DC power supply and distribution system insulation weak point positioning method and readable storage medium

Abstract

The invention relates to the technical field of rail transit fault detection, and discloses a DC power supply and distribution system insulation weak point positioning method and a readable storage medium, and the method comprises the steps: obtaining the current signals and potential signals of all fixed test points under the drive of a fixed DC current based on the actual operation scene of a DC power supply and distribution system, constructing sample data, and carrying out the detection of the DC power supply and distribution system; a multi-source domain adaptive auto-encoder model is utilized, and a target insulation weak point positioning model is obtained through a model Call loss function, model reconstruction loss and model classifier loss; in the positioning stage, a fixed direct current injection system which is the same as that in the training stage is adopted, current and potential signals at a fixed test point are collected, and then insulation weak point positioning can be carried out, so that the positioning time is greatly shortened, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit fault detection, and in particular to a method for locating insulation weak points in a direct current power supply and distribution system and a readable storage medium. Background Art

[0002] Currently, DC power distribution systems offer advantages such as low line losses, high power supply reliability, a simple system structure, and easy integration of distributed power sources and energy storage devices. They have become a popular power supply method in various scenarios, such as rail transit and ships. For example, in rail transit DC power supply systems, the traction current originates from the positive terminal of the rectifier unit in the traction substation, passes through the catenary (positive power path) to the train, and then flows back to the negative terminal of the rectifier unit via the running rails (negative power path). In this power supply path, insulation between the positive and negative current paths and the ground must be ensured to prevent current leakage and ground short circuits. However, as the system ages and the environment degrades, weak or even damaged insulation points inevitably develop in the power supply path. This causes some current to leak through these weak points into the surrounding ground, generating stray currents. This can also cause the supply voltage to drop, triggering system protection and causing power outages. This stray current leaking into the surrounding area can accelerate electrochemical corrosion of buried metal structures in urban areas, damage infrastructure, disrupt the normal operation of power equipment, and even pose a threat to personnel. In order to effectively prevent and control stray currents and improve the safety and stability of DC power supply and distribution systems, it is crucial to accurately locate insulation weak points in DC power supply and distribution systems.

[0003] However, current methods for detecting weak insulation in DC power supply and distribution systems have significant limitations. Taking rail transit as an example, the traditional method for detecting the transition conductivity of running rails to the ground has obvious limitations. According to current standards, the current mainstream detection technology mainly applies current between the track and the ground, calculates the leakage current in the tested section, and uses the average voltage of the potential to be measured to calculate the transition conductivity value. Although this method reflects the insulation condition of the track to a certain extent, its testing process is offline, which leads to problems such as high workload and limited testing time. More importantly, due to the long current flow path, the test process can only detect the average transition resistance of a section of the path, making it difficult to directly locate the specific location of weak insulation.

[0004] Currently, offline detection can be performed to locate weak insulation points in DC power supply and distribution systems. Taking the DC power supply system for rail transit as an example, in order to locate weak insulation points in running rails, ZL202410226006.4 "A device and method for locating insulation damage in rail transit running rails" proposes a device and method for locating insulation damage in rail transit running rails. The device includes a first detection device and a second detection device separated by a preset distance. There is a roughly located insulation damage point between the first detection device and the second detection device. The insulation damage point is located on the first running rail or the second running rail of the rail transit. The method includes a method for locating insulation damage in the running rail when there are no rail welds on the rails corresponding to the first detection device and the second detection device, and a method for locating insulation damage in the running rail when there are rail welds on the rails corresponding to the first detection device or the second detection device. The method proposed in this application can accurately locate the insulation damage point of the running rail, but the positioning process requires traversing the entire return path and the device needs to be continuously moved on the running rail. The power supply length of rail transit can usually reach tens of kilometers. The above method of locating insulation damage will consume a lot of manpower and material resources.

[0005] At present, online detection can be performed to locate weak points in the insulation of DC power supply and distribution systems. For example, CN202410226005.X "A method and system for diagnosing grounding faults in rail transit running rails" proposes a method and system for diagnosing grounding faults in rail transit running rails, wherein the method includes: during the test phase of the rail transit line, dividing the line into several sections, and setting grounding faults in different sections of the line; obtaining several characteristic quantities of each section and performing data preprocessing to form a source domain working condition data set; during the operation phase of the rail transit line, obtaining several characteristic quantities of each section and performing data preprocessing to form a target domain working condition data set; training the neural network architecture through the source domain working condition data set and the target domain working condition data set to obtain a trained neural network architecture; re-obtaining several characteristic quantities of each section and performing data preprocessing to form a data set to be detected, performing fault diagnosis on the data set to be detected through the trained neural network architecture, and determining the section where the grounding fault is located. This application can effectively detect grounding faults on the running rails, but due to the use of section data for prediction, there is a problem of low accuracy. During the operation of multiple trains, the operating conditions are changeable and the traction current changes are complex. Therefore, the trained diagnostic model is difficult to apply in the actual operation stage, making it difficult to locate weak insulation points.

[0006] To sum up, the existing rail transit running rail grounding fault detection method requires offline detection to traverse the entire power supply path, which is labor-intensive, time-consuming, and consumes a lot of manpower and material resources; online test sample data and operation data are difficult to obtain, and the system current changes are complex, which seriously affects the system's positioning accuracy; therefore, it is impossible to achieve low-cost and high-precision positioning of insulation weak points in rail transit running rail connections, resulting in the inability to timely and effectively discover hidden dangers in the rail transit DC power supply system, affecting the safe and stable operation of the power supply system. Summary of the Invention

[0007] Therefore, the technical problem to be solved by the present invention is to overcome the problems of low efficiency and poor positioning accuracy in the prior art for detecting insulation weak points in rail transit DC power supply systems.

[0008] To solve the above technical problems, the present invention provides a method for locating insulation weak points in a DC power supply and distribution system, comprising: Multiple fixed test points are set up on the power supply circuit of the DC power supply and distribution system, and insulation faults are set up at different locations to generate insulation weak points. When the insulation weak points are at different locations, the current signal and potential signal of each fixed test point under the drive of a fixed DC current are collected to form a sample array corresponding to each insulation weak point. The current signal and potential signal of each fixed test point in the sample array are fused and spliced ​​to obtain initial features. The training set is constructed based on the initial features of all sample arrays and their corresponding true labels representing the locations of the insulation weak points. Using the training set, the autoencoder including the feature encoder and decoder is trained to obtain the target insulation weak point location model, including: The training set is divided into multiple source domains and one target domain, and the initial features are input into the feature encoder to obtain the corresponding encoding features. For all source domains, the covariance matrix between their encoding features and the encoding features of the target domain is aligned, and the model Coral loss function is constructed. Each encoded feature is input into the decoder for reconstruction to obtain the corresponding reconstructed feature; for each source domain, the Frobenius norm of its initial feature and the reconstructed feature is calculated, the reconstruction loss corresponding to each source domain is constructed, and the sum is obtained to obtain the model reconstruction loss; Input the reconstructed features of each source domain into the classifier, obtain the predicted label and its corresponding true label, and calculate the model classifier loss; The weighted sum of the model Coral loss function, model reconstruction loss, and model classifier loss is used as the total loss function. Training is performed until the total loss function converges to obtain the target insulation weak point positioning model. A fixed DC current is applied to the current path to be tested, and the initial features constructed based on the current signals and potential signals at all fixed test points are input into the target insulation weak point positioning model to locate the insulation weak point.

[0009] Preferably, the sample array includes: When the power supply circuit operates normally, the current signals and potential signals of all fixed test points driven by a fixed DC current are used as a normal sample array; Insulation faults are set at different locations in the power supply circuit. The current signals and potential signals of all fixed test points driven by a fixed DC current are collected in real time when the insulation weak points are at different locations, as the fault sample array corresponding to the insulation weak points at different locations.

[0010] Preferably, obtaining the sample array includes: After the DC power supply and distribution system is built, an actual normal sample array and all actual fault sample arrays are collected; and / or, Build a simulation model of the DC power supply and distribution system and obtain the simulated normal sample array and all simulated fault sample arrays.

[0011] Preferably, after the DC power supply and distribution system is built, an array of actual normal samples and arrays of all actual fault samples are collected, including: Build a DC power supply and distribution system and preset multiple fixed test points on its positive and negative paths; A fixed DC current is injected into the DC power supply and distribution system through a DC power supply, and current signals and potential signals of all fixed test points driven by the fixed DC current are collected as an actual normal sample array when the DC power supply and distribution system is operating normally; Install a short-circuit cable for weak insulation faults, one end of which is connected to the negative path and the other end is connected to the ground grid; A fixed DC current is injected into the DC power supply and distribution system through a DC power supply, and the connection position of the insulation-weak fault short-circuit cable on the negative path and the ground network is changed. The current signal and potential signal of multiple fixed test points under the current insulation-weak fault short-circuit cable connection position are collected in real time as an actual fault sample array; Get the array of actual normal samples and the array of all actual fault samples.

[0012] Preferably, a simulation model of a DC power supply and distribution system is constructed to obtain a simulated normal sample array and all simulated fault sample arrays, including: Use MATLAB / Simulink to establish a simulation model of DC power supply and distribution system; Preset multiple fixed test points on the positive path and the negative path of the simulation model; In the simulation model, a fixed DC current is injected, and the current signal and potential signal of all fixed test points under the normal working condition of the simulation model are collected as the simulation normal sample array; Conduct fault simulation based on actual insulation fault types and fault-prone locations; In the simulation model, a fixed DC current is injected to obtain the current signal and potential signal of all fixed test points under different fault simulation modes as a simulation fault sample array; Get the simulation normal sample array and all simulation fault sample arrays.

[0013] Preferably, the initial features are input into the feature encoder respectively to obtain the corresponding encoded features, including: The initial features are input into the multi-scale local attention module, and after three parallel 1×1, 3×1 and 5×1 depth-wise separable convolutions, they are concatenated along the channel dimension to output multi-scale fusion features; Perform global average pooling and global maximum pooling on the multi-scale fusion features respectively to obtain the corresponding average channel vector and maximum channel vector, input them into the fully connected layer, add them together and activate them, and output the channel attention weight; Multiply the channel attention weights and the multi-scale fusion features channel by channel to obtain enhanced features; Perform residual connection between the enhanced features and the initial features, and output the encoded features corresponding to the initial features.

[0014] Preferably, the model Coral loss function is expressed as: ; in, Represents the Coral loss function value, represents the number of source domains, represents the dimension of the covariance matrix, and Respectively represent The source domain and The covariance matrix of the features of the source domain, represents the covariance matrix of the target domain features, represents the Frobenius norm.

[0015] Preferably, the source domain reconstruction loss is expressed as: ; in, represents the source domain reconstruction loss value, represents the initial features of the source domain, represents the reconstructed features of the source domain, represents the Frobenius norm.

[0016] Preferably, the model classifier loss is expressed as: ; in, represents the model classifier loss value, represents the number of source domains, Indicates the The total number of sample data in the source domain, Indicates the The true labels of the source domain, Indicates the The predicted labels of the source domain.

[0017] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for locating insulation weak points in a DC power supply and distribution system as described above are implemented.

[0018] The above technical solution of the present invention has the following beneficial effects compared with the prior art: The method for locating insulation weak points in a DC power supply and distribution system described in the present invention is based on the actual operating scenarios of the DC power supply and distribution system. It obtains current signals and potential signals of all fixed test points under fixed DC current drive, constructs sample data, trains the autoencoder, and obtains a target insulation weak point positioning model. In the positioning stage, the same fixed DC current injection system as the training stage is used to collect current and potential signals at fixed test points to locate insulation weak points, which greatly shortens the positioning time and improves detection efficiency.

[0019] The present invention provides two methods: insulation fault experimental simulation and system simulation model fault simulation. It adopts a fixed DC current injection method and presets fixed test points to obtain sample data under weak insulation conditions. It constructs a sample data set covering multiple fault types and locations to avoid the deviation of a single data source. The samples constructed by the two sample construction methods are combined with autoencoder training to further enhance the comprehensiveness of model training and improve the model positioning accuracy.

[0020] The present invention utilizes a multi-source domain adaptive autoencoder model and constructs a model Coral loss function to align the feature distributions of the source domain and the target domain, effectively solving the model generalization problem caused by the variability of actual working conditions. The model reconstruction loss is used to minimize the mean square error between the initial features and the reconstructed features, thus avoiding excessive discarding of useful information by the encoder and ensuring the integrity of feature extraction. The model classifier loss is used to minimize the cross-entropy loss between the predicted label and the true label, thereby optimizing the feature discrimination capability and improving the fault location accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings, wherein: Figure 1 It is a flowchart of the steps of the method for locating insulation weak points in a DC power supply and distribution system of the present invention; Figure 2 It is a structural diagram of a DC power supply and distribution system; Figure 3 This is the insulation weak point location model architecture diagram; Figure 4 It is a training and testing flow chart of the insulation weak point location model. DETAILED DESCRIPTION

[0022] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0023] Reference Figure 1 As shown in the flowchart of the method for locating insulation weak points in a DC power supply and distribution system of the present invention, the specific steps include: S101: multiple fixed test points are set on the power supply circuit of the DC power supply and distribution system, and insulation faults are set at different locations to generate insulation weak points; current signals and potential signals of each fixed test point driven by a fixed DC current are collected when the insulation weak points are at different locations, forming a sample array corresponding to each insulation weak point; the current signals and potential signals of each fixed test point in the sample array are respectively fused and spliced ​​to obtain initial features; a training set is constructed using the initial features of all sample arrays and their corresponding true labels representing the locations of the insulation weak points; S102: Using the training set, train an autoencoder including a feature encoder and a decoder to obtain a target insulation weak point location model, including: The training set is divided into multiple source domains and one target domain, and the initial features are input into the feature encoder to obtain the corresponding encoding features. For all source domains, the covariance matrix between their encoding features and the encoding features of the target domain is aligned, and the model Coral loss function is constructed, which is expressed as: ;in, Represents the Coral loss function value, represents the number of source domains, represents the dimension of the covariance matrix, and Respectively represent The source domain and The covariance matrix of the features of the source domain, represents the covariance matrix of the target domain features, represents the Frobenius norm.

[0024] Each encoded feature is input into the decoder for reconstruction to obtain the corresponding reconstructed feature; for each source domain, the Frobenius norm of its initial feature and the reconstructed feature is calculated, and the reconstruction loss corresponding to each source domain is constructed. The sum is used to obtain the model reconstruction loss; the source domain reconstruction loss is expressed as: ;in, represents the source domain reconstruction loss value, represents the initial features of the source domain, represents the reconstructed features of the source domain, represents the Frobenius norm.

[0025] The reconstructed features of each source domain are input into the classifier to obtain the predicted label and its corresponding true label, and the model classifier loss is calculated, which is expressed as: ;in, represents the model classifier loss value, represents the number of source domains, Indicates the The total number of sample data in the source domain, Indicates the The true labels of the source domain, Indicates the The predicted labels of the source domain; The weighted sum of the model Coral loss function, model reconstruction loss, and model classifier loss is used as the total loss function. Training is performed until the total loss function converges to obtain the target insulation weak point positioning model. S103: Apply a fixed DC current to the current path to be detected, input the initial features constructed based on the current signals and potential signals at all fixed test points into the target insulation weak point location model, and locate the insulation weak point.

[0026] The method for locating insulation weak points in a DC power supply and distribution system described in the present invention is based on the actual operating scenarios of the DC power supply and distribution system. It obtains current signals and potential signals of all fixed test points under fixed DC current drive, constructs sample data, trains the autoencoder, and obtains a target insulation weak point positioning model. In the positioning stage, the same fixed DC current injection system as the training stage is used to collect current and potential signals at fixed test points to locate insulation weak points, which greatly shortens the positioning time and improves detection efficiency.

[0027] Among them, the sample array includes: When the power supply circuit operates normally, the current signals and potential signals of all fixed test points driven by a fixed DC current are used as a normal sample array; Insulation faults are set at different locations in the power supply circuit. The current signals and potential signals of all fixed test points driven by a fixed DC current are collected in real time when the insulation weak points are at different locations, as the fault sample array corresponding to the insulation weak points at different locations.

[0028] Specifically, this embodiment provides two methods for obtaining sample arrays. When constructing a training set, the present application can select a sample array constructed using one of the two methods, or can use a combination of the two methods to construct a sample array. The two methods include: After the DC power supply and distribution system is built, an actual normal sample array and all actual fault sample arrays are collected; and / or, Build a simulation model of the DC power supply and distribution system and obtain the simulated normal sample array and all simulated fault sample arrays.

[0029] ① After the DC power supply and distribution system is built, collect the actual normal sample array and all actual fault sample arrays, including: Build a DC power supply and distribution system and preset multiple fixed test points on its positive and negative paths; A fixed DC current is injected into the DC power supply and distribution system through a DC power supply, and current signals and potential signals of all fixed test points driven by the fixed DC current are collected as an actual normal sample array when the DC power supply and distribution system is operating normally; Install a short-circuit cable for weak insulation faults, one end of which is connected to the negative path and the other end is connected to the ground grid; A fixed DC current is injected into the DC power supply and distribution system through a DC power supply, and the connection position of the insulation-weak fault short-circuit cable on the negative path and the ground network is changed. The current signal and potential signal of multiple fixed test points under the current insulation-weak fault short-circuit cable connection position are collected in real time as an actual fault sample array; Get the array of actual normal samples and the array of all actual fault samples.

[0030] ② Build a simulation model of the DC power supply and distribution system and obtain the simulated normal sample array and all simulated fault sample arrays, including: Use MATLAB / Simulink to establish a simulation model of DC power supply and distribution system; Preset multiple fixed test points on the positive path and the negative path of the simulation model; In the simulation model, a fixed DC current is injected, and the current signal and potential signal of all fixed test points under the normal working condition of the simulation model are collected as the simulation normal sample array; Conduct fault simulation based on actual insulation fault types and fault-prone locations; In the simulation model, a fixed DC current is injected to obtain the current signal and potential signal of all fixed test points under different fault simulation modes as a simulation fault sample array; Get the simulation normal sample array and all simulation fault sample arrays.

[0031] The present invention provides two methods: insulation fault experimental simulation and system simulation model fault simulation. It adopts a fixed DC current injection method and presets fixed test points to obtain sample data under weak insulation conditions. It constructs a sample data set covering multiple fault types and locations to avoid the deviation of a single data source. The samples constructed by the two sample construction methods are combined with autoencoder training to further enhance the comprehensiveness of model training and improve the model positioning accuracy.

[0032] Specifically, the initial features are input into the feature encoder to obtain the corresponding encoded features, including: The initial features are input into the multi-scale local attention module, and after three parallel 1×1, 3×1 and 5×1 depth-wise separable convolutions, they are concatenated along the channel dimension to output multi-scale fusion features; Perform global average pooling and global maximum pooling on the multi-scale fusion features respectively to obtain the corresponding average channel vector and maximum channel vector, input them into the fully connected layer, add them together and activate them, and output the channel attention weight; Multiply the channel attention weights and the multi-scale fusion features channel by channel to obtain enhanced features; Perform residual connection between the enhanced features and the initial features, and output the encoded features corresponding to the initial features.

[0033] The method for locating insulation weak points in a DC power supply and distribution system provided by the present invention obtains data samples of insulation weak points in the DC power supply and distribution system through two methods: system-set insulation fault experimental simulation and system simulation model fault simulation. The model is trained based on a multi-scale attention multi-source domain adaptive autoencoder model to obtain an intelligent model for locating insulation weak points in the DC power supply and distribution system. In different stages of subsequent system operation, it is only necessary to apply a current to the system once and detect data during the operation of the DC power supply and distribution system to directly locate insulation weak points in the power supply path. The insulation weak points of the DC power supply and distribution system can be quickly located, and accurate and reliable data support and guarantee for the safe operation of the DC power supply and distribution system can be provided.

[0034] Based on the above embodiment, in an embodiment of the present invention, the insulation weak point location method of the DC power supply and distribution system provided by the present invention is used to locate the insulation weak point, including: S201: Fault data sample acquisition: According to the DC power supply and distribution system structure, fault data samples are obtained in two ways: ① After the DC power supply and distribution system is built, a fixed DC current is injected through the V1 DC power supply (this current is consistent with the injection current for later weak point location). The current injection time is determined by the DC circuit breaker K1; Reference Figure 2 Figure 1 is a schematic diagram of the DC power supply and distribution system structure. The system includes: a positive path C1 of the power supply system, a negative path C2 of the power supply system, a ground grid C3, a shorting wire D1 between the positive and negative paths, and a shorting cable D2 for weak insulation faults. D2 is connected between different locations of C1 and C2 and the ground, simulating a weak insulation fault between the power supply return and the ground. After an insulation fault occurs at different locations, fixed test points P1, P2, P3, and P4 in the current path are selected to obtain fault sample data of D2 at different locations, including current signals and potential signals at fixed test points P1, P2, P3, and P4, to construct fault sample data. ② Based on the DC power supply and distribution system power supply circuit structure shown in the figure, a corresponding simulation model of the system was established using MATLAB / Simulink. A fixed DC current was injected into the system simulation model (this current was consistent with the injection current used for later weak point location). Simulations were performed based on common insulation fault types and fault-prone locations in actual DC power supply and distribution systems. Fixed test points P1, P2, P3, and P4 in the current path were selected to obtain fault sample data at different insulation fault locations, including current and potential signals at the test points. S202: Establish an intelligent location model for insulation weaknesses in the DC power supply and distribution system. This model is trained based on a multi-scale attention multi-source domain adaptive autoencoder model and fault sample data, including: S202-1: Fusion of the above two types of fault sample data: The current and voltage signals at test points P1, P2, P3, and P4 are fused, normalized, and then concatenated to obtain the fused initial feature x of the insulation fault location. Fault samples under different operating conditions are simulated for model training to obtain a robust insulation fault location model. This invention uses two methods: system-configured insulation fault experimental simulation and system simulation model fault simulation to obtain data samples under weak insulation conditions. A fixed DC current is injected during acquisition. This same DC current injection is also used during later location of weak insulation points, improving their accuracy.

[0035] S202-2: Construction of insulation weak point location model: Reference Figure 3 The figure shows the architecture of the insulation weak point location model. The insulation weak point location model of this embodiment is an autoencoder, which includes an encoder Φe Used to extract insulation fault characteristic information v = Φ e ( x ), a decoder Φ d Used to reconstruct fault information x ¢= Φ d ( v ); The feature extraction process of the encoder adopts the form of convolution. l The features of the layer can be expressed as v l = f ( v l-1 * k l + b l ), k l and b l Respectively represent l Convolution and bias of the layer, f represents the activation function; The decoder adopts the form of transposed convolution. In order to adapt to the input requirements of the decoder, the step size of 1 is set as padding during the encoder convolution process.

[0036] To enhance the recognition accuracy of the encoder, the convolutional attention module for data classification is integrated, and the multi-scale local attention module LMSA is introduced into the encoder. The hierarchical feature enhancement mechanism is used to improve the extraction capability of insulation failure features. The weights of different data are extracted by using pooling and convolution operations to increase the accuracy of insulation fault location.

[0037] LMSA employs parallel 1×1, 3×1, and 5×1 depthwise separable convolutional structures, cascaded after each convolutional layer, to capture fault features at different scales within the initial signal feature x. LMSA dynamically fuses multi-scale features through a channel-wise attention mechanism, where the dual-path aggregation of average and maximum pooling effectively highlights fault-sensitive responses. This design not only preserves the local perception characteristics of the convolution operation but also achieves adaptive enhancement of key fault features through attention weighting.

[0038] The classifier is implemented by a fully connected layer.

[0039] S202-3: Insulation weak point location model training: Reference Figure 4The figure shows the training and testing flow chart of the insulation weak point location model; the initial fault feature data under different working conditions are divided into source domain and target domain, where the source domain and target domain contain normal samples and multiple insulation weak fault samples in different sections. The labels of the source domain are accessible during the model training process, and the labels of the target domain are not accessible during the training process. In order to obtain more features, multiple source domain data are input into the model to extract the features of each fault sample data. The fault features of each source domain data and target domain data are extracted through the encoder, and the features output by the encoder of multiple source domains and a single target domain are aligned, and the distance metric Coral loss is used to narrow the difference between domains. The Coral loss is calculated as follows: ; in, Represents the Coral loss function value, represents the number of source domains, represents the dimension of the covariance matrix, and Respectively represent The source domain and The covariance matrix of the features of the source domain, represents the covariance matrix of the target domain features, represents the Frobenius norm.

[0040] The decoder is used to reconstruct the source and target domain data, and the reconstruction loss is calculated using the original data and the reconstructed data. The reconstruction loss is calculated as follows: ; in, represents the source domain reconstruction loss value, represents the initial features of the source domain, represents the reconstructed features of the source domain, represents the Frobenius norm.

[0041] In addition, for all source domain data, the extracted features are input into the classifier to output the predicted sample label value, which is compared with the true label to obtain the classifier loss. The classifier loss is calculated as follows: ; in, represents the model classifier loss value, represents the number of source domains, Indicates the The total number of sample data in the source domain, Indicates the The true labels of the source domain, Indicates the The predicted labels of the source domain.

[0042] By reducing the reconstruction loss, classifier loss, and distance metric loss and performing model training, a model with the ability to generalize to different working conditions can be obtained, thereby realizing the intelligent location of weak insulation faults.

[0043] S203: Fault diagnosis: When intelligent positioning of insulation weaknesses is required during different stages of DC power supply and distribution system operation, a fixed DC current is applied to obtain current and potential signals at fixed test points along the current path. This signal is then fed into the intelligent positioning model for insulation weaknesses in the DC power supply and distribution system to pinpoint the specific fault point.

[0044] The present invention can locate insulation weaknesses in the positive and negative return paths of a DC power supply and distribution system. This method simply injects a fixed DC current into the system and collects current and voltage data during system operation to quickly locate insulation weaknesses.

[0045] Based on the above embodiments, an embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the steps of the method for locating insulation weak points in a DC power supply and distribution system as described above are implemented.

[0046] The method for locating insulation weak points in a DC power supply and distribution system described in the present invention is based on the actual operating scenario of the DC power supply and distribution system, obtains the current signals and potential signals of all fixed test points under the drive of a fixed DC current, constructs sample data, trains the autoencoder, and obtains a target insulation weak point positioning model; in the positioning stage, the same fixed DC current injection system as the training stage is used to collect current and potential signals at fixed test points, so that insulation weak points can be located, which greatly shortens the positioning time and improves the detection efficiency. The present invention sets up two methods, insulation fault experimental simulation and system simulation model fault simulation, adopts a fixed DC current injection method, presets fixed test points, to obtain sample data under insulation weak conditions, constructs a sample data set covering a variety of fault types and locations, and avoids the deviation of a single data source; the samples constructed by the two sample construction methods are combined with the training of the autoencoder to further improve the comprehensiveness of the model training and improve the model positioning accuracy. The present invention utilizes a multi-source domain adaptive autoencoder model and constructs a model Coral loss function to align the feature distributions of the source domain and the target domain, effectively solving the model generalization problem caused by the variability of actual working conditions. The model reconstruction loss is used to minimize the mean square error between the initial features and the reconstructed features, thus avoiding excessive discarding of useful information by the encoder and ensuring the integrity of feature extraction. The model classifier loss is used to minimize the cross-entropy loss between the predicted label and the true label, thereby optimizing the feature discrimination capability and improving the fault location accuracy.

[0047] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, 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.

[0048] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), 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.

[0049] 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.

[0050] 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.

[0051] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A method for locating insulation weak points in a DC power supply and distribution system, characterized in that: include: Set up multiple fixed test points on the DC power supply and distribution system power supply circuit, and set insulation faults at different locations to generate insulation weak points; When insulation weak points are located at different locations, the current signals and potential signals of each fixed test point driven by a fixed DC current are collected to form a sample array corresponding to each insulation weak point. The current signals and potential signals of each fixed test point in the sample array are fused and then spliced ​​to obtain the initial features. Construct a training set using the initial features of all sample arrays and their corresponding true labels representing the locations of insulation weak points; Using the training set, the autoencoder including the feature encoder and decoder is trained to obtain the target insulation weak point location model, including: The training set is divided into multiple source domains and one target domain, and the initial features are input into the feature encoder to obtain the corresponding encoding features. For all source domains, the covariance matrix between their encoding features and the encoding features of the target domain is aligned, and the model Coral loss function is constructed. Each encoded feature is input into the decoder for reconstruction to obtain the corresponding reconstructed feature; for each source domain, the Frobenius norm of its initial feature and the reconstructed feature is calculated, the reconstruction loss corresponding to each source domain is constructed, and the sum is obtained to obtain the model reconstruction loss; Input the reconstructed features of each source domain into the classifier, obtain the predicted label and its corresponding true label, and calculate the model classifier loss; The weighted sum of the model Coral loss function, model reconstruction loss, and model classifier loss is used as the total loss function. Training is performed until the total loss function converges to obtain the target insulation weak point positioning model. A fixed DC current is applied to the current path to be tested, and the initial features constructed based on the current signals and potential signals at all fixed test points are input into the target insulation weak point positioning model to locate the insulation weak point.

2. The method for locating insulation weak points in a DC power supply and distribution system according to claim 1, characterized in that: Sample array, including: When the power supply circuit operates normally, the current signals and potential signals of all fixed test points driven by a fixed DC current are used as a normal sample array; Insulation faults are set at different locations in the power supply circuit. The current signals and potential signals of all fixed test points driven by a fixed DC current are collected in real time when the insulation weak points are at different locations, as the fault sample array corresponding to the insulation weak points at different locations.

3. The method for locating insulation weak points in a DC power supply and distribution system according to claim 2, characterized in that: Acquisition of sample arrays, including: After the DC power supply and distribution system is built, an actual normal sample array and all actual fault sample arrays are collected; and / or, Build a simulation model of the DC power supply and distribution system and obtain the simulated normal sample array and all simulated fault sample arrays.

4. The method for locating insulation weak points in a DC power supply and distribution system according to claim 3, characterized in that: After the DC power supply and distribution system is built, collect the actual normal sample array and all actual fault sample arrays, including: Build a DC power supply and distribution system and preset multiple fixed test points on its positive and negative paths; A fixed DC current is injected into the DC power supply and distribution system through a DC power supply, and current signals and potential signals of all fixed test points driven by the fixed DC current are collected as an actual normal sample array when the DC power supply and distribution system is operating normally; Install a short-circuit cable for weak insulation faults, one end of which is connected to the negative path and the other end is connected to the ground grid; A fixed DC current is injected into the DC power supply and distribution system through a DC power supply, and the connection position of the insulation-weak fault short-circuit cable on the negative path and the ground network is changed. The current signal and potential signal of multiple fixed test points under the current insulation-weak fault short-circuit cable connection position are collected in real time as an actual fault sample array; Get the array of actual normal samples and the array of all actual fault samples.

5. The method for locating insulation weak points in a DC power supply and distribution system according to claim 3, characterized in that: Build a simulation model of the DC power supply and distribution system and obtain the simulated normal sample array and all simulated fault sample arrays, including: Use MATLAB / Simulink to establish a simulation model of DC power supply and distribution system; Preset multiple fixed test points on the positive path and the negative path of the simulation model; In the simulation model, a fixed DC current is injected, and the current signal and potential signal of all fixed test points under the normal working condition of the simulation model are collected as the simulation normal sample array; Conduct fault simulation based on actual insulation fault types and fault-prone locations; In the simulation model, a fixed DC current is injected to obtain the current signal and potential signal of all fixed test points under different fault simulation modes as a simulation fault sample array; Get the simulation normal sample array and all simulation fault sample arrays.

6. The method for locating insulation weak points in a DC power supply and distribution system according to claim 1, characterized in that: Input the initial features into the feature encoder to obtain the corresponding encoded features, including: The initial features are input into the multi-scale local attention module, and after three parallel 1×1, 3×1 and 5×1 depth-wise separable convolutions, they are concatenated along the channel dimension to output multi-scale fusion features; Perform global average pooling and global maximum pooling on the multi-scale fusion features respectively to obtain the corresponding average channel vector and maximum channel vector, input them into the fully connected layer, add them together and activate them, and output the channel attention weight; Multiply the channel attention weights and the multi-scale fusion features channel by channel to obtain enhanced features; Perform residual connection between the enhanced features and the initial features, and output the encoded features corresponding to the initial features.

7. The method for locating insulation weak points in a DC power supply and distribution system according to claim 1, characterized in that: The Coral loss function of the model is expressed as: ； in, Represents the Coral loss function value, represents the number of source domains, represents the dimension of the covariance matrix, and Respectively represent The source domain and The covariance matrix of the features of the source domain, represents the covariance matrix of the target domain features, represents the Frobenius norm.

8. The method for locating insulation weak points in a DC power supply and distribution system according to claim 1, characterized in that: The source domain reconstruction loss is expressed as: ； in, represents the source domain reconstruction loss value, represents the initial features of the source domain, represents the reconstructed features of the source domain, represents the Frobenius norm.

9. The method for locating insulation weak points in a DC power supply and distribution system according to claim 1, characterized in that: The model classifier loss is expressed as: ； in, represents the model classifier loss value, represents the number of source domains, Indicates the The total number of sample data in the source domain, Indicates the The true labels of the source domain, Indicates the The predicted labels of the source domain.

10. 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 for locating insulation weak points in a DC power supply and distribution system according to any one of claims 1 to 9 are implemented.

Images