Transformer intelligent fault diagnosis method based on analysis of gas dissolved in oil
Through the transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, through data preprocessing and feature selection, combined with the improved sparrow algorithm to optimize the support vector machine, the problems of imperfect encoding and single data information in transformer fault diagnosis are solved, and the problem of imperfect coding and single data information is improved. The accuracy of fault identification ensures the reliability and safety of the power grid.
Patent Information
- Application Number
- CN202411097135.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art has problems such as imperfect encoding, absolute judgment boundaries, single and unstructured data information in the transformer fault diagnosis, resulting in low accuracy of fault identification and unable to meet the safe and stable operation needs of the power system.
Using intelligent fault diagnosis method based on dissolved gas analysis in oil, the original data is obtained through oil-chemical tests, normalized processing and encoding, the best feature set is selected using orthogonal tests, and the support vector machine is optimized through the improved sparrow algorithm. Parameters, establish an intelligent transformer fault diagnosis model.
It improves the accuracy of transformer fault diagnosis, overcomes the shortcomings of traditional methods, and ensures the reliability and safe and stable operation of the power grid.
Smart Images

Figure CN119004233A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of transformer fault diagnosis, and in particular relates to an intelligent transformer fault diagnosis method based on analysis of dissolved gas in oil. Background Art
[0002] As a key link in the power transmission and transformation system, the safe and stable operation of the transformer is directly related to the stability of the power system. If the transformer equipment is not properly maintained, the transformer will operate in an unhealthy or sub-healthy state for a long time, which may lead to accidents such as transformer emergency stop and bus tripping, directly affecting the economic operation of the power system and social production and life. Due to the randomness and unpredictability of power transformer failures, coupled with the low early transformer fault diagnosis recognition rate and untimely diagnosis. Therefore, it is necessary to study the accurate identification of power transformer faults.
[0003] At present, traditional oil-dissolved gas analysis methods such as characteristic gas method, David triangle method, and three-ratio method are mainly used for fault diagnosis and analysis of oil-immersed transformers. However, these methods generally have problems such as imperfect coding, absolute judgment boundaries, and single and unstructured collected data information. These problems not only increase the difficulty of data processing, but also lead to low fault identification accuracy.
[0004] With the rapid development of machine learning technology, a variety of intelligent optimization algorithms have been applied to power transformer fault diagnosis, such as BP neural network, support vector machine, extreme learning machine, etc. Artificial intelligence algorithms have improved the accuracy of transformer fault diagnosis to a certain extent, and at the same time meet the requirements of rapidity, flexibility and high precision of power transformer fault diagnosis. Therefore, it is necessary to study the intelligent algorithm model of power transformer fault diagnosis to achieve efficient identification of fault types and improve power supply reliability.
[0005] CN115577639A discloses a transformer fault diagnosis method based on sparrow search method to optimize support vector machine, including: collecting concentration data of dissolved gas in transformer oil, establishing original data set M; generating a candidate feature data set N, divided into training set N1 and test set N2; calculating the importance value of each candidate feature in training set N1 by random forest algorithm; selecting the optimal feature data set corresponding to the highest fault diagnosis accuracy as the optimal feature training set; optimizing the parameters of support vector machine by sparrow search method; and using support vector machine SVM with optimal parameter combination to diagnose transformer fault. The present invention calculates the feature importance of all features in data samples based on random forest algorithm, selects features by using out-of-bag error, and effectively eliminates redundant features; optimizes the parameters of support vector machine by sparrow optimization algorithm, speeds up the optimization speed, thereby obtaining a support vector machine classification model with optimal parameters, and improving the accuracy of transformer fault diagnosis.
[0006] CN114925612A discloses a transformer fault diagnosis method based on sparrow search algorithm optimization hybrid core extreme learning machine, including the steps of: 1) collecting data of transformer operation in six states and preprocessing the data; 2) building a hybrid core extreme learning machine prediction model; 3) determining the population size, discoverer ratio, warning ratio and objective function parameters in SSA, and initializing the population; 4) selecting the error rate of argmin (TrainErrorRate) training samples as the fitness function of SSA; 5) passing the population state information to the HKELM network to obtain different HKELM networks; 6) training the HKELM network to obtain fitness values; 7) comparing fitness values and updating the population state; 8) judging whether the update process meets the requirements; 9) outputting the optimal parameters and bringing them into the HKELM network, and using the optimized network to diagnose transformer faults. This method can greatly improve the stability and learning speed of the forward neural network while ensuring that the network has good generalization performance.
[0007] Although the above method has a certain improvement in diagnostic accuracy compared with the traditional method of transformer fault diagnosis, it is limited by the optimization algorithm, and the diagnostic accuracy needs to be improved, which cannot meet the needs of existing actual production work. Therefore, it is necessary to study an improved intelligent optimization algorithm to improve the accuracy of transformer fault diagnosis. Summary of the invention
[0008] To solve the above problems, the purpose of the present invention is to disclose a transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, so as to improve the accuracy of power transformer fault diagnosis and ensure the safe and stable operation of the power grid. It is achieved by adopting the following technical solutions.
[0009] The technical solutions adopted in this application are as follows: The present invention provides a transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, comprising the following steps: (1) Obtaining the original data of gas in oil through oilification test; (2) Preprocessing of power transformer oil and gas data: using the normalization formula to normalize the characteristic gases dissolved in the power transformer oil; (3) According to the IEC60599 standard and some actual operating data of power transformers, power transformer faults are divided into five categories and coded; (4) Generate multiple feature sets based on the three ratio method, David triangle method, uncoded ratio method and other methods, and encode the data; (5) Use orthogonal test to analyze the feature vector set to be selected and select the best feature set; (6) The general improved intelligent optimization algorithm optimizes the penalty parameter p and kernel function parameter g of the support vector machine, and establishes an intelligent fault diagnosis model for transformers based on dissolved gas analysis in oil.
[0010] The beneficial effects of the present invention are as follows: (1) The present invention obtains a characteristic gas ratio set based on traditional methods such as the characteristic gas method, David triangle method, and three ratio method; and selects characteristic vectors through orthogonal experiments to obtain oil and gas characteristics that are more closely related to power transformer faults.
[0011] (2) The intelligent transformer fault diagnosis method based on dissolved gas analysis in oil of the present invention overcomes the problems of imperfect coding, absolute judgment boundary, single and unstructured collected data information, etc., improves the accuracy of power transformer fault diagnosis, and ensures the reliability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 Schematic diagram of support vector machine classification.
[0013] Figure 2 Optimize the flow chart for improving the sparrow algorithm.
[0014] Figure 3 This is the flow chart of transformer intelligent fault diagnosis. DETAILED DESCRIPTION
[0015] In order to enable those skilled in the art to better understand and implement this patent, the specific implementation methods of this application are now described in detail in conjunction with the drawings in the specification.
[0016] See attached Figures 1 to 3 .
[0017] 1. Support Vector Machine With the continuous development of machine learning technology, a variety of machine learning methods have been applied to transformer fault diagnosis, such as support vector machine (SVM), neural network, etc. Among them, having a large amount of sample data is an important prerequisite for the neural network to show excellent recognition performance. However, in the actual power grid, the probability of transformer failure is relatively low, and the collected data is often relatively small, with only a few to dozens of each type. If neural networks are used for classification, they often fail to achieve the expected results. Therefore, it is difficult to accurately identify the fault type of power transformers with only small sample data.
[0018] Support vector machines have outstanding generalization capabilities, do not require a large amount of data for training, and can solve small sample and nonlinear problems such as transformer fault diagnosis. They perform well in transformer fault diagnosis. Support vector machines (SVM) were originally proposed to study linearly separable binary classification problems. The core idea is to achieve data set classification through hyperplanes and interval maximization. The schematic diagram of the classification idea of support vector machines is shown in the attached figure. Figure 1 shown.
[0019] However, since most of the practical applications are nonlinear, it is necessary to select a kernel function to complete the classification prediction. Commonly used kernel functions include polynomial kernel function, linear kernel function, Gaussian radial basis kernel function, Sigmoid kernel function, etc. However, in solving the multi-classification problem of power transformer fault diagnosis, it is necessary to reasonably set the penalty parameter p and kernel function parameter g to construct the classifier. Therefore, it is necessary to study the key parameter setting method of support vector machine for adaptive transformer fault diagnosis.
[0020] 2. Improved sparrow algorithm Existing research mainly uses intelligent optimization algorithms for optimization. As an algorithm evolved from the foraging process and anti-predation process of sparrows, the sparrow algorithm (SSA) has stronger global optimization and local search capabilities than traditional genetic algorithms (GA) and particle swarm algorithms (PSO), but there are problems such as uneven initial population distribution and slow speed. In order to solve the problem of empirical setting of key parameters of support vector machines, this application introduces cubic chaos mapping and adaptive weights to optimize the sparrow algorithm (SSA) to form an improved sparrow algorithm (ISSA). The optimization flow chart is attached. Figure 2 As shown, the specific steps are as follows: (1) Generate sequence y according to cubic chaotic map i , map the cubic sequence to sparrows to improve the population distribution: Where: y i is a cubic sequence; i = 1, 2, ..., N, y 0 ≠0, X i is the population number; b min and b max are the lower and upper bounds of the search space respectively.
[0021] (2) Introduce adaptive weights and update the discoverer position to: Where: is the position of the ith sparrow in the nth generation; n is the current iteration number; T maxis the maximum number of iterations, set to 50; γ 1 and γ 2 They are random numbers (0,1) and random numbers that obey the normal distribution; is the position of the sparrow with the best global fitness; L is a multi-dimensional all-one matrix; R 2 is a warning value of [0, 1]; ST is a safety threshold of [0.5, 1]. 2 )<safety threshold (ST), the discoverer continues to search for food in the original place, guiding the population to obtain better fitness; when the warning value (R 2 )≥safety threshold (ST), the discoverer needs to move to a safe area immediately to prevent the appearance of predators.
[0022] (3) During the foraging process, if the discoverer finds a better food source, the joiner will go there immediately and its position will be updated to: Where: N is the population size, which is 100; and are the optimal and worst positions for individual sparrows respectively. When i>N / 2, the ith joiner will leave the original foraging ground and move to other food producing areas. When i≤N / 2, the ith joiner will forage for food in the original place.
[0023] (4) When danger occurs during foraging, the alerter leaves the original location and moves closer to other safe food sources or other sparrows. Its position is updated as follows: Where: γ 3 and γ 4 are random numbers that obey the normal distribution and random numbers in [-1,1] respectively; ε is a constant, usually 10 -50 ; C i is the fitness of the current sparrow individual, C w is the global worst fitness, and both change with the number of iterations.
[0024] 3. Intelligent fault diagnosis model for transformers based on dissolved gas analysis in oil In order to solve the problems of incomplete coding and insufficient diagnostic accuracy of the traditional ratio method and improve the accuracy of transformer fault diagnosis, this application proposes a transformer intelligent fault diagnosis model based on oil dissolved gas analysis. The transformer intelligent fault diagnosis flow chart is shown in the attached figure. Figure 3 As shown, the specific steps are as follows: (1) Preprocessing of power transformer oil and gas data. The characteristic gas dissolved in the power transformer oil is normalized according to formula (6): In the formula, x imin 、x imax is the minimum and maximum value before normalization, x i is the value before normalization; (2) According to the IEC 60599 standard and some actual operating data of power transformers, power transformer faults are divided into five types, and the specific coding table is shown in Figure 1. At the same time, the characteristic gases dissolved in the power transformer oil are coded.
[0025] Table 1 Equipment status code table (3) Based on five gases, including hydrogen (H2), methane (CH4), acetylene (C2H2), ethylene (C2H4) and ethane (C2H6), a total of 21 gas ratios are generated as the candidate feature set based on the IEC three-ratio method, David triangle method, and uncoded ratio method, and the transformer oil and gas data are encoded. The transformer fault feature set is shown in Table 2.
[0026] Table 2 Transformer fault feature set Among them, total hydrocarbons (THC) = methane (CH4) + acetylene (C2H2) + ethylene (C2H4) + ethane (C2H6); total gas (SUM) = hydrogen (H2) + methane (CH4) + acetylene (C2H2) + ethylene (C2H4) + ethane (C2H6); (4) Use orthogonal experiments to obtain the best feature set for power transformer classification. Through multiple simulations, the best feature sets are determined to be CH4 / C2H6, C2H6 / CH4, CH4 / THC, C2H2 / THC, C2H6 / THC, (CH4+C2H4) / THC; (5) The oil and gas data of power transformers are divided into a training group and a prediction group in proportion (8 / 2). With accuracy as the fitness function, the penalty parameter p and kernel function parameter g of the support vector machine are trained and tested using the training group, the prediction group and the improved sparrow algorithm to obtain a transformer intelligent fault diagnosis model.
[0027] Overall summary: A transformer intelligent fault diagnosis method based on dissolved gas analysis in oil includes the following steps: Step 1: obtaining raw data of gas in oil through oiling test; Step 2: Preprocessing of oil and gas data of power transformers, using normalization formula to normalize characteristic gases dissolved in power transformer oil; Step 3: According to the IEC60599 standard and some actual operating data of power transformers, the power transformer faults are divided into 5 categories and coded; Step 4: Generate multiple feature sets based on the three ratio method, David triangle method, and uncoded ratio method, and encode the data; Step 5: Use orthogonal test to analyze the feature vector set to be selected and select the best feature set; Step 6: The general improved intelligent optimization algorithm optimizes the penalty parameter p and kernel function parameter g of the support vector machine to establish an intelligent fault diagnosis model for transformers based on dissolved gas analysis in oil.
[0028] The present application has the following major beneficial technical effects: (1) The present invention obtains a characteristic gas ratio set based on traditional methods such as the characteristic gas method, David triangle method, and three ratio method; and selects characteristic vectors through orthogonal experiments to obtain oil and gas characteristics that are more closely related to power transformer faults.
[0029] (2) The intelligent transformer fault diagnosis method based on dissolved gas analysis in oil of the present invention overcomes the problems of imperfect coding, absolute judgment boundary, single and unstructured collected data information, etc., improves the accuracy of power transformer fault diagnosis, and ensures the reliability of power grid operation.
[0030] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including the equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, Features The steps include: Step 1: Get the original data of gas in oil through oilification test; Step 2: Preprocessing of power transformer oil and gas data, using the normalization formula to normalize the characteristic gases dissolved in the power transformer oil; Step 3: According to the IEC60599 standard and some actual operating data of power transformers, the power transformer faults are divided into 5 categories and coded; Step 4: Generate multiple feature sets based on the three ratio method, David triangle method, and uncoded ratio method, and encode the data; Step 5: Use orthogonal test to analyze the feature vector set to be selected and select the best feature set; Step 6: The general improved intelligent optimization algorithm optimizes the penalty parameter p and kernel function parameter g of the support vector machine to establish an intelligent fault diagnosis model for transformers based on dissolved gas analysis in oil.
2. According to claim 1, a transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, Features: In the second step, the characteristic gases dissolved in the power transformer oil are normalized according to the following formula: In the formula, x imin 、x imax is the minimum and maximum value before normalization, x i is the value before normalization.
3. According to claim 2, a transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, Features: In the third step, the power transformer faults are divided into five types, and the characteristic gases dissolved in the power transformer oil are coded: The corresponding relationship is as above.
4. According to claim 3, a transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, Features: In the fourth step, based on the five gases of hydrogen (H2), methane (CH4), acetylene (C2H2), ethylene (C2H4) and ethane (C2H6), a total of 21 gas ratios are generated as the candidate feature set based on the IEC three-ratio method, David triangle method and uncoded ratio method, and the transformer oil and gas data are encoded. The transformer fault feature set is shown in the following table: Among them, total hydrocarbons (THC) = methane (CH4) + acetylene (C2H2) + ethylene (C2H4) + ethane (C2H6); total gas (SUM) = hydrogen (H2) + methane (CH4) + acetylene (C2H2) + ethylene (C2H4) + ethane (C2H6).
5. According to claim 4, a transformer intelligent fault diagnosis method based on dissolved gas analysis in oil, Features: In the fifth step, the orthogonal test is used to obtain the best feature set for power transformer classification. Through multiple simulations, the best feature sets are determined to be CH4 / C2H6, C2H6 / CH4, CH4 / THC, C2H2 / THC, C2H6 / THC, and (CH4+C2H4) / THC.
6. The intelligent transformer fault diagnosis method based on dissolved gas analysis in oil according to claim 5, Features: In the sixth step, the oil and gas data of the power transformer are divided into a training group and a prediction group, with the ratio of the training group to the prediction group being 8:
2. Taking the accuracy as the fitness function, the penalty parameter p and the kernel function parameter g of the support vector machine are trained and tested using the training group, the prediction group and the improved sparrow algorithm to obtain the transformer intelligent fault diagnosis model.
7. The intelligent transformer fault diagnosis method based on dissolved gas analysis in oil according to claim 6, Features: The specific steps to improve the sparrow algorithm are as follows: Step 1: Generate sequence y according to cubic chaotic map i , map the cubic sequence to sparrows to improve the population distribution: Where: y i is a cubic sequence; i = 1, 2, ..., N, y 0 ≠0, X i is the population number; b min and b max are the lower and upper bounds of the search space respectively; Step 2: Introduce adaptive weights and update the finder position to: Where: is the position of the ith sparrow in the nth generation; n is the current iteration number; T max is the maximum number of iterations, set to 50; γ 1 and γ 2 They are random numbers (0,1) and random numbers that obey the normal distribution; is the position of the sparrow with the best global fitness; L is a multi-dimensional all-one matrix; R 2 is a warning value of [0, 1]; ST is a safety threshold of [0.5, 1]. 2 )<safety threshold (ST), the discoverer continues to search for food in the original place, guiding the population to obtain better fitness; when the warning value (R 2 )≥safety threshold (ST), the discoverer needs to move to a safe area immediately to prevent the appearance of predators; Step 3: During the foraging process, if the discoverer finds a better food source, the joiner will go there immediately and its position will be updated to: Where: N is the population size, which is 100; and are the best and worst positions of individual sparrows respectively; when i>N / 2, the i-th joiner will leave the original foraging area and move to other food producing areas; when i≤N / 2, the i-th joiner will forage in the original place; Step 4: When danger occurs during foraging, the alarm will leave the original location and move closer to other safe food sources or other sparrows. Its position will be updated as follows: Where: γ 3 and γ 4 are random numbers that obey the normal distribution and random numbers in [-1,1], respectively. ε is a constant, and ε is 10 -50 , C i is the fitness of the current sparrow individual, C w is the global worst fitness.
Citation Information
Patent Citations
Transformer fault diagnosis method for optimizing mixed kernel extreme learning machine based on sparrow search algorithm
CN114925612A
Transformer fault diagnosis method for optimizing support vector machine based on sparrow search method
CN115577639A
Unit combination output distribution method and system based on improved whale algorithm
CN112865195A
Cloud center resource load prediction method based on variational mode and ISSA-LSSVM
CN115934324A
High-voltage isolation switch fault diagnosis method
CN117388681A
Cited By
Joint feature-based model fusion transformer fault detection method and system
CN120011933A
