Wire and cable fault diagnosis model training method and device, medium and product

By synchronously training a generative adversarial network model for cable fault diagnosis, the problem of local optima in the training of cable fault diagnosis models is solved, the generator is fully trained, and the accuracy of cable fault diagnosis is improved.

CN118940111AInactive Publication Date: 2024-11-12GUANGDONG POWER GRID CO LTD +1

Patent Information

Application Number
CN202410989269.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing fault diagnosis models for wires and cables are prone to getting stuck in local optima during training, resulting in insufficient optimization of model parameters and affecting diagnostic performance.

Method used

A generative adversarial network model for cable fault diagnosis is adopted. By synchronously training the generator and discriminator in parallel, the model parameters of the generator and discriminator are updated using the loss function and historical batch model parameters, thus avoiding local optima and achieving full training of the generator.

Benefits of technology

The performance of the cable fault diagnosis model has been improved, ensuring that the generator model parameters are closer to the global optimum, thus enhancing the accuracy of fault diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940111A_ABST
    Figure CN118940111A_ABST
Patent Text Reader

Abstract

The invention discloses a wire and cable fault diagnosis model training method and device, a medium and a product. In a training set based on a cable fault sample data set, at least two parallel generators in a cable fault diagnosis generative adversarial network model are synchronously trained in any batch of training processes; updating the current model parameter of any generator based on the loss function corresponding to any generator and the model parameters respectively corresponding to the at least two generators; the current model parameters of the discriminator are updated based on the loss function corresponding to the discriminator and the model parameters of the discriminator in the historical batch, and the at least two generators are multi-classification models with the same model structure; determining a candidate cable fault diagnosis model in the trained at least two generators; and determining the candidate cable fault diagnosis model meeting the preset performance index threshold as a target cable fault diagnosis model, so that the target cable fault diagnosis model is fully optimized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a method, equipment, medium and product for training a wire and cable fault diagnosis model. Background Art

[0002] In the power transmission and distribution network of the power system, power cables are the main medium for power transmission. By detecting the type of fault that has occurred in the power cable based on the classification model, the local power grid with the fault can be quickly powered. Among them, the accuracy of the classification model is affected by the training process.

[0003] At present, the existing technology usually inputs sample data into the wire and cable fault diagnosis model to obtain the prediction results corresponding to the sample data, and determines the target loss function based on the prediction results and classification labels corresponding to the sample data, and updates the parameters of the classification model by back-propagating the target loss function.

[0004] However, there is randomness in the training process, such as random initialization and random seeds, which leads to randomness in the performance of the trained wire and cable fault diagnosis model. The model parameters of the trained wire and cable fault diagnosis model are prone to fall into local optimality, and it is difficult to determine the global optimal solution of the model parameters of the wire and cable fault diagnosis model, resulting in the trained wire and cable fault diagnosis model not being fully optimized. Summary of the invention

[0005] The present invention provides a wire and cable fault diagnosis model training method, equipment, medium and product to solve the problem that the generator is easily trapped in the local optimum during training, thereby achieving full training of the generator and improving the performance of the cable fault diagnosis model.

[0006] According to one aspect of the present invention, a method for training a wire and cable fault diagnosis model is provided, comprising:

[0007] Obtain a cable fault sample data set, where the cable fault sample data set includes cable fault sample data and fault classification labels corresponding to the cable fault sample data;

[0008] Constructing a cable fault diagnosis generative adversarial network model, the cable fault diagnosis generative adversarial network model includes a generative network model and a discriminative network model; the generative network model includes at least two generators in parallel, and the discriminative network model includes a discriminator; at least two generators are multi-classification models with the same model structure;

[0009] Based on the training set in the cable fault sample data set, the randomly initialized cable fault diagnosis generative adversarial network model is trained to obtain a trained generative network model; in any batch training process of synchronous training of at least two parallel generators, the current model parameters corresponding to any generator are updated based on the loss function corresponding to any generator and the model parameters corresponding to at least two generators; and the current model parameters corresponding to the discriminator are updated based on the loss function corresponding to the discriminator and the model parameters of the discriminator in the historical batches;

[0010] Based on the validation set in the cable fault sample data set, determine the optimal first performance index corresponding to at least two generators in the trained generative network model, and determine the generator corresponding to the optimal first performance index as a candidate cable fault diagnosis model;

[0011] Based on the test set in the cable fault sample data set, the performance index of the candidate cable fault diagnosis model is determined, and when the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold, the candidate cable fault diagnosis model is determined as the target cable fault diagnosis model.

[0012] According to another aspect of the present invention, there is provided a method for diagnosing a fault in a wire or cable, comprising:

[0013] Obtain equipment operating parameters of target wires and cables;

[0014] The equipment operating parameters are input into the target cable fault diagnosis model to obtain the fault type information of the target wire and cable; wherein the target cable fault diagnosis model is obtained based on the wire and cable fault diagnosis model training method provided in any embodiment of the present invention.

[0015] According to another aspect of the present invention, there is provided a wire and cable fault diagnosis model training device, comprising:

[0016] A cable fault sample data set acquisition module is used to acquire a cable fault sample data set, where the cable fault sample data set includes cable fault sample data and fault classification labels corresponding to the cable fault sample data;

[0017] A model building module is used to build a cable fault diagnosis generative adversarial network model, the cable fault diagnosis generative adversarial network model includes a generative network model and a discriminative network model; the generative network model includes at least two generators in parallel, and the discriminative network model includes a discriminator; at least two generators are multi-classification models with the same model structure;

[0018] A model training module is used to train a randomly initialized cable fault diagnosis generative adversarial network model based on a training set in a cable fault sample data set to obtain a trained generative network model; in any batch training process of synchronous training of at least two parallel generators, update the current model parameters corresponding to any generator based on the loss function corresponding to any generator and the model parameters corresponding to at least two generators; and update the current model parameters corresponding to the discriminator based on the loss function corresponding to the discriminator and the model parameters of the discriminator in historical batches;

[0019] A candidate cable fault diagnosis model determination module is used to determine the optimal first performance index corresponding to at least two generators in the trained generative network model based on the verification set in the cable fault sample data set, and determine the generator corresponding to the optimal first performance index as the candidate cable fault diagnosis model;

[0020] The target cable fault diagnosis model determination module is used to determine the performance index of the candidate cable fault diagnosis model based on the test set in the cable fault sample data set, and determine the candidate cable fault diagnosis model as the target cable fault diagnosis model when the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold.

[0021] According to another aspect of the present invention, there is provided a wire and cable fault diagnosis device, comprising:

[0022] An equipment working parameter acquisition module is used to obtain equipment working parameters of target wires and cables;

[0023] The fault type information determination module is used to input the equipment operating parameters into the target cable fault diagnosis model to obtain the fault type information of the target wire and cable; wherein the target cable fault diagnosis model is obtained based on the wire and cable fault diagnosis model training method provided in any embodiment of the present invention.

[0024] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0025] at least one processor; and

[0026] a memory communicatively connected to at least one processor; wherein,

[0027] The memory stores a computer program that can be executed by at least one processor, and the computer program is executed by at least one processor so that the at least one processor can execute the wire and cable fault diagnosis model training method of any embodiment of the present invention.

[0028] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions, and the computer instructions are used to implement the wire and cable fault diagnosis model training method of any embodiment of the present invention when executed by a processor.

[0029] According to another aspect of the present invention, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, it implements the wire and cable fault diagnosis model training method according to any embodiment of the present invention.

[0030] The technical solution of the embodiment of the present invention solves the problem of easily falling into local optimal solution during generator training by updating the current model parameters corresponding to any generator based on the loss function corresponding to any generator and the model parameters corresponding to at least two generators respectively during any batch training process of synchronous training of at least two parallel generators; and updating the current model parameters corresponding to the discriminator based on the loss function corresponding to the discriminator and the model parameters of the discriminator in historical batches, instead of updating the current model parameters corresponding to any generator only based on the loss function corresponding to any generator. By synchronously training at least two generators and learning the relationship between the model parameters corresponding to at least two generators respectively, falling into local optimal solution is avoided, so that the model parameters corresponding to at least two generators respectively are closer to the global optimal solution, the generator is fully trained, and the performance of the target cable fault diagnosis model is improved.

[0031] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 It is a flow chart of a wire and cable fault diagnosis model training method provided by the first embodiment of the present invention;

[0034] Figure 2 It is a flow chart of a wire and cable fault diagnosis model training method provided by Embodiment 2 of the present invention;

[0035] Figure 3 It is a flow chart of a wire and cable fault diagnosis method provided by an embodiment of the present invention;

[0036] Figure 4 It is a structural schematic diagram of a wire and cable fault diagnosis model training device provided by Embodiment 4 of the present invention;

[0037] Figure 5 It is a structural schematic diagram of a wire and cable fault diagnosis device provided by Embodiment 5 of the present invention;

[0038] Figure 6 It is a structural schematic diagram of an electronic device for implementing the wire and cable fault diagnosis model training method or the wire and cable fault diagnosis method of an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0040] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] Embodiment 1

[0042] Figure 1 This is a flow chart of a wire and cable fault diagnosis model training method provided by the first embodiment of the present invention. This embodiment can be applied to wire and cable fault detection based on a deep learning model. The method can be executed by a wire and cable fault diagnosis model training device. The wire and cable fault diagnosis model training device can be implemented in the form of hardware and / or software. The wire and cable fault diagnosis model training device can be configured in electronic devices such as computers and servers. Figure 1 As shown, the method includes:

[0043] S110: Obtain a cable fault sample data set, where the cable fault sample data set includes cable fault sample data and fault classification labels corresponding to the cable fault sample data.

[0044] In this embodiment, the cable fault sample data set is a data set obtained based on wires and cables, and the cable fault sample data set includes multiple sample data items, each of which includes a set of corresponding cable fault sample data and a fault classification label. The cable fault sample data includes multiple characteristic parameter items, wherein the characteristic parameter items are physical parameter items that characterize the working state of the wires and cables, for example, the characteristic parameter items include but are not limited to current, voltage, inductance and capacitance. The fault classification label is used to characterize the fault type to which the cable fault sample data belongs, and the fault classification label includes the probability (0 or 100%) that the cable fault sample data belongs to multiple fault type information respectively, and there is only one 100% among the probabilities that the fault sample data included in the fault classification label belongs to multiple fault type information respectively.

[0045] Specifically, the cable fault sample data set is obtained by calling the data set of sample data items stored locally and / or in the server. In some embodiments, the original sample data set is preprocessed, wherein the preprocessing includes normalization processing.

[0046] For example, the cable fault sample dataset is represented as in, Characterizes the i-th sample data item in the cable fault sample data set, and They are Included cable fault sample data and fault classification labels, among which, Among them, t in Characterization The cable fault sample data includes N characteristic parameter items, n∈[1,N], Among them, there are V fault types in a single-line cable, b iv Characterization The probability of belonging to the vth fault type, v∈[1,V],b iv is 0 or 1 (i.e., 100%), In the case of the vth fault type,

[0047] S120. Construct a cable fault diagnosis generative adversarial network model. The cable fault diagnosis generative adversarial network model includes a generative network model and a discriminative network model. The generative network model includes at least two parallel generators, and the discriminative network model includes a discriminator. At least two generators are multi-classification models with the same model structure.

[0048] In this embodiment, the cable fault diagnosis generative adversarial network model is a network model used to perform generative adversarial training on the generator. The generative network model is used to generate pseudo fault classification labels corresponding to the cable fault sample data, and the discriminative network model is used to discriminate the authenticity of pseudo fault classification labels and fault classification labels.

[0049] Specifically, two or more generators are constructed based on a preset generator architecture, and a discriminator is constructed based on a preset discriminator model. By setting two or more generators in parallel, different generators have the same input and parallel output, a generation network model is obtained, and the discriminator is set as a discriminant network model. Among them, the preset generator architecture is the architecture of a multi-classification model. For example, the preset generator architecture includes one or more combinations of a convolutional neural network module (for example, a BP neural network module), a machine learning module (for example, a support vector machine module), and a fully connected layer. The preset discriminator architecture is the architecture of a binary classification model. For example, the preset discriminator architecture includes one or more combinations of a convolutional neural network module and a fully connected layer.

[0050] Exemplarily, assuming that the generated network model includes N generators, each generator includes P model parameters, then the generated network model includes N×P model parameters. After a cable fault sample data is input into the generated network model, the generated network model can output N pseudo fault classification labels.

[0051] S130. Based on the training set in the cable fault sample data set, the randomly initialized cable fault diagnosis generative adversarial network model is trained to obtain a trained generative network model; in any batch training process of synchronous training of at least two parallel generators, the current model parameters corresponding to any generator are updated based on the loss function corresponding to any generator and the model parameters corresponding to at least two generators; and the current model parameters corresponding to the discriminator are updated based on the loss function corresponding to the discriminator and the model parameters of the discriminator in historical batches.

[0052] In this embodiment, the training set is a data set consisting of multiple sample data items used for training in the cable fault sample data set. The current model parameters corresponding to any generator are the model parameters corresponding to the generator in the current training batch. The current model parameters corresponding to the discriminator are the model parameters corresponding to the discriminator in the current training batch.

[0053] Specifically, the cable fault diagnosis generative adversarial network model is initialized based on a random initialization algorithm to obtain two or more generators and discriminators after random initialization. It should be noted that random initialization leads to the randomness of the model parameters of the generator initialization, and the model parameters of different generators are at least partially different.

[0054] The training process of the cable fault diagnosis generative adversarial network model includes multiple batches of training processes, and each batch of training processes includes a discriminator training stage and a generator training stage that are executed sequentially.

[0055] It should be noted that the previous training batch of the first training batch is the zeroth training batch, and the model parameters in the zeroth training batch are the initialized model parameters. The current model parameters corresponding to the discriminator (or generator) are the model parameters of the discriminator (or generator) after being updated in the previous training batch.

[0056] In the discriminator training phase, the loss function corresponding to the discriminator is determined based on the training set and at least two generators in the untrainable state. Among them, the model parameters of at least two generators in the untrainable state are initialized model parameters or model parameters updated in the previous training batch. Optionally, the process of determining the loss function corresponding to the discriminator includes: inputting the cable fault sample data in the training set into the at least two generators in the untrainable state to obtain the fault prediction results corresponding to the at least two generators respectively; constructing a fault prediction result data set based on the fault prediction results corresponding to the at least two generators respectively, and constructing a fault classification label data set based on the fault classification labels in the training set; wherein the fault prediction result data set includes a fault prediction result and a first authenticity label; the fault classification label data set includes a fault classification label and a second authenticity label; inputting the fault prediction result in the fault prediction result data set into the discriminator to obtain the first authenticity prediction result corresponding to the fault prediction result; inputting the fault classification label in the fault classification label data set into the discriminant network model to obtain the second authenticity prediction result corresponding to the fault classification label; determining a first discriminant loss function based on the first authenticity prediction result and the first authenticity label; and determining a second discriminant loss function based on the second authenticity prediction result and the second authenticity label; determining the loss function corresponding to the discriminator based on the first discriminant loss function and the second discriminant loss function.

[0057] The first authenticity label is a label used to represent unauthenticity; the second authenticity label is a label used to represent authenticity.

[0058] Specifically, based on at least two generators synchronously performing multi-classification processing on the input cable fault sample data, fault prediction results output by at least two generators are obtained. It can be understood that the fault prediction result is a pseudo-fault classification label corresponding to the cable fault sample data, and the fault prediction result includes the probabilities (0 to 100%) of multiple fault type information predicted by the generator in an untrainable state.

[0059] Exemplarily, assuming that the generation network model includes N generators, the cable fault sample data Input the nth generator in the untrainable state, and the fault prediction result corresponding to the nth generator is expressed as:

[0060]

[0061] Among them, Biagnostic n,i,v Represents the prediction of the nth generator The probability of belonging to the vth fault type.

[0062] By combining the fault prediction result corresponding to each generator obtained based on each cable fault sample data in the training set and the first authenticity label into a pseudo-label data item corresponding to a generator, and by collecting multiple pseudo-label data items corresponding to multiple generators obtained based on multiple cable fault sample data in the training set, a fault prediction result data set is formed. By combining the fault classification label corresponding to each cable fault sample data and the second authenticity label into a true label data item, a fault classification label data set is formed by collecting multiple true label data items. Among them, the first authenticity label represents the authenticity as unauthentic; the second authenticity label represents the authenticity as authentic.

[0063] For example, assuming that the first authenticity label is a logical identifier 0; the second authenticity label is a logical identifier 1; and the training set includes H sample data items, the training set is represented as The fault prediction result dataset includes H×N pseudo-label data items. The fault prediction result dataset is expressed as The fault classification label dataset includes H true label data items. The fault classification label dataset is expressed as

[0064] Based on the discriminator, the fault prediction results in the fault prediction result data set are predicted for authenticity, and a first authenticity prediction result corresponding to the fault prediction result is obtained. Based on the discriminator, the fault classification label in the fault classification label data set is predicted for authenticity, and a second authenticity prediction result corresponding to the fault classification label is obtained. The first discriminant loss function is obtained by performing a distance calculation between the first authenticity prediction result corresponding to the fault prediction result and the first authenticity label, and the second discriminant loss function is obtained by performing a distance calculation between the second authenticity prediction result corresponding to the fault classification label and the second authenticity label. The loss function corresponding to the discriminant is obtained by summing the first discriminant loss function and the second discriminant loss function. Exemplarily, it is assumed that the first authenticity prediction result and the second authenticity prediction result can be logical identifiers 0 or 1.

[0065] The first discriminant loss function is expressed as:

[0066] The second discriminant loss function is expressed as:

[0067] The loss function corresponding to the discriminator is expressed as loss D =(loss D1 +loss D2 ) / 2.

[0068] It should be noted that the model parameters of the discriminator in the historical batch include the model parameters of the discriminator corresponding to at least one training batch before the current training batch.

[0069] In some embodiments, the zeroth training batch to the previous training batch are all determined as historical batches. Optionally, based on the loss function corresponding to the discriminator, the current model parameters corresponding to the discriminator are updated to obtain third current updated model parameters; when the current training batch is the first training batch, the third current updated model parameters are determined as the model parameters corresponding to the discriminator in the second training batch; when the current training batch is not the first training batch, the fourth performance indicators corresponding to the model parameters corresponding to the discriminator are determined, and the best fourth performance indicator and the worst fourth performance indicator are determined among the multiple fourth performance indicators; the model parameters corresponding to the best fourth performance indicator are determined as the best model parameters corresponding to the discriminator, and the model parameters corresponding to the worst fourth performance indicator are determined as the worst model parameters corresponding to the discriminator; based on the third current updated model parameters and the best model parameters corresponding to the discriminator, the third current space is determined. distance; and based on the third currently updated model parameter and the worst model parameter corresponding to the discriminator, determine the fourth current spatial distance; determine the ratio of the third current spatial distance to the fourth current spatial distance as the second target weight; determine the fourth data item based on the update speed corresponding to the discriminator in the previous training batch and the fourth hyperparameter; determine the difference between the best model parameters corresponding to the discriminator and the third currently updated model parameter as the third factor; determine the fifth data item based on the third factor, the fifth hyperparameter and the third random number; determine the current update speed corresponding to the discriminator based on the fourth data item, the fifth data item and the second target weight; determine the sum of the current update speed corresponding to the discriminator and the third currently updated model parameter as the fourth currently updated model parameter; determine the fourth currently updated model parameter as the model parameter corresponding to the discriminator in the next training batch.

[0070] Among them, the model parameters corresponding to the discriminator include the current model parameters corresponding to the discriminator and the model parameters of the discriminator in the historical batch.

[0071] Specifically, the third current updated model parameters are obtained by inputting the loss function corresponding to the discriminator and the current model parameters corresponding to the discriminator into the traditional model parameter update formula. Exemplarily, when the current training batch is the kth training batch, the current model parameters of the discriminator are expressed as:

[0072]

[0073] Among them, the lth current model parameter in the discriminator is W D,k,l , then the current model parameters after the discriminator is updated are expressed as in, η1 is the discriminator learning rate.

[0074] If the current training batch is the first training batch, the first training batch is the current training batch, and the historical batch only includes the zeroth training batch, then the current model parameters of the discriminator are updated to the third current updated model parameters, and the updated current updated model parameters are set to the model parameters corresponding to the second training batch. Exemplarily, in the case of the second training batch, the current model parameters of the discriminator are the third current updated model parameters of the first training batch, that is,

[0075] If the current training batch is not the first training batch, the first training batch is not the current training batch, and the historical batches include one or more training batches except the zeroth training batch, then the performance index of the discriminator in each historical batch is calculated based on the training set to obtain the fourth performance index corresponding to the model parameters of the discriminator in the historical batch, and the performance index of the discriminator in the current training batch is calculated based on the training set to obtain the fourth performance index corresponding to the current model parameters of the discriminator in the current training batch.

[0076] The fourth performance indicator that best represents the performance of the discriminator among multiple fourth performance indicators is determined as the optimal fourth performance indicator, the fourth performance indicator that worst represents the performance of the discriminator among multiple fourth performance indicators is determined as the worst fourth performance indicator, the model parameters corresponding to the discriminator and the optimal fourth performance indicator are determined as the best model parameters corresponding to the discriminator, and the model parameters corresponding to the worst fourth performance indicator of the discriminator in the current training batch and the historical batches are determined as the worst model parameters corresponding to the discriminator. Exemplarily, assuming that the current training batch is the kth training batch, the best model parameters corresponding to the current training batch of the discriminator are expressed as The worst model parameters of the discriminator in the kth training batch are expressed as k1 represents the training batches of the discriminator with the best model parameters as the model parameters, and k2 represents the training batches of the discriminator with the worst model parameters as the model parameters, where k1∈[0,k], k2∈[0,k].

[0077] The third current spatial distance is obtained by performing distance calculation on the third current updated model parameter and the previous optimal model parameter based on the preset distance calculation formula, and the fourth current spatial distance is obtained by performing distance calculation on the third current updated model parameter and the previous worst model parameter based on the preset distance calculation formula, and the third current spatial distance is divided by the fourth current spatial distance to obtain the second target weight. Exemplarily, the second target weight of the kth training batch is represented by D 3,k / D 4,k , where the third current spatial distance is represented by D 3,k , the fourth current spatial distance is expressed as D 3,k , the preset distance calculation formula can be a Euclidean distance calculation formula.

[0078] The update speed is the speed at which the model parameters are updated. After the model parameters of the discriminator are updated in each training batch, the update speed of the discriminator in the training batch is stored. In the current training batch, the update speed corresponding to the discriminator in the previous training batch is read. It should be noted that when the current training batch is the second training batch, the update speed corresponding to the previous training batch is zero. The fourth data item is obtained by multiplying the update speed corresponding to the discriminator in the previous training batch by the fourth hyperparameter. Exemplarily, the fourth data item in the kth training batch is represented as is the update speed of the discriminator corresponding to the k-1th training batch, and μ4 is the fourth hyperparameter.

[0079] The third factor is obtained by performing a difference process on the optimal model parameters corresponding to the discriminator and the third currently updated model parameters. The fifth data item is obtained by performing a product process on the third factor, the fifth hyperparameter and the third random number. Exemplarily, the fifth data item in the kth training batch is represented as:

[0080] in, is the third factor, μ5 is the fifth hyperparameter, and γ3 is the third random number.

[0081] The fourth data item and the fifth data item are summed to obtain the sum of the fourth data item and the fifth data item, and the sum of the fourth data item and the fifth data item is multiplied by the second target weight to obtain the current update speed corresponding to the discriminator. The current update speed is the update speed of the current training batch. Exemplarily, the current update speed corresponding to the discriminator in the kth training batch is expressed as:

[0082] Understandably,

[0083] The fourth current updated model parameters corresponding to the discriminator are obtained by summing the current update speed corresponding to the discriminator and the third current updated model parameters, and the current model parameters corresponding to the discriminator are updated based on the fourth current updated model parameters to obtain the updated current model parameters in the current training batch, and the updated current model parameters in the current training batch are used as the model parameters corresponding to the discriminator in the next training batch. Exemplarily, the fourth current updated model parameters corresponding to the discriminator in the kth training batch are expressed as It is understandable that the model parameters corresponding to the discriminator in the k+1th training batch are Among them, k>1.

[0084] The technical solution of this embodiment can improve the training speed of the discriminator by determining the current update speed corresponding to the discriminator based on the fourth data item, the fifth data item and the second target weight when the current training batch is not the first training batch, and determining the model parameters corresponding to the discriminator in the next training batch based on the current update speed corresponding to the discriminator and the sum of the third currently updated model parameters.

[0085] In the generator training phase, based on the training set and the discriminator in the untrainable state, the loss functions corresponding to at least two generators in the current training batch are determined. Optionally, the process of determining the loss function corresponding to any generator includes: inputting the fault prediction result corresponding to any generator into the discriminator in the untrainable state to obtain the third authenticity prediction result corresponding to the fault prediction result; determining the first generation loss function corresponding to any generator based on the third authenticity prediction result and the second authenticity label corresponding to the fault prediction result; determining the second generation loss function corresponding to any generator based on the fault classification label in the training set and the fault prediction result; determining the loss function corresponding to any generator based on the first generation loss function and the second generation loss function.

[0086] Specifically, for any generator, based on a discriminator in an untrainable state, the fault prediction result output by the generator is subjected to authenticity discrimination to obtain a third authenticity prediction result corresponding to the fault prediction result. Exemplarily, the third authenticity prediction result can be a logical identifier 0 or 1. The first generation loss function is obtained by performing a distance calculation between the third authenticity prediction result and the second authenticity label. The first generation loss function is used to characterize the degree of difference between the fault prediction result generated by the generator and the true fault classification label in terms of data distribution. The first generation loss function is obtained by performing a distance calculation between the third authenticity prediction result and the second authenticity label. The second generation loss function is obtained by performing a loss calculation on the fault classification label and the fault prediction result corresponding to the cable fault sample data. The second generation loss function is used to characterize the degree of difference between the fault prediction result generated by the generator and the true fault classification label. The loss function corresponding to the generator is obtained by summing the first generation loss function and the second generation loss function.

[0087] Exemplarily, the first generation loss function of the nth generator is expressed as:

[0088] The second generation loss function of the nth generator is expressed as

[0089] The loss function corresponding to the nth generator is expressed as loss G,n =(loss G1,n +loss G2,n ) / 2.

[0090] The technical solution of this embodiment obtains the loss function corresponding to the discriminator by summing the first discriminant loss function and the second discriminant loss function, and obtains the loss function corresponding to the generator by summing the first generation loss function and the second generation loss function, so that the loss function corresponding to the discriminator can accurately characterize the degree of difference between the authenticity prediction result predicted by the discriminator and the authenticity label, and the loss function corresponding to the generator can accurately characterize the degree of difference between the fault prediction result generated by the generator and the fault classification label, which helps to improve the reliability of generator training and discriminator training.

[0091] In any training batch, the deviation data item of any generator is used to characterize the deviation between the model parameters corresponding to the generator in the training batch and the model parameters corresponding to the next training batch of the training batch. The deviation data item is determined based on the loss function corresponding to the generator and the model parameters corresponding to at least two generators. Among them, the model parameters corresponding to each generator include but are not limited to the model parameters corresponding to the generator in the current training batch. For example, the model parameters corresponding to the nth generator include the model parameters corresponding to the nth generator in the zeroth training batch to the current training batch. By summing the current model parameters corresponding to the generator and the deviation data item, the updated current model parameters corresponding to the generator are obtained.

[0092] Exemplarily, when the current training batch is the kth training batch, the current model parameters of the nth generator are expressed as: Among them, the pth current model parameter in the generator is W G,n,k,p , then the updated current model parameters corresponding to the nth generator are expressed as:

[0093]

[0094] Among them, W′ G,n,k,p =W G,n,k,p -θ G,n,k,p ;

[0095] η2 is the generator learning rate, θ G,n,k,p is the deviation data item of the nth generator in the k+1th training batch. In the case of the k+1th training batch, the current model parameters of the generator are the current model parameters updated after the kth training batch, that is,

[0096] When the training end condition is not met, the iterative parameters of the fault diagnosis generative adversarial network model are adjusted so that the values ​​of the loss functions corresponding to the two or more generators and the value of the loss function corresponding to the discriminator gradually decrease until the training end condition is met, and the training of the fault diagnosis generative adversarial network model is stopped to obtain a trained generative network model. For example, the training end condition may be the convergence of the loss function value and / or the average performance index and the reaching of a fixed number of training rounds, etc., which is not limited in this embodiment.

[0097] S140. Based on the validation set in the cable fault sample data set, determine the optimal first performance index corresponding to at least two generators in the trained generative network model, and determine the generator corresponding to the optimal first performance index as a candidate cable fault diagnosis model.

[0098] In this embodiment, the validation set is a data set consisting of multiple sample data items in the cable fault sample data set for selecting a candidate cable fault diagnosis model. The candidate cable fault diagnosis model is the generator with the best performance in the trained generative network model.

[0099] Specifically, based on the validation set, the performance index of at least two generators in the trained generative network model is calculated to obtain the first performance indexes corresponding to the at least two generators. The best first performance index is determined among the multiple first performance indexes. The generator corresponding to the best first performance index among the at least two generators is determined as the candidate cable fault diagnosis model. Exemplarily, the first performance index and the fourth performance index may be the same or different, and this embodiment does not limit this.

[0100] S150. Determine the performance index of the candidate cable fault diagnosis model based on the test set in the cable fault sample data set, and determine the candidate cable fault diagnosis model as the target cable fault diagnosis model when the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold.

[0101] In this embodiment, the test set is a data set composed of multiple sample data items in the cable fault sample data set for testing the performance of the candidate cable fault diagnosis model. The preset performance index threshold is a preset threshold of the performance index of the candidate cable fault diagnosis model, which is used to determine that the candidate cable fault diagnosis model is a target cable fault diagnosis model. The target cable fault diagnosis model is a cable fault diagnosis model that can be used to automatically detect faults on wires and cables and obtain reliable detection results.

[0102] Specifically, based on the test set, the performance index of the candidate cable fault diagnosis model is calculated to obtain the performance index of the candidate cable fault diagnosis model. In the case where the larger (or smaller) the performance index of the candidate cable fault diagnosis model is, the better the performance of the candidate cable fault diagnosis model is, if the performance index of the candidate cable fault diagnosis model is greater than (or less than) the preset performance index threshold, it can be considered that the performance of the candidate cable fault diagnosis model meets the actual needs, then it is determined that the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold, it can be considered that the performance of the candidate cable fault diagnosis model meets the actual needs, and the candidate cable fault diagnosis model is determined as the target cable fault diagnosis model. If the performance index of the candidate cable fault diagnosis model does not meet the preset performance index threshold, the cable fault diagnosis generative adversarial network model is re-initialized randomly, and the cable fault diagnosis generative adversarial network model after the re-initialization is trained until the target cable fault diagnosis model is obtained.

[0103] Exemplarily, assuming that the generated network model includes N generators, the candidate cable fault diagnosis model is the nth generator, and when the performance index of the nth generator meets the preset performance index threshold, the nth generator is determined as the target cable fault diagnosis model.

[0104] The technical solution of this embodiment solves the problem of easily falling into local optimum during the generator training process by updating the current model parameters corresponding to any generator based on the loss function corresponding to any generator and the model parameters corresponding to at least two generators respectively during any batch training process of synchronous training of at least two parallel generators; and updating the current model parameters corresponding to the discriminator based on the loss function corresponding to the discriminator and the model parameters of the discriminator in historical batches, instead of updating the current model parameters corresponding to any generator only based on the loss function corresponding to any generator. This avoids falling into local optimum during the generator training process, avoids falling into local optimum solution by synchronously training at least two generators and learning the relationship between the model parameters corresponding to at least two generators respectively, and makes the model parameters corresponding to at least two generators closer to the global optimum solution, thereby achieving sufficient training of the generator and improving the performance of the target cable fault diagnosis model.

[0105] Embodiment 2

[0106] Figure 2 This is a flow chart of a wire and cable fault diagnosis model training method provided by Embodiment 2 of the present invention. The technical solution of the embodiment of the present invention is improved on the basis of the above embodiment. The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.

[0107] In this embodiment, the model parameters corresponding to the at least two generators respectively include the current model parameters corresponding to the at least two generators respectively.

[0108] like Figure 2 As shown, the method includes:

[0109] S210: Acquire a cable fault sample data set, where the cable fault sample data set includes cable fault sample data and fault classification labels corresponding to the cable fault sample data.

[0110] S220. Construct a cable fault diagnosis generative adversarial network model. The cable fault diagnosis generative adversarial network model includes a generative network model and a discriminative network model. The generative network model includes at least two parallel generators, and the discriminative network model includes a discriminator. At least two generators are multi-classification models with the same model structure.

[0111] S230. Based on the training set in the cable fault sample data set, the randomly initialized cable fault diagnosis generative adversarial network model is trained to obtain a trained generative network model; in any batch training process of synchronous training of at least two parallel generators, for any generator, the current model parameters of any generator are updated based on the loss function corresponding to any generator to obtain the first current updated model parameters corresponding to any generator; when the current training batch is the first training batch, the first current updated model parameters corresponding to any generator are determined as the model parameters corresponding to any generator in the second training batch; when the current training batch is not the first training batch, the second performance indicators corresponding to at least two generators are determined based on the training set, the best second performance indicator and the worst second performance indicator are determined among multiple second performance indicators, and the best second performance indicator is used as the model parameter corresponding to any generator in the second training batch. The current model parameters of the generator corresponding to the performance index are determined as the current optimal model parameters, and the current model parameters of the generator corresponding to the worst second performance index are determined as the current worst model parameters; based on the current model parameters, current associated parameters and the first currently updated model parameters corresponding to any generator, the current update speed corresponding to any generator is determined; the current associated parameters include the current optimal model parameters, the current worst model parameters and the update speed corresponding to any generator in the previous training batch; the second currently updated model parameters corresponding to any generator are determined based on the sum of the current update speed corresponding to any generator and the first currently updated model parameters; the second currently updated model parameters corresponding to any generator are determined as the model parameters corresponding to any generator in the next training batch; and the current model parameters of the discriminator are updated based on the loss function corresponding to the discriminator and the model parameters of the discriminator in the historical batches.

[0112] Specifically, for any generator, the first current updated model parameters are obtained by inputting the loss function corresponding to the generator and the current model parameters corresponding to the generator into the traditional model parameter updating formula.

[0113] Exemplarily, the first current updated model parameter corresponding to the nth generator in the kth training batch is expressed as in, p∈[1,P].

[0114] If the current training batch is the first training batch, the first training batch is the current training batch, and the historical batches only include the zeroth training batch, then the current model parameters corresponding to the generator are updated to the first current updated model parameters, and the updated current updated model parameters are set to the model parameters corresponding to the second training batch.

[0115] If the current training batch is not the first training batch, the first training batch is not the current training batch, and the historical batches include one or more training batches except the zeroth training batch, then performance indicators of at least two generators of the current training batch are calculated based on the training set to obtain second performance indicators corresponding to current model parameters of at least two generators of the current training batch.

[0116] By determining the second performance indicator that best represents the discriminator performance among multiple second performance indicators as the optimal second performance indicator, determining the second performance indicator that worst represents the discriminator performance among multiple second performance indicators as the worst second performance indicator, determining the model parameters corresponding to the best second performance indicator among the current model parameters of at least two generators in the current training batch as the current optimal model parameters, and determining the model parameters corresponding to the worst second performance indicator among the current model parameters of at least two generators in the current training batch as the current worst model parameters. Exemplarily, assuming that the current training batch is the kth training batch, the current optimal model parameters are expressed as The current worst model parameters are expressed as The n1th generator is the generator with the current optimal model parameters as the current model parameters among the N generators in the current training batch, and the n2th generator is the generator with the current worst model parameters as the current model parameters among the N generators in the current training batch, n1∈[1,N], n2∈[1,N].

[0117] After the model parameters of the generator are updated in each training batch, the update speed corresponding to the generator in the training batch is stored. In the current training batch, the update speed corresponding to the generator in the previous training batch is read. It should be noted that when the current training batch is the second training batch, the update speed corresponding to the previous training batch is zero. By determining the current optimal model parameters, the current worst model parameters and the update speed corresponding to any generator in the previous training batch as the current associated parameters of the generator, the current model parameters, current associated parameters and the first current updated model parameters corresponding to the generator are calculated and processed to obtain the current update speed corresponding to the generator. Exemplarily, assuming that the current training batch is the kth training batch, the current update speed corresponding to the nth generator is expressed as The update speed of the nth generator in the previous training batch is expressed as Understandably,

[0118] Optionally, based on the current model parameters and current optimal model parameters corresponding to any generator, a first current spatial distance is determined; and based on the current model parameters and current worst model parameters corresponding to any generator, a second current spatial distance is determined; the ratio of the first current spatial distance to the second current spatial distance is determined as the first target weight; the first data item is determined based on the update speed corresponding to any generator in the previous training batch and the first hyperparameter; the difference between the current optimal model parameters and the first current updated model parameters corresponding to any generator is determined as the first factor; the second data item is determined based on the first factor, the second hyperparameter and the first random number; the current update speed corresponding to any generator is determined based on the first data item, the second data item and the first target weight.

[0119] Specifically, the first current spatial distance is obtained by performing distance calculation between the current model parameters corresponding to the generator and the current optimal model parameters based on a preset distance calculation formula, and the second current spatial distance is obtained by performing distance calculation between the current model parameters corresponding to the generator and the current worst model parameters based on a preset distance calculation formula, and the first current spatial distance is divided by the second current spatial distance to obtain the first target weight. Exemplarily, the second target weight of the nth generator in the kth training batch is expressed as D 1,n,k / D 2,n,k , where the first current spatial distance is represented by D 1,n,k , the second current spatial distance is expressed as D 2,n,k .

[0120] The first data item is obtained by multiplying the update speed corresponding to the generator in the previous training batch and the first hyperparameter. For example, the first data item of the nth generator in the kth training batch is represented as μ1 is the first hyperparameter.

[0121] The first factor is obtained by performing a difference process between the current optimal model parameter and the first current updated model parameter corresponding to the generator. The second data item is obtained by performing a product process of the first factor, the second hyperparameter and the first random number. Exemplarily, the second data item of the nth generator in the kth training batch is represented as in, is the first factor, μ2 is the second hyperparameter, and γ1 is the first random number.

[0122] The first data item and the second data item are summed to obtain the sum of the first data item and the second data item, and the sum of the first data item and the second data item is multiplied by the first target weight to obtain the current update speed corresponding to the generator.

[0123] For example, the current update speed of the nth generator corresponding to the kth training batch is expressed as:

[0124] The technical solution of this embodiment is as follows: when the current training batch is not the first training batch, the first current spatial distance is determined based on the current model parameters and the current optimal model parameters corresponding to any generator; and the second current spatial distance is determined based on the current model parameters and the current worst model parameters corresponding to any generator; the ratio of the first current spatial distance to the second current spatial distance is determined as the first target weight; the first data item is determined based on the update speed corresponding to any generator in the previous training batch and the first hyperparameter; the difference between the current optimal model parameter and the first current updated model parameter corresponding to any generator is determined as the first factor; the second data item is determined based on the first factor, the second hyperparameter and the first random number; the current update speed corresponding to any generator is determined based on the first data item, the second data item and the first target weight, so that the accurate current update speed corresponding to any generator can be obtained based on multiple generators trained synchronously, which helps to dynamically adjust the current update speed of the generator so that the generator is fully trained.

[0125] In some embodiments, optionally, the model parameters corresponding to at least two generators respectively also include the model parameters of at least two generators in historical batches respectively; the current associated parameters also include the optimal model parameters of any generator in historical batches; accordingly, based on the current model parameters, current associated parameters and the first currently updated model parameters corresponding to any generator, the current update speed corresponding to any generator is determined, and it also includes: determining the third performance indicators corresponding to the model parameters corresponding to any generator based on the training set, and determining the optimal third performance indicator among multiple third performance indicators; determining the model parameters corresponding to any generator and the optimal third performance indicator as the optimal model parameters corresponding to any generator; determining the difference between the optimal model parameters corresponding to any generator and the first currently updated model parameters as the second factor; determining the third data item based on the second factor, the third hyperparameter and the second random number; accordingly, determining the current update speed corresponding to any generator based on the first data item, the second data item and the first target weight, including: determining the current update speed corresponding to any generator based on the first data item, the second data item, the third data item and the first target weight.

[0126] It should be noted that the model parameters of any generator in the historical batch include the model parameters corresponding to at least one training batch before the current training batch of the generator.

[0127] Specifically, for any generator, the performance index of the generator in each historical batch is calculated based on the training set to obtain the third performance index corresponding to the model parameters of the generator in the historical batch, and the performance index of the generator in the current training batch is calculated based on the training set to obtain the third performance index corresponding to the current model parameters of the generator in the current training batch. The third performance index that best characterizes the performance of the generator among the multiple third performance indicators of the generator is determined as the optimal third performance index, and the model parameters of the generator corresponding to the optimal third performance index are determined as the optimal model parameters corresponding to the generator. Exemplarily, assuming that the current training batch is the kth training batch, the optimal model parameters of the nth generator in the current training batch are expressed as k3 represents the training batch of the nth generator with the best model parameters as model parameters, k3∈[0,k].

[0128] The second factor is obtained by subtracting the previous optimal model parameters corresponding to the generator from the first currently updated model parameters. The third data item is obtained by multiplying the second factor, the third hyperparameter and the second random number. For example, the third data item of the nth generator in the kth training batch is represented as in, is the second factor, μ3 is the second hyperparameter, and γ2 is the second random number.

[0129] The first data item, the second data item and the third data item are summed to obtain the sum of the first data item, the second data item and the third data item, and the sum of the first data item, the second data item and the third data item is multiplied by the first target weight to obtain the current update speed corresponding to the generator.

[0130] For example, the current update speed of the nth generator corresponding to the kth training batch is expressed as:

[0131] The technical solution of this embodiment determines the optimal model parameters of all time corresponding to any generator based on the training set; determines the difference between the optimal model parameters of all time corresponding to any generator and the first currently updated model parameters as the second factor; determines the third data item based on the second factor, the third hyperparameter and the second random number; determines the current update speed of any generator based on the first data item, the second data item, the third data item and the first target weight, which can further improve the accuracy of the current update speed of any generator based on the model parameters of any generator in the historical batches, help to improve the training speed of the generator, avoid the training of the generator from falling into the local optimal solution, and ensure the sufficient training of the generator.

[0132] The second current updated model parameter corresponding to the generator is obtained by summing the current update speed corresponding to the generator and the first current updated model parameter. The current model parameter corresponding to the generator is updated based on the second current updated model parameter to obtain the updated current model parameter of the generator in the current training batch, and the updated current model parameter of the generator in the current training batch is used as the model parameter corresponding to the generator in the next training batch. Exemplarily, the second current updated model parameter corresponding to the nth generator in the kth training batch is expressed as:

[0133] It is understandable that the model parameters corresponding to the nth generator in the k+1th training batch are Among them, k>1.

[0134] S240. Based on the validation set in the cable fault sample data set, determine the optimal first performance indicator corresponding to at least two generators in the trained generative network model, and determine the generator corresponding to the optimal first performance indicator as a candidate cable fault diagnosis model.

[0135] S250. Based on the test set in the cable fault sample data set, determine the performance index of the candidate cable fault diagnosis model, and when the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold, determine the candidate cable fault diagnosis model as the target cable fault diagnosis model.

[0136] The technical solution of this embodiment is to determine the current update speed corresponding to any generator based on the current model parameters, current associated parameters and first current updated model parameters corresponding to any generator when the current training batch is not the first training batch; the current associated parameters include the current optimal model parameters, the current worst model parameters and the update speed corresponding to any generator in the previous training batch; the second current updated model parameters corresponding to any generator are determined based on the sum of the current update speed corresponding to any generator and the first current updated model parameters; the second current updated model parameters corresponding to any generator are determined as the model parameters corresponding to any generator in the next training batch, which can make the model parameters corresponding to at least two generators closer to the global optimal solution, realize the full training of the generator, and improve the performance of the target cable fault diagnosis model.

[0137] Embodiment 3

[0138] Figure 3 is a flow chart of a wire and cable fault diagnosis method provided by an embodiment of the present invention. The technical solution of the embodiment of the present invention is the application of the target cable fault diagnosis model trained in any of the above embodiments. Figure 3 As shown, the method includes:

[0139] S310, obtaining equipment operating parameters of the target wire and cable.

[0140] In this embodiment, the target wires and cables are wires and cables that need to be fault-detected, and the equipment operating parameters include multiple characteristic parameter items that are the same as the cable fault sample data. The equipment operating parameters are used to characterize the working status of the target wires and cables.

[0141] Specifically, the device operating parameters of the target wire and cable are obtained by calling the pre-stored or real-time collected physical parameter items characterizing the working state of the target wire and cable. Exemplarily, the physical parameter items collected by the parameter collection device on the target wire and cable are received in real time to obtain the device operating parameters of the transformer to be detected, wherein the parameter collection device includes but is not limited to an ammeter, a voltmeter and an intelligent cloud platform.

[0142] S320. Input the equipment operating parameters into the target cable fault diagnosis model to obtain the fault type information of the target wire and cable; wherein the target cable fault diagnosis model is obtained based on the wire and cable fault diagnosis model training method provided in any of the above embodiments.

[0143] Specifically, based on the target cable fault diagnosis model, the device operating parameters are subjected to multi-classification processing to obtain prediction results, wherein the prediction results include the fault type probabilities (0 to 100%) of the device operating parameters belonging to multiple fault type information, and the fault type information is generated based on the fault type corresponding to the maximum fault type probability among the multiple fault type probabilities. The fault type information is information characterizing a fault type to which the device operating parameters belong, and exemplary, the fault type information includes but is not limited to a logical identifier and a text identifier.

[0144] The technical solution of this embodiment obtains the fault type information of the target wire and cable by inputting the equipment operating parameters into the target cable fault diagnosis model, and can obtain accurate fault type information.

[0145] Embodiment 4

[0146] Figure 4 Schematic diagram of the structure of a wire and cable fault diagnosis model training device provided by the fourth embodiment of the present invention. Figure 4 As shown, the device comprises:

[0147] The cable fault sample data set acquisition module 410 is used to acquire a cable fault sample data set, where the cable fault sample data set includes cable fault sample data and fault classification labels corresponding to the cable fault sample data;

[0148] The model building module 420 is used to build a cable fault diagnosis generative adversarial network model, which includes a generative network model and a discriminative network model; the generative network model includes at least two generators in parallel, and the discriminative network model includes a discriminator; at least two generators are multi-classification models with the same model structure;

[0149] The model training module 430 is used to train the randomly initialized cable fault diagnosis generative adversarial network model based on the training set in the cable fault sample data set to obtain a trained generative network model; in any batch training process of synchronous training of at least two parallel generators, update the current model parameters corresponding to any generator based on the loss function corresponding to any generator and the model parameters corresponding to at least two generators respectively; and update the current model parameters corresponding to the discriminator based on the loss function corresponding to the discriminator and the model parameters of the discriminator in the historical batch;

[0150] A candidate cable fault diagnosis model determination module 440 is used to determine the optimal first performance index corresponding to at least two generators in the trained generation network model based on the verification set in the cable fault sample data set, and determine the generator corresponding to the optimal first performance index as a candidate cable fault diagnosis model;

[0151] The target cable fault diagnosis model determination module 450 is used to determine the performance index of the candidate cable fault diagnosis model based on the test set in the cable fault sample data set, and determine the candidate cable fault diagnosis model as the target cable fault diagnosis model when the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold.

[0152] The technical solution of this embodiment solves the problem of easily falling into local optimum during the generator training process by updating the current model parameters of any generator based on the loss function corresponding to any generator and the model parameters corresponding to at least two generators respectively during any batch training process of synchronous training of at least two parallel generators; and updating the current model parameters of the discriminator based on the loss function corresponding to the discriminator and the model parameters of the discriminator in historical batches, instead of updating the current model parameters of any generator only based on the loss function corresponding to any generator. By synchronously training at least two generators and learning the relationship between the model parameters corresponding to at least two generators respectively, it is avoided to fall into the local optimal solution, so that the model parameters corresponding to at least two generators respectively are closer to the global optimal solution, thereby achieving sufficient training of the generator and improving the performance of the target cable fault diagnosis model.

[0153] On the basis of the above embodiment, optionally, the model parameters corresponding to at least two generators respectively include current model parameters corresponding to at least two generators respectively; accordingly, the model training module 430 is specifically used to update the current model parameters of any generator based on the loss function corresponding to any generator, and obtain the first current updated model parameters corresponding to any generator; when the current training batch is the first training batch, the first current updated model parameters corresponding to any generator are determined as the model parameters corresponding to any generator in the second training batch; when the current training batch is not the first training batch, the second performance indicators corresponding to at least two generators are determined based on the training set, and the best second performance indicator and the worst second performance indicator are determined among multiple second performance indicators. The second performance indicator, determines the current model parameters of the generator corresponding to the best second performance indicator as the current best model parameters, and determines the current model parameters of the generator corresponding to the worst second performance indicator as the current worst model parameters; based on the current model parameters, current associated parameters and the first currently updated model parameters corresponding to any generator, determines the current update speed corresponding to any generator; the current associated parameters include the current best model parameters, the current worst model parameters and the update speed corresponding to any generator in the previous training batch; determines the second currently updated model parameters corresponding to any generator based on the sum of the current update speed corresponding to any generator and the first currently updated model parameters; determines the second currently updated model parameters corresponding to any generator as the model parameters corresponding to any generator in the next training batch.

[0154] On the basis of the above embodiments, optionally, the model training module 430 is also used to determine a first current spatial distance based on the current model parameters and the current optimal model parameters corresponding to any generator; and to determine a second current spatial distance based on the current model parameters and the current worst model parameters corresponding to any generator; determine the ratio of the first current spatial distance to the second current spatial distance as the first target weight; determine a first data item based on the update speed corresponding to any generator in the previous training batch and the first hyperparameter; determine the difference between the current optimal model parameter and the first current updated model parameter corresponding to any generator as the first factor; determine the second data item based on the first factor, the second hyperparameter and the first random number; determine the current update speed corresponding to any generator based on the first data item, the second data item and the first target weight.

[0155] On the basis of the above embodiment, optionally, the model parameters corresponding to at least two generators respectively also include the model parameters of at least two generators in historical batches; accordingly, the current associated parameters also include the optimal model parameters of any generator in historical batches; accordingly, the model training module 430 is also used to determine the third performance indicators corresponding to the model parameters corresponding to any generator based on the training set, and determine the optimal third performance indicator among multiple third performance indicators; the model parameters corresponding to any generator and the optimal third performance indicator are determined as the optimal model parameters of any generator; the difference between the optimal model parameters of any generator and the first currently updated model parameters is determined as the second factor; the third data item is determined based on the second factor, the third hyperparameter and the second random number; accordingly, the model training module 430 is also used to determine the current update speed corresponding to any generator based on the first data item, the second data item, the third data item and the first target weight.

[0156] On the basis of the above embodiment, optionally, the model training module 430 is specifically used to update the current model parameters corresponding to the discriminator based on the loss function corresponding to the discriminator to obtain the third current updated model parameters; when the current training batch is the first training batch, the third current updated model parameters are determined as the model parameters corresponding to the discriminator in the second training batch; when the current training batch is not the first training batch, the fourth performance indicators corresponding to the corresponding model parameters of the discriminator are determined, and the best fourth performance indicator and the worst fourth performance indicator are determined among the multiple fourth performance indicators; wherein the model parameters corresponding to the discriminator include the current model parameters corresponding to the discriminator and the model parameters of the discriminator in the historical batches; the model parameters corresponding to the best fourth performance indicator are determined as the best model parameters corresponding to the discriminator, and the model parameters corresponding to the worst fourth performance indicator are determined as the worst model parameters corresponding to the discriminator; based on The third current spatial distance is determined based on the third currently updated model parameters and the optimal model parameters of the discriminator; and the fourth current spatial distance is determined based on the third currently updated model parameters and the worst model parameters of the discriminator; the ratio of the third current spatial distance to the fourth current spatial distance is determined as the second target weight; the fourth data item is determined based on the update speed corresponding to the discriminator in the previous training batch and the fourth hyperparameter; the difference between the optimal model parameters of the discriminator and the third currently updated model parameters is determined as the third factor; the fifth data item is determined based on the third factor, the fifth hyperparameter and the third random number; the current update speed corresponding to the discriminator is determined based on the fourth data item, the fifth data item and the second target weight; the sum of the current update speed corresponding to the discriminator and the third currently updated model parameter is determined as the fourth currently updated model parameter; the fourth currently updated model parameter is determined as the model parameter corresponding to the discriminator in the next training batch.

[0157] On the basis of the above embodiment, optionally, the model training module 430 includes a discriminator loss function determination module; wherein the discriminator loss function determination module is used to input the cable fault sample data in the training set into at least two generators in an untrainable state to obtain fault prediction results corresponding to at least two generators respectively; construct a fault prediction result data set based on the fault prediction results corresponding to at least two generators respectively, and construct a fault classification label data set based on the fault classification labels in the training set; wherein the fault prediction result data set includes a fault prediction result and a first authenticity label; the fault classification label data set includes a fault classification label and a second authenticity label; the first authenticity label is a label used to characterize an unauthentic label; the second authenticity label is a label used to characterize an authentic label; the fault prediction result in the fault prediction result data set is input into the discriminator to obtain a first authenticity prediction result corresponding to the fault prediction result; the fault classification label in the fault classification label data set is input into the discriminant network model , obtain the second authenticity prediction result corresponding to the fault classification label; determine the first discriminant loss function based on the first authenticity prediction result and the first authenticity label; and determine the second discriminant loss function based on the second authenticity prediction result and the second authenticity label; determine the loss function corresponding to the discriminator based on the first discriminant loss function and the second discriminant loss function; accordingly, the model training module 430 also includes a generator loss function determination module; wherein the generator loss function determination module is used to input the fault prediction result corresponding to any generator into the discriminator in an untrainable state to obtain the third authenticity prediction result corresponding to the fault prediction result; determine the first generation loss function corresponding to any generator based on the third authenticity prediction result and the second authenticity label corresponding to the fault prediction result; determine the second generation loss function corresponding to any generator based on the fault classification label and the fault prediction result in the training set; determine the loss function corresponding to any generator based on the first generation loss function and the second generation loss function.

[0158] The wire and cable fault diagnosis model training device provided in the embodiment of the present invention can execute the wire and cable fault diagnosis model training method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0159] Embodiment 5

[0160] Figure 5 Schematic diagram of the structure of a wire and cable fault diagnosis device provided by Embodiment 5 of the present invention. Figure 5 As shown, the device comprises:

[0161] The device operating parameter acquisition module 510 is used to acquire the device operating parameters of the target wire and cable;

[0162] The fault type information determination module 520 is used to input the equipment operating parameters into the target cable fault diagnosis model to obtain the fault type information of the target wire and cable; wherein the target cable fault diagnosis model is obtained based on the wire and cable fault diagnosis model training method provided in any embodiment of the present invention.

[0163] The technical solution of this embodiment obtains the fault type information of the target wire and cable by inputting the equipment operating parameters into the target cable fault diagnosis model, and can obtain accurate fault type information.

[0164] The wire and cable fault diagnosis device provided in the embodiment of the present invention can execute the wire and cable fault diagnosis method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0165] Embodiment 6

[0166] Figure 6 It is a structural diagram of an electronic device that implements the wire and cable fault diagnosis model training method or wire and cable fault diagnosis method of an embodiment of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0167] like Figure 6 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0168] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0169] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as a wire and cable fault diagnosis model training method or a wire and cable fault diagnosis method.

[0170] In some embodiments, the wire and cable fault diagnosis model training method or the wire and cable fault diagnosis method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the wire and cable fault diagnosis model training method or the wire and cable fault diagnosis method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the wire and cable fault diagnosis model training method or the wire and cable fault diagnosis method by any other appropriate means (e.g., by means of firmware).

[0171] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0172] The computer program for implementing the wire and cable fault diagnosis model training method or wire and cable fault diagnosis method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program can be executed entirely on the machine, partially on the machine, partially on the machine as an independent software package and partially on a remote machine, or completely on a remote machine or server.

[0173] Embodiment 7

[0174] Embodiment 7 of the present invention further provides a computer-readable storage medium, the computer-readable storage medium storing computer instructions, the computer instructions being used to cause a processor to execute a wire and cable fault diagnosis model training method, the method comprising:

[0175] A cable fault sample data set is obtained, the cable fault sample data set includes cable fault sample data and fault classification labels corresponding to the cable fault sample data; a cable fault diagnosis generative adversarial network model is constructed, the cable fault diagnosis generative adversarial network model includes a generative network model and a discriminative network model; the generative network model includes at least two generators in parallel, and the discriminative network model includes a discriminator; at least two generators are multi-classification models with the same model structure; based on the training set in the cable fault sample data set, the randomly initialized cable fault diagnosis generative adversarial network model is trained to obtain a trained generative network model; in any batch training process of synchronous training of at least two parallel generators, based on the loss function corresponding to any generator and at least The current model parameters corresponding to any generator are updated based on the model parameters corresponding to at least two generators respectively; and the current model parameters corresponding to the discriminator are updated based on the loss function corresponding to the discriminator and the model parameters of the discriminator in the historical batches; based on the validation set in the cable fault sample data set, the optimal first performance index corresponding to at least two generators in the trained generative network model is determined, and the generator corresponding to the optimal first performance index is determined as the candidate cable fault diagnosis model; based on the test set in the cable fault sample data set, the performance index of the candidate cable fault diagnosis model is determined, and when the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold, the candidate cable fault diagnosis model is determined as the target cable fault diagnosis model.

[0176] Alternatively, the computer instructions are used to cause the processor to execute a wire and cable fault diagnosis method, the method comprising:

[0177] Obtain equipment operating parameters of the target wire and cable; input the equipment operating parameters into a target cable fault diagnosis model to obtain fault type information of the target wire and cable; wherein the target cable fault diagnosis model is obtained based on the wire and cable fault diagnosis model training method provided in any embodiment of the present invention.

[0178] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0179] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0180] The systems and techniques described herein may be implemented in a computing system that includes a backend (e.g., as a data server), or a computing system that includes middleware (e.g., an application server), or a computing system that includes a frontend (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backends, middleware, or frontends. The systems may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0181] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0182] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0183] Embodiment 8

[0184] Embodiment 8 of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements a wire and cable fault diagnosis model training method or a wire and cable fault diagnosis method according to any embodiment of the present invention.

[0185] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A wire and cable fault diagnosis model training method, characterized in that: include: Acquire a cable fault sample data set, wherein the cable fault sample data set includes cable fault sample data and fault classification labels corresponding to the cable fault sample data; Constructing a cable fault diagnosis generative adversarial network model, the cable fault diagnosis generative adversarial network model includes a generative network model and a discriminative network model; the generative network model includes at least two generators in parallel, and the discriminative network model includes a discriminator; the at least two generators are multi-classification models with the same model structure; Based on the training set in the cable fault sample data set, the randomly initialized cable fault diagnosis generative adversarial network model is trained to obtain the trained generative network model; in any batch training process of synchronous training of at least two parallel generators, the current model parameters corresponding to any generator are updated based on the loss function corresponding to any generator and the model parameters corresponding to the at least two generators respectively; And, updating the current model parameters corresponding to the discriminator based on the loss function corresponding to the discriminator and the model parameters of the discriminator in the historical batches; Based on the validation set in the cable fault sample data set, determine the optimal first performance index corresponding to the at least two generators in the trained generative network model, and determine the generator corresponding to the optimal first performance index as a candidate cable fault diagnosis model; Based on the test set in the cable fault sample data set, the performance index of the candidate cable fault diagnosis model is determined, and when the performance index of the candidate cable fault diagnosis model meets the preset performance index threshold, the candidate cable fault diagnosis model is determined as the target cable fault diagnosis model.

2. The method according to claim 1, characterized in that: The model parameters corresponding to the at least two generators include the current model parameters corresponding to the at least two generators; accordingly, The updating of the current model parameters of any generator based on the loss function corresponding to any generator and the model parameters respectively corresponding to the at least two generators comprises: For any of the generators, update the current model parameters of any of the generators based on the loss function corresponding to any of the generators to obtain first current updated model parameters corresponding to any of the generators; In the case where the current training batch is the first training batch, determining the first current updated model parameter corresponding to any one of the generators as the model parameter corresponding to any one of the generators in the second training batch; In the case where the current training batch is not the first training batch, second performance indicators corresponding to the at least two generators are determined based on the training set, an optimal second performance indicator and a worst second performance indicator are determined among the plurality of second performance indicators, a current model parameter of the generator corresponding to the optimal second performance indicator is determined as the current optimal model parameter, and a current model parameter of the generator corresponding to the worst second performance indicator is determined as the current worst model parameter; Based on the current model parameters, current associated parameters and first current updated model parameters corresponding to any one of the generators, determining the current update speed corresponding to any one of the generators; the current associated parameters include the current optimal model parameters, the current worst model parameters and the update speed corresponding to any one of the generators in the previous training batch; Determine a second current update model parameter corresponding to any one of the generators based on the sum of the current update speed corresponding to any one of the generators and the first current update model parameter; The second currently updated model parameters corresponding to any one of the generators are determined as the model parameters corresponding to any one of the generators in the next training batch.

3. The method according to claim 2, characterized in that The determining the current update speed corresponding to any one of the generators based on the current model parameters, the current association parameters and the first current update model parameters corresponding to any one of the generators comprises: Determine a first current spatial distance based on the current model parameters corresponding to any one of the generators and the current optimal model parameters; and determine a second current spatial distance based on the current model parameters corresponding to any one of the generators and the current worst model parameters; determine a ratio of the first current spatial distance to the second current spatial distance as a first target weight; Determine a first data item based on an update speed corresponding to any one of the generators in a previous training batch and a first hyperparameter; Determine the difference between the current optimal model parameter and the first current updated model parameter corresponding to any one of the generators as a first factor; determine a second data item based on the first factor, a second hyperparameter and a first random number; The current update speed corresponding to any one of the generators is determined based on the first data item, the second data item and the first target weight.

4. The method according to claim 3, characterized in that: The model parameters corresponding to the at least two generators respectively also include the model parameters of the at least two generators in the historical batches; correspondingly, the current associated parameters also include the previous optimal model parameters corresponding to any one of the generators in the historical batches; The determining of the current update speed corresponding to any one of the generators based on the current model parameters, the current association parameters and the first current update model parameters corresponding to any one of the generators further includes: Determine the third performance indicators corresponding to the model parameters corresponding to any one of the generators based on the training set, and determine the optimal third performance indicator among the plurality of the third performance indicators; Determine the model parameters corresponding to any one of the generators and the optimal third performance indicator as the previous optimal model parameters corresponding to any one of the generators; Determine the difference between the previous optimal model parameters corresponding to any one of the generators and the first currently updated model parameters as the second factor; determining the third data item based on the second factor, a third hyperparameter, and a second random number; Accordingly, determining the current update speed corresponding to any one of the generators based on the first data item, the second data item and the first target weight includes: The current update speed corresponding to any one of the generators is determined based on the first data item, the second data item, the third data item and the first target weight.

5. The method according to claim 1, characterized in that: The updating of the current model parameters corresponding to the discriminator based on the loss function corresponding to the discriminator and the model parameters of the discriminator in the historical batches includes: Based on the loss function corresponding to the discriminator, the current model parameters corresponding to the discriminator are updated to obtain third current updated model parameters; In a case where the current training batch is the first training batch, determining the third current updated model parameter as the model parameter corresponding to the discriminator in the second training batch; In the case where the current training batch is not the first training batch, determining the fourth performance indicators respectively corresponding to the corresponding model parameters of the discriminator, and determining the best fourth performance indicator and the worst fourth performance indicator among the plurality of the fourth performance indicators; wherein the model parameters corresponding to the discriminator include the current model parameters corresponding to the discriminator and the model parameters of the discriminator in the historical batches; Determine the model parameters corresponding to the best fourth performance index as the best model parameters corresponding to the discriminator, and determine the model parameters corresponding to the worst fourth performance index as the worst model parameters corresponding to the discriminator; Determine a third current spatial distance based on the third current updated model parameter and the best model parameters of the discriminator; and determine a fourth current spatial distance based on the third current updated model parameter and the worst model parameters of the discriminator; and determine a ratio of the third current spatial distance to the fourth current spatial distance as a second target weight; Determine a fourth data item based on an update speed corresponding to the discriminator in a previous training batch and a fourth hyperparameter; Determine the difference between the previous optimal model parameters corresponding to the discriminator and the third currently updated model parameters as a third factor; determine a fifth data item based on the third factor, a fifth hyperparameter and a third random number; Determine a current update speed corresponding to the discriminator based on the fourth data item, the fifth data item and the second target weight; Determine the sum of the current update speed corresponding to the discriminator and the third current update model parameter as the fourth current update model parameter; The fourth currently updated model parameter is determined as the model parameter corresponding to the discriminator in the next training batch.

6. The method according to claim 1, characterized in that The process of determining the loss function corresponding to the discriminator includes: Inputting the cable fault sample data in the training set into the at least two generators in a non-trainable state to obtain fault prediction results corresponding to the at least two generators respectively; A fault prediction result data set is constructed based on the fault prediction results corresponding to the at least two generators respectively, and a fault classification label data set is constructed based on the fault classification labels in the training set; wherein the fault prediction result data set includes the fault prediction result and a first authenticity label; the fault classification label data set includes the fault classification label and a second authenticity label; the first authenticity label is a label used to represent an unauthentic label; the second authenticity label is a label used to represent an authentic label; Inputting the fault prediction result in the fault prediction result data set into the discriminator to obtain a first authenticity prediction result corresponding to the fault prediction result; Inputting the fault classification label in the fault classification label data set into the discriminant network model to obtain a second authenticity prediction result corresponding to the fault classification label; Determining a first discriminant loss function based on the first authenticity prediction result and the first authenticity label; and determining a second discriminant loss function based on the second authenticity prediction result and the second authenticity label; Determine a loss function corresponding to the discriminator based on the first discriminant loss function and the second discriminant loss function; Accordingly, the process of determining the loss function corresponding to any generator includes: Inputting the fault prediction result corresponding to any one of the generators into the discriminator in a non-trainable state to obtain a third authenticity prediction result corresponding to the fault prediction result; Determine a first generation loss function corresponding to any one of the generators based on a third authenticity prediction result and a second authenticity label corresponding to the fault prediction result; Determine a second generation loss function corresponding to any one of the generators based on the fault classification label in the training set and the fault prediction result; Determine the loss function corresponding to any one of the generators based on the first generation loss function and the second generation loss function.

7. A method for diagnosing faults of electric wires and cables, characterized in that: include: Obtain equipment operating parameters of target wires and cables; The equipment operating parameters are input into a target cable fault diagnosis model to obtain the fault type information of the target wire and cable; wherein the target cable fault diagnosis model is obtained based on the wire and cable fault diagnosis model training method described in any one of claims 1-6.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the wire and cable fault diagnosis model training method described in any one of claims 1-6, or the wire and cable fault diagnosis method described in claim 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the wire and cable fault diagnosis model training method described in any one of claims 1-6, or the wire and cable fault diagnosis method described in claim 7 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the wire and cable fault diagnosis model training method according to any one of claims 1 to 6, or the wire and cable fault diagnosis method according to claim 7.

Citation Information

Patent Citations

  • Power transmission line fault detection method based on auxiliary classification generative adversarial network

    CN113406437A

  • Power grid fault positioning method

    CN113920080A

  • Power transmission line state monitoring method and device and storage medium

    CN118362162A

Cited By

  • Rotating machine fault diagnosis method based on confrontation noise reduction convolutional neural network

    CN119803899A