Fault diagnosis method, device and equipment of transformer and storage medium

Abstract

The invention relates to the technical field of power systems, in particular to a transformer fault diagnosis method, device and equipment and a storage medium, and the method comprises the steps: obtaining DGA sample data according to preset transformer characteristic parameters; training a preset PNN network based on the DGA sample data to obtain a fault diagnosis model; the preset PNN is a model used for determining the fault information of the transformer according to the DGA data; a smoothing factor in the preset PNN network is determined according to an improved sea gull optimization algorithm; and inputting the DGA data corresponding to the target transformer into the fault diagnosis model to obtain fault information corresponding to the target transformer. According to the invention, the fault information of the target transformer can be comprehensively and accurately diagnosed, so that a guiding effect is provided for operation maintenance and state maintenance of the transformer.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of power systems, and in particular, to a fault diagnosis method, device, equipment and storage medium for a transformer. Background Technique

[0002] With the rapid development of modern society, the demand for and quality of electric energy in various fields have been significantly improved. The safe, stable and reliable operation of the power system is the basis for ensuring the rapid development of the national economy. As a key equipment in the power system and the core of energy conversion and transmission in the power grid, the transformer is a key hub equipment in the first line of defense system for power grid safety. At present, many transformers have been in operation for more than twenty years. These aging transformers face many problems such as weakened insulation performance and equipment failures during operation, which greatly increases the probability of accidents. Timely fault diagnosis and maintenance are necessary measures to ensure the safe and stable operation of the power system.

[0003] Although, to improve power supply reliability, fault diagnosis and maintenance of substations can be carried out through a preventive and regular maintenance approach. However, if the maintenance frequency is too low, it cannot effectively prevent the occurrence of faults; on the contrary, if the maintenance frequency is too high, it will waste a large amount of financial and material resources. And due to the lack of pertinence during the maintenance process, powering off all equipment for maintenance will greatly reduce the actual usage time of the equipment.

[0004] Therefore, it is of great theoretical and practical significance to conduct effective state evaluation and in-depth fault diagnosis research on the early warning of the power grid after a transformer fault, guide the operation and maintenance and condition-based maintenance of the transformer, and prevent and reduce the probability of faults. Summary of the Invention

[0005] The present application provides a fault diagnosis method, device, equipment and storage medium for a transformer, which can comprehensively and accurately diagnose the fault information of the target transformer to provide guidance for the operation and maintenance and condition-based maintenance of the transformer.

[0006] In a first aspect, a fault diagnosis method for a transformer is provided, including:

[0007] Obtain DGA sample data according to preset transformer characteristic parameters;

[0008] Train a preset PNN network based on the DGA sample data to obtain a fault diagnosis model; the preset PNN network is a model for determining the fault information of the transformer according to DGA data; the smoothing factor in the preset PNN network is determined according to an improved seagull optimization algorithm;

[0009] Input the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

[0010] Optionally, DGA sample data is obtained according to preset transformer characteristic parameters, including:

[0011] According to the absolute content of the fault characteristic gas of the transformer, determine the relative concentration content between two fault characteristic gases, and the ratio between each fault characteristic gas and the total hydrocarbon gas;

[0012] Based on the absolute content of various fault characteristic gases, the relative concentration content between each two fault characteristic gases, and the ratio between various fault characteristic gases and the total hydrocarbon gas, determine the preset transformer characteristic parameters.

[0013] Optionally, the improved seagull optimization algorithm is obtained by the following method:

[0014] Initialize the seagull population in the seagull optimization algorithm based on the Tent mapping strategy, so that the seagull individuals in the seagull population are evenly distributed in the search space;

[0015] Nonlinearly improve the control quantity that controls the movement behavior of seagull individuals to balance the global search and local search capabilities of the seagull optimization algorithm;

[0016] Based on the mutation mechanism, add a perturbation term to the optimal seagull individual in the seagull population to complete the improvement of the seagull optimization algorithm.

[0017] Optionally, before training the preset PNN network based on the DGA sample data to obtain a fault diagnosis model, it further includes:

[0018] Use kernel principal component analysis to reduce the dimension of the DGA sample data to obtain the characteristic data corresponding to the DGA sample data;

[0019] Divide the characteristic data corresponding to the DGA sample data into a training set and a test set;

[0020] Train the preset PNN network based on the training set to obtain an initial fault diagnosis model; the preset PNN network is a neural network determined according to the Bayes classification rule and the Parzen window probability density function;

[0021] Based on the test set, perform model testing on the initial fault diagnosis model to obtain the model test result;

[0022] If the model test result is passed, determine the initial fault diagnosis model as the final fault diagnosis model.

[0023] Optionally, input the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer, including:

[0024] The DGA data corresponding to the target transformer is processed by kernel principal component analysis for dimensionality reduction to obtain the diagnostic feature data corresponding to the DGA data;

[0025] The diagnostic feature data is input into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

[0026] Optionally, after inputting the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer, it further includes:

[0027] According to the severity corresponding to the fault information of the target transformer, the faults of the target transformer are fuzzily classified to obtain the fault levels corresponding to each fault information;

[0028] The original time series corresponding to the trigger events of each fault information is calculated by multi-order differencing according to the autoregressive integrated moving average model to obtain the stationary time series corresponding to multiple trigger events;

[0029] For the time information corresponding to each trigger event in the stationary time series, a logical operation set is established; the logical operation set is used to display the sequence rules of established operations between multiple trigger events; the established operations include logical AND, OR, and NOT operations;

[0030] The most important trigger event among multiple trigger events is placed at the position corresponding to the leader node, and the remaining trigger events are placed at the positions corresponding to the member nodes;

[0031] In response to the feedback signal sent by the leader node, it is judged whether the member node can feedback information to the leader node within the preset time;

[0032] If a member node fails to feedback information to the leader node within the preset time, the classification result of the fault information is determined according to the trigger event corresponding to the member node and the fault level corresponding to the fault information corresponding to the trigger event.

[0033] Optionally, if a member node fails to feedback information to the leader node within the preset time, determining the classification result of the fault information according to the trigger event corresponding to the member node and the fault level corresponding to the fault information corresponding to the trigger event includes:

[0034] According to the classification result of the fault information, the alarm module is controlled to perform a fault alarm; the alarm module includes one or more of a screen alarm module, a voice reminder alarm module, and a short message alarm module.

[0035] In a second aspect, a fault diagnosis device for a transformer is provided, including:

[0036] A sample data acquisition module for acquiring DGA sample data according to preset transformer characteristic parameters;

[0037] A model training module for training a preset PNN network based on the DGA sample data to obtain a fault diagnosis model; the preset PNN network is a model for determining the fault information of a transformer according to DGA data; the smoothing factor in the preset PNN network is determined according to an improved seagull optimization algorithm;

[0038] A fault diagnosis module for inputting the DGA data corresponding to a target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

[0039] In a third aspect, an electronic device is provided, including: a processor and a memory, where the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the method in the first aspect or its various implementation manners.

[0040] In a fourth aspect, a computer-readable storage medium is provided for storing a computer program, and the computer program causes a computer to execute the method in the first aspect or its various implementation manners.

[0041] Through the technical solution provided by this application, by acquiring DGA sample data based on preset transformer characteristic parameters, it is possible to avoid the problem of incomplete diagnosis of fault information by the fault diagnosis model trained by the DGA sample data due to local optimality; then, an improved seagull optimization algorithm is used to optimize the smoothing factor in the preset PNN network, which can improve the accuracy of the fault diagnosis model trained by the preset PNN network; finally, by inputting the DGA data corresponding to the target transformer into the fault diagnosis model, the comprehensive and accurate fault information corresponding to the target transformer can be obtained; thus, it can be seen that the fault diagnosis model provided by this application can comprehensively and accurately diagnose the DGA data corresponding to the target transformer, and can thus guide the operation and maintenance and condition-based maintenance of the transformer, which has important theoretical and practical significance for fault prevention and reduction of the occurrence probability of faults in the substation.

[0042] Other features and advantages of the present disclosure will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of this application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of this application, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0044] Figure 1 A scenario diagram provided by an embodiment of the present application;

[0045] Figure 2 A flowchart of a fault diagnosis method for a transformer provided by an embodiment of the present application;

[0046] Figure 3 A trend diagram of the control quantity A varying with the number of iterations provided by an embodiment of the present application;

[0047] Figure 4 A cumulative contribution rate diagram of the principal components provided by an embodiment of the present application;

[0048] Figure 5 A diagram of the running time and accuracy rate of the fault diagnosis model provided by an embodiment of the present application; (a is the accuracy rate and running time of the fault diagnosis model trained with the DGA sample data without dimensionality reduction; B is the accuracy rate and running time of the fault diagnosis model trained with the DGA sample data after dimensionality reduction);

[0049] Figure 6 A schematic diagram of a fault diagnosis device for a transformer provided by an embodiment of the present application;

[0050] Figure 7 A schematic block diagram of an electronic device provided by an embodiment of the present application. Specific implementation manners

[0051] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server including a series of steps or units does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0053] As described above, in the event of a failure in a transformer, especially an oil-immersed transformer, the insulating materials inside will crack under the action of electrothermal energy to generate characteristic gases, which will dissolve in the oil. The content and type of the characteristic gases can, to a certain extent, reflect the type of transformer failure. The method of analyzing dissolved gases in oil is based on this principle to determine the type of transformer failure. This method has the advantages of low economic cost and little influence from the external environment compared with electrical test methods, so it is widely used in the fault analysis of transformers in the power system.

[0054] Although, based on the DGA technology, traditional transformer fault diagnosis methods such as characteristic gas identification method, IEC three-ratio method, and four-ratio method have been gradually proposed, the fault of an oil-immersed transformer is a complex non-linear problem. These traditional methods have disadvantages such as fuzzy judgment boundaries, incomplete coding, and over-reliance on manual experience, resulting in a relatively low accuracy rate of fault diagnosis.

[0055] To solve the above technical problems, the inventive concept of this application is as follows: Obtain DGA sample data according to preset transformer characteristic parameters; train a preset PNN network based on the DGA sample data to obtain a fault diagnosis model; the preset PNN network is a model used to determine the fault information of a transformer according to DGA data; the smoothing factor in the preset PNN network is determined according to an improved seagull optimization algorithm; input the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer. The fault diagnosis model provided by this application can comprehensively and accurately diagnose the DGA data corresponding to the target transformer, and thus can guide the operation and maintenance and condition-based maintenance of the transformer, which has important theoretical and practical significance for fault prevention and reduction of the probability of faults in substations.

[0056] It should be understood that the technical solution of this application can be applied to the following scenarios, but is not limited to:

[0057] In some realizable ways, Figure 1 This is an application scenario diagram provided by an embodiment of this application. As Figure 1 shown, this application scenario may include an electronic device 110 and a network device 120. The electronic device 110 can establish a connection with the network device 120 through a wired network or a wireless network.

[0058] Exemplarily, the electronic device 110 can be a desktop computer, a laptop computer, a tablet computer, etc., but not limited thereto. The network device 120 can be a terminal device or a server, but not limited thereto. In an embodiment of the present application, the electronic device 110 can send a request message to the network device 120, and the request message can be used to request to obtain DGA sample data. Further, the electronic device 110 can receive a response message sent by the network device 120, and the response message includes the obtained DGA sample data.

[0059] In addition, Figure 1 Exemplarily, one electronic device 110 and one network device 120 are given. In fact, other numbers of electronic devices and network devices can be included, and the present application does not limit this.

[0060] In some other realizable manners, the technical solution of the present application can also be executed by the above-mentioned electronic device 110, or the technical solution of the present application can also be executed by the above-mentioned network device 120, and the present application does not limit this.

[0061] After introducing the application scenarios of the embodiments of the present application, the technical solution of the present application will be elaborated in detail below:

[0062] Figure 2 is a flowchart of a fault diagnosis method for a transformer provided in an embodiment of the present application. This method can be executed by the electronic device 110 as shown in Figure 1 shown, but not limited thereto. As shown in Figure 2 shown, this method can include the following steps:

[0063] S210. Obtain DGA sample data according to preset transformer characteristic parameters.

[0064] Among them, the transformer can be an oil-immersed transformer.

[0065] Here, according to the preset transformer characteristic parameters, the faults of the transformer can be comprehensively judged to avoid the problem of local optimality of the obtained DGA sample data, which leads to incomplete diagnosis of fault information by the trained fault diagnosis model, and effectively improve the accuracy of the trained fault diagnosis model in diagnosing fault information.

[0066] S220. Train a preset PNN network based on the DGA sample data to obtain a fault diagnosis model.

[0067] Here, the preset PNN network is a model for determining the fault information of the transformer according to DGA data; the smoothing factor in the preset PNN network is determined according to an improved seagull optimization algorithm.

[0068] Among them, by using an improved seagull optimization algorithm to optimize the smoothing factor in the preset PNN network, the accuracy, convergence speed, convergence accuracy, etc. of the fault diagnosis model trained by the DGA sample data for the preset PNN network can be improved.

[0069] S230. Input the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

[0070] Here, the fault information may include fault type information, fault status information, and fault location information.

[0071] The method provided in this embodiment can avoid the problem of incomplete diagnosis of fault information by the fault diagnosis model trained by the DGA sample data due to local optimality of the DGA sample data by obtaining the DGA sample data based on the preset transformer characteristic parameters; then, by using an improved seagull optimization algorithm to optimize the smoothing factor in the preset PNN network, the accuracy of the fault diagnosis model trained by the preset PNN network can be improved; finally, by inputting the DGA data corresponding to the target transformer into the fault diagnosis model, the comprehensive and accurate fault information corresponding to the target transformer can be obtained; thus, it can be seen that the fault diagnosis model provided in this application can comprehensively and accurately diagnose the DGA data corresponding to the target transformer, and further can guide the operation and maintenance and condition-based maintenance of the transformer, which has important theoretical and practical significance for fault prevention and reduction of the occurrence probability of faults in the substation.

[0072] In some possible implementation embodiments, obtaining DGA sample data according to the preset transformer characteristic parameters may include the following steps:

[0073] S310. Determine the relative concentration between two fault characteristic gases and the ratio between each fault characteristic gas and the total hydrocarbon gas according to the absolute content of the fault characteristic gas of the transformer.

[0074] Here, the selection of the fault characteristic gas can be based on the statistical analysis of transformer faults. It is determined that overheating faults and discharge faults are the main faults that occur in the transformer, and H 2 、CH 4 、C 2 H 6 、C 2 H 4 、C 2 H 2 are the main fault characteristic gases related to these two types of faults. Therefore, the faults of the transformer are judged by the above 5 fault characteristic gases; since the traditional H 2 、CH 4 、C 2 H 6 、C2 H 4 、C 2 H 2 Using these 5 fault characteristic gases as the DGA input variables will result in the problem of local optimality. At the same time, the fault information contained in these 5 characteristic parameters is incomplete, which is also likely to lead to a low accuracy of transformer fault diagnosis. Therefore, this application improves the triple ratio method, namely C 2 H 2 / C 2 H 4 、CH 4 / H 2 、C 2 H 4 / C 2 H 6 On the basis of the three-dimensional fault data, it is expanded to 16-dimensional fault data, and then combined with the original H 2 、CH 4 、C 2 H 6 、C 2 H 4 、C 2 H 2 five-dimensional characteristic data, which can effectively improve the accuracy of transformer fault diagnosis; among them, the multi-dimensional fault characteristic data is shown in Table 1.

[0075] Table 1 Multi-dimensional fault characteristic data

[0076]

[0077]

[0078] Among them, the absolute content of the 5 gas concentrations is represented by S 1 to S 5 ; the relative content of the concentration between every two of the 5 gases is represented by S6 to S15; the ratio between the 5 gases and the total hydrocarbon gas TH(CH 4 +C 2 H 6 +C 2 H 4 +C 2 H 2 ) is represented by S16 - S21.

[0079] S320. Determine the preset transformer characteristic parameters based on the absolute content of various fault characteristic gases, the relative content of the concentration between every two fault characteristic gases, and the ratio between various fault characteristic gases and the total hydrocarbon gas.

[0080] Here, based on the absolute content of various fault characteristic gases, the relative concentration content between every two fault characteristic gases, and the ratio between various fault characteristic gases and total hydrocarbon gases, the determined preset transformer characteristic parameters are used as DGA input variables to avoid the problem of local optimum. At the same time, it can also ensure that the fault information contained in the preset transformer characteristic parameters is complete, so that the fault diagnosis model trained according to the DGA sample data determined by the preset transformer characteristic parameters can accurately diagnose the fault information of the target transformer.

[0081] Using the above method, the DGA sample data obtained based on the preset transformer characteristic parameters can contain the characteristics corresponding to complete fault information, thereby improving the accuracy of the fault information diagnosis of the target transformer by the fault diagnosis model trained based on the DGA sample data.

[0082] In some possible implementation embodiments, the improved seagull optimization algorithm is obtained by the following method:

[0083] S410. Initialize the seagull population in the seagull optimization algorithm based on the Tent mapping strategy, so that the seagull individuals in the seagull population are evenly distributed in the search space.

[0084] Here, the initial positions of the seagull individuals in the seagull population initialized based on the Tent mapping strategy make the seagull population evenly and flatly distributed in the search space.

[0085] Among them, the formula applied by the Tent mapping strategy is as follows:

[0086]

[0087] In the formula, x n is the seagull individual before the Tent mapping strategy; x n+1 is the seagull individual after the Tent mapping strategy; μ is the chaos parameter, which affects the mapping uniformity. Research shows that when μ = 0.5, the mapping uniformity of the Tent mapping strategy is the best. Therefore, the above formula can be written as:

[0088]

[0089] Taking the number of seagull individuals in the seagull population as N and the space dimension as D as an example, the process of initializing the seagull population in the seagull optimization algorithm using the Tent mapping strategy includes the following steps:

[0090] 1) Randomly generate a D-dimensional vector [x 1 = x 11 , x 12 ,... x 1DAs the first seagull individual.

[0091] 2) From the above formula, by x 1 Iteratively obtain the remaining individuals to form matrix X:

[0092]

[0093] 3) Map the above formula to each seagull individual in the seagull population; among them, the calculation formula for the position of the seagull individual is as follows:

[0094] H i (t) = (UB - LB)X i + LB;

[0095] In the formula, H i (t) is the position of the i-th seagull individual; UB is the upper bound of the position search of the seagull individual; LB is the lower bound of the position search of the seagull individual.

[0096] Here, by introducing the Tent mapping strategy to initialize the seagull population, the diversity of the seagull population can be increased, the initial seagull individuals can be optimized, the speed of the early iteration of the seagull optimization algorithm can be increased, and at the same time, the probability of the seagull optimization algorithm falling into the local optimum can be reduced.

[0097] S420. Nonlinear improvement is made to the control quantity that controls the movement behavior of seagull individuals to balance the global search and local search capabilities of the seagull optimization algorithm.

[0098] In the implementation process of the seagull optimization algorithm, while calculating the fitness value of seagull individuals, that is, the objective function value, the current position of seagull individuals is also calculated, and the optimal individual in the seagull population is marked. Specifically, when the seagull optimization algorithm performs the selection calculation, the algorithm imitates the seagull individual flying from one position to another position. During the migration process, the seagull population should meet 3 conditions to avoid collisions; here, in order to prevent the seagull individual from colliding with other seagull individuals, a control quantity A is added when calculating the initial new position of the seagull individual; among them, the calculation formula for the initial new position of the seagull individual is: C s (t) = AP S (t); in the formula, C s (t) is the initial new position of the seagull individual; A is the control quantity; P S (t) is the current position of the seagull individual.

[0099] Here, during the seagull migration process, the control quantity A can effectively avoid collisions between adjacent seagulls. It can be seen that its magnitude linearly decreases from 2 to 0. Due to its constant change rate, the convergence ability of the seagull optimization algorithm is reduced to a certain extent. For this reason, a non-inertial weight strategy is proposed, and the expression of the control quantity A is: In the formula, t represents the current iteration number, and t max represents the maximum iteration number; the curve of its control variable A changing with the iteration number is as shown in Figure 3 .

[0100] Among them, in the process of determining the optimal position direction of seagull individuals, while avoiding collisions with other seagulls of the same kind, seagull individuals will automatically move forward to the optimal position of the target position; the calculation formula for the optimal position of seagull individuals is: M s (t) = B(P bs (t) - P s (t)); where M s (t) is the seagull individual approaching its corresponding optimal position; B is the fitness value of the seagull individual; P bs (t) is the position of the optimal individual in the seagull population; when the seagull individual approaches its corresponding optimal position, the seagull individual moves forward in the direction of the optimal position and reaches the final new position; the calculation formula for the final new position is: D s (t) = |C s (t) + M s (t)|; where D s (t) is the final new position of the seagull individual.

[0101] Among them, in the local search process of the seagull optimization algorithm, that is, when the attacking seagull individuals migrate, they maintain an appropriate height through their wings and body weight to continuously change the angle and speed of the seagull individuals during the attack; when the seagull individuals forage and attack prey, they move in a spiral manner, and the attack position formula of the seagull individuals can be obtained: P s (t) = D s (t)·x·y·z + p bs (t).

[0102] In addition, aiming at the problems of single search method and dependence on the optimal individual in the seagull optimization algorithm, a seagull population evolution strategy based on multiple influencing factors is proposed. An attraction and repulsion strategy is adopted to update the distance D S (t) of the optimal individual, so that the position update of the seagull individual is affected by the attraction of the global optimal value and the repulsion of the global worst value. Its position update formula is as follows:

[0103]

[0104] Among them: represents the updated position, X b，i and X w，i , respectively represent the global optimal individual and the global worst individual; r 1 and r 2 are both random numbers uniformly distributed in the range of [0, 1]. In the above formula +r1 (X b，i -|X j，i |) represents the attracting effect of the optimal individual, -r 2 (X w，i -|X j，i |) represents the repulsive effect of the worst individual.

[0105] It can be seen that by nonlinearly improving the control quantity that controls the movement behavior of seagull individuals, the global search and local search capabilities of the seagull optimization algorithm can be balanced.

[0106] In addition, when nonlinearly improving the control quantity that controls the movement behavior of seagull individuals, the control factor f corresponding to the control quantity can be improved through the parameter control method of the cosine function; among them, the method for improving the control factor f c is as follows: c The method for improving the control factor f

[0107] (1) Introduce a parameter control method based on the cosine function to improve the control factor f c is as follows:

[0108] Its formula is as follows:

[0109]

[0110] (2) Use the relevant parameters of the control factor of the seagull optimization algorithm to replace the above formula, and the improved expression of the control factor can be obtained:

[0111]

[0112] Compared with the original linear change, the improved non-linear change method has a larger slope in the early stage, that is, the parameter f c decreases faster, which can reduce the large fluctuations of seagull individuals during position adjustment in the early stage and improve the search efficiency and stability of the population.

[0113] S430. Add a perturbation term to the optimal seagull individual in the seagull population based on the mutation mechanism to complete the improvement of the seagull optimization algorithm.

[0114] Here, the mutation mechanism can include the Gaussian distribution mutation algorithm and the Cauchy distribution mutation algorithm, and the new positions of seagull individuals are calculated through parallel search optimization of the Gaussian distribution mutation algorithm and the Cauchy distribution mutation algorithm. The specific process is as follows:

[0115] A new position update formula is proposed for the Seagull Optimization Algorithm. This position update method is not affected by the globally optimal individual. When the historical optimal solution and the globally worst solution of the seagull individuals are introduced as influencing factors, the seagull individuals move randomly under the attraction of the historical optimal solution and the repulsion of the globally worst solution, thus realizing the update of the position. The update formula is: HS i (t + 1) = r d ×HS i (t)+(1 - r d )×p b,i -(1 - r d )×HS w (t); where HS i (t + 1) represents the position of the i-th seagull individual after update, HS i (t) is the position before update, P b,i is the historical optimal position of the i-th seagull individual, HS w (t) is the globally worst position of the seagull population, and r d is a random number uniformly distributed in the range [0, 1].

[0116] Seagull individuals perform parallel search, that is, seagull individuals randomly select one of these two search methods in each iteration, with equal probability.

[0117]

[0118] Among them, the Gaussian distribution mutation algorithm is often used for the mutation operation of optimization algorithms, and its probability density expression is as follows:

[0119]

[0120] Among them: μ is the expectation, and σ 2 is the variance. When μ = 0 and σ 2 = 1, it is called the standard normal distribution. The formula for Gaussian mutation operation is as follows:

[0121] P i G = P i (1 + Gauss(0, σ 2 ))

[0122] The Cauchy distribution mutation algorithm is also a commonly used mutation operator. The probability density function of the Cauchy distribution mutation algorithm is as follows:

[0123]

[0124] Taking x 0 = 1 and ρ = 1, at this time the Cauchy distribution is the standard distribution, and the formula for the Cauchy distribution mutation algorithm can be obtained as follows:

[0125] p i C = p i (1 + Cauchy(0, ρ));

[0126] The Cauchy distribution mutation algorithm in the above formula can generate a random number that follows the standard Cauchy distribution. The Cauchy distribution mutation algorithm is similar to the Gaussian distribution mutation algorithm. The operation of the Cauchy distribution mutation algorithm can enrich the diversity of the population, making the algorithm easier to jump out of local traps and thus find better solutions, avoiding premature convergence. Therefore, the Gaussian distribution mutation algorithm has a strong local search ability, a relatively weak global search ability, and a lack of ability to jump out of local traps. The Cauchy distribution mutation algorithm has better global search ability due to its wider distribution range and is easier to jump out of local traps. Now, these two mutation methods are combined to introduce an adaptive mutation strategy, which acts on the optimal seagull:

[0127] HS best.v (t) = HS best (t)[1 + γGauss(0, 1) + (1 - γ)Gauss(0, 1)];

[0128] Among them, HS best.v (t) represents the position of the optimal seagull after mutation, and HS best (t) is the position of the optimal seagull before mutation; Cauchy(0, 1) and Causs(0, 1) are the standard Gaussian mutation and the standard Cauchy mutation respectively; γ = 1 - t 2 / MaxT 2 is the mutation control coefficient. γ is adaptively adjusted with the number of iterations to coordinate the local development and global exploration capabilities of the algorithm. In the early stage of iteration, the value of γ is larger, and the Cauchy distribution mutation algorithm dominates. At this time, the mutation range is larger, and the algorithm has better global exploration ability. As the iteration progresses, γ gradually decreases, the proportion of the Gaussian distribution mutation algorithm increases, the mutation range becomes smaller, and the local search ability of the algorithm is enhanced.

[0129] The position of the mutated seagull individual is not necessarily better than the original position. A greedy selection strategy needs to be used to define a good evolution direction for the optimal seagull. The steps of its position update are as follows: Obtain the position information of the optimal seagull HS best (t), and perform mutation operations on HS best (t) according to the adaptive mutation seagull optimization algorithm and its optimization test to obtain a new position HS best.v (t), and compare the fitness values of the two positions and select the better one for update. Thus, the position update formula of the optimal seagull can be obtained:

[0130]

[0131] The mutation mechanism endows the optimal seagull with a local search method. This local search is usually carried out within a small range near its original position, and there is also a certain probability of jumping to an area far from the original position. By combining the Gaussian distribution mutation algorithm and the Cauchy distribution mutation algorithm, and introducing the mutation control coefficient γ for adaptive adjustment of the mutation range, the mutation range is made more compatible with the evolutionary situation of the population, and the probability that the optimal seagull can find a better position after mutation is increased.

[0132] Using the above method, in view of the limitations of the traditional seagull optimization algorithm in dealing with complex problems, three strategies are adopted to improve the traditional seagull optimization algorithm. First, the Tent mapping strategy is introduced to initialize the seagull population, so that the population is evenly distributed in the search space, improving the optimization speed of the seagull optimization algorithm in the early stage; second, the control quantity A that controls the movement behavior of the seagull is non-linearly improved to balance the global search and local search capabilities of the algorithm; finally, the mutation mechanism is introduced to add a perturbation term based on the optimal individual, avoiding the situation that seagull individuals fall into local optima, so that the optimization performance of the improved seagull optimization algorithm is greatly improved.

[0133] In some possible embodiments, training a preset PNN network based on DGA sample data to obtain a fault diagnosis model may include the following steps:

[0134] S510. Use kernel principal component analysis to perform dimensionality reduction on the DGA sample data to obtain the characteristic data corresponding to the DGA sample data.

[0135] It should be noted that DGA sample data has its own characteristics. First, the same gas varies greatly in quantity; second, this distribution is not uniform. Often, a large number of DGA sample data are distributed in a small interval, while a small number of DGA sample data are distributed in a large interval; third, different types of DGA sample data vary greatly. Some fault type data has a large amount, while some fault type data has a small amount. Therefore, these DGA sample data need to be preprocessed to achieve a better classification effect.

[0136] Taking the preset transformer characteristic parameters in step S320 as an example, since the preset transformer characteristic parameters are 21-dimensional data, the DGA sample data obtained based on the preset transformer characteristic parameters is prone to redundancy due to its large number of dimensions, which in turn affects the training efficiency of the DGA sample data for the preset PNN network. Therefore, kernel principal component analysis is used here to reduce the dimensionality of the DGA sample data and generate the characteristic data corresponding to the DGA sample data to remove redundant data and retain most of the original characteristic information in the DGA sample data.

[0137] Among them, the process of dimensionality reduction of DGA sample data by kernel principal component analysis is as follows:

[0138] S511. Determine the number of principal components k.

[0139] Here, assume there are m samples x 1 , x 2 , … x n ∈R, each sample has a dimension of n, and the input matrix X formed by m samples is standardized by the following formula to eliminate the influence of parameter dimensions and different orders of magnitude:

[0140]

[0141] In the formula, i = 1, 2…n.j = 1, 2…n; S j is the sample standard deviation; is the sample mean.

[0142] The matrix X after standardization can be expressed as:

[0143]

[0144] Let the dimensionality-reduced space be Γ(u 1 , u 2 ,... u n ), and the covariance matrix of matrix X is:

[0145]

[0146] The following relationship exists between the eigenvectors and eigenvalues of the input samples:

[0147] λ i u i = Vu i , i = 1, 2,... n;

[0148] Then the transformation matrix can be denoted as:

[0149] U = [u 1 , u 2 ,... u n ;

[0150] Sort λ from largest to smallest to obtain the principal elements:

[0151] S i = u i ′x i , i = 1, 2,... n

[0152] (3) Finally, determine the number of principal components k through the cumulative contribution rate:

[0153]

[0154] where Conyr(y i ) is the contribution rate of the i-th principal component.

[0155] S512: Use the preset transformer characteristic parameters as input variables, and calculate the cumulative interpretable variance ratio of the input variables.

[0156] Here, from the cumulative interpretable variance contribution rate, it can be known that when the dimension of the fault features after dimensionality reduction is 7, the interpretable variance contribution rate can reach more than 95%. After that, the interpretable variance contribution rate tends to be stable. Therefore, the dimensionality reduction dimension of the used kernel principal component analysis method is selected as 7. Since the DGA sample data of the transformer is non-linear data with redundant information, if it is directly used for training and testing the fault diagnosis model, it will greatly affect the convergence speed and convergence accuracy of the model. Therefore, it is necessary to use the kernel principal component analysis method to perform dimensionality reduction processing on the DGA sample data. Here, the contribution rate and cumulative contribution rate of the principal components are as Figure 4 shown.

[0157] S520: Divide the characteristic data corresponding to the DGA sample data into a training set and a test set.

[0158] Here, dividing the characteristic data after dimensionality reduction processing of the DGA sample data into a training set for training the preset PNN network and a test set for testing the preset PNN network can effectively improve the training efficiency and test efficiency of the PNN network.

[0159] S530: Train the preset PNN network based on the training set to obtain an initial fault diagnosis model.

[0160] In this step, by using the characteristic data corresponding to the DGA sample data obtained in step S510 as the training set, the average accuracy and operating efficiency of the initial fault diagnosis model can be improved; from Figure 5 it can be known that the average accuracy of the initial fault diagnosis model trained with the DGA sample data before dimensionality reduction is 73.78%, and the average operating time is 6.21 s; the average accuracy of the initial model trained with the characteristic data corresponding to the DGA sample data after dimensionality reduction is 77.33, and the average operating time is 4.99 s. Thus, it can be seen that compared with the initial fault diagnosis model trained with the DGA sample data before dimensionality reduction, the initial model trained with the characteristic data corresponding to the DGA sample data after dimensionality reduction has an accuracy improvement of 3.55% and a reduction in operating time of 1.22 s. This can verify the effectiveness of the characteristic data extracted by the principal components obtained by the kernel component analysis method.

[0161] Here, the preset PNN network is a neural network determined according to the Bayes classification rule and the Parzen window probability density function.

[0162] Among them, before training the preset PNN network based on the training set, the number of neurons in each layer of the preset PNN network and the smoothing factor are initialized.

[0163] Since the preset PNN network is a neural network based on the Bayes classification rule and the Parzen window probability density function, the preset PNN network has the advantages of simple structure, few parameters, high fault tolerance, good robustness, and strong classification ability. Therefore, the initial fault diagnosis model obtained by training the preset PNN network has great advantages in the field of transformer fault diagnosis.

[0164] Suppose there are ω in the pattern set in the Bayes classification rule i (i = 1, 2,... d), for any x ∈ R k , the prior probability of each pattern is the conditional P(w i ) probability is P(w i |x); the formula for obtaining the posterior probability in this pattern according to the Bayes classification rule is:

[0165]

[0166] Here, for any j = 1, 2,... d, i ≠ j, if P(ω i ∣x) > P(ω i ∣x), then If P(ω i ) = P(ω j ), then the above formula can be rewritten as

[0167] In addition, usually the posterior probability obtained based on the above formula can be used for multi-pattern classification. Additionally, for a certain misclassification risk, a cost function is introduced, that is, the Bayes minimum average risk criterion, namely:

[0168]

[0169] In the above Bayes classification rule, P(x∣ω i ) is generally unknown. The Parzen window method can accurately estimate the probability density function under a large number of training samples. Here, assume that the number of samples belonging to type ω i is N i in total, then the density is estimated by the following formula:

[0170]

[0171] where x ij is the i-th training sample in the j-th class, m is the dimension of the feature vector, and δ is the smoothing factor. Among them, according to probability theory analysis, assuming that the probability of the sample x falling into the region D is:

[0172] Assuming that n samples satisfying P(x) are randomly selected, the probability that b samples fall into the region D is: Its expectation is: E(b) = np.

[0173] There will be a peak near the mean of the binomial distribution of b. When the number of samples n is large enough, it can be considered that For estimating the probability, if P(x) is continuous and the region D is small enough, then where x is the point in the sample and V is the volume of the region D range. Construct the region R, and expand it from R 1 to obtain R n , and by combining the above three equations, we can get:

[0174] Select a small region D that can enclose x to estimate P(x), and establish an m-dimensional cube with side length mh, whose volume is: Count the number of samples falling into the region D, and define the following function:

[0175]

[0176] where ψ is a cube with unit length centered at the origin. If the number of times ix falls into this region is incremented by 1, then the number of samples finally falling into the cube is denoted as:

[0177]

[0178] By combining the above two equations, we can get:

[0179]

[0180] Since the above equation is a conventional probability estimation method, the magnitude of P(x n ) is proportional to the number of samples in the neighborhood set. On this basis, the Parzen window probability density function estimation is:

[0181]

[0182] where m is the dimension of the feature vector, δ is the smoothing factor, and x ji is the i-th training sample in the j-th class. The topological structure of the probabilistic neural network includes an input layer, a hidden layer, a summation layer, and an output layer.

[0183] Since the number of neurons in the input layer is the same as the dimension of the input features, it receives the sample data, does not perform calculations on the data, and transmits the feature vector to the hidden layer. The number of neurons in the hidden layer is the same as the number of training samples. Each neuron has a center, and the distance between the input vector and the center is calculated as the activation function. The input-output relationship of this layer can be simplified as follows:

[0184]

[0185] In the formula, X is the input vector, and W i is the connection weight between the input layer and the hidden layer. Both are normalized to unit length. δ is a smoothing parameter that affects the classification accuracy.

[0186] The number of neurons in the summation layer is the same as the number of categories of the training samples. The probabilities of the same-category patterns in the hidden layer are added up. The calculation formula is:

[0187]

[0188] The output layer receives the output of the summation layer. The number of neurons is the same as that of the hidden layer, and there is a one-to-one mapping relationship. It outputs the category with the largest calculation result in the summation layer. The calculation formula is: y = argmax(ν i ).

[0189] The specific implementation process of the PNN algorithm includes the following steps:

[0190] (1) Since the values of the feature quantities may vary in magnitude under different patterns, the data samples need to be normalized through the following formula: Among them, the data samples are the feature data corresponding to the DGA sample data;

[0191]

[0192] In the formula, x ij [k] is the k-th feature quantity in the j-th sample of the i-th category.

[0193] (2) Transmit the normalized data samples from the input layer to the hidden layer. One training sample corresponds to one hidden layer neuron, and calculate the Euclidean distance as the following formula:

[0194]

[0195] In the formula, P t [k] is the k-th eigenvalue of the t-th sample matrix.

[0196] Activate the hidden layer neurons to obtain the probability density matrix as the following formula:

[0197]

[0198] (3) The summation layer calculates the mean by adding the initial probabilities output by the hidden layer according to the following formula:

[0199]

[0200] S540. Test the initial fault diagnosis model based on the test set to obtain the model test results.

[0201] Here, the probability that the test sample t in the test set belongs to the i-th class is: maxpros ti = P ωi (p t ).

[0202] S550. If the model test result is a pass, determine the initial fault diagnosis model as the final fault diagnosis model.

[0203] Using the above method, a transformer fault diagnosis model based on the PNN network was established. In view of the nonlinearity and redundancy of the DGA sample data, the kernel principal component analysis method was used to reduce the dimension of the DGA sample data, retaining the main information of the DGA sample data, which can effectively improve the operation speed and accuracy of the fault diagnosis model.

[0204] In some possible embodiments, inputting the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer may include the following steps:

[0205] S610. Use the kernel principal component analysis method to reduce the dimension of the DGA data corresponding to the target transformer to obtain the diagnostic feature data corresponding to the DGA data.

[0206] S620. Input the diagnostic feature data into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

[0207] Using the above method, by reducing the dimension of the DGA data corresponding to the target transformer using the kernel principal component analysis method, redundant data in the DGA data corresponding to the target transformer can be removed, and the diagnostic feature data can retain most of the original feature information in the DGA data corresponding to the target transformer; then inputting the diagnostic feature data obtained by reducing the dimension of the DGA data corresponding to the target transformer into the fault diagnosis model can effectively improve the efficiency of diagnosing the fault information of the target transformer by the fault diagnosis model.

[0208] In some possible embodiments, after inputting the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer, the following steps may further be included:

[0209] S710. Fuzzily classify the faults of the target transformer according to the severity corresponding to the fault information of the target transformer, and obtain the fault levels corresponding to each fault information.

[0210] It should be noted that before fuzzily classifying the faults of the target transformer according to the severity corresponding to the fault information of the target transformer and obtaining the fault levels corresponding to each fault information, it also includes building a fault classification system for the transformer; among them, building a fault classification system for the transformer may include the following steps:

[0211] S711. Construct a data acquisition layer, a data transmission layer, and a central control layer. Among them, the data acquisition layer is used to control the gas sensor to collect the content information of 5 fault characteristic gases, namely H 2 , CH 4 , C 2 H 6 , C 2 H 4 , C 2 H 2 5 in the transformer oil, and after A / D conversion of the content information of the above 5 fault characteristic gases, transmit it to the STM32 minimum system; the data transmission layer includes the STM32 minimum system and the MCGS touch screen. The STM32 minimum system is used to transmit the content information of the above 5 fault characteristic gases to the MCGS touch screen for real-time display through RS485; the central control layer includes the upper computer. The MCGS touch screen transmits the content information of the above 5 fault characteristic gases to the upper computer through the Ethernet, and the upper computer can perform fault diagnosis according to the content information of the above 5 fault characteristic gases to realize real-time monitoring of the transformer status;

[0212] S712. Determine the data acquisition unit according to the system function requirements. The data acquisition unit can be divided into a data acquisition module and a solenoid valve control module according to the system function; among them, the data acquisition module realizes the functions of data acquisition and analog-to-digital conversion; the solenoid valve control module realizes the function of solenoid valve control.

[0213] S713. Set the input variable as 7-dimensional variable characteristic data, and select 7 transformer fault categories, namely low-energy discharge fault (D 1 ), high-energy discharge fault (D 2 ), normal state (N), partial discharge fault (PD), low-temperature overheat fault (T 1 ), medium-temperature overheat fault (T 2 ), and high-temperature overheat fault (T 3 ) as the output variables.

[0214] Here, input the fault information corresponding to the target transformer into the fault classification system, and the fault levels corresponding to each fault information can be obtained.

[0215] When the fault classification system classifies each fault information, it can determine the fault level of the fault in the target transformer based on the severity corresponding to the fault information. For example, if the severity corresponding to the fault information includes 4 status levels, namely level I, level II, level III, and level IV, by analyzing the static stability degree of the power grid in the target transformer, the severity of the power grid fault problem can be calculated from three aspects: overload, low voltage, and loss of load. Based on this, the fuzzy level of the fault severity of the target transformer is divided.

[0216] S720. Perform multi-order difference calculations on the original time series corresponding to the trigger events of each fault information according to the autoregressive integrated moving average model to obtain the stationary time series corresponding to multiple trigger events.

[0217] If the fault classification result obtained based on a single trigger event corresponding to the fault information has a large error, therefore, here the trigger events corresponding to all fault information are arranged in chronological order to generate the original time series, and then the autoregressive integrated moving average model is used for multi-order difference operations to transform the original time series into the stationary time series corresponding to multiple trigger events.

[0218] S730. Establish a logical operation set for the time information corresponding to each trigger event in the stationary time series; the logical operation set is used to display the sequence rules of established operations between multiple trigger events; the established operations include logical AND, OR, and NOT operations.

[0219] For the time information corresponding to each trigger event in the stationary time series containing multiple trigger events, introduce the time information of each trigger event through AND, OR, and NOT operations, establish a logical operation set, and display the sequence rules of established operations between multiple trigger events.

[0220] S740. Place the most important trigger event among multiple trigger events at the position corresponding to the leader node, and place the remaining trigger events at the positions corresponding to the member nodes.

[0221] S750. In response to the feedback signal sent by the leader node, determine whether the member node can feedback information to the leader node within the preset time.

[0222] S760. If a member node fails to feedback information to the leader node within the preset time, then determine the classification result of the fault information according to the trigger event corresponding to the member node and the fault level corresponding to the fault information corresponding to the trigger event.

[0223] Since the timing change of the triggering event will affect the result of the fault classification response, the leadership synthesis algorithm is used here for nested operations. The most important triggering event is selected and placed at the position corresponding to the leader node, and the remaining triggering events are placed at the positions corresponding to the member nodes. Using the leadership synthesis algorithm, the leader node sends a signal to the member nodes, and the remaining member nodes need to send feedback information to the leader node. When the member node does not send the corresponding feedback information within the preset time, the fault classification response result corresponding to the fault information is determined in combination with the fault level corresponding to the fault information.

[0224] By using the above method, the fault information corresponding to the target transformer can be accurately classified, so that the power system can quickly make a response according to the classification result of the fault information.

[0225] In some possible embodiments, if a member node fails to feedback information to the leader node within the preset time, the classification result of the fault information is determined according to the triggering event corresponding to the member node and the fault level corresponding to the fault information corresponding to the triggering event, which may include: controlling the alarm module to perform a fault alarm according to the classification result of the fault information; the alarm module includes one or more of a screen alarm module, a voice reminder alarm module, and a short message alarm module.

[0226] By using the above method, the management personnel can intuitively understand the level of the fault information through the alarm module, and the maintenance personnel can be notified in time when a fault occurs, improving the efficiency of the power supply system maintenance.

[0227] Figure 6 It is a schematic diagram of a fault diagnosis device 800 of a transformer according to an embodiment provided by the present invention. The device 800 includes:

[0228] A sample data acquisition module 810, configured to acquire DGA sample data according to preset transformer characteristic parameters;

[0229] A model training module 820, configured to train a preset PNN network based on the DGA sample data to obtain a fault diagnosis model; the preset PNN network is a model for determining the fault information of the transformer according to the DGA data; the smoothing factor in the preset PNN network is determined according to an improved seagull optimization algorithm;

[0230] A fault diagnosis module 830, configured to input the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

[0231] In some implementable manners, the sample data acquisition module 810 includes:

[0232] A gas determination unit, configured to determine the relative concentration between two fault characteristic gases and the ratio between each fault characteristic gas and the total hydrocarbon gas according to the absolute content of the fault characteristic gases of the transformer;

[0233] A transformer characteristic parameter determination unit, configured to determine preset transformer characteristic parameters based on the absolute content of various fault characteristic gases, the relative concentration between each two fault characteristic gases, and the ratio between various fault characteristic gases and the total hydrocarbon gas.

[0234] In some realizable manners, the fault diagnosis device 800 of the transformer further includes:

[0235] An initialization module, configured to initialize the seagull population in the seagull optimization algorithm based on the Tent mapping strategy, so that the seagull individuals in the seagull population are evenly distributed in the search space;

[0236] A non-linear improvement module, configured to perform non-linear improvement on the control quantity that controls the movement behavior of seagull individuals, so as to balance the global search and local search capabilities of the seagull optimization algorithm;

[0237] A perturbation term addition module, configured to add a perturbation term to the optimal seagull individual in the seagull population based on the mutation mechanism, so as to complete the improvement of the seagull optimization algorithm.

[0238] In some realizable manners, the model training module includes:

[0239] A sample feature data acquisition unit, configured to perform dimensionality reduction processing on the DGA sample data by using kernel principal component analysis to obtain the feature data corresponding to the DGA sample data;

[0240] A data partitioning unit, configured to partition the feature data corresponding to the DGA sample data into a training set and a test set;

[0241] A model training unit, configured to train a preset PNN network based on the training set to obtain an initial fault diagnosis model; the preset PNN network is a neural network determined according to the Bayes classification rule and the Parzen window probability density function;

[0242] A model testing unit, configured to perform model testing on the initial fault diagnosis model based on the test set to obtain a model testing result;

[0243] A model determination unit, configured to, if the model testing result is passed, determine the initial fault diagnosis model as the final fault diagnosis model.

[0244] In some realizable manners, the fault diagnosis module includes:

[0245] A diagnostic feature data acquisition unit, which is used to perform dimensionality reduction processing on the DGA data corresponding to the target transformer by using kernel principal component analysis to obtain the diagnostic feature data corresponding to the DGA data;

[0246] A fault diagnosis unit, which is used to input the diagnostic feature data into a fault diagnosis model to obtain the fault information corresponding to the target transformer.

[0247] In some implementable ways, the fault diagnosis device 800 of the transformer further includes:

[0248] A fault fuzzy classification module, which is used to perform fuzzy classification on the faults of the target transformer according to the severity corresponding to the fault information of the target transformer to obtain the fault levels corresponding to each fault information;

[0249] A multi-order difference calculation module, which is used to perform multi-order difference calculation on the original time series corresponding to the trigger events of each fault information according to the autoregressive integrated moving average model to obtain the stationary time series corresponding to multiple trigger events;

[0250] A logical operation set establishment module, which is used to establish a logical operation set for the time information corresponding to each trigger event in the stationary time series; the logical operation set is used to display the sequence rules of established operations between multiple trigger events; the established operations include logical AND, OR, and NOT operations;

[0251] A trigger event placement module, which is used to place the most important trigger event among multiple trigger events at the position corresponding to the leader node and place the remaining trigger events at the positions corresponding to the member nodes;

[0252] A judgment module, which is used to judge whether a member node can feedback information to the leader node within a preset time in response to a feedback signal sent by the leader node;

[0253] A classification result determination module, which is used to, if a member node fails to feedback information to the leader node within a preset time, determine the classification result of the fault information according to the trigger event corresponding to the member node and the fault level corresponding to the fault information corresponding to the trigger event.

[0254] In some implementable ways, the fault diagnosis device 800 of the transformer further includes:

[0255] An alarm control module, which is used to control an alarm module to perform fault alarm according to the classification result of the fault information; the alarm module includes one or more of a screen alarm module, a voice reminder alarm module, and a short message alarm module.

[0256] It should be understood that the embodiments of the transformer fault diagnosis device and the embodiments of the transformer fault diagnosis method can correspond to each other, and similar descriptions can refer to the embodiments of the transformer fault diagnosis method. To avoid repetition, it will not be elaborated here. Specifically, Figure 3 The illustrated apparatus 800 can execute the above-described embodiments of the transformer fault diagnosis method, and the foregoing and other operations and / or functions of each module in the apparatus 800 respectively implement the corresponding processes in the above-described transformer fault diagnosis method. For the sake of brevity, it will not be elaborated here.

[0257] In the foregoing, the apparatus 800 of the embodiments of the present invention has been described from the perspective of functional modules. It should be understood that the functional modules can be implemented in the form of hardware, can also be implemented by instructions in the form of software, and can also be implemented by a combination of hardware and software modules. Specifically, each step of the transformer fault diagnosis method and the detection method embodiments in the embodiments of the present invention can be completed by the integrated logic circuit in the hardware in the processor and / or instructions in the form of software. The steps of the transformer fault diagnosis method and the detection method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by a combination of the hardware and software modules in the decoding processor. Optionally, the software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps in the above-described embodiments of the transformer fault diagnosis method and the detection method.

[0258] Figure 7 is a schematic block diagram of an electronic device 110 according to an embodiment provided by the present invention.

[0259] As Figure 7 shown, the electronic device 110 may include:

[0260] A memory 111 and a processor 112. The memory 111 is used to store a computer program and transmit the program code to the processor 112. In other words, the processor 112 can call and run the computer program from the memory 111 to implement the method in the embodiments of the present invention.

[0261] For example, the processor 112 can be used to execute the above method embodiments according to the instructions in the computer program.

[0262] In some embodiments of the present invention, the electronic device 110 may include but is not limited to:

[0263] General-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the like.

[0264] In some embodiments of the present invention, the memory 111 includes, but is not limited to:

[0265] Volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synch link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0266] In some embodiments of the present invention, the computer program can be divided into one or more modules, and the one or more modules are stored in the memory 111 and executed by the processor 112 to complete the method provided by the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the controller.

[0267] As Figure 7 shown, the electronic device 110 may further include:

[0268] A transceiver 113, which can be connected to the processor 112 or the memory 111.

[0269] Among them, the processor 112 can control the transceiver 113 to communicate with other devices. Specifically, it can send information or data to other devices, or receive information or data sent by other devices. The transceiver 113 can include a transmitter and a receiver. The transceiver 113 can further include an antenna, and the number of antennas can be one or more.

[0270] It should be understood that each component in the electronic device is connected through a bus system. Among them, the bus system includes, in addition to the data bus, a power bus, a control bus, and a status signal bus.

[0271] The present invention also provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a computer, the computer can execute the methods in the above method embodiments. Or rather, an embodiment provided by the present invention also provides a computer program product containing instructions. When the instructions are executed by a computer, the computer executes the methods in the above method embodiments.

[0272] When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or a wireless manner (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a Digital Video Disc (DVD)), or a semiconductor medium (such as a Solid State Disk (SSD)), etc.

[0273] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0274] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of devices or modules can be in an electrical, mechanical, or other form.

[0275] The modules described as separate components may or may not be physically separated. The components displayed as modules may or may not be physical modules, that is, they can be located in one place, or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. For example, in each embodiment of the present application, the functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0276] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A transformer fault diagnosis method, characterized in that: include: Obtain DGA sample data according to preset transformer characteristic parameters; The preset PNN network is trained based on the DGA sample data to obtain a fault diagnosis model; the preset PNN network is a model for determining the fault information of the transformer according to the DGA data; the smoothing factor in the preset PNN network is determined according to the improved Seagull optimization algorithm; The DGA data corresponding to the target transformer is input into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

2. The method according to claim 1, characterized in that The step of obtaining DGA sample data according to preset transformer characteristic parameters includes: Determine the relative concentration of two fault characteristic gases and the ratio of each fault characteristic gas to the total hydrocarbon gas according to the absolute content of the fault characteristic gas of the transformer; The preset transformer characteristic parameters are determined based on the absolute contents of the various fault characteristic gases, the relative concentration contents between two of the fault characteristic gases, and the ratio between the various fault characteristic gases and the total hydrocarbon gas.

3. The method according to claim 1, characterized in that The improved Seagull optimization algorithm is obtained by the following method: Initializing the seagull population in the seagull optimization algorithm based on the Tent mapping strategy so that the seagull individuals in the seagull population are evenly distributed in the search space; A nonlinear improvement is performed on the control quantity for controlling the individual motion behavior of the seagull to balance the global search and local search capabilities of the seagull optimization algorithm; Based on the mutation mechanism, a disturbance term is added to the optimal seagull individual in the seagull population to improve the seagull optimization algorithm.

4. The method according to claim 1, characterized in that: The method of training a preset PNN network based on the DGA sample data to obtain a fault diagnosis model includes: Performing dimensionality reduction processing on the DGA sample data by using a kernel principal component analysis method to obtain feature data corresponding to the DGA sample data; Dividing the feature data corresponding to the DGA sample data into a training set and a test set; The preset PNN network is trained based on the training set to obtain an initial fault diagnosis model; the preset PNN network is a neural network determined according to the Bayes classification rule and the Parzen window probability density function; Performing a model test on the initial fault diagnosis model based on the test set to obtain a model test result; If the model test result is that the test passes, the initial fault diagnosis model is determined as the final fault diagnosis model.

5. The method according to claim 4, characterized in that The step of inputting the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer includes: Using the kernel principal component analysis method to perform dimensionality reduction processing on the DGA data corresponding to the target transformer to obtain diagnostic feature data corresponding to the DGA data; The diagnostic feature data is input into the fault diagnosis model to obtain fault information corresponding to the target transformer.

6. The method according to claim 1, characterized in that After the DGA data corresponding to the target transformer is input into the fault diagnosis model to obtain the fault information corresponding to the target transformer, the method further includes: According to the severity of the fault information corresponding to the target transformer, the fault of the target transformer is fuzzily graded to obtain the fault level corresponding to each fault information; Perform multi-order difference calculation on the original time series corresponding to the triggering events of each of the fault information according to the autoregressive integrated moving average model to obtain the stationary time series corresponding to the multiple triggering events; For the time information corresponding to each trigger event in the stationary time series, a logic operation set is established; the logic operation set is used to display the sequence rules of the predetermined operations between the multiple trigger events; the predetermined operations include logical AND, OR, and NOT operations; Placing the most important trigger event among the plurality of trigger events at a position corresponding to the leader node, and placing the remaining trigger events at positions corresponding to the member nodes; In response to the feedback signal sent by the leader node, determining whether the member node can feedback information to the leader node within a preset time; If a member node fails to feed back information to the leader node within a preset time, the classification result of the fault information is determined according to the trigger event corresponding to the member node and the fault level corresponding to the fault information corresponding to the trigger event.

7. The method according to claim 6, characterized in that If a member node fails to feed back information to the leader node within a preset time, after determining the classification result of the fault information according to the trigger event corresponding to the member node and the fault level corresponding to the fault information corresponding to the trigger event, the method further includes: According to the classification result of the fault information, the alarm module is controlled to issue a fault alarm; the alarm module includes one or more of a screen alarm module, a voice reminder alarm module, and a short message alarm module.

8. A transformer fault diagnosis device, characterized in that: include: A sample data acquisition module is used to acquire DGA sample data according to preset transformer characteristic parameters; A model training module, used for training a preset PNN network based on the DGA sample data to obtain a fault diagnosis model; the preset PNN network is a model for determining the fault information of the transformer according to the DGA data; the smoothing factor in the preset PNN network is determined according to the improved Seagull optimization algorithm; The fault diagnosis module is used to input the DGA data corresponding to the target transformer into the fault diagnosis model to obtain the fault information corresponding to the target transformer.

9. An electronic device, characterized in that: include: A processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein the computer program enables a computer to execute the method according to any one of claims 1 to 7.

Images