A method for generating an operation and maintenance strategy for a power transformer and a related device

Abstract

The present application discloses a method for generating an operation and maintenance strategy for a power transformer and a related device. The method comprises: respectively obtaining status data of different components of the power transformer and basic information of the transformer; extracting comprehensive data information of the power transformer based on the status data and basic information of the transformer, and constructing a two-dimensional image training data set based on the comprehensive data information and time series; using the two-dimensional image training data set to train a preset convolutional neural network model to obtain an optimized convolutional neural network model, wherein the optimized model parameters of the preset convolutional neural network model are obtained by optimization training based on the artificial fish swarm-firefly algorithm; inputting the currently obtained real-time comprehensive data information into the optimized convolutional neural network model to perform operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy. The present application can solve the technical problem that the existing technology cannot design an operation and maintenance strategy based on the health status of the components of the power transformer, resulting in the actual operation and maintenance strategy lacking pertinence and having poor effect.

Description

Technical Field

[0001] The present application relates to the technical field of power transformers, and in particular to a method for generating an operation and maintenance strategy for a power transformer and related devices. Background Art

[0002] The power industry is a fundamental energy sector that is crucial to the national economy and people's livelihoods, and is of vital importance to the healthy development of the national economy and the improvement of people's living standards. Power transformers are the hub of electrical energy transmission and conversion in the power system. The safety and reliability of transformers are fundamental to the safe and stable operation of the entire power system. For a long time, power grid companies both domestically and internationally have invested significant effort in developing power transformer operation and maintenance strategies.

[0003] The current differentiated O&M solution primarily involves dedicated O&M data collectors analyzing collected data, classifying defective components, and then assessing the equipment's health. This then determines the control level based on the equipment's health and importance, and ultimately formulates and implements an O&M strategy based on that control level. However, this approach fails to tailor O&M strategies to the detailed health status of each power transformer component, resulting in a lack of targeted O&M strategies. This leads to both over- and under-O&M, waste of O&M resources, and poor O&M performance. Summary of the Invention

[0004] The present application provides a method and related device for generating an operation and maintenance strategy for a power transformer, which is used to solve the technical problem that the existing technology cannot design an operation and maintenance strategy based on the health status of the components of the power transformer, resulting in the actual operation and maintenance strategy lacking pertinence and having poor effect.

[0005] In view of this, a first aspect of the present application provides a method for generating an operation and maintenance strategy for a power transformer, comprising:

[0006] Obtain status data of different components of the power transformer and basic information of the transformer respectively;

[0007] Extracting comprehensive data information of the power transformer according to the state quantity data and the transformer basic information, and constructing a two-dimensional image training data set according to the comprehensive data information and a time series;

[0008] Using the two-dimensional image training data set to train a preset convolutional neural network model to obtain an optimized convolutional neural network model, wherein the optimized model parameters of the preset convolutional neural network model are obtained by optimization training based on an artificial fish swarm-firefly algorithm;

[0009] The currently acquired real-time comprehensive data information is input into the optimized convolutional neural network model to perform operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy.

[0010] Preferably, extracting comprehensive data information of the power transformer based on the state quantity data and the transformer basic information, and constructing a two-dimensional image training data set based on the comprehensive data information and time series, includes:

[0011] Calculating the deduction value of each component according to the state quantity data, and extracting the weighted deduction value of each component based on a preset weight;

[0012] Performing a status evaluation on each component according to the deduction value and preset evaluation rules to obtain a status evaluation result, and calculating the equipment failure probability by combining the status evaluation result and the deduction value;

[0013] Obtain equipment value, load level, and equipment status based on equipment price, load importance, and equipment importance, respectively;

[0014] Calculate the extent of equipment loss based on equipment cost, personal safety factor, and power safety factor;

[0015] Building comprehensive data information based on the equipment failure probability, the equipment value, the load level, the equipment status, the equipment loss level, and a preset operation and maintenance strategy;

[0016] The one-dimensional data information is converted into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set.

[0017] Preferably, the step of converting the one-dimensional data information into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set further includes:

[0018] The comprehensive data information is subjected to abnormality elimination, normalization and one-hot encoding processing to achieve data information standardization.

[0019] Preferably, the two-dimensional image training data set is used to train a preset convolutional neural network model to obtain an optimized convolutional neural network model, and the optimized model parameters of the preset convolutional neural network model are obtained by optimization training according to the artificial fish school-firefly algorithm, and the method further includes:

[0020] Build an initial convolutional neural network model based on the attention mechanism;

[0021] Using an artificial fish swarm-firefly algorithm to perform joint optimization training based on individual sharing on initial model parameters in the initial convolutional neural network model to obtain optimized model parameters;

[0022] The initial convolutional neural network model is adjusted by optimizing the model parameters to obtain a preset convolutional neural network model.

[0023] A second aspect of the present application provides a device for generating an operation and maintenance strategy for a power transformer, comprising:

[0024] An information acquisition unit, used to respectively acquire status data of different components of the power transformer and basic information of the transformer;

[0025] An information processing unit, configured to extract comprehensive data information of the power transformer based on the state quantity data and the transformer basic information, and construct a two-dimensional image training data set based on the comprehensive data information and a time series;

[0026] A model training unit is used to train a preset convolutional neural network model using the two-dimensional image training data set to obtain an optimized convolutional neural network model, wherein the optimized model parameters of the preset convolutional neural network model are obtained according to the artificial fish swarm-firefly algorithm optimization training;

[0027] The strategy generation unit is used to input the currently acquired real-time comprehensive data information into the optimized convolutional neural network model to perform operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy.

[0028] Preferably, the information processing unit is specifically used to:

[0029] Calculating the deduction value of each component according to the state quantity data, and extracting the weighted deduction value of each component based on a preset weight;

[0030] Performing a status evaluation on each component according to the deduction value and preset evaluation rules to obtain a status evaluation result, and calculating the equipment failure probability by combining the status evaluation result and the deduction value;

[0031] Obtain equipment value, load level, and equipment status based on equipment price, load importance, and equipment importance, respectively;

[0032] Calculate the extent of equipment loss based on equipment cost, personal safety factor, and power safety factor;

[0033] Building comprehensive data information based on the equipment failure probability, the equipment value, the load level, the equipment status, the equipment loss level, and a preset operation and maintenance strategy;

[0034] The one-dimensional data information is converted into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set.

[0035] Preferably, it also includes:

[0036] The standardization unit is used to perform abnormal elimination, normalization and one-hot encoding processing on the comprehensive data information to achieve data information standardization.

[0037] Preferably, it also includes:

[0038] Model building unit, used to build the initial convolutional neural network model based on the attention mechanism;

[0039] a parameter optimization unit, configured to perform individual-sharing-based joint optimization training on initial model parameters in the initial convolutional neural network model using an artificial fish swarm-firefly algorithm to obtain optimized model parameters;

[0040] A model optimization unit is used to adjust the initial convolutional neural network model by optimizing the model parameters to obtain a preset convolutional neural network model.

[0041] A third aspect of the present application provides a device for generating an operation and maintenance strategy for a power transformer, the device comprising a processor and a memory;

[0042] The memory is used to store program code and transmit the program code to the processor;

[0043] The processor is used to execute the method for generating an operation and maintenance strategy for a power transformer described in the first aspect according to the instructions in the program code.

[0044] In a fourth aspect, the present application provides a computer-readable storage medium, which is used to store program code, and the program code is used to execute the method for generating an operation and maintenance strategy for a power transformer described in the first aspect.

[0045] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0046] In the present application, a method for generating an operation and maintenance strategy for a power transformer is provided, comprising: respectively obtaining status data of different components of the power transformer and basic information of the transformer; extracting comprehensive data information of the power transformer based on the status data and basic information of the transformer, and constructing a two-dimensional image training data set based on the comprehensive data information and a time series; using the two-dimensional image training data set to train a preset convolutional neural network model to obtain an optimized convolutional neural network model, wherein the optimized model parameters of the preset convolutional neural network model are obtained by optimization training based on an artificial fish swarm-firefly algorithm; and inputting the currently obtained real-time comprehensive data information into the optimized convolutional neural network model for operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy.

[0047] The method for generating an operation and maintenance strategy for a power transformer provided in the present application performs operation and maintenance data analysis by acquiring status data of different components of the power transformer and basic information of the transformer, and then trains an optimized neural network model based on comprehensive data information composed of data of multiple components and equipment, so that the model has stronger expressive power in generating operation and maintenance strategies and can ensure the accuracy of the generated optimized operation and maintenance strategies; optimizing and training the model parameters using a specific algorithm can improve the learning performance of the model and thereby ensure the reliability of the model. Therefore, the present application can solve the technical problem that the existing technology cannot design an operation and maintenance strategy based on the health status of the components of the power transformer, resulting in the actual operation and maintenance strategy being lacking in specificity and having poor results. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of a method for generating an operation and maintenance strategy for a power transformer provided in an embodiment of the present application;

[0049] Figure 2 A schematic structural diagram of a device for generating an operation and maintenance strategy for a power transformer provided in an embodiment of the present application;

[0050] Figure 3 This is an example diagram of the component status deduction standard provided in the embodiment of the present application;

[0051] Figure 4 A schematic diagram of the comprehensive data information processing flow provided in the embodiment of the present application;

[0052] Figure 5 Schematic diagram of the working principle of the attention mechanism provided in the embodiment of this application;

[0053] Figure 6 A schematic diagram of the structure of the initial convolutional neural network model provided in the embodiment of the present application;

[0054] Figure 7 A schematic diagram of the data processing flow of the preset convolutional neural network model provided in an embodiment of the present application;

[0055] Figure 8 Schematic diagram of the optimal individual sharing process of the artificial fish swarm algorithm and the firefly algorithm provided in the embodiments of the present application. DETAILED DESCRIPTION

[0056] In order to help those skilled in the art better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0057] For easier understanding, see Figure 1 , an embodiment of a method for generating an operation and maintenance strategy for a power transformer provided by the present application includes:

[0058] Step 101: Obtain status data of different components of the power transformer and basic transformer information respectively.

[0059] In this embodiment, the status data of six components, namely the power transformer body, bushing, tap changer, cooling system, non-electrical protection system and online monitoring device, are mainly obtained; the basic information of the transformer mainly refers to the equipment price, load status or importance, importance of the equipment relative to the power grid and fault conditions of the power transformer, which is mainly used to extract deeper representative information; other information can also be selected according to actual conditions, which is not limited here.

[0060] In addition, the status data of the power transformer itself mainly includes operating conditions, inspections and checks, high-voltage tests, chromatographic analysis, and simplified tests. Operating conditions can be further divided into short-circuit conditions, transformer overload, and neutral point DC current. Inspections and checks mainly check oil leakage, terminal clamps, main body oil level, tap changer oil level, noise and vibration, foundation subsidence, operating oil temperature, gas relay operation, oil conservator, respirator, secondary terminals, rust, and core insulation. High-voltage tests mainly include infrared detection, DC resistance of windings and bushings, winding dielectric loss, winding deformation testing, and insulation resistance testing. Chromatographic analysis includes total hydrocarbons, acetylene, and hydrogen. Simplified tests include furfural content in oil, oil dielectric loss factor, oil breakdown voltage, moisture, gas content in oil, oil particle size test, and corrosive sulfur.

[0061] Bushings are mainly divided into porcelain bushings, appearance, tests and bushing current transformers; porcelain bushings are divided into porcelain bushing creepage, external insulation configuration, porcelain insulation damage, composite external insulation cracking and hydrophobicity; appearance includes oil level indication, oil leakage, bushing wiring and end screen lead-out wire; tests include infrared detection, dielectric loss and capacitance, partial discharge live test, insulating oil and insulation resistance; bushing current transformers include oil leakage, DC resistance and insulation resistance.

[0062] The tap changer can be divided into appearance, maintenance and testing. Appearance mainly refers to the tap position, oil tank and oil leakage; maintenance mainly refers to the number of switching times or maintenance intervals, transmission mechanism, filter element, light gas relay inspection and control circuit; testing mainly refers to dynamic characteristics, DC resistance, oil breakdown voltage, oil chromatography detection, water content and oil pressure resistance.

[0063] The cooling system includes submersible oil pump, cooling fan, radiator, cooler and others; submersible oil pump includes oil leakage, refusing to move and motor operation; cooling fan includes refusing to move, fan blades and abnormal noise; radiator includes oil leakage and radiator heat dissipation effect; cooler includes oil leakage and cooler electric valve; others include oil flow sensor, water flow sensor and air pump.

[0064] The non-electrical protection system includes a cooling control box, a gas relay, a thermometer, a pressure relief valve, a pressure regulating switch oil flow (pressure) relay, a quick-acting oil pressure relay and a discharge gap; the cooling control box includes a contactor, a power supply, a cooling system maintenance, a cooler control system and a starting method; the gas relay includes oil leakage, a sight glass, relay action, a rain cover and periodic calibration; the thermometer includes a temperature indication and periodic calibration; the pressure relief valve includes oil leakage, an action indication, a secondary circuit and periodic calibration; the pressure regulating switch oil flow relay includes oil leakage and a sight glass; the quick-acting oil pressure relay includes oil leakage and relay action; the discharge gap includes inspection and testing.

[0065] The online monitoring devices include transformer oil chromatography online monitoring devices, bushing online monitoring devices, neutral point DC monitoring devices, leakage detection devices and other monitoring devices; these state quantity information include two types of state quantity data: detection data and device data.

[0066] Step 102: extract comprehensive data information of the power transformer based on the state quantity data and basic information of the transformer, and construct a two-dimensional image training data set based on the comprehensive data information and time series.

[0067] Furthermore, step 102 includes:

[0068] Calculate the deduction value of each component based on the state quantity data, and extract the weighted deduction value of each component based on the preset weight;

[0069] Evaluate the status of each component based on the deduction value and preset evaluation rules to obtain the status evaluation results, and calculate the equipment failure probability by combining the status evaluation results and the deduction value;

[0070] Obtain equipment value, load level, and equipment status based on equipment price, load importance, and equipment importance, respectively;

[0071] Calculate the extent of equipment loss based on equipment cost, personal safety factor, and power safety factor;

[0072] Build comprehensive data information based on equipment failure probability, equipment value, load level, equipment status, equipment loss level and preset operation and maintenance strategy;

[0073] The one-dimensional data information is converted into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set.

[0074] It's important to note that all of the aforementioned state information is configured with corresponding deduction criteria. This deduction criteria allows you to determine the basic deduction value for each component's corresponding sub-item, and then calculate the weighted deduction value based on the weight coefficient. That is, each component's sub-item has a corresponding deduction value; each component has multiple deduction values, and the total deduction value for that component can be calculated. You can even calculate the total deduction value for the entire device for subsequent data analysis and calculations.

[0075] The process of evaluating the status of each component based on the deduction value and the preset evaluation rules can be described as follows: when the deduction points of any single state quantity and the total deduction points of the component are both within the normal state range of the preset evaluation rules, it is considered to be in a normal state; when the deduction points of any single state quantity or the total deduction points of all state quantities of the component reach the attention state range specified in the preset evaluation rules, it is considered to be in a caution state; when the deduction points of any single state quantity reach the abnormal state or serious state range specified in the preset evaluation rules, it is considered to be in an abnormal state or serious state; the specific status area rules are expressed as follows Figure 3 shown.

[0076] The average failure probability of the equipment is calculated as:

[0077]

[0078] Where l is the status evaluation result, ranging from 1 to 4, 1 represents normal status, 2 represents caution status, 3 represents abnormal status, and 4 represents severe status; P l It is the probability of component failure under a certain evaluation state, that is, the average failure probability of the equipment. l 、N l They are the number of faulty devices and the total number of devices under a certain evaluation status.

[0079] Specifically, we first need to define the defect index, a parameter used to describe the severity of component defects. The higher the defect index, the more serious the component defect. This index is obtained by converting the status grade and deduction value obtained from the component evaluation process. The calculation process of the equipment defect index d is:

[0080]

[0081] Among them, S i is the evaluation status index of the component, M is the actual total deduction value of the equipment, and M i It is the maximum deduction value for the grade of this type of component. i The value of changes according to the status evaluation results of the component. It is 0 in normal state, 1 in caution state, 2 in abnormal state, and 3 in severe state. i The value of is obtained based on the statistical evaluation results of the equipment status.

[0082] Furthermore, the relationship between the equipment failure probability P and the equipment defect index d follows an exponential distribution. In practice, the defect index is divided into several intervals, and the number of equipment failures within each defect index interval within the target area is counted. The median of each defect index interval is selected to calculate the average failure probability within that interval, which represents the equipment failure probability. Theoretically, dividing the interval into more intervals will yield more fitting points and more accurate results, but this requires sufficient sample data. To correspond to the four status levels of the condition evaluation results (normal, caution, abnormal, and severe), the defect index interval is divided into four intervals: 0-1, 1-2, 2-3, and 3-4. The relationship between defect index and equipment failure probability is shown in Table 1.

[0083] Table 1 Defect index-equipment failure probability list

[0084] Status Level Defect Index Number of faulty devices Total number of devices Failure probability normal 0≤d≤1 n1 N1 n1 / N1 Notice 1<d≤2 n2 N2 n2 / N2 abnormal 2<d≤3 n3 N3 n3 / N3 serious 3<d≤4 n4 N4 n4 / N4

[0085] Equipment value, load level, and equipment status can be directly calculated based on equipment price, load importance, and equipment importance. For example, equipment value is described in levels (see Table 2). Load levels can also be described in three levels based on load importance: tertiary load, secondary load, and primary load. Equipment status is divided into hub substations, key substations, and general substations based on the importance of the substation where the equipment is located in the power grid. This is also considered based on whether the substation grid structure meets the N-1 requirement. Equipment status is dynamically determined annually based on the operating mode of the power grid under the jurisdiction of each power supply bureau, and can also be quantified in levels (see Table 3 for details).

[0086] Table 2 Equipment value level list

[0087] Equipment value Device level value range Less than 100,000 yuan 1 More than 100,000 yuan but less than 200,000 yuan 2 More than 200,000 yuan but less than 300,000 yuan 3 More than 300,000 yuan but less than 500,000 yuan 4 More than 500,000 yuan but less than 800,000 yuan 5 More than 800,000 yuan but less than 1 million yuan 6 More than 1 million yuan but less than 10 million yuan 7 More than 10 million yuan but less than 50 million yuan 8 More than 50 million yuan but less than 100 million yuan 9 More than 100 million yuan 10

[0088] Table 3 Equipment status level list

[0089]

[0090] Calculate the extent of equipment loss based on equipment cost, personal safety factor, and power safety factor:

[0091]

[0092] Wherein, j ranges from 1 to 3, with a value of 1 indicating equipment cost, a value of 2 indicating personal safety, and a value of 3 indicating power safety; W Lj is the weight of the degree of loss caused by the loss factors, with reference values ​​as follows: 1-equipment cost, take 0.3; 2-personal safety, take 0.3; 3-power safety, take 0.4; L jis the loss degree of a certain factor; the loss degree of a certain factor L j Calculate as follows:

[0093]

[0094] Among them, K = 1~n, which represents the loss level of the loss factor. For equipment cost factor, n is 9; for personal safety factor, n is 9; for power safety factor, n is 9; for IOF jk The loss value of the loss factor at a certain level; POF jk is the probability of occurrence of a loss factor at a certain level; the probability of occurrence of a loss factor is calculated according to the following formula:

[0095]

[0096] Where j is the loss factor; k is the loss level of the loss factor; n is the total number of failures; n jk is the number of failures for a loss factor at a certain level. The loss factor, i.e., the level and value range of the equipment loss degree, is described in Table 4. The loss factor can be obtained by statistically analyzing the number of failures, i.e., the number of failures for each loss factor at different accident event levels.

[0097] Table 4 List of equipment loss levels

[0098]

[0099] The O&M strategy is a collection of component O&M strategies, primarily including routine O&M strategies, ad hoc O&M strategies, and outage O&M strategies. For example, a specific O&M strategy specifies O&M frequency as once a month, once a quarter, once every six months, and once a year. The definition of "once a month" as "1," "once a quarter" as "2," "once every six months" as "3," and "once a year" as "4" is unclear.

[0100] Combining all the above information, we can construct comprehensive data information, or what is called an operation and maintenance data set. The operation and maintenance data set can be divided into a data ratio of 8:1:1 based on subsequent training requirements, corresponding to the training set, validation set, and test set respectively; the training set is used to train the model, the validation set is used to tune the model parameters, and the test set is used to evaluate the model's performance.

[0101] This embodiment converts one-dimensional data information, namely comprehensive data information, into a two-dimensional image format, which not only enhances data representation but also facilitates data analysis by the model. Operation and maintenance strategies can be set to meet periodicity, with one cycle being an annual or semi-annual period. Time is used as the horizontal variable to form a time series. The comprehensive data information is used as the vertical variable to form a periodic two-dimensional image, thereby obtaining a two-dimensional image training dataset. The same operation can be performed on the validation and test sets, resulting in data information of the same format for unified processing.

[0102] Furthermore, the one-dimensional data information is converted into a two-dimensional image based on the comprehensive data information and the time series to obtain a two-dimensional image training data set, which also includes:

[0103] The comprehensive data information is processed by outlier elimination, normalization and one-hot encoding to achieve data information standardization.

[0104] It should be noted that since the dimensions of the comprehensive data information in this embodiment are not uniform and there are many information levels, some preprocessing operations can be performed before processing the data information, such as abnormal data removal, dimensional normalization, and one-hot encoding, etc., which can convert the data information into standardized data for subsequent analysis.

[0105] Abnormal elimination targets, such as missing values, are typically caused by temporary system failures or storage system write errors, resulting in data anomalies. These typically include constant or null values ​​and are relatively easy to identify. For long periods of missing data, the sequence segment is discarded. For short periods of missing data, data is filled and corrected based on data feature analysis.

[0106] The normalization operation in this embodiment is to map different types of comprehensive data information to the range [0, 1], which can be specifically expressed as:

[0107]

[0108] Among them, x is the original data, x max 、x min Indicates the maximum and minimum values ​​of each type of information, x * The data are obtained after normalization.

[0109] This example uses a one-hot encoding mechanism to encode the normalized data, and represents all data information in binary vector form, thereby quantifying and distinguishing different data volumes. This not only better expresses data differences, but also facilitates data analysis of the model. Figure 4,The comprehensive data information as the initial set can be dimensionally transformed after ,standardization processing, and then a two-dimensional image training data set is obtained.

[0110] Step 103: Use the two-dimensional image training data set to train the preset convolutional neural network model to obtain an optimized convolutional neural network model. The optimized model parameters of the preset convolutional neural network model are obtained by optimization training based on the artificial fish swarm-firefly algorithm.

[0111] The preset convolutional neural network model in this embodiment can be understood as a pre-built model with determined model parameters. By training the model with a two-dimensional image training data set, the model performance can be optimized, and the model's analysis and expression capabilities for operation and maintenance strategy-related data can be improved, thereby obtaining prediction results that are more in line with actual conditions. The model parameters in the preset convolutional neural network model in this embodiment are obtained by optimizing the improved fusion algorithm, namely the artificial fish swarm-firefly algorithm, which is an optimization algorithm after the fusion of the artificial fish swarm algorithm and the firefly algorithm. Since the population evolution of the two optimization algorithms is carried out synchronously, it is possible to achieve the optimal individual sharing of the population, realize joint optimization iteration, and improve the global search capability of the population. Therefore, the optimized model parameters obtained based on this fusion algorithm optimization can improve the learning ability of the model to a certain extent and ensure the training effect of the model.

[0112] Furthermore, before step 103, the following steps are also included:

[0113] Build an initial convolutional neural network model based on the attention mechanism;

[0114] The artificial fish swarm-firefly algorithm is used to perform joint optimization training based on individual sharing on the initial model parameters in the initial convolutional neural network model to obtain the optimized model parameters;

[0115] The initial convolutional neural network model is adjusted by optimizing the model parameters to obtain a preset convolutional neural network model.

[0116] The loss function of the initial convolutional neural network model in this embodiment is the center loss function, which can be expressed as:

[0117]

[0118] Among them, x i is the input feature vector, c y is the center of category y corresponding to x, c y Each iteration update is implemented through mini-batch. Mini-batch is a parameter update method for small batch gradient descent. In order to avoid large disturbances caused by a few sample labeling errors, this embodiment uses the scalar α c To control c y The update rate of αc ∈[0,1]. In addition, in order to alleviate the overfitting of the model, this embodiment adopts Dropout to reduce the dependency between neurons and speed up the convergence of the model.

[0119] In this embodiment, the attention mechanism is introduced. Two attention mechanisms (Attention-left and Attention-right) are used to perform attention calculations on the output vector and internal features of the convolutional neural network layer, thereby enhancing the weights of important time step vectors and important features in each time step vector, thereby improving the accuracy of operation and maintenance strategy selection. For an example of the principle of the attention mechanism, please refer to Figure 5 .

[0120] The network structure of the initial convolutional neural network model in this embodiment can be found in Figure 6 and Figure 7 , mainly including 9 convolutional layers, 3 pooling layers, and 3 fully connected layers. The number of filters in the first convolutional layer is 256, and the final output size through the fully connected layer is 2. Except for the Sigmoid activation function used in the last fully connected layer, the other convolutional layers and fully connected layers all use the ReLU function, and Flatten one-dimensionalization is used between the last maximum pooling layer and the fully connected layer. In addition, the network parameters of the initial convolutional neural network model of this embodiment are shown in Table 5.

[0121] Table 5 Network parameter list of the initial convolutional neural network model

[0122] Serial number Network layer Filter size step length Number of filters Activation Function 1 Convolutional layer 1 3×3 1×1 256 Relu 2 Convolutional layer 2 3×3 1×1 128 Relu 3 Convolutional layer 3 3×3 1×1 128 Relu 4 Max pooling layer 1 2×2 2×2 / / 5 Convolutional layer 4 3×3 1×1 128 Relu 6 Convolutional layer 5 3×3 1×1 128 Relu 7 Convolutional layer 6 3×3 1×1 128 Relu 8 Max Pooling Layer 2 2×1 2×1 / / 9 Convolutional layer 7 3×3 1×1 64 Relu 10 Convolutional layer 8 3×3 1×1 64 Relu 11 Convolutional layer 9 3×3 2×2 16 Relu 12 Max pooling layer 3 2×2 2×2 / /

[0123] The core of the artificial fish swarm and firefly algorithms' individual-sharing joint optimization training of the initial model parameters in the initial convolutional neural network model is the fusion training of the artificial fish swarm algorithm and the firefly algorithm. While both algorithms have certain advantages when solving specific problems, they still exhibit problems such as insufficient adaptability, slow convergence, and being trapped in local optima in certain circumstances. Therefore, this embodiment sets a rule that, at regular intervals, the optimal individuals from a population in one algorithm are shared with a population in the other algorithm, improving the global search capabilities of both populations.

[0124] Specifically, the parameters of the initialization population of the two algorithms can be set to be consistent, and the initial model parameters are the initialization population. In addition, the fitness values ​​of the two populations can be calculated based on the central loss function in the convolutional neural network model of this embodiment; the global optimal positions searched by the two populations are recorded respectively, and the individual search results are sorted from best to worst according to the fitness value to form a sorted set. Observe the algebra to see if sharing is required. If so, share the optimal individual; take out the first m results and the last m results of the fitness sorted sets of the artificial fish swarm algorithm and the firefly algorithm, and then exchange them with each other. Please refer to Figure 8 To prevent the value from not being updated and returning to the original path in subsequent iterations, the initial value of m is set to no more than half of the population. As the number of iterations increases, it may also fall into the "premature maturity" situation, so m is set to increase with the number of iterations:

[0125]

[0126] Where N is the population size, t and t max are the current number of iterations and the maximum number of iterations respectively, r0 is the proportional control coefficient, which can be set in the range [0.51,1]. is a floor function. During the swap, the firefly algorithm population and the artificial fish swarm algorithm population update their positions and fitness values. The global optimal value, including both the artificial fish swarm algorithm and the firefly algorithm, is recorded. Determine whether the condition for stopping the loop is met. If so, output the global optimal value and global optimal position.

[0127] Step 104: Input the currently acquired real-time comprehensive data information into the optimized convolutional neural network model to perform operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy.

[0128] The real-time, comprehensive data currently being acquired is data on power transformers that require analysis, and the corresponding operation and maintenance strategies for these components are unclear. Using this real-time, comprehensive data as input and the predicted, optimized operation and maintenance strategies as output demonstrates highly efficient information processing capabilities. Furthermore, it is understood that this real-time, comprehensive data also requires the aforementioned data processing to be integrated into a data format acceptable to the model. The specific process will not be detailed here.

[0129] In addition, the daily operation and maintenance strategy of the power transformer is the minimum value among the corresponding values ​​of the daily operation and maintenance strategies of each component of the power transformer; the special patrol maintenance strategy of the power transformer is the set of special patrol maintenance strategies of each component of the power transformer; and the power outage maintenance strategy of the power transformer is the set of power outage maintenance strategies of each component of the power transformer.

[0130] The method for generating an operation and maintenance strategy for a power transformer provided in an embodiment of the present application performs operation and maintenance data analysis by acquiring status data of different components of the power transformer and basic information of the transformer, and then trains an optimized neural network model based on comprehensive data information composed of data of multiple components and equipment, so that the model has stronger expressive power in generating operation and maintenance strategies and can ensure the accuracy of the generated optimized operation and maintenance strategies; optimizing and training the model parameters using a specific algorithm can improve the learning performance of the model and thus ensure the reliability of the model. Therefore, the embodiment of the present application can solve the technical problem that the existing technology cannot design an operation and maintenance strategy based on the health status of the components of the power transformer, resulting in the actual operation and maintenance strategy lacking pertinence and having poor results.

[0131] For easier understanding, see Figure 2 The present application provides an embodiment of a device for generating an operation and maintenance strategy for a power transformer, comprising:

[0132] The information acquisition unit 201 is used to respectively acquire status data of different components of the power transformer and basic information of the transformer;

[0133] The information processing unit 202 is used to extract comprehensive data information of the power transformer based on the state quantity data and the basic information of the transformer, and to construct a two-dimensional image training data set based on the comprehensive data information and the time series;

[0134] A model training unit 203 is configured to train a preset convolutional neural network model using a two-dimensional image training dataset to obtain an optimized convolutional neural network model, wherein the optimized model parameters of the preset convolutional neural network model are obtained by optimization training based on an artificial fish swarm-firefly algorithm;

[0135] The strategy generation unit 204 is used to input the currently acquired real-time comprehensive data information into the optimized convolutional neural network model to perform operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy.

[0136] Furthermore, the information processing unit 202 is specifically configured to:

[0137] Calculate the deduction value of each component based on the state quantity data, and extract the weighted deduction value of each component based on the preset weight;

[0138] Evaluate the status of each component based on the deduction value and preset evaluation rules to obtain the status evaluation results, and calculate the equipment failure probability by combining the status evaluation results and the deduction value;

[0139] Obtain equipment value, load level, and equipment status based on equipment price, load importance, and equipment importance, respectively;

[0140] Calculate the extent of equipment loss based on equipment cost, personal safety factor, and power safety factor;

[0141] Build comprehensive data information based on equipment failure probability, equipment value, load level, equipment status, equipment loss level and preset operation and maintenance strategy;

[0142] The one-dimensional data information is converted into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set.

[0143] Furthermore, it also includes:

[0144] The standardization unit 205 is used to perform abnormal elimination, normalization and one-hot encoding processing on the comprehensive data information to achieve data information standardization.

[0145] Furthermore, it also includes:

[0146] A model building unit 206 is used to build an initial convolutional neural network model based on the attention mechanism;

[0147] a parameter optimization unit 207 for performing individual-sharing joint optimization training on initial model parameters in the initial convolutional neural network model using an artificial fish swarm-firefly algorithm to obtain optimized model parameters;

[0148] The model optimization unit 208 is used to adjust the initial convolutional neural network model by optimizing model parameters to obtain a preset convolutional neural network model.

[0149] The present application also provides an operation and maintenance strategy generation device for a power transformer, the device including a processor and a memory;

[0150] The memory is used to store program codes and transmit the program codes to the processor;

[0151] The processor is used to execute the method for generating an operation and maintenance strategy for a power transformer in the above method embodiment according to the instructions in the program code.

[0152] The present application also provides a computer-readable storage medium, which is used to store program code, and the program code is used to execute the method for generating an operation and maintenance strategy for a power transformer in the above method embodiment.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0154] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0155] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0156] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or 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 and includes a number of instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (English full name: Read-Only Memory, English abbreviation: ROM), random access memory (English full name: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program code.

[0157] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for generating an operation and maintenance strategy for a power transformer, characterized in that: include: Obtain status data of different components of the power transformer and basic information of the transformer respectively; The comprehensive data information of the power transformer is extracted according to the state quantity data and the basic information of the transformer, and a two-dimensional image training data set is constructed according to the comprehensive data information and the time series. The specific process is as follows: Calculating the deduction value of each component according to the state quantity data, and extracting the weighted deduction value of each component based on a preset weight; Performing a status evaluation on each component according to the deduction value and preset evaluation rules to obtain a status evaluation result, and calculating the equipment failure probability by combining the status evaluation result and the deduction value; Obtain equipment value, load level, and equipment status based on equipment price, load importance, and equipment importance, respectively; Calculate the extent of equipment loss based on equipment cost, personal safety factor, and power safety factor; Building comprehensive data information based on the equipment failure probability, the equipment value, the load level, the equipment status, the equipment loss level, and a preset operation and maintenance strategy; Converting the one-dimensional data information into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set; Build an initial convolutional neural network model based on the attention mechanism; Using an artificial fish swarm-firefly algorithm to perform joint optimization training based on individual sharing on initial model parameters in the initial convolutional neural network model to obtain optimized model parameters; Adjusting the initial convolutional neural network model by optimizing the model parameters to obtain a preset convolutional neural network model; The preset convolutional neural network model is trained using the two-dimensional image training data set to obtain an optimized convolutional neural network model, wherein the optimized model parameters of the preset convolutional neural network model are obtained by optimization training using an artificial fish swarm-firefly algorithm; The currently acquired real-time comprehensive data information is input into the optimized convolutional neural network model to perform operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy.

2. The method for generating an operation and maintenance strategy for a power transformer according to claim 1, wherein: The step of converting the one-dimensional data information into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set may also include: The comprehensive data information is subjected to abnormality elimination, normalization and one-hot encoding processing to achieve data information standardization.

3. A device for generating an operation and maintenance strategy for a power transformer, characterized in that: include: An information acquisition unit, used to respectively acquire status data of different components of the power transformer and basic information of the transformer; An information processing unit is used to extract comprehensive data information of the power transformer based on the state quantity data and the transformer basic information, and to construct a two-dimensional image training data set based on the comprehensive data information and time series. The information processing unit is specifically used to: Calculating the deduction value of each component according to the state quantity data, and extracting the weighted deduction value of each component based on a preset weight; Performing a status evaluation on each component according to the deduction value and preset evaluation rules to obtain a status evaluation result, and calculating the equipment failure probability by combining the status evaluation result and the deduction value; Obtain equipment value, load level, and equipment status based on equipment price, load importance, and equipment importance, respectively; Calculate the extent of equipment loss based on equipment cost, personal safety factor, and power safety factor; Building comprehensive data information based on the equipment failure probability, the equipment value, the load level, the equipment status, the equipment loss level, and a preset operation and maintenance strategy; Converting the one-dimensional data information into a two-dimensional image according to the comprehensive data information and the time series to obtain a two-dimensional image training data set; Model building unit, used to build the initial convolutional neural network model based on the attention mechanism; a parameter optimization unit, configured to perform individual-sharing-based joint optimization training on initial model parameters in the initial convolutional neural network model using an artificial fish swarm-firefly algorithm to obtain optimized model parameters; A model optimization unit, configured to adjust the initial convolutional neural network model by optimizing the model parameters to obtain a preset convolutional neural network model; A model training unit is used to train the preset convolutional neural network model using the two-dimensional image training data set to obtain an optimized convolutional neural network model, wherein the optimized model parameters of the preset convolutional neural network model are obtained by optimization training based on the artificial fish swarm-firefly algorithm; The strategy generation unit is used to input the currently acquired real-time comprehensive data information into the optimized convolutional neural network model to perform operation and maintenance strategy analysis to obtain an optimized operation and maintenance strategy.

4. The device for generating an operation and maintenance strategy for a power transformer according to claim 3, characterized in that: Also includes: The standardization unit is used to perform abnormal elimination, normalization and one-hot encoding processing on the comprehensive data information to achieve data information standardization.

5. A device for generating an operation and maintenance strategy for a power transformer, characterized in that: The device includes a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the method for generating an operation and maintenance strategy for a power transformer according to any one of claims 1-2 according to the instructions in the program code.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store program code, and the program code is used to execute the method for generating an operation and maintenance strategy for a power transformer according to any one of claims 1-2.

Images