Multi-component mixed insulating oil transformer fault diagnosis method and system

Abstract

The invention discloses a multi-component mixed insulating oil transformer fault diagnosis method and system, and relates to the technical field of transformer fault diagnosis, and the method comprises the steps: determining the state characteristic parameters of a fault intelligent diagnosis model; expanding state characteristic parameter data based on an SMOTE algorithm; constructing a fault intelligent diagnosis model based on a random forest algorithm; and optimizing the fault intelligent diagnosis model based on a grid search algorithm. According to the multi-component mixed insulating oil transformer fault diagnosis method provided by the invention, efficient and accurate fault identification is realized by combining the dissolved gas in the oil and the partial discharge characteristic parameters. By synthesizing a small number of oversampling technology extended data, the problem of data imbalance is solved, and the stability and reliability of the diagnosis model are ensured. According to the method, power failure operation is not needed, the fault screening process is simplified, and the diagnosis efficiency and real-time performance are improved. The optimization algorithm further enhances the adaptability of the model, is widely suitable for different types of transformers, and meets the requirements of intelligent maintenance and safe operation.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer fault diagnosis, and in particular to a multi-component mixed insulating oil transformer fault diagnosis method and system. Background Art

[0002] Insulating oil with excellent performance is crucial to ensure the safe operation of transformers in oil-immersed power grids. Currently, mineral oil transformers account for more than 90% of oil-immersed transformers. Traditional mineral oil transformers have been used in power grids for more than a hundred years, but the mineral oil-paper insulation system has poor thermal stability and low explosion temperature, which seriously restricts the safe operation level of transformers. Compared with mineral oil, natural esters have good physical, chemical and electrical properties, and perform outstandingly in terms of life extension, renewability, fire resistance, and environmental protection, but they also have shortcomings such as high kinematic viscosity and poor oxidation stability.

[0003] Mixed insulating oil can complement the performance advantages of different oils and is an important direction for the development of new liquid dielectrics. By compounding multiple types of insulating oils in a certain proportion, the resulting mixed insulating oil has the special advantage of taking into account the advantages of different insulating oils.

[0004] In recent years, domestic and foreign scholars have conducted research on mixed insulating oils that combine the advantages of multiple oil products, and distribution transformers using or replacing mixed insulating oils have been put into operation in some areas.

[0005] However, there are few studies on fault diagnosis methods for mixed oil transformers. In order to ensure the safe and stable operation of insulating oil transformers, it is urgent to carry out key research on the operation and maintenance of multi-component mixed insulating oil transformers based on dissolved gas analysis and partial discharge in oil. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is that during operation, the insulation system of the oil-immersed transformer will undergo irreversible aging under the action of thermal stress and electrical stress. The oil-paper insulation effect is significantly reduced after aging, and may even cause insulation failure, posing a serious safety hazard.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: a multi-component mixed insulating oil transformer fault diagnosis method, comprising:

[0009] Determine the state characteristic parameters of the fault intelligent diagnosis model;

[0010] Expand state feature parameter data based on SMOTE algorithm;

[0011] Build an intelligent fault diagnosis model based on the random forest algorithm;

[0012] Optimize the intelligent fault diagnosis model based on grid search algorithm.

[0013] As a preferred solution of the multi-component mixed insulating oil transformer fault diagnosis method described in the present invention, the state characteristic parameters include dissolved gas parameters in the oil and partial discharge characteristic parameters.

[0014] As a preferred embodiment of the multi-component mixed insulating oil transformer fault diagnosis method of the present invention, wherein: the dissolved gas parameters in the oil include CH4, C2H2, C2H4, C2H6 and total hydrocarbons;

[0015] The local discharge characteristic parameters include the number of discharge pulses, the maximum discharge amount and the average discharge amount.

[0016] As a preferred solution of the multi-component hybrid insulating oil transformer fault diagnosis method of the present invention, wherein: the SMOTE algorithm is expressed as:

[0017] p j =x+rand(0,1)×(y j -x),j=1,2,...,N

[0018] Among them, rand(0,1) represents a random number in the interval (0,1).

[0019] As a preferred solution of the multi-component hybrid insulating oil transformer fault diagnosis method of the present invention, wherein: the expanded state characteristic parameter data includes expanding the data of each fault type to the same number;

[0020] Calculate the distance between each minority sample and other similar samples, and take the K samples with the smallest distance as the nearest neighbors;

[0021] Randomly select one of the K nearest neighbors, connect it to the minority sample, calculate the difference between them, and synthesize a new sample on the connection line according to the difference ratio;

[0022] Repeat the above steps of selecting nearest neighbors and linear interpolation to generate a specified number of synthetic samples.

[0023] As a preferred solution of the multivariate mixed insulating oil transformer fault diagnosis method of the present invention, wherein: the random forest algorithm is expressed as:

[0024]

[0025] Δi(t)=i(t p )-(P L ·i(t l )+P R ·i(t R ))

[0026] Among them, i(t) represents the best segmentation metric in the classification tree construction, which is to calculate the impurity function, p(j|t) represents the conditional probability that the sample in node t is of class j, M represents the total number of sample categories, and P L is the left child node t l For partition node t p The probability, P R is the right child node t R For partition node t p probability.

[0027] As a preferred solution of the multivariate hybrid insulating oil transformer fault diagnosis method of the present invention, wherein: the optimized fault intelligent diagnosis model includes the number of trees in the custom forest, the maximum depth of the tree, and the minimum number of samples required to split the internal nodes;

[0028] Create an initial model based on the task type and select evaluation indicators based on the task type;

[0029] Traverse all custom hyperparameter combinations and evaluate them to obtain the best hyperparameter combination, retrain the model, and evaluate the optimized performance based on the test set.

[0030] Another object of the present invention is to provide a multi-component mixed insulating oil transformer fault diagnosis system, which can construct a multi-component mixed insulating oil transformer fault diagnosis system, integrate the characteristic parameters of dissolved gases in oil such as CH4, C2H2, C2H4, C2H6 and total hydrocarbons, and the local discharge characteristic parameters including the number of discharge pulses, maximum discharge amount and average discharge amount, and perform intelligent diagnosis of transformer faults based on the GS-RF deep learning algorithm to ensure the safe operation of the mixed oil transformer.

[0031] In order to solve the above technical problems, the present invention provides the following technical solutions: a multi-component hybrid insulating oil transformer fault diagnosis system, comprising: a feature extraction module, a feature expansion module, a fault diagnosis module and a model optimization module;

[0032] The feature extraction module is used to determine the state feature parameters of the fault intelligent diagnosis model;

[0033] The feature expansion module is used to expand the state feature parameter data based on the SMOTE algorithm;

[0034] The fault diagnosis module is used to build a fault intelligent diagnosis model based on the random forest algorithm;

[0035] The model optimization module is used to optimize the fault intelligent diagnosis model based on a grid search algorithm.

[0036] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned multi-component hybrid insulating oil transformer fault diagnosis method when executing the computer program.

[0037] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the multi-component hybrid insulating oil transformer fault diagnosis method as described above.

[0038] Beneficial effects of the present invention: The multi-component mixed insulating oil transformer fault diagnosis method provided by the present invention realizes more accurate fault identification and improves the diagnosis efficiency by combining the dissolved gas in the oil and the characteristic parameters of local discharge.

[0039] In data processing, the synthetic minority oversampling technology was introduced to overcome the data imbalance problem in traditional methods and ensure the stability and reliability of the diagnostic model.

[0040] The entire diagnostic process is simple and quick, and fault screening can be completed without power outage, reducing dependence on experimental data and significantly improving the convenience and real-time nature of fault diagnosis.

[0041] By optimizing the algorithm and tuning the parameters of the diagnostic model, the model becomes more adaptable and can cope with complex actual operating conditions. It is widely applicable to different types of transformers and can meet the needs of intelligent maintenance and safety assurance during the operation of transformer equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0043] Figure 1 An overall flow chart of a multi-component mixed insulating oil transformer fault diagnosis method provided by an embodiment of the present invention.

[0044] Figure 2 A ratio diagram of different faults in original data of a multi-component mixed insulating oil transformer fault diagnosis method provided by an embodiment of the present invention.

[0045] Figure 3 A curve chart showing the change in accuracy of different models of a multi-component mixed insulating oil transformer fault diagnosis method as a function of the proportion of training sets provided in an embodiment of the present invention.

[0046] Figure 4A grid search optimized random forest accuracy change curve of a multi-component hybrid insulating oil transformer fault diagnosis method provided by an embodiment of the present invention.

[0047] Figure 5 A grid search optimized random forest confusion matrix diagram of a multi-component hybrid insulating oil transformer fault diagnosis method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] Example 1, reference Figure 1-Figure 3 , as an embodiment of the present invention, provides a multi-component mixed insulating oil transformer fault diagnosis method, comprising:

[0051] Step 1: Determine the state characteristic parameters for the fault intelligent diagnosis model. Based on the existing research on gas production and fault diagnosis methods of transformers under different faults at home and abroad, combined with the gas production obtained by experimental measurement of the oil used in the present invention under different faults, CH4, C2H2, C2H4, C2H6 and total hydrocarbon dissolved in oil are selected as characteristic parameters. According to the relevant standards of GB / T 6451-2023 and IEC 60076 and the current research status at home and abroad, the number of discharge pulses, maximum discharge amount and average discharge amount are selected as partial discharge characteristic parameters.

[0052] The main insulating materials of oil-immersed transformers are insulating oil and insulating paper. During long-term operation, they will gradually deteriorate under the influence of electric fields, moisture, temperature and mechanical forces, and eventually cause failures and end the life of the transformer. When thermal failures or discharge failures occur inside the equipment, various gases will be dissolved in the oil. Different types of gases and concentrations can reflect different types of failures. The fault diagnosis method based on dissolved gas in oil has the advantages of not requiring the transformer to be powered off and can be operated. Analyzing the composition and content of dissolved gas in oil is one of the most effective measures to monitor the safe operation of oil-filled electrical equipment. It has been recognized worldwide as a method for detecting initial transformer failures.

[0053] Existing studies on the gas production of transformers under different faults at home and abroad have found that the main aging gas product of mineral oil is CH4, while C2H6 is the main gas product shared by various types of vegetable oils in the process and is the low-temperature thermal fault product of various vegetable oils. The study also found that a certain amount of C2H2 will be produced in the oil regardless of whether it is mineral or plant insulation and whether there is insulating paper or not. The characteristic gases mainly used in the existing transformer fault diagnosis methods at home and abroad are H2, C2H4, C2H2, CH4, C2H6, etc. Combined with the gas production obtained by experimental measurement of the oil used in the present invention under different faults, the dissolved gases in the oil of CH4, C2H2, C2H4, C2H6 and total hydrocarbons are finally selected as characteristic parameters.

[0054] The insulation system of power transformers is complex, and insulation defects are easily formed during the operation of the equipment, resulting in uneven local electric field distribution and local discharge on the surface, inside or metal tip of the insulating medium. Local discharge causes continuous damage to the insulating material, resulting in a decrease in local insulation performance, threatening the safe operation of the transformer. Therefore, the study of the local discharge characteristics of the multi-component mixed insulating oil-paper composite system is of practical significance for the development and operation and maintenance of mixed oil transformers. According to the relevant standards of GB / T 6451-2023 and IEC 60076 and the current research status at home and abroad, the number of discharge pulses, the maximum discharge amount and the local average discharge amount are selected as the discharge characteristic parameters.

[0055] Table 1 Relationship between discharge fault types of transformers with multi-component mixed insulating oil and dissolved gas components in the oil

[0056]

[0057] Table 2 Relationship between overheating fault type of transformer with multi-component mixed insulating oil and dissolved gas components in oil

[0058]

[0059]

[0060] Step 2: Expand the state feature parameter data based on the SMOTE algorithm. The data set for machine training in the present invention is derived from real test data. The amount of data is not enough to support machine learning training, so the data set needs to be expanded first. For the data used for machine training in this patent, the synthetic minority over-sampling technique (SMOTE) is the most appropriate.

[0061] Since the data set for machine training in the present invention is derived from real test data, the number of direct use is not enough to support machine learning training, and the data set needs to be expanded first. Common data set expansion methods include: random oversampling method, which directly randomly selects minority class samples and copies them so that their number is equivalent to the number of majority class samples; synthetic minority oversampling technology, which is different from the random oversampling method in that it directly copies existing minority class samples, but interpolates and synthesizes new samples between multiple minority class samples; adaptive synthetic sampling technology, which takes into account the difficulty of each minority class sample, that is, its closeness to the majority class sample, and dynamically determines how many synthetic samples to generate. For the data used for machine training in this patent, it is most appropriate to select the synthetic minority oversampling technique (Synthetic Minority Over-sampling Technique, SMOTE). The SMOTE algorithm is an oversampling method that mainly inserts new samples into some minority class samples with similar positions in order to achieve the purpose of balancing samples. The characteristic of the SMOTE algorithm is that it does not simply copy samples according to the random oversampling method, but adds new non-existent samples, so it can avoid overfitting of the classifier to a certain extent.

[0062] Assume that there are minority class samples. For each sample x, search its k (usually 5) minority class nearest neighbor samples. If the upsampling factor is N, randomly select N samples from its k nearest neighbor samples, denoted as y1, y2, y3..., y n ; Perform random linear interpolation between minority class samples x and yj (j=1,2,…,n) to construct new minority class samples pj.

[0063] p j =x+rand(0,1)×(y j -x), j=1,2,...,N (1)

[0064] In the formula, x is each sample in the original minority class sample, rand(0,1) represents a random number in the interval (0,1), and y j are some of the nearest neighbor samples of the minority class of the randomly selected sample x. By merging these newly synthesized minority class sample points into the original data set, a new training set can be generated.

[0065] In combination with the method of the present invention, the SMOTE method is used to expand the data of each fault type to the same number (100). The main steps are as follows:

[0066] ① Calculate the distance between each minority sample and other similar samples, and take the K samples (usually 5) with the smallest distance as the nearest neighbors.

[0067] ② Randomly select one of the K nearest neighbors, connect it to the minority sample, calculate the difference between them, and synthesize a new sample located on the connection line based on the difference ratio. The newly generated feature vector retains some characteristics of the original sample and introduces some characteristics of the neighbor sample.

[0068] ③ Repeat the above steps of selecting nearest neighbors and linear interpolation to generate a specified number of synthetic samples.

[0069] Step 3: Construct an intelligent fault diagnosis model based on the random forest algorithm. Based on the content of dissolved hydrocarbon gas in oil, the diagnostic effects of several machine learning methods were preliminarily compared, and it was found that the random forest method had the best effect. When the training set accounted for 80%, the accuracy was close to 90%. Therefore, the random forest method was selected as the transformer fault assessment and diagnosis method in the present invention.

[0070] Random forest is an ensemble learning method. It is implemented on the basis of decision trees. It is a supervised ensemble learning method of the bagging type. It improves the prediction accuracy by integrating multiple decision trees. It is the most commonly used ensemble learning model. Decision tree is a tree structure based on information entropy. It divides a feature into different subtrees according to the amount of information. When all samples in the subtree belong to the same category, the division is stopped. Common feature division algorithms include ID3, C4.5, CART, etc. The amount of information is used to measure the amount of information contained in an event. i Point to x i The probability of occurrence, as shown in formula (2), the smaller the probability of an event occurring, the greater the amount of information it contains.

[0071] I(X=x i )=-logp i (2)

[0072] Information entropy is used to measure the uncertainty of variables, as shown in formula (3), where p i is the probability of the occurrence of the ith event, and k is the total number of events. The larger the information entropy value, the greater the amount of information and the higher the value of the information.

[0073]

[0074] The core of the random forest model is composed of multiple decision trees. It improves prediction accuracy and prevents overfitting by constructing multiple decision trees (weak classifiers) and aggregating their prediction results. Each decision tree corresponds to a training set. To build N decision trees, it is necessary to generate a corresponding number of training sets. Generating N training subsets from the original training set involves statistical sampling technology. When constructing each tree, the random forest will randomly select a part of the original features to split the nodes, and at the same time perform replacement sampling on the training data. Sampling with replacement means that when extracting samples, a sampling method is used in which the extraction results are replaced after extraction, so that the size of the data set remains unchanged each time sampling occurs, and duplicate samples appear in the extracted sample set. Binary classification is repeatedly performed on each non-leaf node of the decision tree until the sample cannot be divided any further. The best segmentation measurement method in the classification tree construction is to calculate the impurity function i(t), as shown in formula (4), p(j|t) represents the conditional probability that the sample in node t is of class j, M is the total number of sample categories, and the greater the purity of the sample set, the smaller the value of the impurity function.

[0075]

[0076] The impurity function is used to select the split attribute xp and the split value p so that the homogeneity of the left and right nodes in the classification tree is maximized, that is, the maximization formula (5), where PL is the probability of the left child node tl to the split node tp, and PR is the probability of the right child node tR to the split node tp. After the decision tree is built, it needs to be pruned to prevent it from overfitting.

[0077] Δi(t)=i(t p )-(P L ·i(t l )+P R ·i(t R )) (5)

[0078] Random forests provide an integrated prediction result of multiple decision trees to determine the final result: for classification tasks, the prediction results of the tree are determined by voting (classification); for regression tasks, the final prediction is obtained by averaging the prediction results of all trees (regression). Random forests are highly random, and the selection of features for each sampling and each split is different. They can be effectively applied to large-scale samples with high accuracy. They can also process high-dimensional data, are insensitive to missing data, and can evaluate the importance of each feature.

[0079] In order to confirm the accuracy of the random forest method in transformer fault diagnosis, the content of dissolved hydrocarbon gas in oil was used as the criterion for preliminary screening. Figure 2 As shown, the proportions after treatment are the same, both for 100 groups.

[0080] The content of dissolved hydrocarbon gas in oil is used as the criterion for preliminary screening. The study found that based on the standard parameter training of three models in Python, the curves of the accuracy of different models changing with the proportion of training sets are as follows: Figure 3 As shown. When judging only by the dissolved hydrocarbon gas in the oil, it is found that the random forest method is significantly better than other machine learning methods in most cases. And it is found that when the training set accounts for 80%, its accuracy is close to 90%. Among the other methods, the average accuracy of the support vector machine is only kept around 80%, and the neural network algorithm is the worst in this case, less than 60% in most cases. Therefore, a preliminary screening is carried out based on the global accuracy of each method, and the random forest method with the highest accuracy is confirmed as the model for further parameter optimization.

[0081] Step 4: Optimize the intelligent fault diagnosis model based on the grid search algorithm. In machine learning, optimizing model parameters is a key step to improve model performance and ensure that the model is accurately generalized to new data. The core of parameter optimization is to optimize the hyperparameters that control the behavior of the learning algorithm and the model structure. Hyperparameter optimization is a typical combinatorial optimization problem with the characteristics of high dimension and non-convexity. It is usually costly to evaluate the performance of each set of hyperparameters, so efficient optimization strategies and computing resource management are crucial. The present invention uses an efficient grid search algorithm (Grid Search, GS) to optimize the random forest method.

[0082] In machine learning, optimizing model parameters is a key step to improve model performance and ensure that the model accurately generalizes to new data. Parameter optimization includes parameters that can be automatically updated during the training process, as well as hyperparameters that are set before model training to control the behavior of the learning algorithm and the model structure. Different machine learning hyperparameters are also different. Common hyperparameters include: learning rate in neural networks, number of decision trees in random forests, SVM penalty parameters, etc. This series of adjustable hyperparameters cannot be learned directly from training data and need to be set manually. Manually finding the optimal parameter combination is not only time-consuming but also inefficient compared to using optimization algorithms. Hyperparameter optimization is a typical combinatorial optimization problem with high dimensionality and non-convexity. It often requires retraining the model and evaluating its performance on the validation set. Evaluating the performance of each set of hyperparameters is usually costly, so efficient optimization strategies and computing resource management are essential for large-scale machine learning projects. Common hyperparameter optimization methods include random search, Bayesian optimization, simulated annealing, etc. This paper uses grid search to optimize random forests.

[0083] Grid Search (GS):

[0084] This paper uses grid search to optimize the hyperparameters of random forest. Grid search is a method of searching for all parameter combinations within a rectangular range according to a given step size. Based on the idea of ​​this method, this paper selects the value range of the neighborhood factor k and the embedding dimension d for a given sample set, and combines the value ranges of the two into a grid area. The k and d corresponding to the maximum average value of the Silhouette index of the sample set after feature compression in this area are the optimal parameters. According to the parameters, the feature samples after feature compression of the sample set are obtained.

[0085] Grid search refers to gridding the variable region, traversing all grid points, solving the objective function value that satisfies the constraint function, and finally comparing and selecting the optimal point. It provides a systematic way to find the optimal hyperparameter combination and is widely used in the training process of machine learning models. The specific operations are as follows:

[0086] ① Define the hyperparameter space: customize the hyperparameters that need to be optimized and their value ranges. Typical hyperparameters include: the number of trees in the forest, the maximum depth of the tree, the minimum number of samples required to split internal nodes, etc.

[0087] ② Initialize the random forest model: Create an initial model based on the task type.

[0088] ③Select evaluation indicators: Select appropriate evaluation indicators based on the task type. For example, for the classification task in this article, you can choose Accuracy, F1-Score, etc.

[0089] ④ Grid search process:

[0090] 1) Set the range of neighborhood factor k and embedding dimension d in grid search and their corresponding step size. Construct a two-dimensional grid in the two-dimensional space composed of k and d, and use Silhouette to represent the initial value of the average value S of the Silhouette index.

[0091] 2) Select a set of parameter values ​​on the two-dimensional grid constructed by k and d, and use the Laplace eigenmapping algorithm to map the original high-dimensional data set X to the d-dimensional space to obtain the d-dimensional data set Y.

[0092] 3) For data set Y, calculate the corresponding Silhouette index average if And record the current neighborhood factor k and embedding dimension d as the current optimal parameters.

[0093] 4) Repeat steps 2) and 3) until all parameter combinations in the two-dimensional grid are tested.

[0094] 5) Finally, the optimal parameters k and d are obtained, and sensitive features are obtained through Laplace mapping.

[0095] ⑤Perform grid search: Based on the previously defined model instance, hyperparameter grid, number of cross-validation folds (for example, cv=5 means 5-fold cross-validation), scoring indicators, etc., traverse all custom hyperparameter combinations and evaluate them to obtain the best hyperparameter combination.

[0096] ⑥ Apply optimal parameters: Use the best hyperparameter combination obtained in the previous step to retrain the model and evaluate the optimized performance based on the test set.

[0097] Embodiment 2 is an embodiment of the present invention, which provides a multi-component mixed insulating oil transformer fault diagnosis system, including:

[0098] Feature extraction module, feature expansion module, fault diagnosis module and model optimization module;

[0099] The feature extraction module is used to determine the state characteristic parameters of the fault intelligent diagnosis model;

[0100] The feature expansion module is used to expand the state feature parameter data based on the SMOTE algorithm;

[0101] The fault diagnosis module is used to build an intelligent fault diagnosis model based on the random forest algorithm;

[0102] The model optimization module is used to optimize the fault intelligent diagnosis model based on the grid search algorithm.

[0103] Embodiment 3, an embodiment of the present invention, is different from the first two embodiments in that:

[0104] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0105] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0106] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0107] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0108] Example 4, reference Figure 4-Figure 5 , which is an embodiment of the present invention, provides a multi-component mixed insulating oil transformer fault diagnosis method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0109] Based on the characteristic parameters of dissolved hydrocarbon gas and discharge in oil, after systematically optimizing the random forest hyperparameters, a transformer fault diagnosis algorithm with high accuracy in the test set was obtained.

[0110] The grid search method is used to optimize the hyperparameters of the random forest. The 5-fold cross validation (cv=5) is used to evaluate the model performance. The accuracy (scoring='accuracy') is selected as the evaluation indicator. The number of decision trees tried 'n_estimators' (that is, the total number of trees in the random forest) ranges from [10, 50, 100, 200, 400]; the maximum depth of the tree 'max_depth' search range is [None, 10, 20, 30, 50], None means that there is no limit on the depth of the tree, allowing the tree to grow as deep as possible, while other values ​​set a specific depth upper limit to control the complexity of the model; the minimum number of samples required for internal node splitting 'mi n_samples_split' is set to [2,5,10], which means that at least 2, 5 or 10 samples are required to decide whether to further split a node; the minimum number of samples on a leaf node 'min_samples_leaf' is set to [1,2,4], which determines the minimum number of samples required to form a leaf node; the number of features to consider when finding the best split 'max_features' is set to ['log2','sqrt'], 'log2' means that the number of features considered at each split is at most the logarithm of the total number of features (base 2), and 'sqrt' means that it is at most the square root of the number of features, all of which are to reduce computational complexity and prevent overfitting.

[0111] Based on the above grid search settings, the optimized random forest method is used to obtain the best hyperparameter combination {'max_depth':30,'max_features':'sqrt','min_samples_leaf':1,'min_samples_split':2,'n_estimators':10}. With this parameter, 100% accuracy can be achieved in the test set. The accuracy changes with the parameter settings during the optimization process as shown in the figure. Figure 4 As shown in the figure, the parameter setting on the horizontal axis represents the sequence number of the hyperparameter combination in the grid search, and the vertical axis represents the average accuracy of all cross-validation folds under the given parameter setting.

[0112] When only the dissolved hydrocarbon gas in oil is used for single parameter diagnosis training, the random forest method has excellent results. When the training set ratio is 80%, the accuracy rate is close to 90%. Figure 5 The accuracy of grid search optimization random forest method applied to multivariate mixed insulating oil transformer fault diagnosis is determined. The data offset from the diagonal indicates inaccuracy. After combined diagnosis with discharge volume, the optimization result of random forest can obtain 96.67% accuracy in the training set.

[0113] The method of the present invention uses less data than the traditional method, the required experimental data is more streamlined, and the use of the diagnostic method is simpler and faster. While reducing the amount of data used, the fault diagnosis effect of the multi-component mixed insulating oil transformer is also guaranteed, with higher accuracy and better effect than the traditional method.

[0114] 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 preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multi-component mixed insulating oil transformer fault diagnosis method, characterized in that: include: Determine the state characteristic parameters of the fault intelligent diagnosis model; Expand state feature parameter data based on SMOTE algorithm; Build an intelligent fault diagnosis model based on the random forest algorithm; Optimize the intelligent fault diagnosis model based on grid search algorithm.

2. The multi-component mixed insulating oil transformer fault diagnosis method according to claim 1, characterized in that: The state characteristic parameters include dissolved gas parameters in oil and local discharge characteristic parameters.

3. The fault diagnosis method for a multi-component mixed insulating oil transformer according to claim 2, characterized in that: The dissolved gas parameters in the oil include CH4, C2H2, C2H4, C2H6 and total hydrocarbons; The local discharge characteristic parameters include the number of discharge pulses, the maximum discharge amount and the average discharge amount.

4. The multi-component mixed insulating oil transformer fault diagnosis method according to claim 3, characterized in that: The SMOTE algorithm is expressed as, p j =x+rand(0,1)×(y j -x),j=1,2,...,N Among them, rand(0,1) represents a random number in the interval (0,1).

5. The multi-component mixed insulating oil transformer fault diagnosis method according to claim 4, characterized in that: The expanding state characteristic parameter data includes expanding the data of each fault type to the same number; Calculate the distance between each minority sample and other similar samples, and take the K samples with the smallest distance as the nearest neighbors; Randomly select one of the K nearest neighbors, connect it to the minority sample, calculate the difference between them, and synthesize a new sample on the connection line according to the difference ratio; Repeat the above steps of selecting nearest neighbors and linear interpolation to generate a specified number of synthetic samples.

6. The multi-component mixed insulating oil transformer fault diagnosis method according to claim 5, characterized in that: The random forest algorithm is expressed as, Δi(t)=i(t p )-(P L ·i(t l )+P R ·i(t R )) Among them, i(t) represents the best segmentation metric in the classification tree construction, which is to calculate the impurity function, p(j|t) represents the conditional probability that the sample in node t is of class j, M represents the total number of sample categories, and P L is the left child node t l For partition node t p The probability, P R is the right child node t R For partition node t p probability.

7. The multi-component mixed insulating oil transformer fault diagnosis method according to claim 6, characterized in that: The optimized fault intelligent diagnosis model includes the number of trees in the custom forest, the maximum depth of the tree, and the minimum number of samples required to split the internal nodes; Create an initial model based on the task type and select evaluation indicators based on the task type; Traverse all custom hyperparameter combinations and evaluate them to obtain the best hyperparameter combination, retrain the model, and evaluate the optimized performance based on the test set.

8. A system using the multi-component mixed insulating oil transformer fault diagnosis method as claimed in any one of claims 1 to 7, characterized in that: include: Feature extraction module, feature expansion module, fault diagnosis module and model optimization module; The feature extraction module is used to determine the state feature parameters of the fault intelligent diagnosis model; The feature expansion module is used to expand the state feature parameter data based on the SMOTE algorithm; The fault diagnosis module is used to build a fault intelligent diagnosis model based on the random forest algorithm; The model optimization module is used to optimize the fault intelligent diagnosis model based on a grid search algorithm.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the multi-component hybrid insulating oil transformer fault diagnosis method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-component hybrid insulating oil transformer fault diagnosis method according to any one of claims 1 to 7 are implemented.

Images