Transformer multi-parameter fusion state assessment method, device, equipment and medium

Abstract

The invention discloses a transformer multi-parameter fusion state evaluation method, device, equipment and medium, which belongs to the technical field of transformer fault diagnosis. The method comprises the following steps: collecting multiple operating parameters of the transformer and forming a data set; establishing a probability diffusion model based on dynamic conditions, introducing conditional probability distribution to generate target parameters, introducing random processes to construct a continuous function relationship between target parameters and conditional parameters, and obtaining data distribution characteristics of target parameters; constructing evaluation indicators based on the data distribution characteristics of target parameters, and determining abnormal evaluation indicators using KL divergence and multivariate variation contribution measurement; calculating the abnormal contribution degree of each target parameter in the abnormal evaluation indicator based on the abnormal decomposition strategy of multivariate variation contribution measurement, and regarding the target parameter with the largest abnormal contribution as the fault state of the transformer, locating the fault and determining the cause of the fault. The invention improves the accuracy of fault diagnosis and ensures the efficient and stable operation of the transformer.

Description

Technical Field

[0001] The present invention relates to a transformer multi-parameter fusion state assessment method, device, equipment and medium, specifically a transformer multi-parameter fusion state assessment method, device, equipment and medium based on dynamic conditional probability diffusion, belonging to the technical field of transformer state assessment. Background Art

[0002] Transformers are key components in power systems, and their reliable operation is crucial to the stability of the entire system. However, due to the complex internal structure of the transformer and the harsh working environment, various faults may occur after long-term operation, such as partial discharge, winding deformation, insulation aging, etc. If these faults are not discovered and handled in time, they may cause equipment damage, thereby affecting the safety and reliability of power supply.

[0003] Existing transformer fault diagnosis technologies mainly include methods based on oil chromatography analysis, electrical testing, temperature monitoring, etc. Although these methods can reflect the status of the transformer to a certain extent, most of them only focus on a single parameter or a certain aspect of information, and lack a comprehensive assessment of the overall health of the transformer. In addition, with the development of smart grids, the requirements for transformer status monitoring are becoming higher and higher, requiring not only real-time monitoring, but also the ability to predict the occurrence of potential faults.

[0004] Therefore, it is particularly important to develop a transformer fault diagnosis and evaluation technology that can integrate multiple parameter information. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a transformer multi-parameter fusion state assessment method, device, equipment and medium, which can realize a comprehensive assessment of the transformer state, not only improve the accuracy of fault diagnosis, but also provide early warning of potential problems, thereby ensuring efficient and stable operation of the transformer.

[0006] The technical solution adopted by the present invention to solve the technical problem is:

[0007] In a first aspect, an embodiment of the present invention provides a transformer multi-parameter fusion state assessment method, comprising the following steps:

[0008] Step S1, collecting multiple operating parameters of the transformer and forming a data set, wherein the multiple operating parameters include an image, infrared, partial discharge, sound print, and ambient temperature and humidity of the transformer;

[0009] Step S2, establishing a probability diffusion model based on dynamic conditions, introducing conditional probability distribution to generate target parameters, introducing random processes to construct a continuous function relationship between the target parameters and the conditional parameters, and obtaining data distribution characteristics of the target parameters;

[0010] Step S3, constructing an evaluation index based on the data distribution characteristics of the target parameter, and determining an abnormal evaluation index using KL divergence and multivariate variation contribution measurement, wherein the evaluation index is the probability density distribution difference between the generated target parameter and the true target parameter;

[0011] Step S4, based on the abnormal decomposition strategy of multivariate variation contribution measurement, the abnormal contribution degree of each target parameter in the abnormal evaluation index is calculated, and the target parameter with the largest abnormal contribution is regarded as the fault state of the transformer, the fault is located and the cause of the fault is determined.

[0012] As a possible implementation of this embodiment, the probability diffusion model based on dynamic conditions includes a forward diffusion process and a reverse diffusion process.

[0013] The forward diffusion process gradually integrates the noise signal into the model input data, so that the data gradually transforms into pure noise data;

[0014] The reverse diffusion process systematically removes the noise in the data injected during the forward diffusion process, gradually restoring until the original model input data is accurately restored.

[0015] As a possible implementation of this embodiment, the establishment of a probability diffusion model based on dynamic conditions includes the following steps:

[0016] Process from the original data set The forward Markov chain converts this pure input Transformed into a series of perturbed data points , and adopts diffusion conversion kernel;

[0017] The forward diffusion process is defined as:

[0018] ,

[0019] ,

[0020] in, Indicates that when the initial state X0 is known, all states from time 1 to T X t The joint probability distribution of Indicates that the state at the previous moment is known X t-1 In the case of X t The probability distribution of represents the noise level parameter; express X t The distribution follows the mean , the variance is Normal distribution; I is the identity matrix, indicating the size of the variance; is the state variable at time t;

[0021] The sampling is transformed into:

[0022] ,

[0023] in, and ; is a coefficient between 0 and 1;

[0024] It is expressed as:

[0025] ,

[0026] and ;

[0027] The reverse diffusion process is the denoising To restore , given by the Markov chain:

[0028] ,

[0029] ,

[0030] ,

[0031] in , is a trainable denoising function;

[0032] The reverse diffusion process is trained by solving the following optimization problem. Once trained, we can sample from the equation :

[0033] ,

[0034] in represents the noise vector.

[0035] As a possible implementation of this embodiment, the method of introducing conditional probability distribution to generate target parameters and introducing random process to construct a continuous function relationship between the target parameters and the conditional parameters includes the following steps:

[0036] distributed It is defined by expanding the inverse process in the equation to a conditional process:

[0037] ,

[0038] ,

[0039] ;

[0040] Given condition parameters and target parameters , noise target is sampled, and Training is done by minimizing the loss function:

[0041] ,

[0042] in, Represents the noise function Generated by a Gaussian random process, that is ; is a coefficient between 0 and 1 that controls the target parameter and noise Mixing ratio; is a noise term generated by a Gaussian random process, with an expected value of 0 and a variance of Σ; is a noisy target, which consists of the target parameters without noise and Composition, of which Multiply The square root of Multiply by (1- )’s square root; is a noise function, which needs to be trained by minimizing the loss function L(theta); L(theta) is and The sum of the squares of the distances between them, E represents the expected value, and ||·||2 represents the Euclidean norm.

[0043] As a possible implementation of this embodiment, in step S3, the multivariate variation contribution measure is used to evaluate the reliability of data for generating target parameters;

[0044] Using the multivariate variation contribution metric to evaluate the data reliability of generating target parameters includes the following steps:

[0045] Set the set of parameters to be evaluated and the corresponding distribution;

[0046] Calculate the contribution of variation to the data generated by the distribution of each parameter;

[0047] Based on the calculated contribution to variation, the reliability of the data points generated by each distribution is evaluated;

[0048] Outputs the results of the multivariate contribution measure.

[0049] As a possible implementation of this embodiment, in step S3, the KL divergence is used to calculate the anomaly assessment index;

[0050] Calculating the abnormality assessment index using the KL divergence includes the following steps:

[0051] Calculate the difference between the true value and the predicted value of the target parameter;

[0052] Set a sliding time window of length L to continuously intercept KL divergence data;

[0053] The KL divergence data intercepted by the sliding time window is used to calculate the anomaly evaluation index;

[0054] According to the fluctuation range of the abnormal evaluation index value under normal operating conditions, the maximum abnormal evaluation index value is calculated and used as the threshold of the abnormal evaluation index.

[0055] As a possible implementation of this embodiment, for a random sample X with a length of n and a dimension of p, the calculation formula of the multivariate variation contribution measure is:

[0056] ,

[0057] ,

[0058] ,

[0059] in, is the contribution of multivariate variation, is the sample mean of the random sample X, and S is the sample covariance matrix of the random sample X.

[0060] As a possible implementation of this embodiment, the expression using KL divergence is:

[0061] Define the true probability density function of the target parameter and the resulting probability density function ;

[0062] KL divergence is defined as:

[0063] ,

[0064] For the KL divergence calculation in discrete cases, its form becomes:

[0065] ,

[0066] x are all possible states;

[0067] When dealing with multiple parameters, it is necessary to calculate the KL divergence of each parameter separately. If the parameter set is , then we get a series of KL divergence values:

[0068] ,

[0069] Set a sliding time window of length L, continuously intercept KL divergence data, and use the intercepted KL divergence data to calculate the anomaly evaluation index:

[0070] ,

[0071] in, is an abnormal evaluation indicator.

[0072] As a possible implementation of this embodiment, step S4 includes the following steps:

[0073] If the abnormal evaluation index value exceeds the preset threshold, the transformer is judged to be in an abnormal operation state;

[0074] The time period from when the abnormal evaluation index value exceeds the threshold for the first time to when it falls below the threshold is defined as the fault operation stage;

[0075] The abnormal evaluation index value within the stage is calculated by taking the fault operation stage as the time window length;

[0076] Decompose the abnormal evaluation index values ​​in the fault operation phase and calculate the contribution of each state parameter to the abnormal state;

[0077] The component represented by the parameter with the largest abnormal contribution is regarded as the fault state of the transformer.

[0078] As a possible implementation of this embodiment, the specific implementation steps of step S4 are:

[0079] If a fault is detected in the status monitoring, the period from when the abnormal evaluation indicator value first exceeds the threshold to when it falls back below the threshold is considered as the fault operation stage. Ta , and by taking the fault operation phase as the time window length Ta To calculate the abnormal evaluation index value in this stage :

[0080] ,

[0081] During the fault operation phase To break it down:

[0082] ,

[0083] ,

[0084] in, This statistic is composed of all the i The parameters other than the Represents the i The abnormal contribution of parameters, and the highest The value means i The parameter has the greatest correlation with the abnormal operation of the transformer;

[0085] Will The fault represented by the parameter is regarded as the fault state of the transformer.

[0086] In a second aspect, an embodiment of the present invention provides a transformer multi-parameter fusion state assessment device, comprising:

[0087] A data acquisition module, used to collect multiple operating parameters of the transformer and form a data set, wherein the multiple operating parameters include the transformer's image, infrared, partial discharge, sound print, and ambient temperature and humidity;

[0088] The model building module is used to establish a probability diffusion model based on dynamic conditions, introduce conditional probability distribution to generate target parameters, introduce random processes to build a continuous function relationship between target parameters and conditional parameters, and obtain data distribution characteristics of target parameters;

[0089] An indicator construction module is used to construct an evaluation indicator based on the data distribution characteristics of the target parameter, and determine the abnormal evaluation indicator using KL divergence and multivariate variation contribution measurement, where the evaluation indicator is the difference in probability density distribution between the generated target parameter and the true target parameter;

[0090] The fault location module is used to calculate the abnormal contribution degree of each target parameter in the abnormal evaluation index based on the abnormal decomposition strategy of multivariate variation contribution measurement, and regard the target parameter with the largest abnormal contribution as the fault state of the transformer to locate the fault and determine the cause of the fault.

[0091] As a possible implementation of this embodiment, the probability diffusion model based on dynamic conditions includes a forward diffusion process and a reverse diffusion process.

[0092] The forward diffusion process gradually integrates the noise signal into the model input data, so that the data gradually transforms into pure noise data;

[0093] The reverse diffusion process systematically removes the noise in the data injected during the forward diffusion process, gradually restoring until the original model input data is accurately restored.

[0094] As a possible implementation of this embodiment, the multivariate variation contribution measure is used to evaluate the reliability of data for generating target parameters; and the KL divergence is used to calculate an anomaly assessment index.

[0095] As a possible implementation of this embodiment, the fault location module includes:

[0096] The abnormal operation state determination module is used to detect that the abnormal evaluation index value exceeds a preset threshold value, and then determine that the transformer enters an abnormal operation state;

[0097] A fault operation stage determination module is used to define the time period from when the abnormal evaluation index value exceeds the threshold for the first time to when it falls below the threshold as the fault operation stage;

[0098] An abnormal evaluation index value calculation module, used to calculate the abnormal evaluation index value within the stage by taking the fault operation stage as the time window length;

[0099] A contribution degree calculation module is used to decompose the abnormal evaluation index value in the fault operation stage and calculate the contribution degree of each state parameter to the abnormal state;

[0100] The fault state determination module is used to regard the component represented by the parameter with the largest abnormal contribution as the fault state of the transformer.

[0101] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of any of the above-mentioned transformer multi-parameter fusion state assessment methods.

[0102] In a fourth aspect, an embodiment of the present invention provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned transformer multi-parameter fusion state assessment methods are executed.

[0103] The beneficial effects of the technical solution of the embodiment of the present invention are as follows:

[0104] The present invention uses advanced data analysis methods to achieve a comprehensive assessment of the transformer status, which not only improves the accuracy of fault diagnosis, but also can warn of potential problems in advance, thereby providing a basis for maintenance strategies and ensuring the efficient and stable operation of the transformer.

[0105] Through the analysis of the multi-parameter data set of the transformer, the dynamic conditional probability diffusion model proposed in the present invention has the smallest prediction error and the highest accuracy, which verifies the effectiveness and superiority of the method; the dynamic conditional probability diffusion model proposed in the present invention has the lowest average score of continuous sorting probability, the lowest root mean square error and mean absolute error, the highest decision coefficient, and has the smallest prediction error and the highest accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 is a flow chart of a transformer multi-parameter fusion state assessment method according to an exemplary embodiment;

[0107] Figure 2 is a schematic diagram of a bidirectional Markov process of a diffusion model according to an exemplary embodiment;

[0108] Figure 3 is a schematic diagram of a training process of a diffusion model according to an exemplary embodiment;

[0109] Figure 4 is a schematic diagram of transformer status monitoring results according to an exemplary embodiment;

[0110] Figure 5 It is a structural schematic diagram of a transformer multi-parameter fusion state evaluation device according to an exemplary embodiment. DETAILED DESCRIPTION

[0111] In order to more clearly illustrate the technical features of the solution of the present invention, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0112] like Figure 1 As shown, a transformer multi-parameter fusion state assessment method provided by an embodiment of the present invention includes the following steps:

[0113] Step S1, collecting multiple operating parameters of the transformer and forming a data set, wherein the multiple operating parameters include an image, infrared, partial discharge, sound print, and ambient temperature and humidity of the transformer;

[0114] Step S2, establishing a probability diffusion model based on dynamic conditions, introducing conditional probability distribution to generate target parameters, introducing random processes to construct a continuous function relationship between the target parameters and the conditional parameters, and obtaining data distribution characteristics of the target parameters;

[0115] Step S3, constructing an evaluation index based on the data distribution characteristics of the target parameter, and determining an abnormal evaluation index using KL divergence and multivariate variation contribution measurement, wherein the evaluation index is the probability density distribution difference between the generated target parameter and the true target parameter;

[0116] Step S4, based on the abnormal decomposition strategy of multivariate variation contribution measurement, the abnormal contribution degree of each target parameter in the abnormal evaluation index is calculated, and the target parameter with the largest abnormal contribution is regarded as the fault state of the transformer, the fault is located and the cause of the fault is determined.

[0117] This embodiment collects multiple operating parameters of the transformer, such as image, infrared, partial discharge, voiceprint, ambient temperature and humidity, and forms a set to provide basic data support for subsequent state monitoring; this embodiment proposes a dynamic conditional probability diffusion model, which distinguishes the data distribution characteristics of the target parameter through conditional parameters, and forms the basis for subsequent state monitoring analysis through random process diffusion, thereby establishing a model framework; this embodiment proposes a multivariate contribution metric to evaluate and analyze the reliability of data generated from each distribution to ensure the data quality used in subsequent analysis; this embodiment uses KL divergence and multivariate contribution metric to propose an abnormal evaluation index, which integrates information from multiple parameters, integrates the probability density distribution difference between the generated data of the state monitoring target parameter and the real data, realizes comprehensive monitoring of the transformer operation state, and realizes continuous monitoring of the transformer state by quantifying the abnormal state; this embodiment proposes an abnormal decomposition strategy based on multivariate contribution metric, calculates the abnormal contribution degree of each state parameter when the transformer is in an abnormal operation stage, and regards the parameter with the largest abnormal contribution as the fault state of the transformer, thereby locating the fault and determining the cause of the fault.

[0118] As a possible implementation of this embodiment, the probability diffusion model based on dynamic conditions includes a forward diffusion process (from right to left) and a reverse diffusion process (from left to right).

[0119] The forward diffusion process gradually integrates the noise signal into the model input data, so that the data gradually transforms into pure noise data; the role of this process is to simulate the situation where the data is disturbed, thereby providing the necessary conditions for reverse diffusion;

[0120] The reverse diffusion process systematically removes the noise in the data injected during the forward diffusion process, gradually recovering until the original model input data is accurately restored. The key role of this process is to recover useful information from the noise, thereby improving the authenticity and accuracy of the data.

[0121] In the reverse process based on the dynamic conditional probability diffusion model, the reliability of the generated data is improved by introducing a random process to construct a continuous function relationship between the target parameters and the conditional parameters; at the same time, the conditional probability distribution is introduced to enable the model to learn the probability distribution of the conditional parameters to generate the target parameters, integrating the information of multiple parameters, thereby improving the accuracy of the generated data. The main purpose of this step is to distinguish the data distribution characteristics of the target parameters through the conditional parameters, and to form the basis for subsequent state monitoring analysis through random process diffusion.

[0122] As a possible implementation of this embodiment, the establishment of a probability diffusion model based on dynamic conditions includes the following steps:

[0123] Process from the original data set Start with , where the subscript 0 indicates that the data does not contain any noise; Figure 2 As shown, the forward Markov chain transforms this pure input Transformed into a series of perturbed data points , and adopts diffusion conversion kernel;

[0124] The forward diffusion process is defined as:

[0125] ,

[0126] ,

[0127] in, Indicates that when the initial state X0 is known, all states from time 1 to T X t The joint probability distribution of Indicates that the state at the previous moment is known X t-1 In the case of X t The probability distribution of represents the noise level parameter, which controls the correlation between the new state and the old state; express X t The distribution follows the mean , the variance is Normal distribution; I is the identity matrix, indicating the size of the variance; is the state variable at time t;

[0128] The sampling is transformed into:

[0129] ,

[0130] in, and ; is a coefficient between 0 and 1;

[0131] It is expressed as:

[0132] ,

[0133] and ;

[0134] The reverse diffusion process is the denoising To restore , given by the Markov chain:

[0135] ,

[0136] ,

[0137] ,

[0138] in , is a trainable denoising function;

[0139] The reverse diffusion process is trained by solving the following optimization problem. Once trained, we can sample from the equation :

[0140] ,

[0141] in represents the noise vector.

[0142] As a possible implementation of this embodiment, the method of introducing conditional probability distribution to generate target parameters and introducing random process to construct a continuous function relationship between the target parameters and the conditional parameters includes the following steps:

[0143] distributed It is defined by expanding the inverse process in the equation to a conditional process:

[0144] ,

[0145] ,

[0146] ;

[0147] Given condition parameters and target parameters , noise target is sampled, and Training is done by minimizing the loss function:

[0148] ,

[0149] in, Represents the noise function Generated by a Gaussian random process, that is ; is a coefficient between 0 and 1 that controls the target parameter and noise Mixing ratio; is a noise term generated by a Gaussian random process, with an expected value of 0 and a variance of Σ; is a noisy target, which consists of the target parameters without noise and Composition, of which Multiply The square root of Multiply by (1- )’s square root; is a noise function, which needs to be trained by minimizing the loss function L(theta); L(theta) is and The sum of the squares of the distances between them, E represents the expected value, and ||·||2 represents the Euclidean norm.

[0150] The influence of random process diffusion and conditional parameters is taken into account. The Gaussian random process generates a noise function with continuous properties, which enables the model to generate data at any time node. In addition, the information of the conditional parameters is effectively used to improve the quality of the generated target parameter data. The introduction of random processes to construct a continuous function relationship between the target parameters and the conditional parameters improves the reliability of the generated data; the introduction of conditional probability distribution enables the model to learn the probability distribution of the conditional parameters to generate the target parameters, which integrates the information of multiple parameters and improves the accuracy of the generated data.

[0151] As a possible implementation of this embodiment, in step S3, the multivariate variation contribution measure is used to evaluate the reliability of data for generating target parameters;

[0152] Using the multivariate variation contribution metric to evaluate the data reliability of generating target parameters includes the following steps:

[0153] Set the set of parameters to be evaluated and the corresponding distributions; this step is to determine which parameters will be included in the evaluation and prepare for subsequent calculations;

[0154] Calculate the variation contribution of the data generated by the distribution of each parameter; this calculation process involves performing variation analysis on the data points generated by each distribution to obtain the variation contribution; the key to this step is to quantify the extent to which the data point represents the variation of the parameter through the variation contribution metric;

[0155] According to the calculated variation contribution, the reliability of the data points generated by each distribution is evaluated; data points with high variation contribution may indicate unstable data or high noise, whereas data points with low variation contribution may indicate relatively stable and reliable data. The purpose of this step is to select more reliable data points so that unreliable data sources can be excluded in subsequent analysis;

[0156] Output the results of the multivariate contribution measure, which can be used in subsequent steps, such as the calculation of anomaly assessment indicators; in this way, it can be ensured that the data entering the next step has a high reliability, thereby improving the evaluation accuracy of the entire system.

[0157] As a possible implementation of this embodiment, in step S3, the KL divergence is used to calculate the anomaly assessment index;

[0158] Calculating the abnormality assessment index using the KL divergence includes the following steps:

[0159] Calculate the difference between the true value and the predicted value of the target parameter. The calculation of KL divergence can help understand the degree of deviation between the predicted data and the actual data.

[0160] In order to achieve continuous abnormal state monitoring, a sliding time window with a length of L is set to continuously intercept KL divergence data;

[0161] The KL divergence data intercepted by the sliding time window is used to calculate the anomaly evaluation index, which reflects the degree of anomaly of the data during the monitoring period;

[0162] According to the fluctuation range of the abnormal evaluation index value under normal operating conditions, the maximum abnormal evaluation index value is calculated and used as the threshold of the abnormal evaluation index to evaluate the operating status of the transformer.

[0163] As a possible implementation of this embodiment, for a random sample X with a length of n and a dimension of p, the calculation formula of the multivariate variation contribution measure is:

[0164] ,

[0165] ,

[0166] ,

[0167] in, is the contribution of multivariate variation, is the sample mean of the random sample X, and S is the sample covariance matrix of the random sample X.

[0168] As a possible implementation of this embodiment, the expression using KL divergence is:

[0169] Define the true probability density function of the target parameter and the resulting probability density function ;

[0170] KL divergence is defined as:

[0171] ,

[0172] For the KL divergence calculation in discrete cases, its form becomes:

[0173] ,

[0174] x are all possible states;

[0175] When dealing with multiple parameters, it is necessary to calculate the KL divergence of each parameter separately. If the parameter set is , then we get a series of KL divergence values:

[0176] ,

[0177] In order to achieve continuous abnormal state monitoring, a sliding time window of length L is set, KL divergence data is continuously intercepted, and the intercepted KL divergence data is used to calculate the abnormal evaluation index:

[0178] ,

[0179] in, is an abnormal evaluation indicator.

[0180] The sliding step of the time window determines the time resolution of transformer condition monitoring. The data in the normal operating state conforms to the logical relationship of the model, and its generated value is close to the true value. The KL divergence between them is small, and the calculated abnormal evaluation index value is also small; while the data in the abnormal operating state does not conform to the logical relationship of the model, and its generated value deviates greatly from the true value, and the KL divergence is large, so the abnormal evaluation index value calculated using KL divergence will also increase. The calculated maximum abnormal evaluation index value is set as the threshold, which indicates the fluctuation range of the abnormal evaluation index value of the transformer under normal operating conditions, and this is used as a standard for evaluating the operating status of the transformer.

[0181] As a possible implementation of this embodiment, step S4 includes the following steps:

[0182] If the abnormal evaluation index value exceeds the preset threshold, the transformer is judged to be in an abnormal operation state;

[0183] The time period from when the abnormal evaluation index value exceeds the threshold for the first time to when it falls below the threshold is defined as the fault operation stage;

[0184] The abnormal evaluation index value within the stage is calculated by taking the fault operation stage as the time window length;

[0185] Decompose the abnormal evaluation index values ​​in the fault operation phase and calculate the contribution of each state parameter to the abnormal state;

[0186] The component represented by the parameter with the largest abnormal contribution is regarded as the fault state of the transformer.

[0187] As a possible implementation of this embodiment, Figure 4 As shown, the specific implementation steps of step S4 are:

[0188] If a fault is detected in the status monitoring, the period from when the abnormal evaluation indicator value first exceeds the threshold to when it falls back below the threshold is considered as the fault operation stage. Ta , and by taking the fault operation phase as the time window length Ta To calculate the abnormal evaluation index value in this stage :

[0189] ,

[0190] During the fault operation phase To break it down:

[0191] ,

[0192] ,

[0193] in, This statistic is composed of all the i The parameters other than the Represents the i The abnormal contribution of parameters, and the highest The value means i The parameter has the greatest correlation with the abnormal operation of the transformer;

[0194] Will The fault represented by the parameter is regarded as the fault state of the transformer.

[0195] When the transformer detects abnormal operation, the next step is to detect the fault. The method to determine the fault is to decompose the abnormal evaluation index value, then calculate the abnormal contribution of each parameter, and regard the fault represented by the parameter with the largest abnormal contribution as the fault state of the transformer.

[0196] like Figure 5 As shown, an embodiment of the present invention provides a transformer multi-parameter fusion state assessment device, comprising:

[0197] A data acquisition module, used to collect multiple operating parameters of the transformer and form a data set, wherein the multiple operating parameters include the transformer's image, infrared, partial discharge, sound print, and ambient temperature and humidity;

[0198] The model building module is used to establish a probability diffusion model based on dynamic conditions, introduce conditional probability distribution to generate target parameters, introduce random processes to build a continuous function relationship between target parameters and conditional parameters, and obtain data distribution characteristics of target parameters;

[0199] An indicator construction module is used to construct an evaluation indicator based on the data distribution characteristics of the target parameter, and determine the abnormal evaluation indicator using KL divergence and multivariate variation contribution measurement, where the evaluation indicator is the difference in probability density distribution between the generated target parameter and the true target parameter;

[0200] The fault location module is used to calculate the abnormal contribution degree of each target parameter in the abnormal evaluation index based on the abnormal decomposition strategy of multivariate variation contribution measurement, and regard the target parameter with the largest abnormal contribution as the fault state of the transformer to locate the fault and determine the cause of the fault.

[0201] As a possible implementation of this embodiment, the probability diffusion model based on dynamic conditions includes a forward diffusion process and a reverse diffusion process.

[0202] The forward diffusion process gradually integrates the noise signal into the model input data, so that the data gradually transforms into pure noise data;

[0203] The reverse diffusion process systematically removes the noise in the data injected during the forward diffusion process, gradually restoring until the original model input data is accurately restored.

[0204] As a possible implementation of this embodiment, the multivariate variation contribution measure is used to evaluate the reliability of data for generating target parameters; and the KL divergence is used to calculate an anomaly assessment index.

[0205] As a possible implementation of this embodiment, the fault location module includes:

[0206] The abnormal operation state determination module is used to detect that the abnormal evaluation index value exceeds a preset threshold value, and then determine that the transformer enters an abnormal operation state;

[0207] A fault operation stage determination module is used to define the time period from when the abnormal evaluation index value exceeds the threshold for the first time to when it falls below the threshold as the fault operation stage;

[0208] An abnormal evaluation index value calculation module, used to calculate the abnormal evaluation index value within the stage by taking the fault operation stage as the time window length;

[0209] A contribution degree calculation module is used to decompose the abnormal evaluation index value in the fault operation stage and calculate the contribution degree of each state parameter to the abnormal state;

[0210] The fault state determination module is used to regard the component represented by the parameter with the largest abnormal contribution as the fault state of the transformer.

[0211] The multi-parameter data sets of power transformers over a period of time are selected, including the image, infrared, partial discharge, soundprint, ambient temperature and humidity of the transformer. The transformer is considered to be in normal operation. The entire data set is used as a test set. The diffusion steps of the diffusion model T Set to 100, minimum noise level =0.0001, maximum noise level =0.5. For the neural network model, 2 residual layers, 32 residual channels, and 4 attention heads were set. The number of training cycles was set to 600 times, and the Adam optimizer was used. The initial learning rate was 0.001, which decayed to 0.0001 and 0.00001 at 75% and 90% of the total training cycle, respectively. In order to achieve the above purpose, a multi-parameter coupled oil-immersed transformer intelligent sensor device based on micro-electromechanical system was used to collect transformer images, infrared, partial discharge, voiceprint, ambient temperature and humidity and other operating parameters and form a set. A method based on dynamic conditional probability diffusion model is proposed. The model distinguishes the data distribution characteristics of the target parameters through conditional parameters, and forms the basis for subsequent state monitoring analysis through random process diffusion; then, a multivariate variation contribution measure is proposed to evaluate and analyze the reliability of the data generated from each distribution, and an abnormal evaluation index is proposed by using KL divergence and multivariate variation contribution measure. This index integrates the probability density distribution difference between the generated data of the state monitoring target parameters and the real data, and realizes the comprehensive monitoring of the transformer operation state; finally, an abnormal decomposition strategy based on multivariate variation contribution measure is proposed, which is used to calculate the abnormal contribution degree of each state parameter when the transformer has an abnormal operation stage, and the component represented by the parameter with the largest abnormal contribution is regarded as the fault state of the transformer. The experimental results show that the proposed method has the lowest average score of 0.274 for the continuous sorting probability based on the dynamic conditional probability diffusion model proposed by the present invention through the analysis of the multi-parameter data set of the transformer, and the root mean square error (RMSE) and mean absolute error (MAE) are the lowest, the decision coefficient (R) is the highest, and it has the smallest prediction error and the highest accuracy, which verifies the effectiveness and superiority of this method.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A transformer multi-parameter fusion state assessment method, characterized in that: The steps include: Step S1, collecting multiple operating parameters of the transformer and forming a data set, wherein the multiple operating parameters include an image, infrared, partial discharge, sound print, and ambient temperature and humidity of the transformer; Step S2, establishing a probability diffusion model based on dynamic conditions, introducing conditional probability distribution to generate target parameters, introducing random processes to construct a continuous function relationship between the target parameters and the conditional parameters, and obtaining data distribution characteristics of the target parameters; Step S3, constructing an evaluation index based on the data distribution characteristics of the target parameter, using multivariate variation contribution measurement to evaluate the data reliability of the generated target parameter, and using KL divergence to calculate an abnormal evaluation index, where the evaluation index is the probability density distribution difference between the generated target parameter and the true target parameter; Step S4, calculating the abnormal contribution degree of each target parameter in the abnormal evaluation index based on the abnormal decomposition strategy of multivariate variation contribution measurement, and considering the target parameter with the largest abnormal contribution as the fault state of the transformer, locating the fault and determining the cause of the fault; The probability diffusion model based on dynamic conditions includes a forward diffusion process and a reverse diffusion process. The forward diffusion process gradually integrates the noise signal into the model input data, so that the data gradually transforms into pure noise data; The reverse diffusion process systematically removes the noise in the data injected during the forward diffusion process, gradually recovering until the original model input data is accurately recovered; The establishment of a probability diffusion model based on dynamic conditions includes the following steps: Process from the original data set The forward Markov chain converts this pure input Transformed into a series of perturbed data points , and adopts diffusion conversion kernel; The forward diffusion process is defined as: ， ， in, Indicates that when the initial state X0 is known, all states from time 1 to T X t The joint probability distribution of Indicates that the state at the previous moment is known X t-1 In the case of X t The probability distribution of represents the noise level parameter; express X t The distribution of , the variance is Normal distribution; I is the identity matrix, indicating the size of the variance; is the state variable at time t; The sampling is transformed into: ， in, and ; is a coefficient between 0 and 1; It is expressed as: ， and ; The reverse diffusion process is the denoising To restore , given by the Markov chain: ， ， ， in , is a trainable denoising function; The reverse diffusion process is trained by solving the following optimization problem. Once trained, we can sample from the equation : ， in represents the noise vector; The method of introducing conditional probability distribution to generate target parameters and introducing random process to construct a continuous function relationship between the target parameters and the conditional parameters comprises the following steps: distributed It is defined by expanding the inverse process in the equation to a conditional process: ， ， ； Given condition parameters and target parameters , noise target is sampled, and Training is done by minimizing the loss function: ， in, Represents the noise function Generated by a Gaussian random process, that is ; is a coefficient between 0 and 1 that controls the noise target and noise Mixing ratio; is a noise term generated by a Gaussian random process, with an expected value of 0 and a variance of Σ; is a noisy target, which consists of the target parameters without noise and Composition, of which Multiply The square root of Multiply by (1- )’s square root; is a noise function, which needs to be trained by minimizing the loss function L(theta); L(theta) is and The sum of the squares of the distances between them, E represents the expected value, and ||·||2 represents the Euclidean norm.

2. The transformer multi-parameter fusion state assessment method according to claim 1 is characterized in that: In step S3, the use of multivariate contribution metrics to evaluate the data reliability of generated target parameters includes the following steps: Set the set of parameters to be evaluated and the corresponding distribution; Calculate the contribution of variation to the data generated by the distribution of each parameter; Based on the calculated contribution to variation, the reliability of the data points generated by each distribution is evaluated; Outputs the results of the multivariate contribution measure.

3. The transformer multi-parameter fusion state assessment method according to claim 2 is characterized in that: In step S3, the calculation of the abnormality assessment index using KL divergence includes the following steps: Calculate the difference between the true value and the predicted value of the target parameter; Set a sliding time window of length L to continuously intercept KL divergence data; The KL divergence data intercepted by the sliding time window is used to calculate the anomaly evaluation index; According to the fluctuation range of the abnormal evaluation index value under normal operating conditions, the maximum abnormal evaluation index value is calculated and used as the threshold of the abnormal evaluation index.

4. The transformer multi-parameter fusion state assessment method according to claim 1 is characterized in that: For a random sample X with length n and dimension p, the calculation formula of the multivariate variation contribution measure is: ， ， ， in, is the contribution of multivariate variation, is the sample mean of the random sample X, and S is the sample covariance matrix of the random sample X.

5. The transformer multi-parameter fusion state assessment method according to claim 4 is characterized in that: The expression of KL divergence is: Define the true probability density function of the target parameter and the resulting probability density function ; KL divergence is defined as: ， For the KL divergence calculation in discrete cases, its form becomes: ， x are all possible states; When dealing with multiple parameters, it is necessary to calculate the KL divergence of each parameter separately. If the parameter set is , then we get a series of KL divergence values: ， Set a sliding time window of length L, continuously intercept KL divergence data, and use the intercepted KL divergence data to calculate the anomaly evaluation index: ， in, is an abnormal evaluation indicator.

6. The transformer multi-parameter fusion state assessment method according to claim 4 or 5, characterized in that: The step S4 comprises the following steps: If the abnormal evaluation index value exceeds the preset threshold, the transformer is judged to be in an abnormal operation state; The time period from when the abnormal evaluation index value exceeds the threshold for the first time to when it falls below the threshold is defined as the fault operation stage; By taking the fault operation phase as the time window length T a To calculate the abnormal evaluation index value within this stage; Decompose the abnormal evaluation index values ​​in the fault operation phase and calculate the contribution of each state parameter to the abnormal state; The component represented by the parameter with the largest abnormal contribution is regarded as the fault state of the transformer.

7. The transformer multi-parameter fusion state assessment method according to claim 6, characterized in that: The specific implementation steps of step S4 are: If a fault is detected in the condition monitoring, the time period from when the abnormal evaluation indicator value first exceeds the threshold to when it falls back below the threshold is regarded as the fault operation stage, and the fault operation stage is taken as the time window length T a To calculate the abnormal evaluation index value in this stage : ， During the fault operation phase To break it down: ， ， in, This statistic is composed of all the i The parameters other than the Represents the i The abnormal contribution of parameters, and the highest The value means i The parameter has the greatest correlation with the abnormal operation of the transformer; Will The fault represented by the parameter is regarded as the fault state of the transformer.

8. A transformer multi-parameter fusion state assessment device, characterized in that: include: A data acquisition module, used to collect multiple operating parameters of the transformer and form a data set, wherein the multiple operating parameters include the transformer's image, infrared, partial discharge, sound print, and ambient temperature and humidity; The model building module is used to establish a probability diffusion model based on dynamic conditions, introduce conditional probability distribution to generate target parameters, introduce random processes to build a continuous function relationship between target parameters and conditional parameters, and obtain data distribution characteristics of target parameters; An indicator construction module is used to construct an evaluation indicator based on the data distribution characteristics of the target parameter, use multivariate variation contribution measurement to evaluate the data reliability of the generated target parameter, and use KL divergence to calculate the abnormal evaluation indicator, which is the difference in probability density distribution between the generated target parameter and the true target parameter; The fault location module is used to calculate the abnormal contribution degree of each target parameter in the abnormal evaluation index based on the abnormal decomposition strategy of multivariate variation contribution measurement, and regard the target parameter with the largest abnormal contribution as the fault state of the transformer to locate the fault and determine the cause of the fault; The probability diffusion model based on dynamic conditions includes a forward diffusion process and a reverse diffusion process. The forward diffusion process gradually integrates the noise signal into the model input data, so that the data gradually transforms into pure noise data; The reverse diffusion process systematically removes the noise in the data injected during the forward diffusion process, gradually recovering until the original model input data is accurately recovered; The establishment of a probability diffusion model based on dynamic conditions includes the following steps: Process from the original data set The forward Markov chain converts this pure input Transformed into a series of perturbed data points , and adopts diffusion conversion kernel; The forward diffusion process is defined as: ， ， in, Indicates that when the initial state X0 is known, all states from time 1 to T X t The joint probability distribution of Indicates that the state at the previous moment is known X t-1 In the case of X t The probability distribution of represents the noise level parameter; express X t The distribution of , the variance is Normal distribution; I is the identity matrix, indicating the size of the variance; is the state variable at time t; The sampling is transformed into: ， in, and ; is a coefficient between 0 and 1; It is expressed as: ， and ; The reverse diffusion process is the denoising To restore , given by the Markov chain: ， ， ， in , is a trainable denoising function; The reverse diffusion process is trained by solving the following optimization problem. Once trained, we can sample from the equation : ， in represents the noise vector; The method of introducing conditional probability distribution to generate target parameters and introducing random process to construct a continuous function relationship between the target parameters and the conditional parameters comprises the following steps: distributed It is defined by expanding the inverse process in the equation to a conditional process: ， ， ； Given condition parameters and target parameters , noise target is sampled, and Training is done by minimizing the loss function: ， in, Represents the noise function Generated by a Gaussian random process, that is ; is a coefficient between 0 and 1 that controls the noise target and noise Mixing ratio; is a noise term generated by a Gaussian random process, with an expected value of 0 and a variance of Σ; is a noisy target, which consists of the target parameters without noise and Composition, of which Multiply The square root of Multiply by (1- )’s square root; is a noise function, which needs to be trained by minimizing the loss function L(theta); L(theta) is and The sum of the squares of the distances between them, E represents the expected value, and ||·||2 represents the Euclidean norm.

9. The transformer multi-parameter fusion state assessment device according to claim 8, characterized in that: The fault location module comprises: The abnormal operation state determination module is used to detect that the abnormal evaluation index value exceeds a preset threshold value, and then determine that the transformer enters an abnormal operation state; A fault operation stage determination module is used to define the time period from when the abnormal evaluation index value exceeds the threshold for the first time to when it falls back below the threshold as the fault operation stage; An abnormal evaluation index value calculation module, used to calculate the abnormal evaluation index value within the stage by taking the fault operation stage as the time window length; A contribution degree calculation module is used to decompose the abnormal evaluation index value in the fault operation stage and calculate the contribution degree of each state parameter to the abnormal state; The fault state determination module is used to regard the component represented by the parameter with the largest abnormal contribution as the fault state of the transformer.

10. A computer device, characterized in that: The computer device comprises a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the transformer multi-parameter fusion state assessment method as described in any one of claims 1 to 7.

11. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, executes the steps of the transformer multi-parameter fusion state assessment method as claimed in any one of claims 1 to 7.

Images