models. If all classifier results are displayed normally, the current traction converter is considered to work normally, otherwise, if one or more classifiers predict faults, the faults need to be further determined through expert rules. Since abnormal data values of the input model may be caused by burrs, spikes, etc. of data due to sensor problems, an expert generally performs abnormal diagnosis in a manner of "the same fault is continuously reported multiple times".
In the above manner, the following advantages can be possessed:
(1) The abnormal judgment threshold value of the traction converter module is more scientific and reasonable. Based on data driving, analysis modeling is carried out on historical data to obtain a judging rule of faults and equipment abnormality, and compared with the judging mode of setting a threshold value by experience knowledge at present, the judging rule is more scientific and reasonable;
(2) The problems that the data sampling frequency of the traction converter module is too high, the data waveform changes too fast, and effective analysis and diagnosis are difficult to carry out on data at the fault moment and the abnormal moment before the fault are solved. The invention adopts a method for associating traction converter module data with train network data, and the model comprehensively considers the influence of data on two sides on faults. When the abnormal characteristics of the data of the converter module are not obvious, the model mainly analyzes and considers the influence of network data with lower sampling frequency on faults; otherwise, the model mainly considers the data characteristics of the traction converter module;
(3) In modeling aspect, a clustering algorithm is adopted to determine time distribution of abnormal data before failure (e.g. abnormal fluctuation occurs in data of several seconds to several minutes before failure), and compared with the traditional method of determining the abnormal data range by experience knowledge, the method is more accurate and reliable.
In the description, each embodiment is described in a progressive manner, and each embodiment is mainly described by the differences from other embodiments, so that the same similar parts among the embodiments are mutually referred. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative elements and steps are described above generally in terms of functionality in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
Specific examples are set forth herein to illustrate the principles and embodiments of the present application, and the description of the examples above is only intended to assist in understanding the methods of the present application and their core ideas. It should be noted that it would be obvious to those skilled in the art that various improvements and modifications can be made to the present application without departing from the principles of the present application, and such improvements and modifications fall within the scope of the claims of the present application.
It should also be noted that in this specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
Claims (8)
1. The early warning method for the traction converter fault is characterized by comprising the following steps of:
respectively extracting from traction converter historical data and network historical data of a train to obtain rectifying and inverting module operation data and network operation data of the train, and constructing an EOVW energy index by utilizing the rectifying and inverting module operation data to obtain data characteristics and fault information of the rectifying and inverting module; wherein the fault information includes a fault type;
the data characteristics and the network operation data are associated according to different vehicle types, vehicle numbers and time of the train, and associated data are obtained;
performing cluster analysis operation on the associated data according to different vehicle types and fault types to obtain a time range of data abnormality before the fault, and attaching a fault early warning label to the abnormal data in the time range;
introducing the abnormal data and the normal data added with the fault early warning label into a random forest classifier for iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, and obtaining a trained random forest classifier; the random forest classifier is constructed according to the fault types and the dichotomy;
Analyzing actual running data of the train by using the trained random forest classifier, and if the train does not pass through the trained random forest classifier, sending early warning information through a preset path;
performing cluster analysis operation on the associated data according to different vehicle types and the fault types to obtain a time range of data abnormality before the fault, wherein the time range comprises the following steps:
performing hierarchical clustering processing on the associated data according to different vehicle types and different fault types to obtain hierarchical clustered data; the hierarchical clustering processed data comprises cluster information and center point position information;
and carrying out K-means clustering processing by utilizing the cluster information and the central point position information, and determining the time range of data abnormality before failure.
2. The method of claim 1, wherein constructing an EOVW energy index from the rectifier inverter module operational data to obtain data characteristics and fault information for the rectifier inverter module, comprises:
extracting EOVW energy indexes of the sensor information by wavelet transformation from the operation data of the rectifying inversion module in a second unit;
and taking the EOVW energy index as the data characteristic of the rectification inversion module of the current second, and obtaining the fault information.
3. The warning method according to claim 1, further comprising, after the warning information is transmitted through the preset path:
judging whether the same early warning information is received for a plurality of times within a preset time;
if yes, judging that the train is abnormal, and sending a judging result of the train abnormality to a train driver.
4. The method of claim 3, further comprising;
recording and storing all received early warning information, and generating an early warning information log so as to carry out subsequent analysis and investigation according to the early warning information log.
5. An early warning system for traction converter faults, comprising:
the operation data acquisition and analysis unit is used for respectively extracting the operation data of the rectification inversion module and the network operation data of the train from the traction converter historical data and the network historical data of the train, and constructing an EOVW energy index by utilizing the operation data of the rectification inversion module to obtain the data characteristics and the fault information of the rectification inversion module; wherein the fault information includes a fault type;
the data association unit is used for associating the data characteristics with the network operation data according to different vehicle types, vehicle numbers and time of the train to obtain associated data;
The cluster analysis operation execution unit is used for executing cluster analysis operation on the associated data according to different vehicle types and fault types to obtain a time range of data abnormality before the fault, and attaching a fault early warning label to the abnormal data in the time range;
the classifier training unit is used for introducing the abnormal data and the normal data added with the fault early warning label into the random forest classifier for iterative training until the classification accuracy of all the random forest classifiers reach a threshold value, so as to obtain a trained random forest classifier; the random forest classifier is constructed according to the fault types and the dichotomy;
the actual fault prediction unit is used for analyzing the actual running data of the train by utilizing the trained random forest classifier, and if the train does not pass through the trained random forest classifier, early warning information is sent through a preset path;
the cluster analysis operation execution unit includes:
the hierarchical clustering processing subunit is used for performing hierarchical clustering processing on the associated data according to different vehicle types and different fault types to obtain hierarchical clustering processed data; the hierarchical clustering processed data comprises cluster information and center point position information;
And the K-means clustering processing subunit is used for carrying out K-means clustering processing by utilizing the cluster information and the central point position information and determining the time range of data abnormality before failure.
6. The warning system of claim 5 wherein the operational data acquisition and analysis unit comprises:
the wavelet transformation subunit is used for extracting EOVW energy indexes of the sensor information through wavelet transformation by taking the operation data of the rectifying and inverting module as a unit of seconds;
and the processing subunit is used for taking the EOVW energy index as the data characteristic of the rectification inversion module of the current second and obtaining the fault information.
7. The early warning system of claim 5, wherein the actual fault prediction unit further comprises:
the expert rule judging subunit is used for judging whether the same early warning information is received for a plurality of times within the preset time;
and the expert rule judgment pass through the subunit and is used for judging that the train is abnormal and sending the judging result of the train abnormality to a train driver.
8. The warning system of claim 7, further comprising:
and the record storage unit is used for recording and storing all the received early warning information and generating an early warning information log so as to carry out subsequent analysis and investigation according to the early warning information log.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711094900.7A CN109754110B (en) | 2017-11-03 | 2017-11-03 | Early warning method and system for traction converter faults |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711094900.7A CN109754110B (en) | 2017-11-03 | 2017-11-03 | Early warning method and system for traction converter faults |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109754110A CN109754110A (en) | 2019-05-14 |
| CN109754110B true CN109754110B (en) | 2023-07-11 |
Family
ID=66402281
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711094900.7A Active CN109754110B (en) | 2017-11-03 | 2017-11-03 | Early warning method and system for traction converter faults |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109754110B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111985651A (en) * | 2019-05-22 | 2020-11-24 | 中国移动通信集团福建有限公司 | Operation and maintenance method and device for business system |
| CN110991472B (en) * | 2019-08-01 | 2023-10-03 | 南京航空航天大学 | Method for diagnosing minor faults of high-speed train traction system |
| CN110991240B (en) * | 2019-10-31 | 2023-06-09 | 湖南世优电力科技股份有限公司 | Single-phase fault identification and classification method for low-voltage neutral point ungrounded system |
| EP4063875A4 (en) * | 2019-11-18 | 2023-08-09 | Zhuzhou CRRC Times Electric Co., Ltd. | Multi-information fusion-based fault early warning method and device for converter |
| CN111060844A (en) * | 2019-12-09 | 2020-04-24 | 南京航空航天大学 | Interturn short-circuit fault diagnosis method and device for high-speed train traction transmission system |
| CN113071537B (en) * | 2020-01-06 | 2022-11-08 | 株洲中车时代电气股份有限公司 | Early warning method, device and system for cooling system |
| CN111259947A (en) * | 2020-01-13 | 2020-06-09 | 国网浙江省电力有限公司信息通信分公司 | Power system fault early warning method and system based on multi-mode learning |
| CN113809786A (en) * | 2020-07-23 | 2021-12-17 | 广东毓秀科技有限公司 | Method for predicting faults of UPS rectification and inversion module through big data |
| CN112801315A (en) * | 2021-01-28 | 2021-05-14 | 国网河北省电力有限公司电力科学研究院 | State diagnosis method and device for power secondary equipment and terminal |
| CN112819386A (en) * | 2021-03-05 | 2021-05-18 | 中国人民解放军国防科技大学 | Method, system and storage medium for generating time series data with abnormity |
| CN113034732B (en) * | 2021-03-29 | 2022-12-27 | 南京格物智能科技有限公司 | Method and device for diagnosing blockage of filter screen of traction converter |
| CN114492588A (en) * | 2021-12-29 | 2022-05-13 | 中国大唐集团科学技术研究总院有限公司华东电力试验研究院 | Method, system, equipment and storage medium for identifying faults of auxiliary equipment of thermal power plant |
| CN115792476B (en) * | 2023-02-06 | 2023-04-25 | 石家庄科林电气股份有限公司 | Charging pile rectifying module abnormality early warning method, device, terminal and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102221655B (en) * | 2011-06-16 | 2013-08-07 | 河南省电力公司济源供电公司 | Random-forest-model-based power transformer fault diagnosis method |
| CN104236933B (en) * | 2013-06-13 | 2017-12-26 | 同济大学 | A kind of potential faults method for early warning for train traction system |
| CN103901882B (en) * | 2014-04-15 | 2016-08-17 | 北京交通大学 | A kind of system and method for train dynamics system on-line monitoring fault diagnosis |
| CN106338664B (en) * | 2016-08-11 | 2019-03-29 | 中车株洲电力机车研究所有限公司 | A kind of train current transformer method for diagnosing faults and device |
| CN106338406B (en) * | 2016-10-19 | 2018-08-31 | 北京交通大学 | The on-line monitoring of train traction electric drive system and fault early warning system and method |
-
2017
- 2017-11-03 CN CN201711094900.7A patent/CN109754110B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109754110A (en) | 2019-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109754110B (en) | Early warning method and system for traction converter faults | |
| CN110378869B (en) | Steel rail fastener abnormity detection method with automatic sample marking function | |
| Spiegel et al. | Pattern recognition and classification for multivariate time series | |
| CN103699698B (en) | A kind of being based on improves Bayesian rail transit fault identification method and system | |
| US8886574B2 (en) | Generalized pattern recognition for fault diagnosis in machine condition monitoring | |
| CN109974782B (en) | Equipment fault early warning method and system based on big data sensitive characteristic optimization selection | |
| CN112949715A (en) | SVM (support vector machine) -based rail transit fault diagnosis method | |
| CN109947898B (en) | Equipment fault testing method based on intellectualization | |
| CN103630244A (en) | Equipment fault diagnosis method and system of electric power system | |
| US10552511B2 (en) | Systems and methods for data-driven anomaly detection | |
| CN104966076A (en) | Optical fiber intrusion signal classification and identification method based on support vector machine | |
| CN108333468B (en) | The recognition methods of bad data and device under a kind of active power distribution network | |
| Cong et al. | Applying wavelet packet decomposition and one-class support vector machine on vehicle acceleration traces for road anomaly detection | |
| CN103473540A (en) | Vehicle track incremental modeling and on-line abnormity detection method of intelligent traffic system | |
| Mechefske et al. | Fault detection and diagnosis in low speed rolling element bearings Part II: The use of nearest neighbour classification | |
| CN112326280A (en) | Fault detection method and device for train bogie and computer readable medium | |
| Arul et al. | Data anomaly detection for structural health monitoring of bridges using shapelet transform | |
| CN116257663A (en) | Abnormality detection and association analysis method and related equipment for unmanned ground vehicle | |
| Liu et al. | Active incremental learning for health state assessment of dynamic systems with unknown scenarios | |
| CN114563150B (en) | Bridge health online detection module generation method, detection method, tool box and device | |
| Kamranfar et al. | Pavement distress recognition via wavelet-based clustering of smartphone accelerometer data | |
| CN112882898A (en) | Anomaly detection method, system, device and medium based on big data log analysis | |
| CN112906672A (en) | Steel rail defect identification method and system | |
| Aquize et al. | Self-organizing maps for anomaly detection in fuel consumption. Case study: Illegal fuel storage in Bolivia | |
| Mejia et al. | Data quality enhancement and knowledge discovery from relevant signals in acoustic emission |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |