Method, device and storage medium for locating insulation fault of running rail under unknown working conditions

Abstract

The present invention discloses a method, device, and storage medium for locating insulation faults in running rails under unknown operating conditions. These methods relate to the field of urban rail transit technology. The method comprises: acquiring potential signals at multiple preset rail nodes where the fault location is to be determined; inputting the potential signals at these multiple preset rail nodes into a pre-acquired fault location model to obtain fault labels for the potential signals at each node; and determining the fault location based on the fault labels. This invention addresses the technical problem of being unable to accurately and timely locate insulation faults under unknown operating conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of urban rail transit, and in particular to a method, device and storage medium for locating insulation faults in running rails under unknown working conditions. Background Art

[0002] Most existing urban rail transit systems utilize a DC traction power supply system. The overhead catenary provides the traction current, while the running rails serve as the return path for this traction current. Due to the longitudinal resistance of the running rails, when the traction current returns, a potential difference forms between the rails and the ground, known as the rail potential. Since the running rails are not completely insulated from the ground, some of this current flows into the surrounding dielectric medium, generating stray currents. When insulation failure occurs, the magnitude of this stray current increases significantly, seriously impacting the safe operation of the system. Research has shown that stray current flowing into reinforced concrete can weaken the steel and cause electrochemical corrosion in buried metal pipes. Furthermore, stray current flowing into AC transmission systems can increase AC harmonics and induce DC bias in transformers. Measures to reduce stray current magnitude include adding insulation coatings and replacing aging insulation fasteners. However, these measures cannot be effectively applied on-site when the insulation fault location is unknown. Therefore, timely fault location of insulation faults is crucial for the safe operation of DC traction power supply systems.

[0003] Existing data-driven approaches to insulation fault research suffer from a reliance on empirically derived calculation parameters. Furthermore, in urban rail transit, when operating conditions such as load and slope vary, these methods become less robust and inapplicable. Consequently, there is no solution for locating insulation faults in urban rail transit running rails under unknown operating conditions.

[0004] Therefore, there is an urgent need for a method, device and storage medium for locating insulation faults in running rails under unknown working conditions to solve the above technical problems. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, device and storage medium for locating insulation faults in running rails under unknown working conditions, which can solve the technical problem that the insulation fault position cannot be accurately and timely located under unknown working conditions.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a method for locating insulation faults in running rails under unknown working conditions, comprising:

[0008] Obtaining potential signals at multiple preset nodes of the rail where the fault location is to be determined;

[0009] Inputting the potential signals at a plurality of preset nodes of the rail into a pre-acquired fault location model, obtaining a fault label of the potential signal of each node, and determining the fault location according to the fault label;

[0010] The training method of the fault location model includes:

[0011] Obtain the potential signals at multiple preset rail nodes when the train is under different traction conditions and different insulation fault states occur;

[0012] Preprocess the potential signals at multiple preset rail nodes to obtain labeled training datasets in different source domains.

[0013] Extracting fault features from the training data set, obtaining a first fault reconstruction signal based on the fault features, and constructing a first loss function based on the first fault reconstruction signal and the fault features;

[0014] Extracting the operating condition features of the training data set and constructing a second loss function based on the operating condition features of different source domains;

[0015] Calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels;

[0016] Minimize the first loss function, the second loss function and the classifier loss, and determine a trained fault localization model based on the minimum first loss function, the second loss function and the classifier loss.

[0017] Furthermore, the training process of the fault location model further includes:

[0018] Acquire a first working condition reconstruction signal;

[0019] Acquire a second operating condition reconstruction signal and a second fault reconstruction signal;

[0020] A third loss function is constructed based on the first fault reconstruction signal, the first operating condition reconstruction signal, the second fault reconstruction signal, and the second operating condition reconstruction signal. The expression of the third loss function includes:

[0021] ,

[0022] ,

[0023] ,

[0024] ,

[0025] ,

[0026] ,

[0027] ,

[0028] in, represents the third loss function, Represents the source domain index, Indicates the source domain labeled training dataset, Indicates the Normal samples in the source domain labeled training dataset, Indicates the The reconstructed signal of normal samples in the labeled training dataset of the source domain, Indicates the Fault samples in the source domain labeled training dataset, Indicates the The reconstructed signal of the fault sample in the labeled training dataset of the source domain,||·|| F represents the Frobenius norm, For the The number of samples in the source domain labeled training dataset, is the total number of samples in all source domain labeled training datasets, represents the fault encoder extraction operation, represents the fault decoder reconstruction operation, The first The fault characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The first fault reconstruction signal of the normal sample in the source domain labeled training dataset, Indicates the working condition encoder extraction operation, represents the reconstruction operation of the working condition decoder, The first The working condition characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The first working condition reconstructed signal of normal samples in the source domain labeled training dataset, The first The fault features of the fault samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The second fault reconstruction signal of the fault samples in the source domain labeled training dataset, The first The working condition characteristics of the fault samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The second working condition reconstruction signal of the fault sample in the source domain labeled training dataset;

[0029] Minimize the third loss function and determine the trained convolutional autoencoder based on the minimum third loss function.

[0030] Furthermore, calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels includes:

[0031] Based on the labeled training dataset, the predicted label is calculated. The expression includes:

[0032] ,

[0033] The classifier loss is calculated based on the predicted labels. The expression of the classifier loss function includes:

[0034] ,

[0035] in, represents the classifier loss function, Indicates the parameters of the faulty encoder, Represents the classification operation of the classifier, represents the parameters of the classifier, Indicates the The predicted labels of samples in the source domain labeled training dataset, Indicates the The true labels of samples in the source domain labeled training dataset, express The transpose of .

[0036] Furthermore, constructing an expression for a first loss function based on the first fault reconstruction signal and the fault feature includes:

[0037] ,

[0038] in, represents the first loss function, Indicates the parameters of the faulty encoder, Represents the parameters of the fault decoder.

[0039] Furthermore, the expression of the second loss function constructed based on the working condition characteristics of different source domains includes:

[0040] ,

[0041] in, represents the second loss function, Indicates the source domain labeled training dataset, Indicates the first The working condition characteristics of samples in the source domain labeled training dataset, represents the sample index in the labeled training dataset, Indicates the The number of samples in the source domain labeled training dataset, Indicates the first The working condition characteristics of the samples in the source domain labeled training dataset, H represents the sample in the reproducing kernel Hilbert space, Indicates the parameters of the working condition encoder.

[0042] Furthermore, obtaining potential signals at a plurality of preset rail nodes when a train is in different traction conditions and different insulation fault states occur includes:

[0043] A return current system was established based on the actual subway system. Trains draw current from the traction network and use the running rails as the return path to complete the mixed parameter model of the train current return to the traction substation. The data was divided according to the corresponding number of sampling points, and the corresponding normal and fault conditions were marked for data preprocessing.

[0044] The backflow system is equivalent to a double π-type circuit, and the iterative process is carried out based on the simplified node voltage equation GU=I. Since the substation is in a no-load state in the initial state, the initial node voltage is set to , combined with the train operating power in the line, based on calculate ,based on calculate , combined with the formula calculate ,based on Calculate the node current matrix for one iteration ,based on calculate , when the i-th iteration is completed, the node voltage matrix and power satisfy ||U (i+1) −U (i) ||<ε1,ΔP (i) When <ε2, the iteration is terminated, and the train running status and rail potential signals of each node at any time are obtained to realize the power flow calculation of the dynamic operation of the train;

[0045] Where G is the node admittance matrix, U is the node voltage vector to be determined, and I is the node current injection vector. is the initial node voltage matrix with cross section m, is the positive node voltage vector, The second parameter 0 in is the negative node voltage, The other parameters in 0 are the initial voltage of the middle node, [] T represents transpose, is the node current matrix with cross section m at the current moment, is the node admittance matrix of one iteration, is the node voltage matrix with a cross section of m after one iteration, is the node current matrix with a cross section of m for one iteration, is the node admittance matrix after two iterations, U (i +1) is the total node voltage matrix of iteration i, U (i) is the total node voltage matrix of iteration i-1 times, ΔP (i) is the unbalanced power of all nodes in iteration i, that is, the difference between the injected power and the extracted power of the node, ε1 is the preset voltage difference threshold, and ε2 is the preset power difference threshold.

[0046] In a second aspect, the present invention provides a device for locating insulation faults in running rails under unknown working conditions, comprising:

[0047] A signal acquisition module is used to acquire potential signals at multiple preset nodes of the rail where the fault location is to be determined;

[0048] a fault location module, configured to input the potential signals at a plurality of preset nodes of the rail into a pre-acquired fault location model, obtain a fault label of the potential signal at each node, and determine the fault location according to the fault label;

[0049] The fault location model acquisition module is used to acquire the fault location model. The training method of the fault location model includes:

[0050] Obtain the potential signals at multiple preset rail nodes when the train is under different traction conditions and different insulation fault states occur;

[0051] Preprocess the potential signals at multiple preset rail nodes to obtain labeled training datasets in different source domains.

[0052] Extracting fault features from the training data set, obtaining a first fault reconstruction signal based on the fault features, and constructing a first loss function based on the first fault reconstruction signal and the fault features;

[0053] Extracting the operating condition features of the training data set and constructing a second loss function based on the operating condition features of different source domains;

[0054] Calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels;

[0055] Minimize the first loss function, the second loss function and the classifier loss, and determine a trained fault localization model based on the minimum first loss function, the second loss function and the classifier loss.

[0056] In a third aspect, the present invention provides an electronic terminal comprising a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of any of the above methods are performed.

[0057] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when executed by a processor.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] The present invention first proposes a method for locating insulation faults of running rails under unknown working conditions. Based on a trained fault location model, the fault features and working condition features extracted by the fault location model are used to obtain corresponding reconstructed signals. Then, based on the features and the reconstructed signals corresponding to the features, a first loss function and a second loss function are constructed. By minimizing the first loss function and the second loss function, as many working condition features as possible can be extracted, and the fault features extracted from normal samples can be eliminated. The maximum mean difference can also be used to focus more on the fault features, thereby realizing the section positioning of the running rail insulation fault position under unknown working conditions and improving the positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of a method for locating insulation faults in running rails under unknown working conditions provided by an embodiment of the present invention;

[0061] Figure 2 This is a schematic diagram of a rail potential signal generation process based on a dynamic simulation platform of an urban rail power supply system in a method for locating an insulation fault in an unknown operating condition provided by an embodiment of the present invention;

[0062] Figure 3 This is a structural diagram of a fault location model in a method for locating insulation faults in running rails under unknown working conditions provided by an embodiment of the present invention;

[0063] Figure 4 This is a flowchart of model training and use in a method for locating insulation faults in running rails under unknown working conditions provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. Unless there is a conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0065] The term "and / or" in this disclosure simply describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this disclosure generally indicates that the related objects are in an "or" relationship.

[0066] Example 1:

[0067] Figure 1 This is a flow chart of the method for locating insulation faults in unknown working conditions in the first embodiment of the present invention. This flow chart only shows the logical sequence of the method described in this embodiment. In other possible embodiments of the present invention, different methods may be used without conflict. Figure 1 The steps shown or described are accomplished in the order shown.

[0068] The method for locating insulation faults in unknown operating conditions provided in this embodiment can be applied to a terminal and can be executed by a mechanical equipment fault identification device, which can be implemented by software and / or hardware and can be integrated into a terminal, such as any smart phone, tablet computer or computer device with communication function. Figure 1 and Figure 4 As shown, the method of this embodiment specifically includes the following steps:

[0069] Step 1: Obtain potential signals at multiple preset nodes of the rail where the fault location is to be determined.

[0070] Step 2: Input the potential signals at multiple preset nodes of the rail into a pre-acquired fault location model, obtain the fault label of the potential signal of each node, and determine the fault location according to the fault label.

[0071] In this embodiment, the label of a normal sample is 0. Actual insulation failure often occurs at the location of the track insulation fasteners. The interval between the fasteners is set to 0.6m. Insulation failure is set starting at 6200m, and the failure location is set at intervals of +0.6m to 6380m. Every 20m is used as an insulation failure positioning interval. A total of 9 types of insulation failure samples are set with labels 1 to 9. The simulation is run multiple times to ensure that the number of samples in each type is 100, for a total of 1000 samples.

[0072] It should be noted that the insulation failure location interval is the label mentioned in this application. When the fault label is different labels 1~9, it means that the fault occurs at different positions on the insulation section. The appearance of a non-0 label means that the section belongs to the insulation failure section, that is, the fault position. When the fault label is 0, it means that the section belongs to the normal insulation section and no fault occurs.

[0073] In order to verify the effectiveness and superiority of the proposed invention, this experimental case implemented four fault location tasks under different neighborhood generalization diagnostic tasks. There are four data sets in total for dividing the source domain and the target domain, one of which is a domain generalization task that can be represented by S_02, where 0 and 2 represent two different labeled training sets, and the other two domains are test sets. In order to verify the effectiveness of the method, device, and storage medium for locating insulation faults of running rails under unknown working conditions proposed in the present invention, five other advanced diagnostic methods were compared, including M1 and M2: domain generalization of a single source domain, which respectively represent the average test accuracy and the best accuracy of the two models for the two target domains; M3: convolutional autoencoder fused with Coral statistical distance to achieve domain generalization; M4: ResNet fused with MMD statistical distance to achieve domain generalization; M5: domain generalization method MSAC that removes domain-specific features.

[0074] Compared with other domain generalization methods, the method of the present invention has achieved better experimental results in terms of fault location recognition accuracy in most generalization tasks. The comparison results are shown in Table 1.

[0075] Table 1 Experimental results

[0076]

[0077] Figure 3 The specific structure of the fault location model is shown. The model mainly consists of three parts: 1) fault autoencoder; 2) working condition autoencoder; 3) classifier. The rail potential of multiple source domains is the input signal. The autoencoder in the figure is composed of a convolutional neural network and transposed convolution, and the classifier is composed of a fully connected layer. The model structure is shown in Table 2.

[0078] Table 2

[0079]

[0080] like Figure 4 As shown, the training method of the fault location model includes:

[0081] Obtain the potential signals at multiple preset rail nodes when the train is under different traction conditions and different insulation fault states occur, specifically including:

[0082] Based on the actual subway system, a dynamic simulation platform for the urban rail power supply system is established, which includes calculations and ground faults. The train draws current from the traction network and uses the running rails as the return path to complete the mixed parameter model of the train current return to the traction substation. The data is divided according to the corresponding number of sampling points (it should be noted that the method mentioned above for obtaining the potential signals at multiple preset nodes of the rail where the fault location is to be determined is similar to the method for obtaining the rail potential signals under different insulation fault states here. Only the test rail potential signals that are actually input into the fault location model are not marked). The corresponding normal and fault conditions are marked and data preprocessing is performed.

[0083] Regarding the urban rail power supply system dynamic simulation platform, as shown in the attached Figure 2 As shown: Among them, in the output characteristics of the rectifier unit of the traction substation, U max Indicates the maximum voltage, U d0 Indicates no-load voltage, U dN Indicates rated voltage, U min Indicates the minimum voltage, I dN represents the rated current; VLD is the voltage limiting device, DD is the current drain device, U(x,t) is the rail potential at position x at time t, ΔT is the duration, R ov (t) is the equivalent resistance of VLD at time t, R dd (t) is the equivalent resistance of DD at time t; Z cn is the equivalent resistance of the traction network in section n, TS n is the traction output characteristic of section n, Z n is the equivalent resistance matrix of the track and drainage network in section n, Y n is the equivalent resistance matrix of the track to the drainage network and the drainage network to the ground in section n. The construction of the dynamic simulation platform for the urban rail power supply system belongs to the existing technology and will not be repeated here.

[0084] The urban rail power supply system dynamic simulation platform is equivalent to a double-π type circuit, and the potential signals at multiple preset rail nodes when the train has different insulation fault states under different traction conditions are obtained. The specific steps include:

[0085] A return current system is established based on the actual subway system. The train draws current from the traction network and uses the running rails as the return path to complete the mixed parameter model of the train current return to the traction substation. The data is divided according to the corresponding number of sampling points, and the corresponding normal and fault conditions are marked for data preprocessing.

[0086] The backflow system is equivalent to a double π-type circuit, and the iterative process is carried out based on the simplified node voltage equation GU=I. Since the substation is in a no-load state in the initial state, the initial node voltage is set to , combined with the train operating power in the line, based on calculate ,based on calculate , combined with the formula calculate ,based on Calculate the node current matrix for one iteration ,based on calculate , when the i-th iteration is completed, the node voltage matrix and power satisfy ||U (i+1) −U (i) ||<ε1,ΔP (i) When <ε2, the iteration is terminated, and the train running status and rail potential signals of each node at any time are obtained to realize the power flow calculation of the dynamic operation of the train;

[0087] Where G is the node admittance matrix, U is the node voltage vector to be determined, and I is the node current injection vector. is the initial node voltage matrix with cross section m, is the positive node voltage vector, The second parameter 0 in is the negative node voltage, The other parameters in 0 are the initial voltage of the middle node, [] T represents transpose, is the node current matrix with cross section m at the current moment, is the node admittance matrix of one iteration, is the node voltage matrix with a cross section of m after one iteration, is the node current matrix with a cross section of m for one iteration, is the node admittance matrix after two iterations, U (i +1) is the total node voltage matrix of iteration i, U (i) is the total node voltage matrix of iteration i-1 times, ΔP (i) is the unbalanced power of all nodes in iteration i, that is, the difference between the injected power and the extracted power of the node, ε1 is the preset voltage difference threshold, and ε2 is the preset power difference threshold.

[0088] How to obtain rail potential signals under different insulation fault states under different train traction conditions through power flow calculation belongs to the existing technology and is only explained here through the above content without further elaboration.

[0089] In order to better simulate the actual operating conditions of trains, four most common operating conditions are set, namely empty, full, full, and overloaded. The calculation formulas for the traction force F of trains under different loads include:

[0090] ,

[0091] Where G = 300t is the train weight, λ = 0.1 represents the inertia coefficient of the train's rotating parts, and α is the starting acceleration, which is 0.9~1.0m / s 2 , f x , f p , f r and f c , respectively, represent the basic resistance, grade resistance, curve resistance, and tunnel resistance during train traction. Based on the traction calculation, the train traction power P = 3507.39 kW. The corresponding train power is calculated based on different train load conditions. The traction power under no-load, full-load, full-load, and overload conditions is 0.74P, 0.8P, 1.1P, and 1.2P, respectively.

[0092] The potential signals at multiple preset rail nodes are preprocessed to obtain labeled training datasets in different source domains.

[0093] Extracting fault features of the training data set through a fault encoder of the fault localization model, obtaining a first fault reconstruction signal based on the fault features through a fault decoder of the fault localization model, and constructing a first loss function according to the first fault reconstruction signal and the fault features includes:

[0094] ,

[0095] in, represents the first loss function, Indicates the source domains, Indicates the Normal samples in the labeled training dataset of the source domain, Indicates the parameters of the faulty encoder, represents the parameters of the fault decoder, For the The number of samples in the source domain labeled training dataset, is the total number of samples in all source domain labeled training datasets, The first The fault characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The first fault reconstruction signal of normal samples in the labeled training dataset of the source domain.

[0096] The operating condition features of the training data set are extracted by the operating condition encoder of the fault location model, and the expression of the second loss function is constructed based on the operating condition features of different source domains. The expression includes:

[0097] ,

[0098] in, represents the second loss function, Indicates the source domain labeled training dataset, Indicates the first The working condition characteristics of samples in the source domain labeled training dataset, represents the sample index in the labeled training dataset, Indicates the The number of samples in the source domain labeled training dataset, Indicates the first The working condition characteristics of the samples in the source domain labeled training dataset, H represents the sample in the reproducing kernel Hilbert space, Indicates the parameters of the working condition encoder.

[0099] Minimize the first loss function and the second loss function, and by reducing the first loss function as much as possible, the working condition encoder can extract as many working condition features as possible and eliminate the fault features extracted from normal samples; in order to enhance the generalization ability of the model to all domains and suppress overfitting of the source domain, the statistical distance is used to reduce the differences between the working condition features, so that the model focuses more on the fault features. The maximum mean difference method is widely used to evaluate the data distribution differences between domains.

[0100] Based on the labeled training dataset, the predicted label is calculated. The expression includes:

[0101] ,

[0102] The classifier loss is calculated based on the predicted labels. The expression of the classifier loss function includes:

[0103] ,

[0104] Minimize the classifier loss function by shrinking Make the model predict the label value of the labeled training data set as accurately as possible; represents the classifier loss function, Represents the classification operation of the classifier, represents the parameters of the classifier, Indicates the The predicted labels of samples in the source domain labeled training dataset, Indicates the The true labels of samples in the source domain labeled training dataset, express The transpose of .

[0105] Minimize the first loss function, the second loss function and the classifier loss, and determine a trained fault localization model based on the minimum first loss function, the second loss function and the classifier loss.

[0106] In addition, the training process of the fault location model also includes:

[0107] Obtain a first operating condition reconstruction signal, a second operating condition reconstruction signal, and a second fault reconstruction signal; and construct a third loss function based on the first fault reconstruction signal, the first operating condition reconstruction signal, the second fault reconstruction signal, and the second operating condition reconstruction signal, wherein the expression of the third loss function includes:

[0108] ,

[0109] ,

[0110] ,

[0111] ,

[0112] ,

[0113] ,

[0114] ,

[0115] in, represents the third loss function, Indicates the The reconstructed signal of normal samples in the labeled training dataset of the source domain, Indicates the Fault samples in the source domain labeled training dataset, Indicates the The reconstructed signal of the fault sample in the labeled training dataset of the source domain,||·|| F represents the Frobenius norm, represents the fault encoder extraction operation, represents the fault decoder reconstruction operation, The first The fault characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The first fault reconstruction signal of the normal sample in the source domain labeled training dataset, Indicates the working condition encoder extraction operation, represents the reconstruction operation of the working condition decoder, The first The working condition characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The first working condition reconstructed signal of normal samples in the source domain labeled training dataset, The first The fault features of the fault samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The second fault reconstruction signal of the fault samples in the source domain labeled training dataset, The first The working condition characteristics of the fault samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The second working condition reconstructed signal of the fault sample in the source domain labeled training dataset.

[0116] Minimize the third loss function by shrinking the loss function L r , according to the minimum third loss function, the trained convolutional autoencoder is determined to separate the fault characteristics and the working condition characteristics. The convolutional autoencoder includes the following Figure 3 The working condition autoencoder and the fault autoencoder in the embodiment include a fault encoder and a fault decoder, and the working condition autoencoder includes a working condition encoder and a working condition decoder.

[0117] Convolutional autoencoders usually consist of an encoder Φ e and a decoder Φ d , used to extract features and reconstruct signals respectively, for the encoder Φ e The feature extraction process adopts the form of one-dimensional convolution, and the feature v of the input signal of the lth layer l The expressions include:

[0118] ,

[0119] Among them, k l and b l Represent the convolution and bias of the lth layer respectively, Represents the activation function. This paper uses the rectified linear unit activation function. * represents the convolution process. v l-1 Represents the output feature of the l-1 layer for the decoder Φ d In the form of one-dimensional transposed convolution, in order to adapt to Φ d Input requirements, in the decoder Φ e During the convolution process, the step size is set to 1 and padding is performed.

[0120] In the process of training the fault localization model, the actual labels of the labeled training data set can be compared with the predicted labels to verify the accuracy of the prediction results. The fault localization model trained by this application can improve the accuracy of the prediction results.

[0121] Preferably, the total loss function L of the model can be obtained by combining the above four loss terms, as shown in the formula:

[0122] ,

[0123] Among them, α, β, and γ are adjustable parameters. By minimizing L, a model capable of locating insulation failures in unknown working conditions can be trained.

[0124] Example 2:

[0125] A second embodiment of the present invention provides a device for locating insulation faults in running rails under unknown working conditions, comprising:

[0126] A signal acquisition module is used to acquire potential signals at multiple preset nodes of the rail where the fault location is to be determined;

[0127] a fault location module, configured to input the potential signals at a plurality of preset nodes of the rail into a pre-acquired fault location model, obtain a fault label of the potential signal at each node, and determine the fault location according to the fault label;

[0128] The fault location model acquisition module is used to acquire the fault location model. The training method of the fault location model includes:

[0129] Obtain the potential signals at multiple preset rail nodes when the train is under different traction conditions and different insulation fault states occur;

[0130] Preprocess the potential signals at multiple preset rail nodes to obtain labeled training datasets in different source domains.

[0131] Extracting fault features from the training data set, obtaining a first fault reconstruction signal based on the fault features, and constructing a first loss function based on the first fault reconstruction signal and the fault features;

[0132] Extracting the operating condition features of the training data set and constructing a second loss function based on the operating condition features of different source domains;

[0133] Calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels;

[0134] Minimize the first loss function, the second loss function and the classifier loss, and determine a trained fault localization model based on the minimum first loss function, the second loss function and the classifier loss.

[0135] The unknown operating condition running rail insulation fault locating device provided in the second embodiment of the present invention can execute the unknown operating condition running rail insulation fault locating method provided in the first embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0136] Example 3:

[0137] The third embodiment of the present invention further provides an electronic terminal, comprising a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and the processor is configured to operate according to the instructions to execute the steps of the method described in the first embodiment.

[0138] The electronic terminal provided in the third embodiment of the present invention can execute the unknown working condition running rail insulation fault locating method provided in the first embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0139] Example 4:

[0140] Embodiment 4 of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in embodiment 1 are implemented, and the computer program has functional modules and beneficial effects corresponding to the execution method.

[0141] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, apparatuses, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (apparatus), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0143] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0144] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0145] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for locating insulation faults in running rails under unknown working conditions, characterized in that: include: Obtaining potential signals at multiple preset nodes of the rail where the fault location is to be determined; Inputting the potential signals at a plurality of preset nodes of the rail into a pre-acquired fault location model, obtaining a fault label of the potential signal of each node, and determining the fault location according to the fault label; The training method of the fault location model includes: Obtain the potential signals at multiple preset rail nodes when the train is under different traction conditions and different insulation fault states occur; Preprocess the potential signals at multiple preset rail nodes to obtain labeled training datasets in different source domains. Extracting fault features from the training data set, obtaining a first fault reconstruction signal based on the fault features, and constructing a first loss function based on the first fault reconstruction signal and the fault features; Extracting the operating condition features of the training data set and constructing a second loss function based on the operating condition features of different source domains; Calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels; Minimize the first loss function, the second loss function, and the classifier loss, and determine a trained fault localization model based on the minimum first loss function, the second loss function, and the classifier loss; The training method of the fault location model further includes: Acquire a first working condition reconstruction signal; Acquire a second operating condition reconstruction signal and a second fault reconstruction signal; A third loss function is constructed based on the first fault reconstruction signal, the first operating condition reconstruction signal, the second fault reconstruction signal, and the second operating condition reconstruction signal. The expression of the third loss function includes: ， ， ， ， ， ， ， in, represents the third loss function, Represents the source domain index, Indicates the source domain labeled training dataset, Indicates the Normal samples in the source domain labeled training dataset, Indicates the The reconstructed signal of normal samples in the labeled training dataset of the source domain, Indicates the Fault samples in the labeled training dataset of the source domain, Indicates the The reconstructed signal of the fault sample in the labeled training dataset of the source domain,||·|| F represents the Frobenius norm, For the The number of samples in the source domain labeled training dataset, is the total number of samples in all source domain labeled training datasets, represents the fault encoder extraction operation, represents the fault decoder reconstruction operation, The first The fault characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The first fault reconstruction signal of the normal sample in the source domain labeled training dataset, Indicates the working condition encoder extraction operation, represents the reconstruction operation of the working condition decoder, The first The working condition characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The first working condition reconstructed signal of normal samples in the source domain labeled training dataset, The first The fault features of the fault samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The second fault reconstruction signal of the fault samples in the source domain labeled training dataset, The first The working condition characteristics of the fault samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The second working condition reconstruction signal of the fault sample in the source domain labeled training dataset; Minimizing the third loss function, and determining a trained convolutional autoencoder according to the minimum third loss function; Calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels includes: Based on the labeled training dataset, the predicted label is calculated. The expression includes: ， The classifier loss is calculated based on the predicted labels. The expression of the classifier loss function includes: ， in, represents the classifier loss function, Indicates the parameters of the faulty encoder, Represents the classification operation of the classifier, represents the parameters of the classifier, Indicates the The predicted labels of samples in the source domain labeled training dataset, Indicates the The true labels of samples in the source domain labeled training dataset, express The transpose of .

2. The method for locating insulation faults in unknown operating conditions according to claim 1, characterized in that: The expression for constructing the first loss function according to the first fault reconstruction signal and the fault feature includes: ， in, represents the first loss function, Indicates the parameters of the faulty encoder, Represents the parameters of the fault decoder.

3. The method for locating insulation faults in unknown operating conditions according to claim 2, characterized in that: The expressions for constructing the second loss function based on the working condition characteristics of different source domains include: ， in, represents the second loss function, Indicates the source domain labeled training dataset, Indicates the first The working condition characteristics of samples in the source domain labeled training dataset, represents the sample index in the labeled training dataset, Indicates the The number of samples in the source domain labeled training dataset, Indicates the first The working condition characteristics of the samples in the source domain labeled training dataset, H represents the sample in the reproducing kernel Hilbert space, Indicates the parameters of the working condition encoder.

4. The method for locating insulation faults in unknown operating conditions according to claim 1, characterized in that: Obtaining the potential signals at multiple preset rail nodes when the train is under different traction conditions and different insulation fault states occurs includes: A return current system was established based on the actual subway system. Trains draw current from the traction network and use the running rails as the return path to complete the mixed parameter model of the train current return to the traction substation. The data was divided according to the corresponding number of sampling points, and the corresponding normal and fault conditions were marked for data preprocessing. The backflow system is equivalent to a double π-type circuit, and the iterative process is carried out based on the simplified node voltage equation GU=I. Since the substation is in a no-load state in the initial state, the initial node voltage is set to , combined with the train operating power in the line, based on calculate ,based on calculate , combined with the formula calculate ,based on Calculate the node current matrix for one iteration ,based on calculate , when the i-th iteration is completed, the node voltage matrix and power satisfy ||U (i+1) −U (i) ||<ε1,ΔP (i) When <ε2, the iteration is terminated, and the train running status and rail potential signals of each node at any time are obtained to realize the power flow calculation of dynamic train operation; Where G is the node admittance matrix, U is the node voltage vector to be determined, and I is the node current injection vector. is the initial node voltage matrix with cross section m, is the positive node voltage vector, The second parameter 0 in is the negative node voltage, The other parameters in 0 are the initial voltage of the middle node, [] T represents transpose, is the node current matrix with cross section m at the current moment, is the node admittance matrix of one iteration, is the node voltage matrix with a cross section of m after one iteration, is the node current matrix with a cross section of m for one iteration, is the node admittance matrix after two iterations, U (i +1) is the total node voltage matrix of iteration i, U (i) is the total node voltage matrix of iteration i-1 times, ΔP (i) is the unbalanced power of all nodes in iteration i, that is, the difference between the injected power and the extracted power of the node, ε1 is the preset voltage difference threshold, and ε2 is the preset power difference threshold.

5. A device for locating insulation faults in running rails under unknown working conditions, characterized in that: include: A signal acquisition module is used to acquire potential signals at multiple preset nodes of the rail where the fault location is to be determined; a fault location module, configured to input the potential signals at a plurality of preset nodes of the rail into a pre-acquired fault location model, obtain a fault label of the potential signal at each node, and determine the fault location according to the fault label; The fault location model acquisition module is used to acquire the fault location model. The training method of the fault location model includes: Obtain the potential signals at multiple preset rail nodes when the train is under different traction conditions and different insulation fault states occur; Preprocess the potential signals at multiple preset rail nodes to obtain labeled training datasets in different source domains. Extracting fault features from the training data set, obtaining a first fault reconstruction signal based on the fault features, and constructing a first loss function based on the first fault reconstruction signal and the fault features; Extracting the operating condition features of the training data set and constructing a second loss function based on the operating condition features of different source domains; Calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels; Minimize the first loss function, the second loss function, and the classifier loss, and determine a trained fault localization model based on the minimum first loss function, the second loss function, and the classifier loss; The training method of the fault location model further includes: Acquire a first working condition reconstruction signal; Acquire a second operating condition reconstruction signal and a second fault reconstruction signal; A third loss function is constructed based on the first fault reconstruction signal, the first operating condition reconstruction signal, the second fault reconstruction signal, and the second operating condition reconstruction signal. The expression of the third loss function includes: ， ， ， ， ， ， ， in, represents the third loss function, Represents the source domain index, Indicates the source domain labeled training dataset, Indicates the Normal samples in the source domain labeled training dataset, Indicates the The reconstructed signal of normal samples in the labeled training dataset of the source domain, Indicates the Fault samples in the labeled training dataset of the source domain, Indicates the The reconstructed signal of the fault sample in the labeled training dataset of the source domain,||·|| F represents the Frobenius norm, For the The number of samples in the source domain labeled training dataset, is the total number of samples in all source domain labeled training datasets, represents the fault encoder extraction operation, represents the fault decoder reconstruction operation, The first The fault characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The first fault reconstruction signal of the normal sample in the source domain labeled training dataset, Indicates the working condition encoder extraction operation, represents the reconstruction operation of the working condition decoder, The first The working condition characteristics of normal samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The first working condition reconstructed signal of normal samples in the source domain labeled training dataset, The first The fault features of the fault samples in the labeled training dataset of the source domain, Reconstruction of the fault decoder The second fault reconstruction signal of the fault samples in the source domain labeled training dataset, The first The working condition characteristics of the fault samples in the labeled training dataset of the source domain, Reconstruction of the working condition decoder The second working condition reconstruction signal of the fault sample in the source domain labeled training dataset; Minimizing the third loss function, and determining a trained convolutional autoencoder according to the minimum third loss function; Calculating predicted labels based on the labeled training data set, and calculating classifier loss according to the predicted labels includes: Based on the labeled training dataset, the predicted label is calculated. The expression includes: ， The classifier loss is calculated based on the predicted labels. The expression of the classifier loss function includes: ， in, represents the classifier loss function, Indicates the parameters of the faulty encoder, Represents the classification operation of the classifier, represents the parameters of the classifier, Indicates the The predicted labels of samples in the source domain labeled training dataset, Indicates the The true labels of samples in the source domain labeled training dataset, express The transpose of .

6. An electronic terminal, characterized in that: The method comprises a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 4 are executed.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images