CN115266159A - Fault diagnosis method and system for train traction system - Google Patents

Fault diagnosis method and system for train traction system Download PDF

Info

Publication number
CN115266159A
CN115266159A CN202210914232.2A CN202210914232A CN115266159A CN 115266159 A CN115266159 A CN 115266159A CN 202210914232 A CN202210914232 A CN 202210914232A CN 115266159 A CN115266159 A CN 115266159A
Authority
CN
China
Prior art keywords
matrix
value
feature vector
feature
contactor
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.)
Withdrawn
Application number
CN202210914232.2A
Other languages
Chinese (zh)
Inventor
姚汤伟
章国鹏
陈润婧
王艺莹
叶雨欣
赵婉亦
陈灏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Normal University CJNU
Original Assignee
Zhejiang Normal University CJNU
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 Zhejiang Normal University CJNU filed Critical Zhejiang Normal University CJNU
Priority to CN202210914232.2A priority Critical patent/CN115266159A/en
Publication of CN115266159A publication Critical patent/CN115266159A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M17/00—Testing of vehicles
    • G01M17/08—Railway vehicles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Train Traffic Observation, Control, And Security (AREA)

Abstract

The application relates to the field of intelligent fault diagnosis, and particularly discloses a fault diagnosis method and system of a train traction system, wherein dynamic associated feature information of multiple parameters of the train traction system in a time dimension and high-dimensional implicit associated feature information of state information of a pre-charging contactor and a line contactor of the train traction system in the time dimension are respectively extracted by adopting a deep neural network model, and feature information of the pre-charging contactor and the state information of the line contactor of the train traction system in the time dimension are fused based on a cosine compatible loss function training method to diagnose the fault of the train traction system.

Description

Fault diagnosis method and system for train traction system
Technical Field
The present application relates to the field of fault intelligent diagnosis, and more particularly, to a fault diagnosis method and system for a train traction system.
Background
In the running process of the train, if any tiny or potential faults and hidden dangers cannot be diagnosed in time, chain reaction can be caused to cause accidents, and even disastrous results are caused. The traction system is used as a core subsystem of the train, and the reliability and the safety of the traction system are crucial to the normal operation of the train. If the train has faults in the running process, the optimal solution is to locate the fault reason through online diagnosis and execute a proper isolation protection strategy. If the fault cause cannot be diagnosed in time and the fault is eliminated, driving accidents can be caused, the normal operation of trains is delayed, the transportation order of the whole line and the whole line is influenced, and safety accidents can be caused in serious cases.
Contactors are arranged in traction systems of various locomotives or motor train units so as to realize on-off control of related circuits. But in the actual operation process, the faults of clamping and separating of the contactor often occur, so that the whole trolley traction converter is isolated, and the normal operation of the train is seriously influenced. Moreover, the causes of the clamping and separating faults of the contactor are many, the consequences caused by different fault causes are different, but the specific fault cause cannot be accurately positioned at present, so that a single protection and isolation strategy is usually adopted, and the usability of the train is influenced to different degrees.
Therefore, a fault diagnosis method for a train traction system is expected to accurately diagnose the fault of train traction and timely remove the fault to ensure the normal operation of the train.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a fault diagnosis method and a fault diagnosis system of a train traction system, dynamic association characteristic information of multiple parameters of the train traction system in a time dimension and high-dimensional implicit association characteristic information of state information of a pre-charging contactor and a line contactor of the train traction system in the time dimension are respectively extracted by adopting a deep neural network model, and the characteristic information of the pre-charging contactor and the state information of the line contactor of the train traction system in the time dimension are fused based on a cosine-like compatible loss function training method to carry out fault diagnosis of the train traction system.
According to an aspect of the present application, there is provided a fault diagnosis method of a train traction system, including:
a training phase comprising:
acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary side voltage transformer of a traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a partial pressure resistance value of a grounding detection circuit, primary side input voltage of a tractor, secondary side output voltage of the traction transformer and middle direct current loop voltage;
enabling multiple parameters of each preset time point to pass through a context encoder comprising an embedded layer to obtain multiple characteristic vectors, and enabling the multiple characteristic vectors to be two-dimensionally arranged into a characteristic matrix and then pass through a first convolutional neural network to obtain a parameter characteristic matrix;
constructing the plurality of parameter feature matrixes into a three-dimensional input tensor, and then obtaining a first feature vector through a second convolution neural network using a three-dimensional convolution kernel;
acquiring state information of a pre-charging contactor and a line contactor of the train traction system at a plurality of preset time points;
passing state information of the pre-charging contactor and the line contactor at the plurality of preset time points through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second feature vector;
constructing a state association matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, otherwise, the value is zero;
passing the state incidence matrix through a third convolutional neural network to obtain a state topology matrix;
multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector;
calculating a cosine-type compatible loss function value between the first feature vector and the third feature vector, wherein the cosine-type compatible loss function value is a natural exponent function value raised to a power of 1/2 minus two times the cosine similarity between the first feature vector and the third feature vector;
fusing the first feature vector and the third feature vector to obtain a classified feature vector;
passing the classification feature vector through a classifier to obtain a classification loss function value; and
calculating a weighted sum of the classification loss function values and the cosine-like compatible loss function values as loss function values to train the first through third convolutional neural networks and the context encoder; and
an inference phase comprising:
acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary voltage transformer of a traction transformer, a measured value of a primary current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a partial resistance value of a grounding detection circuit, a primary input voltage of the traction transformer, a secondary output voltage of the traction transformer and a middle direct current loop voltage;
obtaining a plurality of characteristic vectors by passing a plurality of parameters of each preset time point through the context encoder containing the embedded layer trained in the training stage, and obtaining a parameter characteristic matrix by passing the first convolutional neural network trained in the training stage after two-dimensionally arranging the plurality of characteristic vectors into a characteristic matrix;
constructing the plurality of parameter feature matrices into a three-dimensional input tensor, and then obtaining a first feature vector through the second convolutional neural network which is trained in a training stage and uses a three-dimensional convolutional kernel;
acquiring state information of a pre-charging contactor and a line contactor of the train traction system at a plurality of preset time points;
passing state information of the pre-charging contactor and the line contactor at the plurality of preset time points through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second feature vector;
constructing a state association matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, otherwise, the value is zero;
passing the state incidence matrix through the third convolutional neural network trained by a training stage to obtain a state topology matrix;
multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector;
fusing the first feature vector and the third feature vector to obtain a classified feature vector;
and passing the classification characteristic vector through a classifier to obtain a classification result, wherein the classification result is whether the train traction system has a fault or not.
According to the fault diagnosis method and the fault diagnosis system of the train traction system, dynamic associated characteristic information of multiple parameters of the train traction system in a time dimension and high-dimensional implicit associated characteristic information of state information of a pre-charging contactor and a line contactor of the train traction system in the time dimension are respectively extracted by adopting a deep neural network model, and the characteristic information of the pre-charging contactor and the state information of the line contactor of the train traction system in the time dimension are fused based on a cosine compatible loss function training method to diagnose the fault of the train traction system, so that the visual point perceptibility of the prediction results of the characteristic distribution in class probability can be improved by reducing the difficulty degree of mutual representation of the characteristic distribution among vectors, and the accuracy of the fault diagnosis is further improved to ensure the normal running of the train.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is a scene schematic diagram of a fault diagnosis method of a train traction system according to an embodiment of the present application.
Fig. 2 is a flowchart of a training phase in a fault diagnosis method of a train traction system according to an embodiment of the present application.
Fig. 3 is a flowchart of an inference phase in a fault diagnosis method of a train traction system according to an embodiment of the present application.
Fig. 4 is a schematic diagram of an architecture of a training phase in a fault diagnosis method of a train traction system according to an embodiment of the present application.
Fig. 5 is a schematic diagram of an architecture of an inference stage in a fault diagnosis method of a train traction system according to an embodiment of the present application.
Fig. 6 is a block diagram of a fault diagnosis system of a train traction system according to an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Overview of a scene
As mentioned above, any minor or potential faults and hidden dangers during the operation of the train may cause a chain reaction to cause an accident and even cause catastrophic consequences if they cannot be diagnosed in time. The traction system is used as a core subsystem of the train, and the reliability and the safety of the traction system are crucial to the normal operation of the train. If the train has faults in the operation process, the optimal solution is to locate the fault cause through online diagnosis and execute a proper isolation protection strategy. If the fault reason can not be diagnosed in time and the fault can not be eliminated, the driving accident can be caused, the normal operation of the train is delayed, the transportation order of the whole line and the whole line is influenced, and the safety accident can be caused in serious cases.
Contactors are arranged in traction systems of various locomotives or motor train units so as to realize on-off control of related circuits. But in the actual operation process, the faults of clamping and separating of the contactor often occur, so that the whole trolley traction converter is isolated, and the normal operation of the train is seriously influenced. Moreover, the causes of the clamping and separating faults of the contactor are many, the consequences caused by different fault causes are different, but the specific fault cause cannot be accurately positioned at present, so that a single protection and isolation strategy is usually adopted, and the usability of the train is influenced to different degrees.
Therefore, a fault diagnosis method for a train traction system is expected to accurately diagnose the fault of train traction and timely remove the fault to ensure the normal operation of the train.
Accordingly, the inventor of the present application considers that if a fault analysis is performed on a contactor in a train traction system to accurately and timely diagnose the fault of train traction, the fault analysis can be performed based on the associated characteristic information of a plurality of parameters of the contactor of the train traction system, and also considers that implicit modes presented by parameters of various components under different working conditions are different, so that the precision of fault diagnosis is further improved by combining with working condition information.
Specifically, in the technical scheme of the application, firstly, a plurality of parameters of the train traction system at a plurality of preset time points are obtained. Here, the plurality of parameters include a measurement value of a primary side voltage transformer of the traction transformer, a measurement value of a primary side current transformer of the traction transformer, a measurement value of a four-quadrant input current sensor, a measurement value of an intermediate dc voltage sensor, a measurement value of a ground detection voltage sensor, a precharge resistance value, a support capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a voltage dividing resistance value of a ground detection circuit, a primary side input voltage of the traction transformer, a secondary side output voltage of the traction transformer, and an intermediate dc loop voltage.
Considering that the plurality of parameters are associated, a context encoder comprising an embedded layer is used for encoding the plurality of parameters at each preset time point so as to extract global high-dimensional semantic features between the parameter data to be more suitable for characterizing the intrinsic fault features of the contactors of the traction system. Specifically, in the encoding process of the context encoder, it firstly uses an embedding layer to map the multiple parameters of each of the predetermined time points into an embedded vector, that is, uses an embedding layer to map the multiple parameters of each of the predetermined time points into the same vector space as an input vector; then, a global context-based semantic encoding is performed on the sequence of obtained input vectors using a converter of the context encoder to generate the plurality of feature vectors. In this way, after the plurality of feature vectors are two-dimensionally arranged into a feature matrix to integrate a plurality of parameter feature information, deep feature mining is performed on the feature matrix by using a convolutional neural network model with excellent performance in the aspect of implicit associated feature extraction, so that associated feature information of a plurality of parameters at each preset time point is mined, and the parameter feature matrix is obtained.
Then, in order to extract the dynamic change characteristics of the plurality of parameters in the time dimension respectively so as to more accurately monitor and diagnose the fault in real time, the plurality of parameter characteristic matrixes are constructed into a three-dimensional input tensor, and then a first characteristic vector is obtained by using a second convolution neural network of a three-dimensional convolution kernel.
In consideration of the fact that the implicit modes presented by the parameters of the components under different working conditions are different, in the technical scheme of the application, the precision of fault diagnosis needs to be improved by combining with working condition information. That is, the state information of the pre-charging contactor and the line contactor of the train traction system at the plurality of preset time points is further obtained. Specifically, here, the state information of the precharge contactor KM1 and the line contactor KM2 includes that KM1 and KM2 are both open, KM1 is closed and KM2 is open, KM1 and KM2 are both closed, and KM1 is open and KM2 is closed. Then, considering the rule that the pre-charging contactor and the line contactor of the train traction system have dynamics in the time dimension, in order to extract the dynamic rule features more fully, the state information of the pre-charging contactor and the line contactor at the plurality of preset time points is passed through a time sequence encoder comprising a one-dimensional convolutional layer and a full link layer to obtain a second feature vector.
Further, a state association matrix of the pre-charging contactor and the line contactor at the plurality of predetermined time points is constructed to express a state association characteristic of the pre-charging contactor and the line contactor. In the embodiment of the present application, if the states of the pre-charging contactor and the line contactor at a predetermined time point are the same, the value of the corresponding position of the state association matrix is one, and otherwise, the value is zero. In this way, a convolutional neural network is also utilized to extract the state association feature distribution in the state association matrix for the pre-charge contactor and the line contactor.
In this way, the state topology matrix is multiplied by the second eigenvector to map the second eigenvector into the high-dimensional eigenvector space of the state topology matrix, thereby obtaining a third eigenvector with the dynamic association implicit characteristics of the state information.
It should be understood that, since the first feature vector contains structural topology information of circuit parameters and the third feature vector contains logical switching topology information of states, which have mutual response degrees, in order to make the responsiveness between the two stronger through training of model parameters, the cosine-like compatibility loss function is expressed as:
Figure BDA0003774980640000071
wherein, V1Representing said first feature vector, V3Represents said third feature vector, cos (V)1,V3) Representing a cosine similarity between the first feature vector and the third feature vector.
In this way, by introducing the loss function, the viewpoint correlation of the perceptual prediction result of the feature distribution under the class probability can be improved by reducing the difficulty degree of mutual representation of the feature distributions of the vectors, so that the parameter update of the model can optimize the first feature vector V1And a third feature vector V3The degree of correlation response between the two groups, and further, the accuracy of classification can be improved.
Based on this, the present application provides a fault diagnosis method for a train traction system, which includes: a training phase and an inference phase. Wherein the training phase comprises the steps of: acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary voltage transformer of a traction transformer, a measured value of a primary current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a partial resistance value of a grounding detection circuit, a primary input voltage of the traction transformer, a secondary output voltage of the traction transformer and a middle direct current loop voltage; enabling a plurality of parameters of each preset time point to pass through a context encoder comprising an embedded layer to obtain a plurality of characteristic vectors, and enabling the plurality of characteristic vectors to pass through a first convolutional neural network after being two-dimensionally arranged into a characteristic matrix to obtain a parameter characteristic matrix; constructing the plurality of parameter feature matrixes into a three-dimensional input tensor, and then obtaining a first feature vector through a second convolution neural network using a three-dimensional convolution kernel; acquiring state information of a pre-charging contactor and a line contactor of the train traction system at a plurality of preset time points; passing state information of the pre-charging contactor and the line contactor at the plurality of preset time points through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second feature vector; constructing a state association matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, and otherwise, the value is zero; passing the state incidence matrix through a third convolutional neural network to obtain a state topology matrix; multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector; calculating a cosine-type compatible loss function value between the first feature vector and the third feature vector, wherein the cosine-type compatible loss function value is a natural exponent function value raised to a power of 1/2 minus two times the cosine similarity between the first feature vector and the third feature vector; fusing the first feature vector and the third feature vector to obtain a classified feature vector; passing the classification feature vector through a classifier to obtain a classification loss function value; and calculating a weighted sum of the classification loss function values and the cosine-like compatible loss function values as loss function values to train the first through third convolutional neural networks and the context encoder. Wherein the inference phase comprises the steps of: acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary side voltage transformer of a traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a partial pressure resistance value of a grounding detection circuit, primary side input voltage of a tractor, secondary side output voltage of the traction transformer and middle direct current loop voltage; obtaining a plurality of characteristic vectors by passing a plurality of parameters of each preset time point through the context encoder containing the embedded layer trained in the training stage, and obtaining a parameter characteristic matrix by passing the first convolutional neural network trained in the training stage after two-dimensionally arranging the plurality of characteristic vectors into a characteristic matrix; constructing the plurality of parameter feature matrices into a three-dimensional input tensor, and then obtaining a first feature vector through the second convolutional neural network which is trained in a training stage and uses a three-dimensional convolutional kernel; acquiring state information of a pre-charging contactor and a line contactor of the train traction system at the plurality of preset time points; passing state information of the pre-charging contactor and the line contactor at the plurality of preset time points through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second feature vector; constructing a state association matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, otherwise, the value is zero; passing the state incidence matrix through the third convolutional neural network trained by a training stage to obtain a state topology matrix; multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector; fusing the first feature vector and the third feature vector to obtain a classified feature vector; and passing the classified characteristic vector through a classifier to obtain a classification result, wherein the classification result is whether the train traction system has a fault or not.
Fig. 1 illustrates a scene schematic diagram of a fault diagnosis method of a train traction system according to an embodiment of the present application. As shown in fig. 1, in the training phase of the application scenario, first, a plurality of parameters of a train traction system (e.g., T as illustrated in fig. 1) at a plurality of predetermined time points, and status information of a pre-charging contactor (e.g., P as illustrated in fig. 1) and a line contactor (e.g., L as illustrated in fig. 1) of the train traction system at the plurality of predetermined time points are obtained through respective sensors (e.g., R1-Rn as illustrated in fig. 1), where the plurality of parameters include a measurement value of a primary side transformer of a traction transformer, a measurement value of a primary side current transformer of the traction transformer, a measurement value of a four-side input current sensor, a measurement value of an intermediate dc voltage sensor, a measurement value of a ground detection voltage sensor, a pre-charge value, a support capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a divider resistance value of a ground detection circuit, a primary side quadrant input voltage of the traction transformer, a secondary side output voltage of the traction transformer, and an intermediate dc loop voltage. Then, the obtained plurality of parameters and the state information are input into a server (for example, S as illustrated in fig. 1) in which a fault diagnosis algorithm of the train traction system is deployed, wherein the server is capable of training the first to third convolutional neural networks and the context encoder of the fault diagnosis method of the train traction system with the plurality of parameters and the state information based on the fault diagnosis algorithm of the train traction system.
After the training is completed, in an inference phase, first, a plurality of parameters of a train traction system (e.g., T as illustrated in fig. 1) at a plurality of predetermined points in time, and status information of a pre-charge contactor (e.g., P as illustrated in fig. 1) and a line contactor (e.g., L as illustrated in fig. 1) of the train traction system at the plurality of predetermined points in time are acquired through various sensors (e.g., R1-Rn as illustrated in fig. 1). Then, the plurality of parameters and the state information are input into a server (for example, S as illustrated in fig. 1) deployed with a fault diagnosis algorithm of the train traction system, wherein the server can process the plurality of parameters and the state information with the fault diagnosis algorithm of the train traction system to generate a classification result for indicating whether the train traction system has a fault.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary method
Fig. 2 illustrates a flowchart of a training phase in a fault diagnosis method of a train traction system according to an embodiment of the present application. As shown in fig. 2, a method for diagnosing a fault of a train traction system according to an embodiment of the present application includes: a training phase comprising the steps of: s110, acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary voltage transformer of a traction transformer, a measured value of a primary current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharging resistance value, a partial pressure resistance value of a grounding detection circuit, a primary input voltage of the traction transformer, a secondary output voltage of the traction transformer and a middle direct current loop voltage; s120, passing the multiple parameters of each preset time point through a context encoder comprising an embedded layer to obtain multiple characteristic vectors, and two-dimensionally arranging the multiple characteristic vectors into a characteristic matrix and then passing the characteristic matrix through a first convolutional neural network to obtain a parameter characteristic matrix; s130, constructing the parameter feature matrixes into a three-dimensional input tensor, and then obtaining a first feature vector through a second convolution neural network using a three-dimensional convolution kernel; s140, acquiring state information of a pre-charging contactor and a line contactor of the train traction system at the plurality of preset time points; s150, passing the state information of the pre-charging contactor and the line contactor at the plurality of preset time points through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second eigenvector; s160, constructing state association matrixes of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, and otherwise, the value is zero; s170, passing the state incidence matrix through a third convolutional neural network to obtain a state topology matrix; s180, multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector; s190, calculating a cosine-like compatible loss function value between the first feature vector and the third feature vector, where the cosine-like compatible loss function value is a natural exponent function value raised to a power of 1/2 times a power of one minus two times a cosine similarity difference between the first feature vector and the third feature vector; s200, fusing the first feature vector and the third feature vector to obtain a classified feature vector; s210, enabling the classification feature vectors to pass through a classifier to obtain a classification loss function value; and S220 calculating a weighted sum of the classification loss function values and the cosine-like compatible loss function values as loss function values to train the first to third convolutional neural networks and the context encoder.
Fig. 3 illustrates a flow chart of an inference phase in a fault diagnosis method of a train traction system according to an embodiment of the present application. As shown in fig. 3, the method for diagnosing a fault of a train traction system according to an embodiment of the present application further includes: an inference phase comprising the steps of: s310, acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary side voltage transformer of a traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharging resistance value, a partial pressure resistance value of a grounding detection circuit, a primary side input voltage of the traction transformer, a secondary side output voltage of the traction transformer and a middle direct current loop voltage; s320, obtaining a plurality of characteristic vectors by passing the plurality of parameters of each preset time point through the context encoder containing the embedded layer trained in the training stage, and obtaining a parameter characteristic matrix by passing the first convolution neural network trained in the training stage after two-dimensionally arranging the plurality of characteristic vectors into a characteristic matrix; s330, constructing the parameter feature matrixes into a three-dimensional input tensor, and then obtaining a first feature vector through the second convolution neural network which is trained in a training stage and uses a three-dimensional convolution kernel; s340, acquiring state information of a pre-charging contactor and a line contactor of the train traction system at the plurality of preset time points; s350, enabling the state information of the pre-charging contactor and the line contactor at the plurality of preset time points to pass through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second feature vector; s360, constructing state association matrixes of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, and if not, the value is zero; s370, passing the state incidence matrix through the third convolutional neural network trained in the training stage to obtain a state topology matrix; s380, multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector; s390, fusing the first feature vector and the third feature vector to obtain a classified feature vector; s400, enabling the classification feature vectors to pass through a classifier to obtain a classification result, wherein the classification result is whether the train traction system has faults or not.
Fig. 4 illustrates an architecture diagram of a training phase in a fault diagnosis method of a train traction system according to an embodiment of the present application. As shown in fig. 4, in the training phase, first, a plurality of parameters (e.g., P1 as illustrated in fig. 4) of each of the predetermined time points are passed through a context encoder (e.g., E1 as illustrated in fig. 4) including an embedded layer to obtain a plurality of feature vectors (e.g., VF1 as illustrated in fig. 4), and the plurality of feature vectors are two-dimensionally arranged into a feature matrix (e.g., MF1 as illustrated in fig. 4) and then passed through a first convolutional neural network (e.g., CNN1 as illustrated in fig. 4) to obtain a parameter feature matrix (e.g., MF2 as illustrated in fig. 4); then, constructing the plurality of parametric eigen matrices as three-dimensional input tensors (e.g., T as illustrated in fig. 4) and then obtaining a first eigenvector (e.g., VF2 as illustrated in fig. 4) by using a second convolutional neural network (e.g., CNN2 as illustrated in fig. 4) of a three-dimensional convolution kernel; then, passing the state information (e.g., P2 as illustrated in fig. 4) of the pre-charging contactor and the line contactor at the plurality of predetermined time points through a timing encoder (e.g., E2 as illustrated in fig. 4) including a one-dimensional convolutional layer and a fully-connected layer to obtain a second eigenvector (e.g., VF3 as illustrated in fig. 4); then, passing the state association matrix (e.g., P3 as illustrated in fig. 4) through a third convolutional neural network (e.g., CNN3 as illustrated in fig. 4) to obtain a state topology matrix (e.g., MF3 as illustrated in fig. 4); then, multiplying the state topology matrix with the second eigenvector to obtain a third eigenvector (e.g., VF4 as illustrated in fig. 4); then, a cosine-like compatibility loss function value (e.g., CCV as illustrated in fig. 4) between the first eigenvector and the third eigenvector is calculated; then, fusing the first feature vector and the third feature vector to obtain a classification feature vector (e.g., VF as illustrated in fig. 4); then, the classification feature vector is passed through a classifier (e.g., circle S as illustrated in fig. 4) to obtain a classification loss function value (e.g., CLV as illustrated in fig. 4); and, finally, calculating a weighted sum of the classification loss function values and the cosine-like compatible loss function values as loss function values to train the first to third convolutional neural networks and the context encoder.
Fig. 5 illustrates an architecture diagram of an inference phase in a fault diagnosis method of a train traction system according to an embodiment of the present application. As shown in fig. 5, in the inference phase, in the network structure, first, a plurality of parameters (e.g., P1 as illustrated in fig. 5) of each of the predetermined time points are passed through the context encoder (e.g., E3 as illustrated in fig. 5) including the embedded layer trained by the training phase to obtain a plurality of feature vectors (e.g., VF1 as illustrated in fig. 5), and the plurality of feature vectors are two-dimensionally arranged into a feature matrix (e.g., MF1 as illustrated in fig. 5), and then the first convolutional neural network (e.g., CN1 as illustrated in fig. 5) trained by the training phase to obtain a parameter feature matrix (e.g., MF2 as illustrated in fig. 5); then, constructing the plurality of parametric eigen matrices as three-dimensional input tensors (e.g., T as illustrated in fig. 5) and then obtaining a first eigenvector (e.g., VF2 as illustrated in fig. 5) through the second convolutional neural network (e.g., CN2 as illustrated in fig. 5) using three-dimensional convolution kernel, which is trained by a training phase; then, passing the state information (e.g., P2 as illustrated in fig. 5) of the pre-charging contactor and the line contactor at the plurality of predetermined time points through a timing encoder (e.g., E2 as illustrated in fig. 5) including a one-dimensional convolutional layer and a fully-connected layer to obtain a second eigenvector (e.g., VF3 as illustrated in fig. 5); then, passing the state association matrix (e.g., P3 as illustrated in fig. 5) through the third convolutional neural network (e.g., CN3 as illustrated in fig. 5) trained by a training phase to obtain a state topology matrix (e.g., MF3 as illustrated in fig. 5); then, multiplying the state topology matrix with the second eigenvector to obtain a third eigenvector (e.g., VF4 as illustrated in fig. 5); then, fusing the first feature vector and the third feature vector to obtain a classification feature vector (e.g., VF as illustrated in fig. 5); finally, the classification feature vector is passed through a classifier (e.g., circle S as illustrated in fig. 5) to obtain a classification result, which is whether the train traction system has a fault.
More specifically, in a training phase, in step S110 and step S120, multiple parameters of the train traction system at multiple predetermined time points are obtained, the multiple parameters at each predetermined time point are passed through a context encoder including an embedded layer to obtain multiple eigenvectors, and the multiple eigenvectors are two-dimensionally arranged into an eigenvector matrix and then passed through a first convolutional neural network to obtain a parameter eigenvector matrix. As described above, it should be understood that, considering that if a fault analysis is performed on a contactor in a train traction system to accurately and timely diagnose a fault in train traction, the fault diagnosis may be performed based on associated characteristic information of a plurality of parameters of the train traction system contactor, and considering that implicit modes presented by parameters of various components under different working conditions are different, the precision of fault diagnosis is further improved by combining with working condition information.
That is, specifically, in the technical solution of the present application, first, a plurality of parameters of the train traction system at a plurality of predetermined time points are obtained. Here, the plurality of parameters include a measurement value of a primary side voltage transformer of the traction transformer, a measurement value of a primary side current transformer of the traction transformer, a measurement value of a four-quadrant input current sensor, a measurement value of an intermediate dc voltage sensor, a measurement value of a ground detection voltage sensor, a precharge resistance value, a support capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a voltage dividing resistance value of a ground detection circuit, a primary side input voltage of the tractor, a secondary side output voltage of the traction transformer, and an intermediate dc loop voltage.
It should be understood that, considering the existence of the correlation among the parameters, the context encoder containing the embedded layer is used for encoding the parameters of each preset time point so as to extract global high-dimensional semantic features among the parameter data to be more suitable for characterizing the intrinsic fault features of the contactors of the traction system. Specifically, in the embodiment of the present application, during the encoding process of the context encoder, it first uses an embedding layer to map multiple parameters of each of the predetermined time points into an embedded vector, that is, uses an embedding layer to map multiple parameters of each of the predetermined time points into the same vector space as an input vector; then, a global context-based semantic encoding is performed on the sequence of obtained input vectors using a converter of the context encoder to generate the plurality of feature vectors.
Further, after the plurality of feature vectors are two-dimensionally arranged into a feature matrix to integrate a plurality of parameter feature information, deep feature mining is performed on the feature matrix by using a convolutional neural network model with excellent performance in the aspect of implicit associated feature extraction, so that associated feature information of a plurality of parameters at each preset time point is mined, and the parameter feature matrix is obtained. Accordingly, in one particular example, the layers of the first convolutional neural network convolve input data in a forward pass of layers, mean pooling along channel dimensions, and activation to generate the parametric feature matrix from a last layer of the first convolutional neural network, wherein the input to the first layer of the first convolutional neural network is the feature matrix.
More specifically, in the training phase, in step S130, the plurality of parametric feature matrices are constructed as three-dimensional input tensors, and then a first feature vector is obtained by using a second convolutional neural network of a three-dimensional convolution kernel. That is, in the technical solution of the present application, in order to extract the dynamic variation characteristics of each of the plurality of parameters in the time dimension to more accurately monitor and diagnose the fault in real time, the plurality of parameter characteristic matrices are further configured as a three-dimensional input tensor, and then a first eigenvector is obtained by using a second convolutional neural network of a three-dimensional convolutional kernel.
Specifically, in this embodiment, the process of constructing the plurality of parametric feature matrices as three-dimensional input tensors and then obtaining a first feature vector by using a second convolutional neural network of a three-dimensional convolution kernel includes: the second convolutional neural network using the three-dimensional convolutional kernel performs, in forward transfer of layers, respectively: performing three-dimensional convolution processing on the input data based on the three-dimensional convolution kernel to obtain a convolution characteristic diagram; performing mean pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the second convolutional neural network is the first eigenvector, and the input of the first layer of the second convolutional neural network is the input tensor.
More specifically, in the training phase, in steps S140 and S150, state information of the pre-charging contactor and the line contactor of the train traction system at the plurality of predetermined time points is obtained, and the state information of the pre-charging contactor and the line contactor at the plurality of predetermined time points is passed through a time sequence encoder comprising a one-dimensional convolutional layer and a full link layer to obtain a second feature vector. It should be understood that, in consideration of the fact that implicit modes presented by parameters of each component under different working conditions are different, in the technical scheme of the present application, the accuracy of fault diagnosis needs to be improved by combining working condition information. That is, the state information of the pre-charging contactor and the line contactor of the train traction system at the plurality of preset time points is further obtained. Specifically, here, the state information of the precharge contactor KM1 and the line contactor KM2 includes that both KM1 and KM2 are open, KM1 is closed and KM2 is open, both KM1 and KM2 are closed, and KM1 is open and KM2 is closed. Then, considering the rule that the pre-charging contactor and the line contactor of the train traction system have dynamics in the time dimension, in order to more fully extract the dynamic rule feature, the state information of the pre-charging contactor and the line contactor at the plurality of predetermined time points is passed through a time sequence encoder comprising a one-dimensional convolutional layer and a full link layer to obtain a second feature vector.
Specifically, in this embodiment of the present application, the process of passing the state information of the pre-charging contactor and the line contactor at the plurality of predetermined time points through a time-sequence encoder including a one-dimensional convolutional layer and a full-link layer to obtain a second feature vector includes: firstly, arranging the state information of the pre-charging contactor and the line contactor at the plurality of preset time points into input vectors according to the time dimension; then, using a full-connection layer of the time sequence encoder to perform full-connection encoding on the input vector by using the following formula to extract high-dimensional implicit features of feature values of all positions in the input vector, wherein the formula is as follows:
Figure BDA0003774980640000151
wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the offset vector,
Figure BDA0003774980640000152
represents a matrix multiplication; finally, the one-dimensional convolution layer of the time sequence encoder is used for carrying out one-dimensional convolution encoding on the input vector by the following formula so as to extract high-dimensional implicit correlation characteristics among characteristic values of all positions in the input vector, wherein the formula is as follows:
Figure BDA0003774980640000153
wherein, a is the width of the convolution kernel in the x direction, F is the parameter vector of the convolution kernel, G is the local vector matrix operated with the convolution kernel function, and w is the size of the convolution kernel.
More specifically, in the training phase, in steps S160, S170 and S180, a state association matrix of the pre-charging contactor and the line contactor at the predetermined time points is constructed, wherein if the states of the pre-charging contactor and the line contactor at the predetermined time points are the same, the value of the corresponding position of the state association matrix is one, otherwise, the corresponding position of the state association matrix is zero, the state association matrix is passed through a third convolutional neural network to obtain a state topology matrix, and then the state topology matrix is multiplied by the second eigenvector to obtain a third eigenvector. That is, in the technical solution of the present application, a state association matrix of the pre-charging contactor and the line contactor at the plurality of predetermined time points is constructed to express a state association characteristic of the pre-charging contactor and the line contactor. Accordingly, in one specific example, if the states of the pre-charge contactor and the line contactor at a predetermined time point are the same, the value of the corresponding position of the state correlation matrix is one, and otherwise is zero. In this way, a convolutional neural network may also be used to extract the state associated feature distributions for the precharge contactor and the line contactor in the state association matrix. Further, the state topology matrix is multiplied by the second eigenvector to map the second eigenvector into a high-dimensional eigenvector space of the state topology matrix, so as to obtain a third eigenvector with the dynamic association implicit characteristic of the state information.
More specifically, in the training phase, in step S190, a cosine-like compatible loss function value between the first feature vector and the third feature vector is calculated, the cosine-like compatible loss function value being a natural exponent function value raised to a power of 1/2 times one minus twice the cosine similarity between the first feature vector and the third feature vector. It should be understood that, since the first feature vector contains structural topology information of circuit parameters, and the third feature vector contains logical switching topology information of states, which have mutual response degrees, in the technical solution of the present application, in order to make the responsiveness between the two stronger through training model parameters, a cosine-like compatibility loss function between the first feature vector and the third feature vector is used for training.
Specifically, in this embodiment, the process of calculating the cosine-like compatible loss function value between the first eigenvector and the third eigenvector includes: calculating a cosine-like compatibility loss function value between the first eigenvector and the third eigenvector according to the following formula;
wherein the formula is:
Figure BDA0003774980640000161
wherein, V1Representing said first feature vector, V3Represents the third feature vector cos (V)1,V3) Representing a cosine similarity between the first feature vector and the third feature vector. It will be appreciated that, in this way, by introducing the penalty function, the viewpoint dependency of the perceptual predictors of the feature distributions under class probability can be enhanced by reducing the difficulty of mutually representing the feature distributions of each other between the vectors, thereby enabling the parameter update of the model to optimize the first feature vector V1And the third feature vector V3The degree of the correlation response therebetween, and in turn,the accuracy of classification can also be improved.
More specifically, in the training phase, in steps S200, S210, and S220, the first feature vector and the third feature vector are fused to obtain a classification feature vector, the classification feature vector is passed through a classifier to obtain a classification loss function value, and a weighted sum of the classification loss function value and the cosine-like compatible loss function value is calculated as a loss function value to train the first to third convolutional neural networks and the context encoder. That is, in the training stage, the first feature vector and the third feature vector are further fused to pass through a classifier to obtain a classification loss function value, so that a weighted sum of the classification loss function value and the cosine-like compatible loss function value can be calculated as a loss function value to train the first to third convolutional neural networks and the context encoder.
Specifically, in this embodiment of the present application, a process of passing the classification feature vector through a classifier to obtain a classification loss function value includes: processing the classification feature vector using the classifier to obtain the classification result with the following formula: softmax { (W)n,Bn):…:(W1,B1) I X }, wherein W1To WnAs a weight matrix, B1To BnIs a bias vector, X is the classification feature vector; and calculating a cross entropy value between the classification result and a real value as the classification loss function value.
After training is completed, the inference phase is entered. The classification feature vector can be obtained as described above. And then, the classified feature vectors pass through a classifier to obtain a classification result for indicating whether the train traction system has faults or not.
In summary, the fault diagnosis method for the train traction system based on the embodiment of the present application is clarified, and dynamic associated feature information of multiple parameters of the train traction system in a time dimension and high-dimensional implicit associated feature information of state information of a pre-charging contactor and a line contactor of the train traction system in the time dimension are respectively extracted by using a deep neural network model, and feature information of the pre-charging contactor and the state information of the line contactor of the train traction system in the time dimension are fused based on a training method of a cosine-like compatible loss function to perform fault diagnosis of the train traction system, so that the viewpoint relevance of a perceivable prediction result of the feature distribution under a class probability can be improved by reducing the difficulty degree of mutual representation of feature distributions among vectors, and the accuracy of the fault diagnosis is further improved to ensure normal operation of the train.
Exemplary System
Fig. 6 illustrates a block diagram of a fault diagnostic system of a train traction system according to an embodiment of the present application. As shown in fig. 6, a fault diagnosis system 600 of a train traction system according to an embodiment of the present application includes: a training module 610 and an inference module 620.
As shown in fig. 6, the training module 610 includes: a parameter data obtaining unit 6101, configured to obtain multiple parameters of the train traction system at multiple predetermined time points, where the multiple parameters include a measurement value of a primary side voltage transformer of the traction transformer, a measurement value of a primary side current transformer of the traction transformer, a measurement value of a four-quadrant input current sensor, a measurement value of a middle dc voltage sensor, a measurement value of a ground detection voltage sensor, a pre-charge resistance value, a support capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a voltage division resistance value of a ground detection circuit, a primary side input voltage of the tractor, a secondary side output voltage of the traction transformer, and a middle dc loop voltage; a first feature extraction unit 6102, configured to pass multiple parameters of each of the predetermined time points through a context encoder including an embedded layer to obtain multiple feature vectors, and two-dimensionally arrange the multiple feature vectors into a feature matrix, and then pass through a first convolutional neural network to obtain a parameter feature matrix; a second encoding unit 6103 configured to obtain the first eigenvector by a second convolutional neural network using a three-dimensional convolutional kernel after configuring the plurality of parametric feature matrices as three-dimensional input tensors; a state data obtaining unit 6104, configured to obtain state information of the pre-charging contactor and the line contactor of the train traction system at the plurality of predetermined time points; a timing encoding unit 6105, configured to pass the state information of the pre-charging contactor and the line contactor at the predetermined time points through a timing encoder comprising a one-dimensional convolution layer and a full link layer to obtain a second feature vector; a correlation matrix constructing unit 6106, configured to construct a state correlation matrix of the pre-charging contactor and the line contactor at the plurality of predetermined time points, where if the states of the pre-charging contactor and the line contactor at the predetermined time points are the same, the value of the corresponding position of the state correlation matrix is one, otherwise, it is zero; a third feature extraction unit 6107, configured to pass the state correlation matrix through a third convolutional neural network to obtain a state topology matrix; a mapping unit 6108, configured to multiply the state topology matrix with the second eigenvector to obtain a third eigenvector; a cosine-type compatible loss function value calculating unit 6109 for calculating a cosine-type compatible loss function value between the first feature vector and the third feature vector, the cosine-type compatible loss function value being a natural exponent function value raised to a power of 1/2 minus twice the cosine similarity between the first feature vector and the third feature vector; a fusion unit 6110, configured to fuse the first feature vector and the third feature vector to obtain a classification feature vector; a classification loss function value calculation unit 6111, configured to pass the classification feature vector through a classifier to obtain a classification loss function value; and a training unit 6112 for calculating a weighted sum of the classification loss function values and the cosine-like compatible loss function values as loss function values to train the first to third convolutional neural networks and the context encoder.
As shown in fig. 6, the inference module 620 includes: the inferred data acquisition unit 6201 is used for acquiring multiple parameters of the train traction system at multiple preset time points, where the multiple parameters include a measured value of a primary side voltage transformer of the traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a ground detection voltage sensor, a pre-charging resistance value, a support capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a partial voltage resistance value of a ground detection circuit, a primary side input voltage of the tractor, a secondary side output voltage of the traction transformer, and a middle direct current loop voltage; a parameter feature matrix generating unit 6202, configured to obtain a plurality of feature vectors by passing the plurality of parameters at each predetermined time point through the context encoder with an embedded layer that is trained in the training stage, and obtain a parameter feature matrix by passing the first convolutional neural network that is trained in the training stage after two-dimensionally arranging the plurality of feature vectors into a feature matrix; a dynamic feature extraction unit 6203, configured to construct the plurality of parameter feature matrices into a three-dimensional input tensor, and then obtain a first feature vector through the second convolutional neural network using a three-dimensional convolutional kernel, which is trained in a training stage; a state information obtaining unit 6204, configured to obtain state information of a pre-charging contactor and a line contactor of the train traction system at the plurality of predetermined time points; a second feature vector generation unit 6205, configured to pass state information of the pre-charge contactor and the line contactor at the plurality of predetermined time points through a time sequence encoder comprising a one-dimensional convolution layer and a full link layer to obtain a second feature vector; a state association unit 6206, configured to construct a state association matrix of the pre-charging contactor and the line contactor at the multiple predetermined time points, where if the states of the pre-charging contactor and the line contactor at the predetermined time points are the same, a value of a corresponding position of the state association matrix is one, and otherwise, the value is zero; a topological feature extraction unit 6207, configured to pass the state association matrix through the third convolutional neural network trained in the training stage to obtain a state topological matrix; a third eigenvector generating unit 6208, configured to multiply the state topology matrix and the second eigenvector to obtain a third eigenvector; a feature fusion unit 6209, configured to fuse the first feature vector and the third feature vector to obtain a classification feature vector; a classifying unit 6210, configured to pass the classified feature vector through a classifier to obtain a classification result, where the classification result indicates whether the train traction system has a fault.
Here, it can be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described fault diagnosis system 600 of the train traction system have been described in detail in the above description of the fault diagnosis method of the train traction system with reference to fig. 1 to 5, and thus, a repetitive description thereof will be omitted.
As described above, the fault diagnosis system 600 of the train traction system according to the embodiment of the present application may be implemented in various terminal devices, such as a server of a fault diagnosis algorithm of the train traction system, and the like. In one example, the fault diagnosis system 600 of the train traction system according to the embodiment of the present application may be integrated into the terminal device as one software module and/or hardware module. For example, the fault diagnosis system 600 of the train traction system may be a software module in the operating device of the terminal device, or may be an application developed for the terminal device; of course, the fault diagnosis system 600 of the train traction system may also be one of many hardware modules of the terminal device.
Alternatively, in another example, the fault diagnosis system 600 of the train traction system and the terminal device may be separate devices, and the fault diagnosis system 600 of the train traction system may be connected to the terminal device through a wired and/or wireless network and transmit the interactive information according to an agreed data format.
The foregoing describes the general principles of the present application in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present application are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.
The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably herein. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".
It should also be noted that in the devices, apparatuses, and methods of the present application, each component or step can be decomposed and/or re-combined. These decompositions and/or recombinations are to be considered as equivalents of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims (10)

1. A fault diagnosis method of a train traction system is characterized by comprising the following steps:
a training phase comprising:
acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary side voltage transformer of a traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a partial pressure resistance value of a grounding detection circuit, primary side input voltage of a tractor, secondary side output voltage of the traction transformer and middle direct current loop voltage;
enabling multiple parameters of each preset time point to pass through a context encoder comprising an embedded layer to obtain multiple characteristic vectors, and enabling the multiple characteristic vectors to be two-dimensionally arranged into a characteristic matrix and then pass through a first convolutional neural network to obtain a parameter characteristic matrix;
constructing the plurality of parameter feature matrixes into a three-dimensional input tensor, and then obtaining a first feature vector through a second convolution neural network using a three-dimensional convolution kernel;
acquiring state information of a pre-charging contactor and a line contactor of the train traction system at a plurality of preset time points;
passing state information of the pre-charging contactor and the line contactor at the plurality of preset time points through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second feature vector;
constructing a state association matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, otherwise, the value is zero;
passing the state incidence matrix through a third convolutional neural network to obtain a state topology matrix;
multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector;
calculating a cosine-type compatible loss function value between the first feature vector and the third feature vector, wherein the cosine-type compatible loss function value is a natural exponent function value raised to a power of 1/2 minus two times the cosine similarity between the first feature vector and the third feature vector;
fusing the first feature vector and the third feature vector to obtain a classified feature vector;
passing the classification feature vector through a classifier to obtain a classification loss function value; and
calculating a weighted sum of the classification loss function values and the cosine-like compatible loss function values as loss function values to train the first through third convolutional neural networks and the context encoder; and
an inference phase comprising:
acquiring multiple parameters of a train traction system at multiple preset time points, wherein the multiple parameters comprise a measured value of a primary side voltage transformer of a traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharge resistance value, a partial pressure resistance value of a grounding detection circuit, primary side input voltage of a tractor, secondary side output voltage of the traction transformer and middle direct current loop voltage;
obtaining a plurality of characteristic vectors by passing a plurality of parameters of each preset time point through the context encoder containing the embedded layer trained in the training stage, and obtaining a parameter characteristic matrix by passing the first convolutional neural network trained in the training stage after two-dimensionally arranging the plurality of characteristic vectors into a characteristic matrix;
constructing the plurality of parameter feature matrices into a three-dimensional input tensor, and then obtaining a first feature vector through the second convolutional neural network which is trained in a training stage and uses a three-dimensional convolutional kernel;
acquiring state information of a pre-charging contactor and a line contactor of the train traction system at the plurality of preset time points;
passing state information of the pre-charging contactor and the line contactor at the plurality of preset time points through a time sequence encoder comprising a one-dimensional convolution layer and a full connection layer to obtain a second feature vector;
constructing a state association matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, otherwise, the value is zero;
passing the state incidence matrix through the third convolutional neural network trained by a training stage to obtain a state topology matrix;
multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector;
fusing the first feature vector and the third feature vector to obtain a classified feature vector;
and passing the classification characteristic vector through a classifier to obtain a classification result, wherein the classification result is whether the train traction system has a fault or not.
2. The method of diagnosing a fault in a train traction system according to claim 1, wherein passing the plurality of parameters at each of the predetermined time points through a context encoder including an embedded layer to obtain a plurality of feature vectors includes:
respectively converting the multiple parameters of each preset time point into input vectors by using the embedding layer of the context encoder containing the embedding layer so as to obtain a sequence of the input vectors; and
globally context-based semantic encoding the sequence of input vectors using the converter of the context encoder including the embedding layer to obtain the plurality of feature vectors.
3. The method for diagnosing the fault of the train traction system according to claim 2, wherein the two-dimensionally arranging the plurality of eigenvectors into the eigenvector matrix and then obtaining the parameter eigenvector matrix through the first convolutional neural network comprises:
each layer of the first convolutional neural network performs convolutional processing, mean pooling along channel dimensions, and activation processing on input data in forward pass of the layer to generate the parameter feature matrix from the last layer of the first convolutional neural network, wherein the input of the first layer of the first convolutional neural network is the feature matrix.
4. The method for diagnosing a fault in a train traction system according to claim 3, wherein constructing the plurality of parametric eigen matrices as three-dimensional input tensors and obtaining a first eigenvector by using a second convolutional neural network of a three-dimensional convolution kernel includes: the second convolutional neural network using the three-dimensional convolutional kernel performs, in forward transfer of layers, respectively:
performing three-dimensional convolution processing on the input data based on the three-dimensional convolution kernel to obtain a convolution characteristic diagram;
performing mean pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and
performing nonlinear activation on the pooled feature map to obtain an activated feature map;
wherein the output of the last layer of the second convolutional neural network is the first eigenvector, and the input of the first layer of the second convolutional neural network is the input tensor.
5. The method for diagnosing a fault in a train traction system according to claim 4, wherein passing the state information of the pre-charging contactor and the line contactor at the plurality of predetermined time points through a time-sequence encoder including a one-dimensional convolutional layer and a full link layer to obtain a second eigenvector comprises:
arranging the state information of the pre-charging contactor and the line contactor at the plurality of preset time points into an input vector according to a time dimension;
performing full-concatenation encoding on the input vector by using a full-concatenation layer of the time sequence encoder according to the following formula to extract high-dimensional implicit features of feature values of each position in the input vector, wherein the formula is as follows:
Figure FDA0003774980630000041
wherein X is the input vector, Y is the output vector, W is the weight matrix, B is the offset vector,
Figure FDA0003774980630000042
represents a matrix multiplication;
performing one-dimensional convolutional coding on the input vector by using a one-dimensional convolutional layer of the time sequence encoder according to the following formula to extract high-dimensional implicit correlation characteristics among characteristic values of all positions in the input vector, wherein the formula is as follows:
Figure FDA0003774980630000043
wherein, a is the width of the convolution kernel in the x direction, F is the parameter vector of the convolution kernel, G is the local vector matrix operated with the convolution kernel function, and w is the size of the convolution kernel.
6. The method for diagnosing a fault in a train traction system according to claim 5, wherein calculating a cosine-like compatibility loss function value between the first eigenvector and the third eigenvector includes:
calculating a cosine-like compatibility loss function value between the first eigenvector and the third eigenvector according to the following formula;
wherein the formula is:
Figure FDA0003774980630000044
wherein, V1Representing said first feature vector, V3Represents said third feature vector, cos (V)1,V3) Representing a cosine similarity between the first feature vector and the third feature vector.
7. The method of fault diagnosis of a train traction system according to claim 6, wherein passing the classification feature vector through a classifier to obtain a classification loss function value comprises:
processing the classification feature vector using the classifier to obtain the classification result with a formula: softmax { (W)n,Bn):…:(W1,B1) I X }, wherein W1To WnAs a weight matrix, B1To BnIs a bias vector, and X is the classification feature vector;
and calculating a cross entropy value between the classification result and the real value as the classification loss function value.
8. A fault diagnostic system for a train traction system, comprising:
a training module comprising:
the system comprises a parameter data acquisition unit, a data acquisition unit and a data acquisition unit, wherein the parameter data acquisition unit is used for acquiring a plurality of parameters of a train traction system at a plurality of preset time points, and the parameters comprise a measured value of a primary side voltage transformer of a traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharging resistance value, a partial pressure resistance value of a grounding detection circuit, a primary side input voltage of a tractor, a secondary side output voltage of the traction transformer and a middle direct current loop voltage;
the first characteristic extraction unit is used for enabling a plurality of parameters of each preset time point to pass through a context encoder comprising an embedded layer so as to obtain a plurality of characteristic vectors, and enabling the plurality of characteristic vectors to be two-dimensionally arranged into a characteristic matrix and then pass through a first convolutional neural network so as to obtain a parameter characteristic matrix;
a second encoding unit configured to construct the plurality of parameter feature matrices into a three-dimensional input tensor and obtain a first feature vector by using a second convolutional neural network of a three-dimensional convolutional kernel;
the state data acquisition unit is used for acquiring the state information of a pre-charging contactor and a line contactor of the train traction system at a plurality of preset time points;
the time sequence coding unit is used for enabling the state information of the pre-charging contactor and the line contactor at the plurality of preset time points to pass through a time sequence coder comprising a one-dimensional convolution layer and a full connection layer so as to obtain a second feature vector;
the correlation matrix construction unit is used for constructing a state correlation matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state correlation matrix is one, and otherwise, the value of the corresponding position of the state correlation matrix is zero;
the third feature extraction unit is used for enabling the state incidence matrix to pass through a third convolutional neural network so as to obtain a state topological matrix;
the mapping unit is used for multiplying the state topology matrix and the second eigenvector to obtain a third eigenvector;
a cosine-type compatible loss function value calculation unit configured to calculate a cosine-type compatible loss function value between the first feature vector and the third feature vector, the cosine-type compatible loss function value being a natural exponent function value raised to a power of 1/2 times a difference between cosine similarities subtracted by two times between the first feature vector and the third feature vector;
a fusion unit for fusing the first feature vector and the third feature vector to obtain a classification feature vector;
the classification loss function value calculation unit is used for enabling the classification characteristic vectors to pass through a classifier so as to obtain classification loss function values; and
a training unit for calculating a weighted sum of the classification loss function values and the cosine-like compatible loss function values as loss function values to train the first to third convolutional neural networks and the context encoder;
an inference module comprising:
the system comprises an inferred data acquisition unit, a data acquisition unit and a data acquisition unit, wherein the inferred data acquisition unit is used for acquiring multiple parameters of a train traction system at multiple preset time points, and the multiple parameters comprise a measured value of a primary side voltage transformer of a traction transformer, a measured value of a primary side current transformer of the traction transformer, a measured value of a four-quadrant input current sensor, a measured value of a middle direct current voltage sensor, a measured value of a grounding detection voltage sensor, a pre-charging resistance value, a supporting capacitor capacitance value, a secondary resonance inductance value, a secondary resonance capacitance value, a fixed discharging resistance value, a partial pressure resistance value of a grounding detection circuit, a primary side input voltage of the traction transformer, a secondary side output voltage of the traction transformer and a middle direct current loop voltage;
a parameter feature matrix generating unit, configured to pass multiple parameters of each of the predetermined time points through the context encoder including the embedded layer trained in the training stage to obtain multiple feature vectors, and obtain a parameter feature matrix through the first convolutional neural network trained in the training stage after two-dimensionally arranging the multiple feature vectors into a feature matrix;
the dynamic feature extraction unit is used for constructing the parameter feature matrixes into a three-dimensional input tensor and then obtaining a first feature vector through the second convolutional neural network which is trained in a training stage and uses a three-dimensional convolutional kernel;
the state information acquisition unit is used for acquiring the state information of a pre-charging contactor and a line contactor of the train traction system at a plurality of preset time points;
a second eigenvector generating unit, configured to pass state information of the pre-charging contactor and the line contactor at the plurality of predetermined time points through a time-sequence encoder including a one-dimensional convolution layer and a full link layer to obtain a second eigenvector;
the state association unit is used for constructing a state association matrix of the pre-charging contactor and the line contactor at the plurality of preset time points, wherein if the states of the pre-charging contactor and the line contactor at the preset time points are the same, the value of the corresponding position of the state association matrix is one, and otherwise, the value of the corresponding position of the state association matrix is zero;
the topological feature extraction unit is used for enabling the state incidence matrix to pass through the third convolutional neural network which is trained in the training stage so as to obtain a state topological matrix;
a third eigenvector generating unit, configured to multiply the state topology matrix and the second eigenvector to obtain a third eigenvector;
a feature fusion unit for fusing the first feature vector and the third feature vector to obtain a classified feature vector;
and the classification unit is used for enabling the classification characteristic vectors to pass through a classifier to obtain a classification result, wherein the classification result is whether the train traction system has faults or not.
9. The fault diagnosis system of a train traction system according to claim 8, wherein the first feature extraction unit is further configured to:
respectively converting the multiple parameters of each preset time point into input vectors by using the embedding layer of the context encoder comprising the embedding layer so as to obtain a sequence of the input vectors; and globally context-based semantic encoding the sequence of input vectors using the converter of the context encoder including the embedded layer to obtain the plurality of feature vectors.
10. The fault diagnosis system of a train traction system according to claim 9, wherein the cosine-like compatible loss function value calculating unit is further configured to: calculating a cosine-like compatibility loss function value between the first eigenvector and the third eigenvector according to the following formula;
wherein the formula is:
Figure FDA0003774980630000071
wherein, V1Representing said first feature vector, V3Represents said third feature vector, cos (V)1,V3) Representing a cosine similarity between the first feature vector and the third feature vector.
CN202210914232.2A 2022-08-01 2022-08-01 Fault diagnosis method and system for train traction system Withdrawn CN115266159A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210914232.2A CN115266159A (en) 2022-08-01 2022-08-01 Fault diagnosis method and system for train traction system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210914232.2A CN115266159A (en) 2022-08-01 2022-08-01 Fault diagnosis method and system for train traction system

Publications (1)

Publication Number Publication Date
CN115266159A true CN115266159A (en) 2022-11-01

Family

ID=83747953

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210914232.2A Withdrawn CN115266159A (en) 2022-08-01 2022-08-01 Fault diagnosis method and system for train traction system

Country Status (1)

Country Link
CN (1) CN115266159A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115993507A (en) * 2023-03-23 2023-04-21 成都交大运达电气有限公司 Fault distance measurement method of electrified railway traction power supply system
CN116127019A (en) * 2023-03-07 2023-05-16 杭州国辰智企科技有限公司 Dynamic parameter and visual model generation WEB 2D automatic modeling engine system

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116127019A (en) * 2023-03-07 2023-05-16 杭州国辰智企科技有限公司 Dynamic parameter and visual model generation WEB 2D automatic modeling engine system
CN116127019B (en) * 2023-03-07 2024-06-11 杭州国辰智企科技有限公司 Dynamic parameter and visual model generation WEB 2D automatic modeling engine system
CN115993507A (en) * 2023-03-23 2023-04-21 成都交大运达电气有限公司 Fault distance measurement method of electrified railway traction power supply system

Similar Documents

Publication Publication Date Title
CN115266159A (en) Fault diagnosis method and system for train traction system
Modi et al. Estimation of energy consumption of electric vehicles using deep convolutional neural network to reduce driver’s range anxiety
US20140324379A1 (en) System and method for predicting temperature of battery
CN115013298B (en) Real-time performance online monitoring system and monitoring method of sewage pump
CN114118251A (en) Fault diagnosis and early warning method based on multi-source data fusion and convolution twin neural network
CN109964182A (en) Method and system for vehicle analysis
Senemmar et al. Deep learning-based fault detection, classification, and locating in shipboard power systems
Zhang et al. Feature engineering and artificial intelligence-supported approaches used for electric powertrain fault diagnosis: A review
CN114529085A (en) Resident income prediction system based on big data and prediction method thereof
CN116086790A (en) Performance detection method and system for high-pressure valve of hydrogen fuel cell
CN116934304A (en) Intelligent power distribution room equipment operation maintenance management system and method thereof
CN115618732A (en) Nuclear reactor digital twin key parameter autonomous optimization data inversion method
CN115805810A (en) Battery failure prediction method, apparatus, device, storage medium, and program product
Yao et al. Fault identification of lithium-ion battery pack for electric vehicle based on ga optimized ELM neural network
Bazhynova et al. Neural Network Model of Assessing the Technical Condition of the Power Unit of a Hybrid Vehicle
CN115146676A (en) Circuit fault detection method and system
CN111547028A (en) Brake intensity fuzzy recognition method considering brake intention
CN113610119B (en) Method for identifying power transmission line development faults based on convolutional neural network
CN111062569A (en) Low-current fault discrimination method based on BP neural network
CN114566047A (en) Early warning method and system based on short-time circuit condition prediction algorithm
CN108510002A (en) The detection method of rewinding material tractive transformer winding resistance to shorting impact capacity
CN115588169A (en) Vehicle overweight running monitoring method and system based on axle data
CN110988631B (en) Partial discharge positioning detection method and device, readable storage medium and electrical equipment
CN114841222B (en) Method for constructing fault detection model of high-voltage circuit breaker, detection method and power distribution system
CN115328088B (en) Automobile fault diagnosis method and system based on cloud edge cooperation and intelligent automobile

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
WW01 Invention patent application withdrawn after publication

Application publication date: 20221101

WW01 Invention patent application withdrawn after publication