Marine high-voltage transformer diagnosis system
By deploying sensors and digital twin technology in marine transformers, building a virtual model and combining the CNN model and Softmax classifier, the real-time diagnosis problem of marine transformers in dynamic marine environments was solved, and rapid fault location and efficient maintenance were achieved.
Patent Information
- Application Number
- CN202510809806.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-17
- Publication Date
- 2025-09-23
AI Technical Summary
The existing marine transformer fault diagnosis system has difficulty in achieving real-time status monitoring and accurate diagnosis in the dynamic marine environment. The model is incomplete, the data is inaccurate, and the interaction between the equipment and the model is insufficient.
By deploying sensors in the transformer, real-time operation data is obtained, a digital twin virtual model is built, a fusion model and a relational database are established, the finite element analysis model is used for data iteration, and the CNN model and Softmax classifier are combined for fault identification.
It realizes real-time health assessment of transformers and rapid fault location, reduces fault inspection time and improves maintenance efficiency.
Smart Images
Figure CN120688319A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of transformers, and in particular relates to a diagnostic system for a marine high-voltage transformer. Background Art
[0002] Marine transformers are electrical equipment designed specifically for the unique operating environments and power requirements of ships. Their primary function is to achieve electrical isolation between the power source and the load, ensuring the safe operation of the ship's power system. With the rapid development of the Internet of Things, artificial intelligence, and communications technologies, and the unique operating environments of marine power equipment, the types and volume of monitoring data are increasing, and the requirements for accuracy and transmission speed are also increasing. However, current fault diagnosis of marine equipment is primarily based on data obtained under ideal operating conditions. In the dynamic and volatile ocean environment, real-time status monitoring and accurate diagnosis of marine equipment remain a significant challenge. Existing fault diagnosis systems suffer from incomplete models, inaccurate data, and insufficient interaction between equipment and models. Summary of the Invention
[0003] Purpose of the invention: To provide a diagnostic system for marine high-voltage transformers, which solves the above-mentioned problems existing in the prior art.
[0004] Technical solution: Marine high-voltage transformer diagnostic system, including the following steps:
[0005] By deploying sensors in the transformer and acquiring sensor signals, we can obtain real-time operating data of the transformer and build a virtual model with the help of digital twin technology.
[0006] Obtain transformer geometric attribute data and real-time operation data to establish a driving data set. Use the virtual model and the driving data set to establish a fusion model. Obtain data from the virtual model and the fusion model to build a relational database. Build a digital finite element analysis model by obtaining data corresponding to the virtual model and the fusion model and the relational database. Use the finite element analysis model to bidirectionally circulate and iterate the real-time operation data and the data in the relational database to obtain real-time monitoring parameters.
[0007] Obtain the historical operating data of the transformer under normal operating conditions, perform statistical analysis on the historical operating data to obtain the average value of the historical operating data, and obtain the threshold value of the corresponding parameter as the health baseline;
[0008] Obtain the operating parameters corresponding to the transformer under different operating conditions, build a matching matrix based on the real-time monitoring parameters and the health baseline, obtain the actual operating conditions under the real-time operating state, and obtain the corresponding health baseline based on the matching matrix. The obtained health baseline is combined with the finite element analysis model to evaluate the health of the transformer under actual operating conditions. The transformer health assessment includes both healthy and faulty states.
[0009] When a fault occurs in the transformer health assessment, the fault data under the fault state is obtained to establish a fault data set. The fault data in the fault data set is input into the initialized CNN model for fault data processing, and the Softmax classifier is used to output the fault classification result for fault identification.
[0010] Preferably, the process of building a matching matrix is as follows:
[0011] According to the hierarchical analysis, the operating parameters of the transformer are divided into target level, criterion level and scheme level according to historical data and actual working conditions, and the matching matrix A is obtained by formula (1), which is as follows:
[0012] A=(k ab ) I×J (1);
[0013] Where k ab Indicates the importance of the ath factor relative to the bth factor in the judgment matrix; I×J indicates that there are I rows and J columns in the matching matrix, where k aa =1,k ab >1, The consistency of the matching matrix A is verified by formula (2), and whether the consistency index is achieved is determined by looking up the table. When the matching matrix A reaches the consistency index, the target weight is calculated. Otherwise, the matching matrix A is calculated again by formula (1) until the matching matrix A reaches the consistency index. Formula (2) is as follows:
[0014]
[0015] Where: q represents the order of the matching matrix.
[0016] Preferably, the target weight is calculated as follows:
[0017] Based on the adaptive weight method, the judgment matrix is established by comparing the input matching matrix A with each other. The subjective weight α of each indicator of the matching matrix A is obtained by calculating the eigenvector and normalizing it. z , and then build an adaptive function based on the historical operation data to calculate the dynamic weight β of each indicator of the historical operation data z , by weighted averaging and normalizing the subjective weight and dynamic weight, the precise weight is obtained as the evaluation standard of the health baseline. The calculation formulas of weighted average and normalization are shown in formulas (3) and (4):
[0018] ω z =h·α z +(1-h)·β z (3);
[0019]
[0020] Where: h represents the proportion of subjective weight, 0<h<1; n represents the number of indicators.
[0021] Preferably, the fault data processing in the CNN model is as follows:
[0022] Using the convolution layer, convolution operations are performed on the data in the relational database to obtain feature data and establish a feature data set. The feature data calculation formula is as follows:
[0023]
[0024] Where: represents the feature map corresponding to the jth neuron in the lth layer; express Input, feature map of l-1 layer; Represents the convolution kernel from the i-th neuron in layer l-1 to the j-th neuron in layer l; Represents the output bias of the j-th neuron in the l-th layer.
[0025] Preferably, a feature dataset is obtained, a Transformer model is built, and the maximum pooling in the feature dataset is calculated using the Transformer model. The calculation process is as follows:
[0026] The pre-planned pooling window is f×f, and the feature data set is divided into several pooling areas of size f×f. The maximum value of all elements in the pooling area is extracted by formula (6) as the value of the corresponding position of the output feature data. By presetting the step size of the pooling window, the feature data set is subjected to non-overlapping pooling to obtain the adopted parameter configuration, where formula (6) is as follows:
[0027]
[0028] Where: Y(i, j) represents the (i, j)th element of the output feature data; c represents the offset in the height direction of the feature data set; e represents the offset in the width direction of the feature data set; s represents the step size; X(i×s+c,j×s+e) represents the element in the local window corresponding to the output position (i, j) in the input feature data.
[0029] Preferably, an optimized sparrow search algorithm is used to search the parameter configuration in the Transformer model, and the parameter configuration is updated to obtain iterative parameters, which are used as input to the Transformer model. The Transformer model is updated, and the output of the Transformer model is converted into a probability distribution of fault classification through a Softmax classifier, and a classification label is output to identify the fault of the transformer.
[0030] Preferably, the output classification process of the Softmax classifier is as follows:
[0031] The loss function is calculated by formula (7), which is as follows:
[0032]
[0033] Where: K{·} represents a one-time function; m represents the total number of input samples; y represents the true label; T represents the transposition operation; x represents the feature vector; θ represents the parameter vector;
[0034] Add rule items, Classify sample m into the probability of class j, calculate the corresponding partial derivative of the loss function, and use the gradient descent method to update the parameters in the iterative process to obtain the classification result and complete the fault diagnosis.
[0035] Beneficial effects: The present invention relates to a diagnostic system for a marine high-voltage transformer. By acquiring the real-time operating data of the transformer, a virtual model is built with the help of digital twin technology, and a finite element analysis model is used to perform bidirectional circulation and update iteration on the real-time operating data and the data in the relational database, real-time monitoring parameters are obtained. A health baseline is obtained by comparing the operating history data with the actual working conditions under the real-time operating state. The health of the transformer is evaluated by combining the health baseline with the finite element analysis model. At the same time, the fault data is processed by the CNN model, and the fault classification result is output in conjunction with the Softmax classifier. By obtaining the output fault classification result, the staff can quickly locate and repair the transformer fault, which reduces the time for fault inspection and the tedious inspection process of the staff, thereby improving the maintenance efficiency of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 This is a system block diagram of the present invention. DETAILED DESCRIPTION
[0037] like Figure 1 As shown, the present invention provides a technical solution: a marine high-voltage transformer diagnostic system, comprising the following steps:
[0038] By deploying sensors in the transformer and acquiring sensor signals, we can obtain real-time operating data of the transformer and build a virtual model with the help of digital twin technology.
[0039] Obtain transformer geometric attribute data and real-time operation data to establish a driving data set. Use the virtual model and the driving data set to establish a fusion model. Obtain data from the virtual model and the fusion model to build a relational database. Build a digital finite element analysis model by obtaining data corresponding to the virtual model and the fusion model and the relational database. Use the finite element analysis model to bidirectionally circulate and iterate the real-time operation data and the data in the relational database to obtain real-time monitoring parameters.
[0040] Obtain the historical operating data of the transformer under normal operating conditions, perform statistical analysis on the historical operating data to obtain the average value of the historical operating data, and obtain the threshold value of the corresponding parameter as the health baseline;
[0041] Obtain the operating parameters corresponding to different operating conditions of the transformer, and build a matching matrix based on the real-time monitoring parameters and health baseline. The process of building the matching matrix is as follows:
[0042] According to the hierarchical analysis, the operating parameters of the transformer are divided into target level, criterion level and scheme level according to historical data and actual working conditions, and the matching matrix A is obtained by formula (1), which is as follows:
[0043] A=(k ab ) I×J (1);
[0044] Where k ab Indicates the importance of the ath factor relative to the bth factor in the judgment matrix; I×J indicates that there are I rows and J columns in the matching matrix, where k aa =1,k ab >1, The consistency of the matching matrix A is verified by formula (2), and whether the consistency index is achieved is determined by looking up the table. When the matching matrix A reaches the consistency index, the target weight is calculated. Otherwise, the matching matrix A is calculated again by formula (1) until the matching matrix A reaches the consistency index. Formula (2) is as follows:
[0045]
[0046] Where q represents the order of the matching matrix. After the matching matrix is constructed, the actual operating conditions under the real-time operating state are obtained. The corresponding health baseline is obtained based on the matching matrix. The obtained health baseline is combined with the finite element analysis model to evaluate the health of the transformer under the actual operating conditions. The transformer health assessment includes both healthy and faulty states.
[0047] When a fault occurs in the transformer health assessment, the fault data in the fault state is obtained to establish a fault data set. The fault data in the fault data set is input into the initialized CNN model for fault data processing, and the Softmax classifier is used to output the fault classification result for fault identification. By obtaining the output fault classification result, the staff can quickly locate and repair the transformer fault, reducing the time for fault inspection and the tedious inspection process of the staff, thereby improving the efficiency of transformer maintenance.
[0048] In a further embodiment, the target weight is calculated as follows:
[0049] Based on the adaptive weight method, the judgment matrix is established by comparing the input matching matrix A with each other. The subjective weight α of each indicator of the matching matrix A is obtained by calculating the eigenvector and normalizing it. z , and then build an adaptive function based on the historical operation data to calculate the dynamic weight β of each indicator of the historical operation data z , by weighted averaging and normalizing the subjective weight and dynamic weight, the precise weight is obtained as the evaluation standard of the health baseline. The calculation formulas of weighted average and normalization are shown in formulas (3) and (4):
[0050] ω z =h·α z +(1-h)·β z (3);
[0051]
[0052] Where: h represents the proportion of subjective weight, 0<h<1; n represents the number of indicators.
[0053] In a further embodiment, fault data processing is performed in the CNN model as follows:
[0054] Using the convolution layer, convolution operations are performed on the data in the relational database to obtain feature data and establish a feature data set. The feature data calculation formula is as follows:
[0055]
[0056] Where: represents the feature map corresponding to the jth neuron in the lth layer; express Input, feature map of l-1 layer; Represents the convolution kernel from the i-th neuron in layer l-1 to the j-th neuron in layer l; Represents the output bias of the jth neuron in the lth layer, obtains the feature data set, builds the Transformer model, and uses the Transformer model to calculate the maximum pooling in the feature data set. The calculation process is as follows:
[0057] The pre-planned pooling window is f×f, and the feature data set is divided into several pooling areas of size f×f. The maximum value of all elements in the pooling area is extracted by formula (6) as the value of the corresponding position of the output feature data. By presetting the step size of the pooling window, the feature data set is subjected to non-overlapping pooling to obtain the adopted parameter configuration, where formula (6) is as follows:
[0058]
[0059] Where: Y(i, j) represents the (i, j)th element of the output feature data; c represents the offset in the height direction of the feature data set; e represents the offset in the width direction of the feature data set; s represents the step size; X(i×s+c, j×s+e) represents the element in the local window corresponding to the output position (i, j) in the input feature data. By calculating the feature data set, the number of parameters required in the finite element analysis model is calculated, thereby improving the calculation of the finite element analysis model.
[0060] In a further embodiment, to address the problems of population diversity and limited accuracy in the Transformer model, an optimized sparrow search algorithm is introduced. The optimized sparrow search algorithm is used to search the parameter configuration in the Transformer model, update the parameter configuration, obtain iterative parameters, and use them as input to the Transformer model. The Transformer model is then updated, and the output of the Transformer model is converted into a probability distribution of fault classification through a Softmax classifier. The output classification process of the Softmax classifier is as follows:
[0061] The loss function is calculated by formula (7), which is as follows:
[0062]
[0063] Where: K{·} represents a one-time function; m represents the total number of input samples; y represents the true label; T represents the transposition operation; x represents the feature vector, θ represents the parameter vector; adding rule terms, Classify sample m into the probability of class j, calculate the corresponding partial derivative of the loss function, and use the gradient descent method to update the parameters in the iterative process to obtain the classification result and complete the fault diagnosis.
[0064] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.
Claims
1. Marine high-voltage transformer diagnostic system, characterized by: The following steps are involved: By deploying sensors in the transformer and acquiring sensor signals, we can obtain real-time operating data of the transformer and build a virtual model with the help of digital twin technology. Obtain transformer geometric attribute data and real-time operation data to establish a driving data set. Use the virtual model and the driving data set to establish a fusion model. Obtain data from the virtual model and the fusion model to build a relational database. Build a digital finite element analysis model by obtaining data corresponding to the virtual model and the fusion model and the relational database. Use the finite element analysis model to bidirectionally circulate and iterate the real-time operation data and the data in the relational database to obtain real-time monitoring parameters. Obtain the historical operating data of the transformer under normal operating conditions, perform statistical analysis on the historical operating data to obtain the average value of the historical operating data, and obtain the threshold value of the corresponding parameter as the health baseline; Obtain the operating parameters corresponding to different operating conditions of the transformer, build a matching matrix based on the real-time monitoring parameters and health baseline, obtain the actual operating conditions under the real-time operating state, and obtain the real-time health baseline based on the matching matrix. The real-time health baseline is combined with the finite element analysis model to evaluate the health of the transformer under actual operating conditions. The transformer health assessment includes both healthy and faulty states. When a fault occurs in the transformer health assessment, the fault data under the fault state is obtained to establish a fault data set. The fault data in the fault data set is input into the initialized CNN model for fault data processing, and the Softmax classifier is used to output the fault classification result for fault identification.
2. The marine high-voltage transformer diagnostic system according to claim 1, characterized in that: The process of building a matching matrix is as follows: According to the hierarchical analysis, the operating parameters of the transformer are divided into target level, criterion level and scheme level according to historical data and actual working conditions, and the matching matrix A is obtained by formula (1), which is as follows: A=(k ab ) I×J (1); Where k ab Indicates the importance of the ath factor relative to the bth factor in the judgment matrix; I×J indicates that there are I rows and J columns in the matching matrix, where k aa =1,k ab >1, The consistency of the matching matrix A is verified by formula (2), and whether the consistency index is achieved is determined by looking up the table. When the matching matrix A reaches the consistency index, the target weight is calculated. Otherwise, the matching matrix A is calculated again by formula (1) until the matching matrix A reaches the consistency index. Formula (2) is as follows: Where: q represents the order of the matching matrix.
3. The marine high-voltage transformer diagnostic system according to claim 2, characterized in that: The target weight is calculated as follows: Based on the adaptive weight method, the judgment matrix is established by comparing the input matching matrix A with each other. The subjective weight α of each indicator of the matching matrix A is obtained by calculating the eigenvector and normalizing it. z , and then build an adaptive function based on the historical operation data to calculate the dynamic weight β of each indicator of the historical operation data z , by weighted averaging and normalizing the subjective weight and dynamic weight, the precise weight is obtained as the evaluation standard of the health baseline. The calculation formulas of weighted average and normalization are shown in formulas (3) and (4): oh z =h·a z +(1-h)·β z (3); Where: h represents the proportion of subjective weight, 0<h<1; n represents the number of indicators.
4. The marine high-voltage transformer diagnostic system according to claim 1, characterized in that: Fault data processing in the CNN model is as follows: Using the convolution layer, convolution operations are performed on the data in the relational database to obtain feature data and establish a feature data set. The feature data calculation formula is as follows: Where: represents the feature map corresponding to the jth neuron in the lth layer; express Input, feature map of l-1 layer; Represents the convolution kernel from the i-th neuron in layer l-1 to the j-th neuron in layer l; Represents the output bias of the j-th neuron in the l-th layer.
5. The marine high-voltage transformer diagnostic system according to claim 4, characterized in that: Obtain a feature dataset, build a Transformer model, and use the Transformer model to calculate the maximum pooling in the feature dataset. The calculation process is as follows: The pre-planned pooling window is f×f, and the feature data set is divided into several pooling areas of size f×f. The maximum value of all elements in the pooling area is extracted by formula (6) as the value of the corresponding position of the output feature data. By presetting the step size of the pooling window, the feature data set is subjected to non-overlapping pooling to obtain the adopted parameter configuration, where formula (6) is as follows: Where: Y(i, j) represents the (i, j)th element of the output feature data; c represents the offset in the height direction of the feature data set; e represents the offset in the width direction of the feature data set; s represents the step size; X(i×s+c,j×s+e) represents the element in the local window corresponding to the output position (i, j) in the input feature data.
6. The marine high-voltage transformer diagnostic system according to claim 5, characterized in that: The optimized sparrow search algorithm is used to search the parameter configuration in the Transformer model, update the parameter configuration, obtain the iterative parameters, and use them as the input of the Transformer model to update the Transformer model. The output of the Transformer model is converted into the probability distribution of fault classification through the Softmax classifier, and the classification label is output to identify the transformer fault.
7. The marine high-voltage transformer diagnostic system according to claim 6, characterized in that: in, The output classification process of the Softmax classifier is as follows: The loss function is calculated by formula (7), which is as follows: Where: K{·} represents a one-time function; m represents the total number of input samples; y represents the true label; T represents the transposition operation; x represents the feature vector; θ represents the parameter vector; Add rule items, Classify sample m into the probability of class j, calculate the corresponding partial derivative of the loss function, and use the gradient descent method to update the parameters in the iterative process to obtain the classification result and complete the fault diagnosis.