A method, system, and storage medium for anti-interference analysis of partial discharge signals.

Abstract

This invention relates to the field of electrical signal analysis technology, and particularly to an anti-interference analysis method, system, and storage medium for partial discharge signals. The method involves acquiring a sample partial discharge signal analysis set from the target device, wherein the sample partial discharge signal analysis set includes associated stored original partial discharge signals, sample partial discharge signal components, sample interference signal components, and sample environmental information; constructing an anti-interference analysis model set, wherein the anti-interference analysis model set includes a target signal prediction model and an interference signal prediction model; setting a joint regression optimization objective, and performing collaborative training on the anti-interference analysis model set based on the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model set is obtained; interacting with the target device to collect real-time original partial discharge signals and real-time environmental information, which are used as input to the anti-interference analysis model set for partial discharge signal analysis. This method can improve the accuracy, stability, and real-time performance of partial discharge signal analysis.

Description

Technical Field

[0001] This invention relates to the field of electrical signal analysis technology, and in particular to an anti-interference analysis method, system and storage medium for partial discharge signals. Background Technology

[0002] In fields such as power equipment and electrical equipment, partial discharge is an important early warning signal of equipment insulation aging and failure. Monitoring and analyzing partial discharge signals is a key link in ensuring the safe operation of equipment. Traditional partial discharge signal analysis methods mainly rely on signal processing techniques such as filtering and spectrum analysis to process the original signal to remove interference and extract the partial discharge signal components.

[0003] Traditional analysis methods are often designed for specific types of interference and are ill-suited to complex and ever-changing environmental interference. When interference signals have similar characteristics to partial discharge signals, misjudgment or signal loss can easily occur. Furthermore, traditional signal processing procedures are complex and inefficient for analyzing real-time acquired signals, failing to meet the real-time requirements of online equipment monitoring. They suffer from technical problems related to poor accuracy, stability, and real-time performance. Summary of the Invention

[0004] This invention addresses the technical problems of poor accuracy, stability, and real-time performance in existing technologies by providing an anti-interference analysis method, system, and storage medium for partial discharge signals.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides an anti-interference analysis method for partial discharge signals, comprising: acquiring a sample partial discharge signal analysis set of a target device, wherein the sample partial discharge signal analysis set includes associated stored sample original partial discharge signals, sample partial discharge signal components, sample interference signal components, and sample environmental information; constructing an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model; setting a joint regression optimization objective, and performing collaborative training on the anti-interference analysis model group according to the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained; interacting with the target device, collecting real-time original partial discharge signals and real-time environmental information, and using them as inputs to the anti-interference analysis model group for partial discharge signal analysis.

[0006] Optionally, a sample partial discharge signal analysis set of the target device is obtained, wherein the sample partial discharge signal analysis set includes associated stored original partial discharge signals, sample partial discharge signal components, sample interference signal components, and sample environment information, including: according to the device model of the target device, calling historical analysis records and performing confidence filtering to obtain historical partial discharge signal analysis samples; performing random crossover mutation on the historical partial discharge signal analysis set to obtain a first expanded partial discharge signal analysis sample; performing adversarial training based on the confidence filtering results to obtain a sample generator and a sample validator, and generating a second expanded partial discharge signal analysis sample through the sample generator; based on the sample validator, performing confidence verification filtering on the first expanded partial discharge signal analysis sample and the second expanded partial discharge signal analysis sample respectively, and merging the confidence verification filtering results with the historical partial discharge signal analysis samples to output the sample partial discharge signal analysis set.

[0007] The target signal prediction model takes the original partial discharge signal and environmental information as input and the partial discharge signal components as output, and is used to predict the partial discharge signal components based on the original partial discharge signal and environmental information.

[0008] The interference signal prediction model takes the original partial discharge signal and environmental information as input and the interference signal components as output, and is used to predict the interference signal components based on the environmental information.

[0009] The process includes setting a joint regression optimization objective and performing collaborative training on the anti-interference analysis model group based on the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained. This includes: defining the prediction output of the target signal prediction model as a first prediction signal and defining the prediction output of the interference signal prediction model as a second prediction signal; defining the signal reconstruction result based on the first prediction signal and the second prediction signal as a first reconstruction signal, and setting the joint regression optimization objective as the mean square error between the first reconstruction signal and the original sample partial discharge signal; using the sample partial discharge signal analysis set as sample input, performing collaborative training on the target signal prediction model and the interference signal prediction model until the joint regression optimization objective converges; and outputting the target signal prediction model and the interference signal prediction model as the anti-interference analysis model group.

[0010] The process involves using the sample partial discharge signal analysis set as input to collaboratively train the target signal prediction model and the interference signal prediction model until the joint regression optimization objective converges. This includes: extracting the original partial discharge signal, the components of the sample partial discharge signal, and the sample environment information from the sample partial discharge signal analysis set to obtain a first sample analysis set; extracting the original partial discharge signal, the components of the sample interference signal, and the sample environment information from the sample partial discharge signal analysis set to obtain a second sample analysis set; independently training the target signal prediction model using the sample partial discharge signal components as supervision, and independently training the interference signal prediction model using the sample interference signal components as supervision, respectively, based on the first and second sample analysis sets; and combining the allocation layer to connect the target signal prediction model and the interference signal prediction model in parallel, and then collaboratively training them using the sample partial discharge signal analysis set until the joint regression optimization objective satisfies a preset target convergence constraint.

[0011] The interactive target device collects real-time raw partial discharge (PD) signals and real-time environmental information, and uses these as inputs to the anti-interference analysis model group for PD signal analysis. This includes: inputting the real-time raw PD signals and real-time environmental information into the target signal prediction model to obtain corresponding predicted PD signal components; inputting the real-time raw PD signals and real-time environmental information into the interference signal prediction model to obtain corresponding predicted interference signal components; superimposing the predicted PD signal components and the predicted interference signal components to obtain a predicted restored signal, and comparing and calculating the fitting error between the predicted restored signal and the real-time raw PD signal; if the fitting error is less than a preset error limit, outputting the predicted PD signal components as the PD signal analysis result.

[0012] Secondly, the present invention provides an anti-interference analysis system for partial discharge signals, comprising: The partial discharge signal analysis set acquisition module is used to acquire the sample partial discharge signal analysis set of the target device, wherein the sample partial discharge signal analysis set includes the associated stored original sample partial discharge signal, sample partial discharge signal components, sample interference signal components and sample environmental information; An anti-interference analysis model group construction module is used to construct an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model; The anti-interference analysis model group training module is used to set a joint regression optimization objective and perform collaborative training on the anti-interference analysis model group according to the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained. The input information real-time acquisition module is used to interact with the target device, acquire real-time raw partial discharge signals and real-time environmental information, and use them as input to the anti-interference analysis model group for partial discharge signal analysis.

[0013] Thirdly, this application provides a storage medium storing a first computer program, which, when executed by a processor, implements an anti-interference analysis method for partial discharge signals as described in the first aspect.

[0014] By implementing this invention, it is possible to obtain a sample partial discharge signal analysis set of a target device, wherein the sample partial discharge signal analysis set includes associated stored original sample partial discharge signals, sample partial discharge signal components, sample interference signal components and sample environmental information, which can solve the problem of insufficient sample size and improve the robustness of model training. By implementing this invention, it is possible to construct an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model. The two models focus on the prediction of partial discharge signals and interference signals, respectively, to achieve accurate decomposition of the original signal, avoid the problem that a single model cannot handle two types of signals at the same time, and incorporate environmental information into the input so that the model can consider the influence of the environment on the two types of signals, thereby improving the adaptability and accuracy of the prediction. By implementing this invention, it is possible to set a joint regression optimization objective and perform collaborative training on the anti-interference analysis model group according to the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained. The training of the two models is linked by the joint objective, ensuring that the prediction results of the two models can be accurately restored after being superimposed, and avoiding the deviation that may occur during the training of a single model. By implementing this invention, it is possible to interact with the target device, collect real-time raw partial discharge signals and real-time environmental information, and use them as input to the anti-interference analysis model group for partial discharge signal analysis, thereby realizing dynamic monitoring and analysis of the device's partial discharge signals and timely acquisition of device status information.

[0015] In summary, by implementing this invention, the accuracy, stability, and real-time performance of partial discharge signal analysis can be improved. Attached Figure Description

[0016] Figure 1 A flowchart illustrating an anti-interference analysis method for partial discharge signals provided by this invention; Figure 2 A schematic diagram of the structure of an anti-interference analysis system for partial discharge signals provided by the present invention; Figure 3 This is a schematic diagram of a storage medium provided by the present invention.

[0017] In the attached diagram, the components represented by each number are as follows: The system includes a partial discharge signal analysis set acquisition module 11, an anti-interference analysis model group construction module 12, an anti-interference analysis model group training module 13, an input information real-time acquisition module 14, a storage medium 400, and a first computer program 410. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0021] Example 1, as Figure 1 As shown, this embodiment of the invention provides an anti-interference analysis method for partial discharge signals, including: S100: Obtain the sample partial discharge signal analysis set of the target device, wherein the sample partial discharge signal analysis set includes the associated stored original sample partial discharge signal, sample partial discharge signal components, sample interference signal components and sample environment information; S200: Construct an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model; S300: Set a joint regression optimization objective, and perform collaborative training on the anti-interference analysis model group according to the joint regression optimization objective and the sample partial discharge signal analysis set until the converged anti-interference analysis model group is obtained; S400: Interactive target device, which collects real-time raw partial discharge signals and real-time environmental information, and uses them as input to the anti-interference analysis model group for partial discharge signal analysis.

[0022] In step S100 of this application embodiment, a sample partial discharge signal analysis set of the target device is obtained. The sample partial discharge signal analysis set includes associated stored sample original partial discharge signals, sample partial discharge signal components, sample interference signal components, and sample environmental information, including: Based on the device model of the target device, retrieve historical analysis records and perform confidence filtering to obtain historical partial discharge signal analysis samples; Random crossover and mutation are performed on the historical partial discharge signal analysis set to obtain the first expanded partial discharge signal analysis sample; Adversarial training is performed based on the confidence screening results to obtain a sample generator and a sample validator, and a second expanded partial discharge signal analysis sample is generated through the sample generator. Based on the sample validator, the first expanded partial discharge signal analysis sample and the second expanded partial discharge signal analysis sample are subjected to credibility verification screening respectively, and the credibility verification screening results are merged with the historical partial discharge signal analysis samples to output the sample partial discharge signal analysis set.

[0023] The purpose of step S100 in this embodiment is to construct a high-quality, highly adaptable, and sufficiently large sample set of partial discharge signals for analysis. Through targeted sample acquisition, expansion, and screening, the sample set is ensured to accurately match the target device model and cover diverse signal scenarios, providing a reliable and rich input foundation for the subsequent training of the anti-interference analysis model group, and ultimately improving the accuracy of the model's anti-interference analysis of the target device's partial discharge signals.

[0024] First, based on the target device's model, historical analysis records need to be retrieved and confidence-filtered to obtain historical partial discharge signal analysis samples. That is, based on the target device's model, the corresponding historical analysis records are retrieved from the database.

[0025] Assuming the target equipment is a "GGD-10kV high-voltage switchgear", it is necessary to retrieve the historical monitoring records of the "GGD-10kV switchgear" from the database. These records include original partial discharge signals such as discharge pulses under different operating conditions, partial discharge signal components such as separated real discharge signals, interference signal components such as electromagnetic interference signals from surrounding motors, and corresponding environmental information such as temperature and humidity.

[0026] Then, the historical records are subjected to confidence screening to remove low-confidence, abnormal, or redundant data, retaining high-quality historical samples. For example, low-confidence samples with signal amplitudes far exceeding the normal discharge range of the equipment and missing environmental information are removed, ultimately retaining 500 high-quality historical samples as the historical partial discharge signal analysis set.

[0027] Next, random crossover and mutation are needed to perform on the historical partial discharge signal analysis set to obtain the first expanded partial discharge signal analysis sample. That is, "random crossover and mutation" is performed on the screened historical partial discharge signal analysis set. For example, the amplitude, frequency and other features of the signal are randomly combined and slightly mutated to generate new samples. If 1000 new samples are generated, they will be the first expanded sample to increase the sample diversity and solve the problem of insufficient historical sample size.

[0028] Furthermore, adversarial training is needed based on the confidence screening results to obtain a sample generator and a sample validator. The sample generator is then used to generate a second expanded partial discharge signal analysis sample. The logic is as follows: based on the confidence screening results from the first step, adversarial training is performed. By constructing a sample generator responsible for generating new samples and a sample validator responsible for judging the authenticity of samples, the two are allowed to compete and optimize each other. Finally, the well-trained sample generator is used to generate new samples, namely the second expanded samples, further expanding the sample size and making the sample features more closely resemble real-world scenarios.

[0029] Specifically, the sample generator can be built using a multilayer perceptron, whose structure includes an input layer, a hidden layer, and an output layer. The input layer dimension is the sum of the original partial discharge signal feature dimension and the sample environmental information feature dimension. For example, if the original partial discharge signal has 100-dimensional time-series features and the environmental information has 5-dimensional features, then the input layer dimension is 105.

[0030] There are three hidden layers. The first layer has 256 neurons and uses ReLU as the activation function; the second layer has 128 neurons and uses ReLU as the activation function; and the third layer has 64 neurons and uses ReLU as the activation function.

[0031] The output layer dimension is consistent with the joint feature dimension of the original partial discharge signal, the sample partial discharge signal components, and the sample interference signal components, and the activation function is a linear activation function.

[0032] During training, the learning rate of the sample generator was set to 0.001; the training batch size was set to 32; and the weight decay coefficient was set to 0.0001.

[0033] The sample validator can also be built using a multilayer perceptron, with the same structure including an input layer, hidden layers, and an output layer. The input layer dimension of the sample validator is the same as the output layer dimension of the sample generator.

[0034] The sample validator has two hidden layers. The first layer has 128 neurons and uses LeakyReLU as the activation function. The second layer has 64 neurons and uses LeakyReLU as the activation function.

[0035] The output layer of the sample validator is 1-dimensional, and the activation function is Sigmoid. The output value is 0 to 1, which is used to represent the probability that the sample is a real sample.

[0036] During training, the learning rate of the sample validator was set to 0.0005; the training batch size was set to 32; and the weight decay coefficient was set to 0.00005.

[0037] The samples used to train the sample validator and sample generator are derived from historical partial discharge signal analysis samples obtained by confidence filtering of historical analysis records retrieved according to the device model of the target device.

[0038] The total number of training rounds was set to 200. During training, the parameters of both the sample generator and the sample validator were updated simultaneously in each round, and the model was optimized through the adversarial game between the two.

[0039] When the accuracy of the sample validator in distinguishing between real and generated samples remains stable at around 50% for 10 consecutive rounds (meaning the sample validator cannot effectively distinguish between real and generated samples), and the samples generated by the sample generator conform to the correlation pattern between partial discharge signals and environmental information after manual sampling, the model training is considered to have converged. The sample generator and sample validator are then obtained.

[0040] Then, a well-trained sample generator is used to generate new samples, i.e., second expanded samples, such as 800 second expanded samples, further expanding the sample size. Next, a sample validator is used to perform credibility verification and screening on the first and second expanded samples respectively, eliminating low-quality expanded samples that do not conform to the characteristics of the real signal. For example, ultimately retaining 800 valid first expanded samples and 600 valid second expanded samples. The filtered credible expanded samples are merged with the original historical partial discharge signal analysis samples, and the final output is a complete sample partial discharge signal analysis set. For example, merging 500 historical partial discharge signal analysis samples and 1400 valid expanded samples yields a "GGD-10kV switchgear sample partial discharge signal analysis set" containing 1900 data points. S200: Construct an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model; In step S200 of this embodiment, the core purpose of constructing the anti-interference analysis model group is to achieve accurate separation of the effective partial discharge component and the interference component in the original partial discharge signal. Through the division of labor and cooperation between the two models, they focus on predicting the partial discharge signal component and the interference signal component respectively. Combined with the correlation characteristics between the original partial discharge signal and environmental information, a model foundation is laid for extracting the real partial discharge signal from the complex original signal, ultimately improving the anti-interference capability of partial discharge signal analysis.

[0041] In step S200 of this application embodiment, the target signal prediction model takes the original partial discharge signal and environmental information as input and the partial discharge signal components as output, and is used to predict the partial discharge signal components based on the original partial discharge signal and environmental information.

[0042] In step S200 of this application embodiment, the interference signal prediction model takes the original partial discharge signal and environmental information as input and the interference signal components as output, and is used to predict the interference signal components based on the environmental information.

[0043] The original partial discharge signal is the unseparated mixed signal, and the environmental information includes temperature, humidity, and other information. For the target signal prediction model, a convolutional neural network can be used. The target signal prediction model mainly consists of five parts: an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer.

[0044] The input layer is used to receive the raw partial discharge signal and environmental information, with an input dimension of 105.

[0045] The target signal prediction model has two convolutional layers. The first layer contains 64 3×3 convolutional kernels with a stride of 1 and the activation function is ReLU. The second layer contains 32 3×3 convolutional kernels with a stride of 1 and the activation function is ReLU.

[0046] The pooling layer of the target signal prediction model corresponds to two convolutional layers, both using 2×2 max pooling with a stride of 2.

[0047] The target signal prediction model has two fully connected layers: the first layer has 128 neurons with the ReLU activation function, and the second layer has 64 neurons with the ReLU activation function.

[0048] The output layer of the target signal prediction model has the same dimension as the feature dimension of the partial discharge signal component, such as 50 dimensions, and the activation function is a linear activation function, outputting the partial discharge signal component.

[0049] In the parameter settings of the target signal prediction model, the learning rate is set to 0.001; the batch size is set to 64; the weight decay coefficient is set to 0.0001; and the dropout ratio is set to 0.2.

[0050] Using the same method, we continued to build an interference signal prediction model, also employing a convolutional neural network, to achieve structural symmetry with the target signal prediction model and adapt to the characteristics of the interference signal.

[0051] The interference signal prediction model also consists of five parts: input layer, convolutional layer, pooling layer, fully connected layer, and output layer.

[0052] The input layer of the interference signal prediction model receives the raw partial discharge signal and environmental information; the example input dimension is 105.

[0053] The interference signal prediction model has two convolutional layers. The first layer contains 64 3×3 convolutional kernels with a stride of 1 and the activation function is ReLU. The second layer contains 32 3×3 convolutional kernels with a stride of 1 and the activation function is ReLU.

[0054] The pooling layer of the interference signal prediction model corresponds to two convolutional layers, both using 2×2 max pooling with a stride of 2.

[0055] The interference signal prediction model has two fully connected layers: the first layer has 128 neurons with the ReLU activation function, and the second layer has 64 neurons with the ReLU activation function.

[0056] The output layer of the interference signal prediction model has the same dimension as the feature dimension of the interference signal components, such as 50 dimensions, and the activation function is a linear activation function, outputting the interference signal components.

[0057] In the parameter settings of the interference signal prediction model, the learning rate is set to 0.001; the batch size is set to 64; the weight attenuation coefficient is set to 0.0001; and the dropout ratio is set to 0.2.

[0058] In training the anti-interference analysis model group, the training samples are derived from the sample partial discharge signal analysis set obtained in step S100, including the original sample partial discharge signal, sample partial discharge signal components, sample interference signal components, and sample environmental information stored therein. The number of training samples for the anti-interference analysis model group needs to match the size of the sample partial discharge signal analysis set, usually no less than 1000, to cover the characteristics of partial discharge signals and interference signals under different environments.

[0059] The initial training rounds for the anti-interference analysis model group were set to 300 rounds. During the collaborative training process, the number of rounds was dynamically adjusted based on the convergence of the joint regression optimization objective to ensure that the model fully learns the signal patterns.

[0060] Then, based on the joint regression optimization objective, namely the mean square error between the first reconstructed signal and the original partial discharge signal of the sample, when the mean square error value is stable within a preset threshold, such as 0.001, for 20 consecutive rounds, and the independent prediction errors of the target signal prediction model and the interference signal prediction model are both lower than their respective thresholds, it is determined that the anti-interference analysis model group has converged.

[0061] The independent prediction error threshold of the target signal prediction model can be set based on the amplitude range and average amplitude of the partial discharge signal components in the sample set. The independent prediction error threshold can be set to 5% to 10% of the average amplitude. For example, if the average amplitude of the partial discharge signal components is 0.5, the independent prediction error threshold of the target signal prediction model can be set to 0.025 to 0.05.

[0062] The independent prediction error threshold of the interference signal prediction model can be set based on the amplitude range and average amplitude of the interference signal components in the sample set. The independent prediction error threshold can be set to 10% to 15% of the average amplitude. For example, if the average amplitude of the interference signal components is 0.3, the prediction error threshold of the independent prediction model can be set to 0.03 to 0.045.

[0063] The specific training method for the anti-interference analysis model group will be explained in detail in subsequent step S300, and will not be repeated here.

[0064] In step S300 of this application embodiment, a joint regression optimization objective is set, and the anti-interference analysis model group is co-trained according to the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained, including: The prediction output of the target signal prediction model is defined as the first prediction signal, and the prediction output of the interference signal prediction model is defined as the second prediction signal. The signal reconstruction result based on the first predicted signal and the second predicted signal is defined as the first reconstructed signal, and the joint regression optimization objective is set as the mean square error between the first reconstructed signal and the original partial discharge signal of the sample. Using the sample partial discharge signal analysis set as sample input, the target signal prediction model and the interference signal prediction model are jointly trained until the joint regression optimization objective converges. The target signal prediction model and the interference signal prediction model are output as the anti-interference analysis model group.

[0065] In this embodiment, the purpose of the above steps is to link the training of the target signal prediction model and the interference signal prediction model by setting a unified optimization objective, thereby ensuring that the prediction results of both can accurately reconstruct the original partial discharge signal. Specifically, by defining "the mean square error between the first reconstructed signal (i.e., the superposition of the first and second predicted signals) and the original partial discharge signal of the sample" as the joint regression optimization objective, the two models are forced to optimize collaboratively during training—ensuring both the accuracy of the partial discharge signal components output by the target signal prediction model and the reasonableness of the interference signal components output by the interference signal prediction model. Ultimately, the superposition of the two signals achieves a high degree of consistency with the original signal, providing constraints for the anti-interference analysis capability of the model group.

[0066] First, the output of the target signal prediction model, i.e., the model's prediction result for the effective partial discharge component, is defined as the first prediction signal. The output of the interference signal prediction model, i.e., the model's prediction result for the invalid interference component, is defined as the second prediction signal.

[0067] Then, the "first reconstructed signal" is calculated, which is to superimpose the first predicted signal and the second predicted signal to simulate the composition logic of the original signal, that is, the original partial discharge signal = partial discharge signal component + interference signal component.

[0068] Next, a joint regression optimization objective is set, which is the mean square error between the first reconstructed signal and the original partial discharge signal of the sample. The smaller this mean square error, the closer the superimposed prediction results of the two models are to the true original signal, that is, the better the synergistic effect of the two.

[0069] During training, the optimization objective will have a reverse effect on the parameter adjustment of the two models: if the mean square error is large, it may indicate that the first prediction signal is inaccurate or the second prediction signal is wrong. The model will optimize the parameters of the two sub-models simultaneously until the mean square error reaches the convergence criterion.

[0070] Through this mechanism, the two models are no longer trained independently, but are constrained by the common goal of "reconstructing the original signal". This avoids signal separation bias caused by over-optimization of a single model, and ultimately improves the overall performance of the anti-interference analysis model group.

[0071] In step S300 of this application embodiment, the sample partial discharge signal analysis set is used as sample input, and the target signal prediction model and the interference signal prediction model are jointly trained until the joint regression optimization objective converges, including: Based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the components of the sample partial discharge signal and the sample environmental information are extracted to obtain the first sample analysis set; Based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the interference signal components of the sample, and the sample environmental information are extracted to obtain a second sample analysis set; Based on the first sample analysis set and the second sample analysis set respectively, the target signal prediction model is independently trained with the sample partial discharge signal component as supervision, and the interference signal prediction model is independently trained with the sample interference signal component as supervision; The target signal prediction model and the interference signal prediction model are combined in parallel with the allocation layer and trained collaboratively with the sample partial discharge signal analysis set until the joint regression optimization objective satisfies the preset target convergence constraint.

[0072] In this embodiment, the purpose of the above steps is to ensure, through a phased training strategy of independent training followed by collaborative optimization, that the anti-interference analysis model group can accurately learn the characteristics of partial discharge signal components and interference signal components respectively, and achieve a high degree of matching between the prediction results of the two and the original signal through collaborative constraints. Ultimately, this enables the model group to form an overall collaborative capability while ensuring the accuracy of its individual predictions, thereby improving the reliability of anti-interference analysis.

[0073] First, based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the components of the sample partial discharge signal, and the sample environmental information need to be extracted to obtain a first sample analysis set. Based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the components of the sample interference signal, and the sample environmental information are extracted to obtain a second sample analysis set.

[0074] Specifically, the original partial discharge (PD) signals, PD signal components, and environmental information are extracted from the sample PD signal analysis set and combined to form the first sample analysis set, which is used for independent training of the target signal prediction model. The original PD signals, interference signal components, and environmental information are extracted from the sample PD signal analysis set and used for independent training of the interference signal prediction model. The purpose of this splitting is to provide targeted supervision data for both models, ensuring input-output matching during the independent training phase.

[0075] Next, the target signal prediction model needs to be independently trained based on the first sample analysis set and the second sample analysis set, with the sample partial discharge signal component as supervision, and the interference signal prediction model needs to be independently trained with the sample interference signal component as supervision.

[0076] Specifically, for the target signal prediction model, the first sample analysis set needs to be used as input, and the sample partial discharge signal components need to be used as supervision labels. The model is trained to learn the mapping relationship between "original partial discharge signal + environmental information" and "partial discharge signal components". For example, by adjusting the model parameters, the predicted partial discharge signal components can be made as close as possible to the real partial discharge signal components in the sample.

[0077] For the interference signal prediction model, the second sample analysis set is required as input, and the sample interference signal components are used as supervision labels to train the model to learn the mapping relationship between "original partial discharge signal + environmental information" and "interference signal components". For example, the predicted interference signal components should be as close as possible to the real interference signal components in the sample.

[0078] The purpose of independent training is to allow the two models to first grasp the basic rules of their respective core tasks, so as to avoid excessive deviation in initial parameters that could affect subsequent collaborative training.

[0079] Furthermore, it is necessary to combine the allocation layer, connect the target signal prediction model and the interference signal prediction model in parallel, and perform collaborative training with the sample partial discharge signal analysis set until the joint regression optimization objective satisfies the preset target convergence constraint.

[0080] First, an allocation layer needs to be constructed as a connection structure between the two models. This layer receives the first predicted signal output by the target signal prediction model and the second predicted signal output by the interference signal prediction model, and performs a signal superposition operation to generate the first reconstructed signal.

[0081] Then, the two models are connected in parallel and trained together: the two models are connected in parallel through an allocation layer, with the sample partial discharge signal analysis set as input and the joint regression optimization objective, namely the mean square error between the first reconstructed signal and the original sample partial discharge signal, as the optimization direction, while adjusting the parameters of the two models.

[0082] Continue training until the joint regression optimization objective meets the preset convergence constraint, that is, the convergence condition set in step S200, to ensure that the prediction results of the anti-interference analysis model group can accurately restore the original signal after superposition.

[0083] This step involves first allowing each model to learn its own task independently, and then optimizing the overall training process through collaborative constraints. This ensures the prediction accuracy of individual models while resolving the overall mismatch that may result from independent training, ultimately enabling the anti-interference analysis model group to possess stable anti-interference analysis capabilities.

[0084] In step S400 of this embodiment, the target device is interacted with to collect real-time raw partial discharge signals and real-time environmental information, which are then used as input to the anti-interference analysis model group for partial discharge signal analysis, including: Input the real-time raw partial discharge signal and the real-time environmental information into the target signal prediction model to obtain the corresponding predicted partial discharge signal components; Input the real-time raw partial discharge signal and the real-time environmental information into the interference signal prediction model to obtain the corresponding predicted interference signal components; The predicted partial discharge signal component and the predicted interference signal component are superimposed to obtain the predicted restored signal, and the fitting error between the predicted restored signal and the real-time original partial discharge signal is calculated by comparison. If the fitting error is less than a preset error limit, the predicted partial discharge signal component is output as the partial discharge signal analysis result.

[0085] The core objective of step S400 in this embodiment is to accurately extract effective partial discharge signal components from the real-time mixed signal of the target device, i.e., the real-time raw partial discharge signal, using a trained anti-interference analysis model set, and to ensure the reliability of the analysis results through fitting error verification. By combining real-time environmental information to separate the partial discharge signal from the interference signal, accurate partial discharge signal data is ultimately provided for equipment status monitoring and fault diagnosis, thereby achieving anti-interference analysis of partial discharge.

[0086] First, the raw, unprocessed partial discharge (PD) signal from the target device, along with real-time environmental information, is simultaneously input into the target signal prediction model to obtain the predicted PD signal components output by the model. Then, the same raw PD signal and environmental information are input into the interference signal prediction model to obtain the predicted interference signal components output by the model.

[0087] Next, the predicted partial discharge signal component and the predicted interference signal component need to be superimposed to obtain the predicted restored signal, and the fitting error between the predicted restored signal and the real-time original partial discharge signal needs to be calculated. That is, the predicted partial discharge signal component and the predicted interference signal component are superimposed to generate the predicted restored signal to simulate the composition of the real-time original partial discharge signal.

[0088] Then, the fitting error between the predicted restored signal and the real-time original partial discharge signal is calculated. This error can be measured by the mean square error, which verifies whether the prediction results of the anti-interference analysis model group are reasonable. If the fitting error is small, it indicates that the separation of the partial discharge signal and the interference signal components is accurate; if the error is large, there may be a prediction bias. A preset error limit is set, which can be determined based on the signal characteristics and engineering accuracy requirements of the sample partial discharge signal analysis set, and is consistent with the threshold setting logic of the convergence standard in step S300.

[0089] If the fitting error is less than the error limit, it indicates that the predicted partial discharge signal component is accurate and reliable, and the predicted partial discharge signal component is output as the final partial discharge signal analysis result; if the error exceeds the threshold, the signal acquisition or model status needs to be checked again to avoid outputting incorrect results.

[0090] Example 2, as Figure 2 As shown, based on the same inventive concept as the anti-interference analysis method for partial discharge signals provided in Embodiment 1, this embodiment of the invention also provides an anti-interference analysis system for partial discharge signals, comprising: The partial discharge signal analysis set acquisition module 11 is used to acquire the sample partial discharge signal analysis set of the target device, wherein the sample partial discharge signal analysis set includes the associated stored original sample partial discharge signal, sample partial discharge signal components, sample interference signal components and sample environmental information; Anti-interference analysis model group construction module 12 is used to construct an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model; The anti-interference analysis model group training module 13 is used to set a joint regression optimization objective and perform collaborative training on the anti-interference analysis model group according to the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained. The input information real-time acquisition module 14 is used to interact with the target device, acquire real-time raw partial discharge signals and real-time environmental information, and use them as input to the anti-interference analysis model group for partial discharge signal analysis.

[0091] Furthermore, the partial discharge signal analysis set acquisition module 11 includes the following execution steps: Based on the device model of the target device, retrieve historical analysis records and perform confidence filtering to obtain historical partial discharge signal analysis samples; Random crossover and mutation are performed on the historical partial discharge signal analysis set to obtain the first expanded partial discharge signal analysis sample; Adversarial training is performed based on the confidence screening results to obtain a sample generator and a sample validator, and a second expanded partial discharge signal analysis sample is generated through the sample generator. Based on the sample validator, the first expanded partial discharge signal analysis sample and the second expanded partial discharge signal analysis sample are subjected to credibility verification screening respectively, and the credibility verification screening results are merged with the historical partial discharge signal analysis samples to output the sample partial discharge signal analysis set.

[0092] Furthermore, the anti-interference analysis model training module 13 includes the following execution steps: Setting a joint regression optimization objective, and performing collaborative training on the anti-interference analysis model set based on the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model set is obtained, including: The prediction output of the target signal prediction model is defined as the first prediction signal, and the prediction output of the interference signal prediction model is defined as the second prediction signal. The signal reconstruction result based on the first predicted signal and the second predicted signal is defined as the first reconstructed signal, and the joint regression optimization objective is set as the mean square error between the first reconstructed signal and the original partial discharge signal of the sample. Using the sample partial discharge signal analysis set as sample input, the target signal prediction model and the interference signal prediction model are jointly trained until the joint regression optimization objective converges. The target signal prediction model and the interference signal prediction model are output as the anti-interference analysis model group.

[0093] Specifically, the sample partial discharge signal analysis set is used as the sample input, and the target signal prediction model and the interference signal prediction model are jointly trained until the joint regression optimization objective converges, including: Based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the components of the sample partial discharge signal and the sample environmental information are extracted to obtain the first sample analysis set; Based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the interference signal components of the sample, and the sample environmental information are extracted to obtain a second sample analysis set; Based on the first sample analysis set and the second sample analysis set respectively, the target signal prediction model is independently trained with the sample partial discharge signal component as supervision, and the interference signal prediction model is independently trained with the sample interference signal component as supervision; The target signal prediction model and the interference signal prediction model are combined in parallel with the allocation layer and trained collaboratively with the sample partial discharge signal analysis set until the joint regression optimization objective satisfies the preset target convergence constraint.

[0094] Furthermore, the real-time input information acquisition module 14 includes the following execution steps: Input the real-time raw partial discharge signal and the real-time environmental information into the target signal prediction model to obtain the corresponding predicted partial discharge signal components; Input the real-time raw partial discharge signal and the real-time environmental information into the interference signal prediction model to obtain the corresponding predicted interference signal components; The predicted partial discharge signal component and the predicted interference signal component are superimposed to obtain the predicted restored signal, and the fitting error between the predicted restored signal and the real-time original partial discharge signal is calculated by comparison. If the fitting error is less than a preset error limit, the predicted partial discharge signal component is output as the partial discharge signal analysis result.

[0095] Example 3, as Figure 3 As shown, based on the same inventive concept as the anti-interference analysis method for partial discharge signals provided in Embodiment 1, this embodiment of the invention also provides a storage medium 400. For example, the storage medium can be a non-transitory computer-readable storage medium, and the storage medium stores a first computer program 410. When the first computer program 410 is executed by a processor, it implements the anti-interference analysis method for partial discharge signals as described in Embodiment 1.

[0096] The non-transitory storage medium refers to a storage medium that can still retain data persistently after power failure, including but not limited to SSDs (solid-state drives), HDDs (hard disk drives), and flash memory devices (USB flash drives, memory cards), etc.

[0097] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0098] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0103] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for anti-interference analysis of partial discharge signals, characterized in that, include: A sample partial discharge signal analysis set of the target device is obtained, wherein the sample partial discharge signal analysis set includes associated stored original sample partial discharge signals, sample partial discharge signal components, sample interference signal components and sample environmental information; Construct an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model; Set a joint regression optimization objective, and perform collaborative training on the anti-interference analysis model group according to the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained; The interactive target device collects real-time raw partial discharge signals and real-time environmental information, which are then used as inputs to the anti-interference analysis model group for partial discharge signal analysis.

2. The anti-interference analysis method for partial discharge signals as described in claim 1, characterized in that, A sample partial discharge signal analysis set of the target device is obtained, wherein the sample partial discharge signal analysis set includes associated stored original sample partial discharge signals, sample partial discharge signal components, sample interference signal components, and sample environmental information, including: Based on the device model of the target device, retrieve historical analysis records and perform confidence filtering to obtain historical partial discharge signal analysis samples; Random crossover and mutation are performed on the historical partial discharge signal analysis set to obtain the first expanded partial discharge signal analysis sample; Adversarial training is performed based on the confidence screening results to obtain a sample generator and a sample validator, and a second expanded partial discharge signal analysis sample is generated through the sample generator. Based on the sample validator, the first expanded partial discharge signal analysis sample and the second expanded partial discharge signal analysis sample are subjected to credibility verification screening respectively, and the credibility verification screening results are merged with the historical partial discharge signal analysis samples to output the sample partial discharge signal analysis set.

3. The anti-interference analysis method for partial discharge signals as described in claim 2, characterized in that, The target signal prediction model takes the original partial discharge signal and environmental information as input and the partial discharge signal components as output, and is used to predict the partial discharge signal components based on the original partial discharge signal and environmental information.

4. The anti-interference analysis method for partial discharge signals as described in claim 3, characterized in that, The interference signal prediction model takes the original partial discharge signal and environmental information as input and the interference signal components as output, and is used to predict the interference signal components based on the environmental information.

5. The anti-interference analysis method for partial discharge signals as described in claim 4, characterized in that, Setting a joint regression optimization objective, and performing collaborative training on the anti-interference analysis model set based on the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model set is obtained, including: The prediction output of the target signal prediction model is defined as the first prediction signal, and the prediction output of the interference signal prediction model is defined as the second prediction signal. The signal reconstruction result based on the first predicted signal and the second predicted signal is defined as the first reconstructed signal, and the joint regression optimization objective is set as the mean square error between the first reconstructed signal and the original partial discharge signal of the sample. Using the sample partial discharge signal analysis set as sample input, the target signal prediction model and the interference signal prediction model are jointly trained until the joint regression optimization objective converges. The target signal prediction model and the interference signal prediction model are output as the anti-interference analysis model group.

6. The anti-interference analysis method for partial discharge signals as described in claim 5, characterized in that, Using the sample partial discharge signal analysis set as sample input, the target signal prediction model and the interference signal prediction model are jointly trained until the joint regression optimization objective converges, including: Based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the components of the sample partial discharge signal and the sample environmental information are extracted to obtain the first sample analysis set; Based on the sample partial discharge signal analysis set, the original partial discharge signal of the sample, the interference signal components of the sample, and the sample environmental information are extracted to obtain a second sample analysis set; Based on the first sample analysis set and the second sample analysis set respectively, the target signal prediction model is independently trained with the sample partial discharge signal component as supervision, and the interference signal prediction model is independently trained with the sample interference signal component as supervision; The target signal prediction model and the interference signal prediction model are combined in parallel with the allocation layer and trained collaboratively with the sample partial discharge signal analysis set until the joint regression optimization objective satisfies the preset target convergence constraint.

7. The anti-interference analysis method for partial discharge signals as described in claim 6, characterized in that, The interactive target device acquires real-time raw partial discharge signals and real-time environmental information, and uses these as inputs to the anti-interference analysis model group for partial discharge signal analysis, including: Input the real-time raw partial discharge signal and the real-time environmental information into the target signal prediction model to obtain the corresponding predicted partial discharge signal components; Input the real-time raw partial discharge signal and the real-time environmental information into the interference signal prediction model to obtain the corresponding predicted interference signal components; The predicted partial discharge signal component and the predicted interference signal component are superimposed to obtain the predicted restored signal, and the fitting error between the predicted restored signal and the real-time original partial discharge signal is calculated by comparison. If the fitting error is less than a preset error limit, the predicted partial discharge signal component is output as the partial discharge signal analysis result.

8. An anti-interference analysis system for partial discharge signals, characterized in that, An anti-interference analysis method for implementing the partial discharge signal according to any one of claims 1 to 7 includes: The partial discharge signal analysis set acquisition module is used to acquire the sample partial discharge signal analysis set of the target device, wherein the sample partial discharge signal analysis set includes the associated stored original sample partial discharge signal, sample partial discharge signal components, sample interference signal components and sample environmental information; An anti-interference analysis model group construction module is used to construct an anti-interference analysis model group, wherein the anti-interference analysis model group includes a target signal prediction model and an interference signal prediction model; The anti-interference analysis model group training module is used to set a joint regression optimization objective and perform collaborative training on the anti-interference analysis model group according to the joint regression optimization objective and the sample partial discharge signal analysis set until a converged anti-interference analysis model group is obtained. The input information real-time acquisition module is used to interact with the target device, acquire real-time raw partial discharge signals and real-time environmental information, and use them as input to the anti-interference analysis model group for partial discharge signal analysis.

9. A storage medium, characterized in that, The storage medium stores a first computer program, which, when executed by a processor, implements the anti-interference analysis method for partial discharge signals as described in any one of claims 1-7.

Images