CN-122020160-A - Submarine cable fault diagnosis method based on countermeasure generation under lean data condition

CN122020160ACN 122020160 ACN122020160 ACN 122020160ACN-122020160-A

Abstract

The invention discloses a submarine cable fault diagnosis method under a lean data condition based on countermeasure generation. The method comprises the steps of constructing MPAGCNN a model comprising an countermeasure generation network and a fault diagnosis network, obtaining a source sample dataset, inputting real data and random noise in the source sample dataset into the countermeasure generation network for training, optimizing the model by minimizing countermeasure generation loss until a generator and a discriminator reach Nash equilibrium, obtaining the generated sample dataset by using a current generator, merging the generated sample dataset and the source sample dataset to obtain a comprehensive sample dataset, inputting the comprehensive sample dataset into the fault diagnosis network for training, optimizing the model by minimizing fault classification positioning loss until a preset performance threshold is met, and outputting the current fault diagnosis network as a final submarine cable fault diagnosis model. The invention can realize the improvement of the reliability of submarine cable fault diagnosis.

Inventors

XIONG HUI
YANG XIBING
YANG CHONG
WEI LI
LIU JINYI
FU LIRONG

Assignees

海南大学

Dates

Publication Date: 20260512
Application Date: 20260119

Claims (6)

1. A method for diagnosing a submarine cable fault under a lean data condition based on countermeasure generation, comprising the steps of: constructing MPAGCNN a model, wherein the MPAGCNN model comprises an countermeasure generation network and a fault diagnosis network; the method comprises the steps of obtaining a source sample data set, inputting real data and random noise in the source sample data set into an countermeasure generation network for training, optimizing a model by minimizing countermeasure generation loss until a generator and a discriminator reach Nash equilibrium, and obtaining a generated sample data set by using a current generator; Inputting the comprehensive sample data set into a fault diagnosis network for training, optimizing a model by minimizing fault classification positioning loss until a preset performance threshold is met, and outputting the current fault diagnosis network as a final submarine cable fault diagnosis model; and inputting the data to be predicted into a final submarine cable fault diagnosis model to perform reasoning and diagnosis, and obtaining a fault diagnosis result.
2. The method for diagnosing the submarine cable fault under the lean data condition based on the countermeasure generation according to claim 1, wherein the countermeasure generation network comprises a generator and a discriminator, the generator comprises a linear layer, a batch normalization layer, an activation function, 5 feature recovery modules 1, 2,3, 4, 5 and a transpose convolution layer and an activation function which are connected in sequence, the feature recovery module 1 comprises the transpose convolution layer, the batch normalization layer and the activation function which are connected in sequence, and the discriminator comprises the convolution layer, the activation function, 5 feature extraction modules A, B, C, D, E and the linear layer which are connected in sequence, and the feature extraction module A comprises the convolution layer, the batch normalization layer and the activation function which are connected in sequence.
3. The method for diagnosing the submarine cable fault under the lean data condition based on the countermeasure generation according to claim 2, wherein the fault diagnosis network comprises a feature extractor and a fault detector, the feature extractor comprises 5 feature extraction modules 1, 2, 3, 4 and 5 which are connected in sequence, the feature extraction module 1 comprises a convolution layer, a batch normalization layer, an activation function and a pooling layer which are connected in sequence, and the fault detector comprises three prediction modules which are respectively arranged behind the feature extraction module 1, the feature extraction module 3 and the feature extraction module 5, and the prediction modules comprise a full connection layer and a Softmax layer.
4. A method of diagnosing a subsea cable fault under conditions of lean data based on countermeasure generation according to claim 1, wherein said acquiring a source sample dataset comprises the steps of: Acquiring a submarine cable data set; Labeling fault types in the submarine cable dataset to obtain a source sample dataset.
5. A method of diagnosing a subsea cable fault under lean data conditions based on countermeasure generation according to claim 1, wherein said countermeasure generation loss is as follows: ; Where λ 1 and λ 2 are hyper-parameters with scheduling policy, L RpGAN is relative match loss, and the formula is: ; Where f is the activation function, D ψ is the arbiter, G θ is the generator, z is the noise distribution, x is the true data distribution, R1 penalizes the gradient of the discriminator on the real data, and the formula is: ; R2 penalizes gradients of the discriminators on the generated data, and the formula is: 。
6. a method of diagnosing a subsea cable fault under conditions of lean data based on countermeasure generation according to claim 1, wherein said fault classification localization loss L y is as follows: ; Where M represents the total number of prediction modules, Representing the loss weight corresponding to the mth prediction module, meeting the following requirements , ; The fault classification locating loss value representing the mth prediction module is defined as: ; wherein K represents the number of prediction frames output by the prediction module; Representing the intersection ratio between the kth prediction frame and the corresponding real frame; A class prediction probability representing a kth prediction frame; representing a real class label corresponding to the kth prediction frame; And (3) with The self-adaptive weight coefficients respectively representing the positive sample and the negative sample satisfy ; Representing a distributed focal point loss function, A weighted classification loss function representing consideration of class imbalance for mitigating different class number imbalance in a fault sample, defined as: ; wherein N represents the number of samples, and C represents the number of fault categories; Representing the prediction probability that the ith sample belongs to the c-th class; Is a class balancing factor; to characterize the spatial offset relationship between the predicted and real frames, a normalized position offset vector is introduced It is defined as: ; In the middle of , The center coordinates of the true fault region are represented, , Representing the width and height of the real fault area; , Representing the center coordinates of the initial prediction block, , H and W respectively represent the height and width of the feature map or the input image; adaptive weighting coefficients for positive and negative samples And (3) with Expressed as: ; In the formula, , Respectively representing positive sample indexes and negative sample indexes corresponding to the current prediction frame; for the weight adjustment function to dynamically adjust the sample weight according to the prediction difficulty, it is defined as: ; the method comprises the steps of obtaining a weight adjustment function, wherein a is the upper limit of the weight adjustment function, c is a curvature adjustment parameter used for controlling the weight change rate, B represents the total number of prediction frames, n represents the spatial offset degree sequencing index of the prediction frames relative to a real frame, and the value range is [0, B ].

Description

Submarine cable fault diagnosis method based on countermeasure generation under lean data condition Technical Field The invention relates to the technical field of cable fault detection, in particular to a submarine cable fault diagnosis method under a lean data condition based on countermeasure generation. Background Submarine cables are the main media for cross-sea power grid interconnection, offshore platform electric energy acquisition and offshore clean energy delivery, and due to the severe and complex working environments, the submarine cables have faults in the working process, so that the fault detection and positioning difficulties are high, the maintenance is difficult and the period is long. The failure of the submarine cable not only affects the normal operation of the submarine cable, but also causes larger economic loss caused by off-grid shutdown of other systems, and even threatens personal safety. Therefore, the working condition of the submarine cable is effectively inspected, potential fault threats and fault positioning of the submarine cable are automatically and accurately identified, and the method has important significance and value for guaranteeing safe and stable operation of an offshore power system. In recent years, deep learning has made a great progress in the field of fault diagnosis, and related scholars have conducted a great deal of research based on a deep learning method in land cable fault diagnosis, have achieved a certain performance, and have published a lot of research results. However, due to poor submarine cable inspection conditions, it is difficult to obtain a large number of trainable samples with good distribution balance, and the large number of balanced distribution samples required by the detection model based on deep learning are very unfavorable, so that the accuracy and generalization of the model obtained by training are poor. Therefore, it is an important challenge how to implement data enhancement under rare sample categories to eliminate serious imbalances in the data so that models can be trained effectively. Disclosure of Invention In order to solve the technical problems, the invention provides a submarine cable fault diagnosis method based on the lean data condition generated by countermeasure, which can improve the reliability of submarine cable fault diagnosis. In order to achieve the above purpose, the technical scheme of the invention is as follows: a method for diagnosing a submarine cable fault under lean data conditions based on countermeasure generation, comprising the steps of: constructing MPAGCNN a model, wherein the MPAGCNN model comprises an countermeasure generation network and a fault diagnosis network; the method comprises the steps of obtaining a source sample data set, inputting real data and random noise in the source sample data set into an countermeasure generation network for training, optimizing a model by minimizing countermeasure generation loss until a generator and a discriminator reach Nash equilibrium, and obtaining a generated sample data set by using a current generator; Inputting the comprehensive sample data set into a fault diagnosis network for training, optimizing a model by minimizing fault classification positioning loss until a preset performance threshold is met, and outputting the current fault diagnosis network as a final submarine cable fault diagnosis model; and inputting the data to be predicted into a final submarine cable fault diagnosis model to perform reasoning and diagnosis, and obtaining a fault diagnosis result. Preferably, the countermeasure generating network comprises a generator and a discriminator, wherein the generator comprises a linear layer, a batch normalization layer, an activating function, 5 feature recovery modules 1,2, 3,4, 5 and a transposition convolution layer and an activating function which are connected in sequence, the discriminator comprises a convolution layer, an activating function, 5 feature extraction modules A, B, C, D, E and a linear layer which are connected in sequence, and the feature extraction module A comprises a convolution layer, a batch normalization layer and an activating function which are connected in sequence. Preferably, the fault diagnosis network comprises a feature extractor and a fault detector, wherein the feature extractor comprises 5 feature extraction modules 1,2,3, 4 and 5 which are connected in sequence, the feature extraction modules 1 comprise a convolution layer, a batch normalization layer, an activation function and a pooling layer which are connected in sequence, the fault detector comprises three prediction modules which are respectively arranged behind the feature extraction modules 1, 3 and 5, and the prediction modules comprise a full connection layer and a Softmax layer. Preferably, the acquiring the source sample dataset comprises the steps of: Acquiring a submarine cable data set; Labeling fault types in the submarine cable da