Insulating oil aging diagnosis method and device based on Raman spectrum, electronic equipment and storage medium

Abstract

The invention discloses a Raman spectrum-based insulating oil aging diagnosis method and device, electronic equipment and a storage medium, and belongs to the field of power equipment safety state assessment, and the method comprises the following steps: obtaining Raman spectrum data of to-be-diagnosed insulating oil; inputting the Raman spectrum data into a preset feature extraction neural network model, so that the feature extraction neural network model generates a corresponding low-dimensional feature vector according to the Raman spectrum data; inputting the low-dimensional feature vector into a preset classifier model, so that the classifier model generates an aging diagnosis result of the to-be-diagnosed insulating oil according to the low-dimensional feature vector; therefore, by implementing the method and the device, the problem that the accuracy and the reliability of diagnosis are influenced due to insufficient distinction degree of different aging stages on the feature level in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of power equipment safety status assessment technology, specifically to a method, device, electronic device, and storage medium for diagnosing the aging of insulating oil based on Raman spectroscopy. Background Technology

[0002] Oil-paper insulation systems are a critical component of high-voltage power equipment such as oil-immersed transformers, and their insulation health directly affects the safe and stable operation of the power grid. Therefore, accurate and timely diagnosis of the aging condition of oil-paper insulation provides important data for equipment condition assessment and lifespan prediction, and is a key technical means to prevent major power accidents.

[0003] Raman spectroscopy, as a non-destructive testing technique capable of reflecting the molecular structure of matter, has shown potential in the diagnosis of insulating oil aging. However, the chemical composition changes of insulating oil during aging are extremely complex. These changes are reflected in Raman spectra as weak "fingerprint" information distributed throughout the high-dimensional data, with very subtle spectral differences between different aging stages. Therefore, the key challenge of existing technologies lies in how to extract a low-dimensional feature from high-dimensional, complex, and noisy Raman spectral data that can effectively characterize and clearly distinguish different aging stages. Conventional feature extraction methods often struggle to learn an optimal feature representation, resulting in insufficient distinguishability between different aging stages at the feature level, thus affecting the accuracy and reliability of the diagnosis. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for aging diagnosis of insulating oil based on Raman spectroscopy, which can solve the problem of insufficient differentiation of different aging stages at the characteristic level in the prior art, thus affecting the accuracy and reliability of the diagnosis.

[0005] An embodiment of the present invention provides a method for diagnosing the aging of insulating oil based on Raman spectroscopy, comprising:

[0006] Obtain the Raman spectral data of the insulating oil to be diagnosed;

[0007] The Raman spectral data is input into a preset feature extraction neural network model, so that the feature extraction neural network model generates a corresponding low-dimensional feature vector based on the Raman spectral data;

[0008] The low-dimensional feature vector is input into a preset classifier model so that the classifier model generates the aging diagnosis result of the insulating oil to be diagnosed based on the low-dimensional feature vector.

[0009] The training of the feature extraction neural network model includes:

[0010] Obtain several triplet samples; each triplet sample is constructed based on known insulating oil aging stage labels, including an anchor sample, a positive sample belonging to the same insulating oil aging stage as the anchor sample, and a negative sample belonging to a different insulating oil aging stage than the anchor sample.

[0011] The triplet samples are divided into several batches of training samples according to a preset batch size.

[0012] Each batch of training samples is sequentially input into the feature extraction neural network model for iterative training until the preset number of training iterations is reached.

[0013] When the feature extraction neural network model receives a batch of training samples, it generates corresponding anchor feature vectors, positive sample feature vectors, and negative sample feature vectors based on the Raman spectral data of the anchor sample, positive sample, and negative sample contained in each triplet sample in the current batch. Based on a triple loss function, with the goal of narrowing the distance between the anchor feature vector and the positive sample feature vector and widening the distance between the anchor feature vector and the negative sample feature vector, a loss function value is calculated. Using a preset optimizer, the learnable parameters in the feature extraction neural network model are updated according to the loss function value, thus completing the training of the feature extraction neural network model.

[0014] Furthermore, the triple loss function is specifically as follows:

[0015] L=max(d(A,P)-d(A,N)+α,0)

[0016] In the formula, L is the loss function value of the triple loss function; A is the anchor sample; P is the positive sample; N is the negative sample; d(A, P) is the anchor-positive sample distance; d(A, N) is the anchor-negative sample distance; α is the preset boundary value; max(·) is the maximum value function.

[0017] Furthermore, the training of the classifier model includes:

[0018] Obtain an insulating oil Raman spectrum training dataset; wherein, the insulating oil Raman spectrum training dataset includes several insulating oil Raman spectrum training samples and known aging stage labels corresponding to each training sample;

[0019] The training samples of the insulating oil Raman spectrum are input into the feature extraction neural network model so that the feature extraction neural network model generates several low-dimensional training feature vectors corresponding to each training sample based on the training samples of the insulating oil Raman spectrum.

[0020] Based on the aforementioned low-dimensional training feature vectors and their corresponding aging stage labels, with the objective of maximizing the geometric margin between feature vectors belonging to different aging stages, an optimal classification hyperplane for dividing the aforementioned low-dimensional training feature vectors is solved and determined, thereby completing the training of the classifier model.

[0021] Furthermore, after acquiring the Raman spectral data of the insulating oil to be diagnosed, the method further includes:

[0022] Baseline correction is performed on the Raman spectral data to generate baseline-corrected spectral data;

[0023] The baseline-corrected spectral data is smoothed to generate smoothed spectral data;

[0024] The smoothed spectral data is normalized to generate normalized spectral data;

[0025] The Raman spectral data are updated based on the normalized spectral data.

[0026] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0027] An embodiment of the present invention provides an insulating oil aging diagnosis device based on Raman spectroscopy, comprising: a Raman spectroscopy data acquisition module, a low-dimensional feature vector generation module, and an aging stage diagnosis result generation module;

[0028] The Raman spectroscopy data acquisition module is used to acquire the Raman spectroscopy data of the insulating oil to be diagnosed.

[0029] The low-dimensional feature vector generation module is used to input the Raman spectral data into a preset feature extraction neural network model, so that the feature extraction neural network model generates a corresponding low-dimensional feature vector based on the Raman spectral data.

[0030] The training of the feature extraction neural network model includes:

[0031] Obtain several triplet samples; each triplet sample is constructed based on known insulating oil aging stage labels, including an anchor sample, a positive sample belonging to the same insulating oil aging stage as the anchor sample, and a negative sample belonging to a different insulating oil aging stage than the anchor sample.

[0032] The triplet samples are divided into several batches of training samples according to a preset batch size.

[0033] Each batch of training samples is sequentially input into the feature extraction neural network model for iterative training until the preset number of training iterations is reached.

[0034] When the feature extraction neural network model receives a batch of training samples, it generates corresponding anchor feature vectors, positive sample feature vectors, and negative sample feature vectors based on the Raman spectral data of the anchor sample, positive sample, and negative sample contained in each triplet sample in the current batch. Based on a triple loss function, with the goal of narrowing the distance between the anchor feature vector and the positive sample feature vector and widening the distance between the anchor feature vector and the negative sample feature vector, a loss function value is calculated. Using a preset optimizer, the learnable parameters in the feature extraction neural network model are updated according to the loss function value, thus completing the training of the feature extraction neural network model.

[0035] The aging stage diagnosis result generation module is used to input the low-dimensional feature vector into a preset classifier model, so that the classifier model generates the aging diagnosis result of the insulating oil to be diagnosed based on the low-dimensional feature vector.

[0036] Furthermore, the triple loss function mentioned in the feature extraction neural network model training module is specifically as follows:

[0037] L = max(d(A,P) - d(A,N) + α,0)

[0038] In the formula, L is the loss function value of the triple loss function; A is the anchor sample; P is the positive sample; N is the negative sample; d(A,P) is the anchor-positive sample distance; d(A,N) is the anchor-negative sample distance; α is the preset boundary value; max(·) is the maximum value function.

[0039] Furthermore, the Raman spectroscopy-based insulating oil aging diagnostic device also includes: a classifier model training module;

[0040] The classifier model training module is also used to obtain the insulating oil Raman spectrum training dataset; wherein, the insulating oil Raman spectrum training dataset includes several insulating oil Raman spectrum training samples and known aging stage labels corresponding to each training sample;

[0041] The training samples of the insulating oil Raman spectrum are input into the feature extraction neural network model so that the feature extraction neural network model generates several low-dimensional training feature vectors corresponding to each training sample based on the training samples of the insulating oil Raman spectrum.

[0042] Based on the aforementioned low-dimensional training feature vectors and their corresponding aging stage labels, with the objective of maximizing the geometric margin between feature vectors belonging to different aging stages, an optimal classification hyperplane for dividing the aforementioned low-dimensional training feature vectors is solved and determined, thereby completing the training of the classifier model.

[0043] Furthermore, the Raman spectroscopy-based insulating oil aging diagnostic device also includes: a pretreatment module;

[0044] The preprocessing module is used to perform baseline correction on the Raman spectral data to generate baseline-corrected spectral data;

[0045] The baseline-corrected spectral data is smoothed to generate smoothed spectral data;

[0046] The smoothed spectral data is normalized to generate normalized spectral data;

[0047] The Raman spectral data are updated based on the normalized spectral data.

[0048] Based on the above method embodiments, the present invention provides corresponding electronic device embodiments.

[0049] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the Raman spectroscopy-based insulating oil aging diagnosis method described in any of the above-described method embodiments.

[0050] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0051] An embodiment of the present invention provides a storage medium storing a computer program thereon, wherein, when the computer program is running, it controls the device where the storage medium is located to execute any of the above-described Raman spectroscopy-based insulating oil aging diagnosis methods in the method embodiments.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] This invention provides a method, apparatus, electronic device, and storage medium for aging diagnosis of insulating oil based on Raman spectroscopy. The method first acquires the Raman spectral data of the insulating oil to be diagnosed and then extracts its low-dimensional feature vector using a preset feature extraction neural network model. Subsequently, the extracted feature vector is input into a classifier model, which outputs the corresponding aging stage diagnosis result. The feature extraction neural network is trained using a contrastive learning approach, calculating a triple loss function based on constructed triplet samples (anchor sample, positive sample, and negative sample), and continuously updating the network parameters with the help of an optimizer to enhance the ability to distinguish different aging stages, thereby improving the accuracy and generalization ability of the diagnostic model.

[0054] This invention processes high-dimensional Raman spectral data using a neural network model trained through contrastive learning based on a triple loss function. The training process actively learns and constructs a novel low-dimensional feature space by continuously comparing triplets composed of anchor points, positive samples, and negative samples. In this feature space, features of samples belonging to the same aging stage are forced closer together, while features of samples belonging to different aging stages are forced further apart. Therefore, the low-dimensional feature vectors generated by this invention have significantly enhanced intra-class compactness and inter-class separation, making subtle aging differences that were previously difficult to distinguish spectrally clearly discernible, thereby greatly improving the accuracy and reliability of subsequent classifier diagnosis. Attached Figure Description

[0055] Figure 1 This is a schematic flowchart of an insulating oil aging diagnosis method based on Raman spectroscopy provided in an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of the structure of an insulating oil aging diagnostic device based on Raman spectroscopy provided in an embodiment of the present invention. Detailed Implementation

[0057] 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.

[0058] like Figure 1 As shown, to address the problem of insufficient differentiation at the characteristic level among different aging stages in existing technologies, which affects the accuracy and reliability of diagnosis, an embodiment of the present invention provides a Raman spectroscopy-based method for diagnosing the aging of insulating oil, comprising at least the following steps:

[0059] Step S1: Obtain the Raman spectral data of the insulating oil to be diagnosed.

[0060] Specifically, the first step is to acquire Raman spectral data of the insulating oil to be diagnosed. This step is achieved using a portable diagnostic device integrating a Raman spectrometer. Specifically, the insulating oil sample to be diagnosed is placed at the detection position of the Raman spectrometer, and the sample is irradiated by the device's built-in excitation light source to generate a Raman scattering signal. The Raman spectrometer acquires this Raman scattering signal and converts it into high-dimensional digital spectral data through its internal spectral analysis unit. To improve the signal-to-noise ratio, multiple measurements can be performed on the same insulating oil sample, and the signals can be integrated. This step allows for the non-destructive and rapid acquisition of raw data reflecting the molecular structure information of the insulating oil, laying the foundation for subsequent feature extraction and diagnosis.

[0061] In a preferred embodiment, after acquiring the Raman spectral data of the insulating oil to be diagnosed, the method further includes:

[0062] Baseline correction is performed on the Raman spectral data to generate baseline-corrected spectral data;

[0063] The baseline-corrected spectral data is smoothed to generate smoothed spectral data;

[0064] The smoothed spectral data is normalized to generate normalized spectral data;

[0065] The Raman spectral data are updated based on the normalized spectral data.

[0066] Specifically, after acquiring the Raman spectral data of the insulating oil to be diagnosed, the process includes a preprocessing step. First, a cubic spline function is used to perform baseline correction on the original Raman spectral data. This step involves fitting the background baseline of the spectral signal and subtracting it from the original signal to generate baseline-corrected spectral data that better reflects the true Raman scattering signal. Next, the baseline-corrected spectral data is smoothed using a five-point cubic smoothing algorithm. This step uses polynomial fitting of local data points to filter out high-frequency random noise in the signal, generating spectral data with a smoother signal curve. Finally, the smoothed spectral data is normalized. This step adjusts the spectral intensities of different samples to a uniform numerical range to eliminate overall intensity variations caused by differences in laser power or detection conditions.

[0067] Step S2: Input the Raman spectral data into a preset feature extraction neural network model so that the feature extraction neural network model generates a corresponding low-dimensional feature vector based on the Raman spectral data.

[0068] It should be noted here that the training of the feature extraction neural network model includes:

[0069] Obtain several triplet samples; each triplet sample is constructed based on known insulating oil aging stage labels, including an anchor sample, a positive sample belonging to the same insulating oil aging stage as the anchor sample, and a negative sample belonging to a different insulating oil aging stage than the anchor sample.

[0070] The triplet samples are divided into several batches of training samples according to a preset batch size.

[0071] Each batch of training samples is sequentially input into the feature extraction neural network model for iterative training until the preset number of training iterations is reached.

[0072] When the feature extraction neural network model receives a batch of training samples, it generates corresponding anchor feature vectors, positive sample feature vectors, and negative sample feature vectors based on the Raman spectral data of the anchor sample, positive sample, and negative sample contained in each triplet sample in the current batch. Based on a triple loss function, with the goal of narrowing the distance between the anchor feature vector and the positive sample feature vector and widening the distance between the anchor feature vector and the negative sample feature vector, a loss function value is calculated. Using a preset optimizer, the learnable parameters in the feature extraction neural network model are updated according to the loss function value, thus completing the training of the feature extraction neural network model.

[0073] In a preferred embodiment, the triple loss function is specifically:

[0074] L=max(d(A,P)-d(A,N)+α,0)

[0075] In the formula, L is the loss function value of the triple loss function; A is the anchor sample; P is the positive sample; N is the negative sample; d(A, P) is the anchor-positive sample distance; d(A, N) is the anchor-negative sample distance; α is the preset boundary value; max(·) is the maximum value function.

[0076] Specifically, when processing the pre-processed Raman spectral data, the data is input into a pre-trained feature extraction neural network model. The core function of this model is to transform high-dimensional, complex spectral data into a low-dimensional feature vector that can effectively characterize the aging state of the sample through nonlinear mapping. This low-dimensional feature vector condenses the key aging-related information in the original spectrum and discards redundant and noisy information, thereby transforming the originally difficult-to-compare spectral data into a feature representation that is mathematically easy to measure and classify.

[0077] It should be noted here that the training process of the feature extraction neural network model is the key to this invention. Based on the principle of contrastive learning, first, a Raman spectroscopy training dataset containing multiple known aging stages needs to be obtained, and according to the aging stage labels corresponding to each sample, several triplet samples composed of anchor samples, positive samples, and negative samples are constructed. During iterative training, these triplet samples are input into the feature extraction neural network model in batches, and the model will generate corresponding feature vectors for each sample in the triplet. Subsequently, based on the triplet loss function, the distance between the anchor feature vector and the positive sample feature vector, as well as the distance between the anchor feature vector and the negative sample feature vector, are calculated. The goal of this loss function is to adjust the internal learnable parameters of the model through an optimizer, so as to continuously拉近 the distance between similar samples (anchor and positive samples) and continuously push away the distance between dissimilar samples (anchor and negative samples) in the output low-dimensional feature space until the training reaches the preset number of times, and finally complete the training of the feature extraction neural network model.

[0078] Through this training method based on metric learning, the feature extraction neural network model is optimized into an efficient feature embedding tool, and the generated low-dimensional feature vectors have a high degree of intra-class compactness and inter-class separability, providing a decisive basis for the subsequent classifier to achieve high-precision and high-robustness diagnosis.

[0079] In a specific embodiment, the known aging stage labels can be divided according to the degree of polymerization (DP) value of the insulating paper, specifically including: the good stage corresponding to DP>500, the initial aging stage corresponding to 250<DP≤500, the intermediate aging stage corresponding to 150<DP≤250, and the final aging stage corresponding to DP≤150.

[0080] Step S3: Input the low-dimensional feature vector into a preset classifier model, so that the classifier model generates an aging diagnosis result of the to-be-diagnosed insulating oil according to the low-dimensional feature vector.

[0081] In a preferred embodiment, the training of the classifier model includes:

[0082] Obtain a Raman spectroscopy training dataset of insulating oil; wherein, the Raman spectroscopy training dataset of insulating oil includes several Raman spectroscopy training samples of insulating oil and the known aging stage labels corresponding to each training sample;

[0083] Input the Raman spectroscopy training samples of insulating oil into the feature extraction neural network model, so that the feature extraction neural network model generates several low-dimensional training feature vectors corresponding to each training sample according to the Raman spectroscopy training samples of insulating oil;

[0084] Based on the aforementioned low-dimensional training feature vectors and their corresponding aging stage labels, with the objective of maximizing the geometric margin between feature vectors belonging to different aging stages, an optimal classification hyperplane for dividing the aforementioned low-dimensional training feature vectors is solved and determined, thereby completing the training of the classifier model.

[0085] Specifically, after obtaining low-dimensional feature vectors through a feature extraction neural network model, the low-dimensional feature vectors are input into a pre-trained classifier model. The classifier model is preferably a Support Vector Machine (SVM), which uses its internally defined classification hyperplane to determine the category to which the input low-dimensional feature vectors belong, and ultimately generates a diagnostic result that characterizes the aging stage of the oil-paper insulation system corresponding to the insulating oil to be diagnosed.

[0086] In a preferred embodiment, the training process of the classifier model is as follows: First, a training dataset of insulating oil Raman spectra containing multiple known aging stages is obtained. This dataset includes several training samples of insulating oil Raman spectra and labels for the known aging stages corresponding to each training sample. Then, the training samples of insulating oil Raman spectra are input into the previously trained feature extraction neural network model to generate several low-dimensional training feature vectors corresponding to each training sample. Finally, based on these low-dimensional training feature vectors and their corresponding aging stage labels, with the objective of maximizing the geometric margin between feature vectors belonging to different aging stages, an optimal classification hyperplane for dividing the low-dimensional training feature vectors is determined through optimization, thereby completing the training of the classifier model.

[0087] By separating the complex feature learning task from the relatively simple classification task, and using the maximum margin classifier to discriminate the optimized features, this invention can achieve high accuracy and robustness in aging diagnosis with low computational cost.

[0088] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0089] like Figure 2 As shown, an embodiment of the present invention provides an insulating oil aging diagnosis device based on Raman spectroscopy, comprising: a Raman spectroscopy data acquisition module, a low-dimensional feature vector generation module, and an aging stage diagnosis result generation module;

[0090] The Raman spectroscopy data acquisition module is used to acquire the Raman spectroscopy data of the insulating oil to be diagnosed.

[0091] The low-dimensional feature vector generation module is used to input the Raman spectral data into a preset feature extraction neural network model, so that the feature extraction neural network model generates a corresponding low-dimensional feature vector based on the Raman spectral data.

[0092] The training of the feature extraction neural network model includes:

[0093] Obtain several triplet samples; each triplet sample is constructed based on known insulating oil aging stage labels, including an anchor sample, a positive sample belonging to the same insulating oil aging stage as the anchor sample, and a negative sample belonging to a different insulating oil aging stage than the anchor sample.

[0094] The triplet samples are divided into several batches of training samples according to a preset batch size.

[0095] Each batch of training samples is sequentially input into the feature extraction neural network model for iterative training until the preset number of training iterations is reached.

[0096] When the feature extraction neural network model receives a batch of training samples, it generates corresponding anchor feature vectors, positive sample feature vectors, and negative sample feature vectors based on the Raman spectral data of the anchor sample, positive sample, and negative sample contained in each triplet sample in the current batch. Based on a triple loss function, with the goal of narrowing the distance between the anchor feature vector and the positive sample feature vector and widening the distance between the anchor feature vector and the negative sample feature vector, a loss function value is calculated. Using a preset optimizer, the learnable parameters in the feature extraction neural network model are updated according to the loss function value, thus completing the training of the feature extraction neural network model.

[0097] The aging stage diagnosis result generation module is used to input the low-dimensional feature vector into a preset classifier model, so that the classifier model generates the aging diagnosis result of the insulating oil to be diagnosed based on the low-dimensional feature vector.

[0098] In a preferred embodiment, the triple loss function in the feature extraction neural network model training module is specifically:

[0099] L=max(d(A,P)-d(A,N)+α,0)

[0100] In the formula, L is the loss function value of the triple loss function; A is the anchor sample; P is the positive sample; N is the negative sample; d(A, P) is the anchor-positive sample distance; d(A, N) is the anchor-negative sample distance; α is the preset boundary value; max(·) is the maximum value function.

[0101] In a preferred embodiment, the Raman spectroscopy-based insulating oil aging diagnostic device further includes: a classifier model training module;

[0102] The classifier model training module is also used to obtain the insulating oil Raman spectrum training dataset; wherein, the insulating oil Raman spectrum training dataset includes several insulating oil Raman spectrum training samples and known aging stage labels corresponding to each training sample;

[0103] The training samples of the insulating oil Raman spectrum are input into the feature extraction neural network model so that the feature extraction neural network model generates several low-dimensional training feature vectors corresponding to each training sample based on the training samples of the insulating oil Raman spectrum.

[0104] Based on the aforementioned low-dimensional training feature vectors and their corresponding aging stage labels, with the objective of maximizing the geometric margin between feature vectors belonging to different aging stages, an optimal classification hyperplane for dividing the aforementioned low-dimensional training feature vectors is solved and determined, thereby completing the training of the classifier model.

[0105] In a preferred embodiment, the Raman spectroscopy-based insulating oil aging diagnostic device further includes: a pretreatment module;

[0106] The preprocessing module is used to perform baseline correction on the Raman spectral data to generate baseline-corrected spectral data;

[0107] The baseline-corrected spectral data is smoothed to generate smoothed spectral data;

[0108] The smoothed spectral data is normalized to generate normalized spectral data;

[0109] The Raman spectral data are updated based on the normalized spectral data.

[0110] It should be noted that the embodiments of the device described above correspond to the embodiments of the present invention described above, and can realize the insulating oil aging diagnosis method based on Raman spectroscopy described in any one of the present invention. Furthermore, the embodiments of the device described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort.

[0111] Based on the above-described method embodiments of the present invention, a corresponding embodiment of an electronic device is provided.

[0112] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the Raman spectroscopy-based insulating oil aging diagnosis method according to any one of the present invention, or, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments.

[0113] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.

[0114] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0115] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0116] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0117] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments;

[0118] Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is executed, the device where the storage medium is located executes any one of the above-described Raman spectroscopy-based methods for diagnosing insulating oil aging.

[0119] The aforementioned storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0120] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0121] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for diagnosing the aging of insulating oil based on Raman spectroscopy, characterized in that, include: Obtain the Raman spectral data of the insulating oil to be diagnosed; The Raman spectral data is input into a preset feature extraction neural network model, so that the feature extraction neural network model generates a corresponding low-dimensional feature vector based on the Raman spectral data; The low-dimensional feature vector is input into a preset classifier model so that the classifier model generates the aging diagnosis result of the insulating oil to be diagnosed based on the low-dimensional feature vector. The training of the feature extraction neural network model includes: Obtain several triplet samples; each triplet sample is constructed based on known insulating oil aging stage labels, including an anchor sample, a positive sample belonging to the same insulating oil aging stage as the anchor sample, and a negative sample belonging to a different insulating oil aging stage than the anchor sample. The triplet samples are divided into several batches of training samples according to a preset batch size. Each batch of training samples is sequentially input into the feature extraction neural network model for iterative training until the preset number of training iterations is reached. When the feature extraction neural network model receives a batch of training samples, it generates corresponding anchor feature vectors, positive sample feature vectors, and negative sample feature vectors based on the Raman spectral data of the anchor sample, positive sample, and negative sample contained in each triplet sample in the current batch. Based on a triple loss function, with the goal of narrowing the distance between the anchor feature vector and the positive sample feature vector and widening the distance between the anchor feature vector and the negative sample feature vector, a loss function value is calculated. Using a preset optimizer, the learnable parameters in the feature extraction neural network model are updated according to the loss function value, thus completing the training of the feature extraction neural network model.

2. The method for diagnosing the aging of insulating oil based on Raman spectroscopy as described in claim 1, characterized in that, The triple loss function is as follows: L = max(d(A,P) - d(A,N) + α,0) In the formula, L is the loss function value of the triple loss function; A is the anchor sample; P is the positive sample; N is the negative sample; d(A, P) is the anchor-positive sample distance; d(A, N) is the anchor-negative sample distance; α is the preset boundary value; max(·) is the maximum value function.

3. The method for diagnosing the aging of insulating oil based on Raman spectroscopy as described in claim 2, characterized in that, The training of the classifier model includes: Obtain an insulating oil Raman spectrum training dataset; wherein, the insulating oil Raman spectrum training dataset includes several insulating oil Raman spectrum training samples and known aging stage labels corresponding to each training sample; The training samples of the insulating oil Raman spectrum are input into the feature extraction neural network model so that the feature extraction neural network model generates several low-dimensional training feature vectors corresponding to each training sample based on the training samples of the insulating oil Raman spectrum. Based on the aforementioned low-dimensional training feature vectors and their corresponding aging stage labels, with the objective of maximizing the geometric margin between feature vectors belonging to different aging stages, an optimal classification hyperplane for dividing the aforementioned low-dimensional training feature vectors is solved and determined, thereby completing the training of the classifier model.

4. The method for diagnosing the aging of insulating oil based on Raman spectroscopy as described in claim 3, characterized in that, After acquiring the Raman spectral data of the insulating oil to be diagnosed, the method further includes: Baseline correction is performed on the Raman spectral data to generate baseline-corrected spectral data; The baseline-corrected spectral data is smoothed to generate smoothed spectral data; The smoothed spectral data is normalized to generate normalized spectral data; The Raman spectral data are updated based on the normalized spectral data.

5. An insulating oil aging diagnostic device based on Raman spectroscopy, characterized in that, include: Raman spectroscopy data acquisition module, low-dimensional feature vector generation module, and aging stage diagnosis result generation module; The Raman spectroscopy data acquisition module is used to acquire the Raman spectroscopy data of the insulating oil to be diagnosed. The low-dimensional feature vector generation module is used to input the Raman spectral data into a preset feature extraction neural network model, so that the feature extraction neural network model generates a corresponding low-dimensional feature vector based on the Raman spectral data. The training of the feature extraction neural network model includes: Obtain several triplet samples; each triplet sample is constructed based on known insulating oil aging stage labels, including an anchor sample, a positive sample belonging to the same insulating oil aging stage as the anchor sample, and a negative sample belonging to a different insulating oil aging stage than the anchor sample. The triplet samples are divided into several batches of training samples according to a preset batch size. Each batch of training samples is sequentially input into the feature extraction neural network model for iterative training until the preset number of training iterations is reached. When the feature extraction neural network model receives a batch of training samples, it generates corresponding anchor feature vectors, positive sample feature vectors, and negative sample feature vectors based on the Raman spectral data of the anchor sample, positive sample, and negative sample contained in each triplet sample in the current batch. Based on a triple loss function, with the goal of narrowing the distance between the anchor feature vector and the positive sample feature vector and widening the distance between the anchor feature vector and the negative sample feature vector, a loss function value is calculated. Using a preset optimizer, the learnable parameters in the feature extraction neural network model are updated according to the loss function value, thus completing the training of the feature extraction neural network model. The aging stage diagnosis result generation module is used to input the low-dimensional feature vector into a preset classifier model, so that the classifier model generates the aging diagnosis result of the insulating oil to be diagnosed based on the low-dimensional feature vector.

6. The Raman spectroscopy-based insulating oil aging diagnostic device as described in claim 5, characterized in that, The triple loss function mentioned in the feature extraction neural network model training module is as follows: L = max(d(A,P) - d(A,N) + α,0) In the formula, L is the loss function value of the triple loss function; A is the anchor sample; P is the positive sample; N is the negative sample; d(A, P) is the anchor-positive sample distance; d(A, N) is the anchor-negative sample distance; α is the preset boundary value; max(·) is the maximum value function.

7. The Raman spectroscopy-based insulating oil aging diagnostic device as described in claim 6, characterized in that, Also includes: Classifier model training module; The classifier model training module is also used to obtain the insulating oil Raman spectrum training dataset; wherein, the insulating oil Raman spectrum training dataset includes several insulating oil Raman spectrum training samples and known aging stage labels corresponding to each training sample; The training samples of the insulating oil Raman spectrum are input into the feature extraction neural network model so that the feature extraction neural network model generates several low-dimensional training feature vectors corresponding to each training sample based on the training samples of the insulating oil Raman spectrum. Based on the aforementioned low-dimensional training feature vectors and their corresponding aging stage labels, with the objective of maximizing the geometric margin between feature vectors belonging to different aging stages, an optimal classification hyperplane for dividing the aforementioned low-dimensional training feature vectors is solved and determined, thereby completing the training of the classifier model.

8. The Raman spectroscopy-based insulating oil aging diagnostic device as described in claim 7, characterized in that, Also includes: Preprocessing module; The preprocessing module is used to perform baseline correction on the Raman spectral data to generate baseline-corrected spectral data; The baseline-corrected spectral data is smoothed to generate smoothed spectral data; The smoothed spectral data is normalized to generate normalized spectral data; The Raman spectral data are updated based on the normalized spectral data.

9. An electronic device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the Raman spectroscopy-based method for diagnosing the aging of insulating oil as described in any one of claims 1 to 4.

10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the storage medium to perform the Raman spectroscopy-based method for diagnosing insulating oil aging as described in any one of claims 1 to 4.

Images