Dry-type transformer turn-to-turn insulation aging degree evaluation method based on improved crayfish optimization algorithm

Abstract

The invention discloses a dry-type transformer turn-to-turn insulation aging degree evaluation method based on an improved crayfish optimization algorithm, and is applied to the technical field of transformer evaluation. Comprising the following steps: constructing a prediction model based on a BP neural network and determining a topological structure; defining population parameters and environment temperature of the improved crayfish optimization algorithm; judging whether to enter a sunstroke prevention stage or a foraging stage according to the environment temperature, and updating individual positions; calculating the fitness of all individuals, judging whether an iteration stop condition is met or not, recording an optimal individual after iteration is finished, introducing a vertical crossover operation, and updating the optimal individual; and endowing the weight and the threshold value of the optimal individual to the aging degree prediction model, and performing dry-type transformer turn-to-turn insulation aging degree prediction on the input data by using the trained aging degree prediction model. According to the method, high-reliability intelligent decision support is provided for power equipment state maintenance, the comprehensive performance of the method is remarkably superior to that of a traditional method, and the method has wide industrial application prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer evaluation, and more specifically, to a method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm. Background Art

[0002] When facing the problem of evaluating the aging degree of the inter-turn insulation of a Nomex dry-type transformer, the existing technical solutions often use BP neural networks. Although BP neural networks perform well, they still have certain limitations, and the most obvious points may affect the ability to solve practical problems: (1) The convergence speed is relatively slow, (2) It is easy to fall into local minima, and (3) It is difficult to determine the number of hidden layers and nodes. Based on the above defects, it is necessary to optimize the parameters of the BP neural network.

[0003] Among the existing parameter optimization algorithms, the meta-heuristic optimization algorithm initializes multiple solutions and relies on the objective function to iteratively solve, avoiding local optimal problems and improving the global search ability. Compared with traditional optimization algorithms, the meta-heuristic algorithm is easier to implement, suitable for complex multi-extremum problems, and widely used in fields such as the multi-traveling salesman problem, image segmentation, ship scheduling, and feature selection.

[0004] Meta-heuristic optimization algorithms are mainly divided into four categories: swarm intelligence algorithms, physics-based algorithms, evolutionary algorithms, and human-inspired algorithms. Swarm intelligence algorithms imitate the behaviors of biological groups. For example, Particle Swarm Optimization (PSO) simulates the foraging behavior of birds, Artificial Bee Colony Algorithm (ABC) draws on the division of labor among bees, Ant Colony Optimization (ACO) uses pheromones to guide the search, and Cuckoo Search (CS) is based on the parasitic behavior of cuckoos. In recent years, new algorithms such as Snake Optimization (SO), White Shark Optimization (WSO), Honey Badger Algorithm (HBA), and Remora Optimization Algorithm (ROA) have emerged. Physics-based optimization algorithms originate from physical phenomena. For example, Gravitational Search Algorithm (GSA) simulates the universal gravitation, Simulated Annealing (SA) is based on the principle of solid annealing, Electromagnetic Field Optimization (EFO) is inspired by electromagnetic forces, Black Hole Algorithm (BH) is based on the black hole swallowing mechanism, Multi-Verse Optimization (MVO) combines the theories of black holes, white holes, and wormholes, and Thermal Exchange Optimization (TEO) originates from Newton's law of cooling. Evolutionary algorithms imitate biological evolution. For example, Genetic Algorithm (GA) is based on Darwin's natural selection, Evolution Strategy (ES) and Genetic Programming (GP) also adopt biological evolution mechanisms, and Differential Evolution (DE) generates new solutions through differential mutation. In addition, there are Virulence Optimization Algorithm (VOA) that simulates the virus invasion process and Evolutionary Programming (EP) which is based on biological genetic behaviors. Human-inspired algorithms draw on human cognition. For example, Harmony Search (HS) simulates music performance to find the best harmony, Teaching-Learning-Based Optimization (TLBO) imitates classroom teaching, Social Group Optimization (SGO) and Group Teaching Optimization Algorithm (GTOA) simulate knowledge dissemination, Brain Storm Optimization (BSO) draws on creative thinking, and Imperial Competitive Algorithm (ICA) is based on the mechanisms of imperial expansion and competition.

[0005] Despite the continuous emergence of new algorithms, according to the No Free Lunch (NFL) theorem, there is no universal optimization algorithm. With the rapid development of swarm intelligence algorithms, the Crayfish Optimization Algorithm (COA), as a new type of nature-inspired algorithm, has been widely used in optimization problems due to its simple and efficient characteristics. However, although the existing Crayfish Optimization Algorithm has good global search performance and a fast convergence speed, it also has the following several disadvantages: (1) Prone to local optimum: In the later stage of the algorithm, due to weak exploration ability, COA may stay at the local optimum solution and it is difficult to further find the global optimum; (2) Imbalance between exploration and exploitation: COA has strong exploitation ability in the early stage, enabling the algorithm to converge quickly, but the exploration ability in the later stage is insufficient, resulting in a limited search range and difficulty in jumping out of the local optimum; (3) Unstable convergence speed: From the convergence curves of benchmark test functions, it can be seen that COA can jump out of the local optimum and converge quickly in some cases, but it may still show slow convergence or stagnation in some problems; (4) High parameter sensitivity: The performance of COA depends on the selection of appropriate parameters. For example, the setting of the temperature parameter will affect the search ability of the algorithm. If the setting is unreasonable, it may lead to unsatisfactory optimization effects. Therefore, how to provide an evaluation method for the inter-turn insulation aging degree of dry-type transformers based on an improved Crayfish Optimization Algorithm is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides an evaluation method for the inter-turn insulation aging degree of dry-type transformers based on an improved Crayfish Optimization Algorithm, an error feedback type BP neural network (SAV-COA-BP) based on the improved Crayfish Optimization Algorithm (SAV-COA), and evaluates the inter-turn insulation aging degree of Nomex dry-type transformers, so as to solve the problems in the prior art.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] An evaluation method for the inter-turn insulation aging degree of dry-type transformers based on an improved Crayfish Optimization Algorithm, comprising the following steps:

[0009] S1. Construct an aging degree prediction model based on a BP neural network;

[0010] S2. Determine the topological structure of the aging degree prediction model;

[0011] S3. Define and initialize the population parameters of the improved Crayfish Optimization Algorithm;

[0012] S4. Define the environmental temperature of the improved Crayfish Optimization Algorithm;

[0013] S5. When the environmental temperature > 30 and the random number < 0.5, the algorithm enters the non - competitive summer - avoiding stage, updating the individual positions; when the environmental temperature > 30 and the random number ≥ 0.5, the algorithm enters the competitive summer - avoiding stage, updating the individual positions;

[0014] S6. When the environmental temperature ≤ 30, the algorithm enters the foraging stage, calculating the food intake and food size, and updating the individual positions;

[0015] S7. Calculate the fitness of all individuals, judge whether the iteration stop condition is met. If it is met, enter S8; if not, return to S4 for the next iteration;

[0016] S8. Record the optimal individual after the iteration ends, introduce the vertical crossover operation, and update the optimal individual;

[0017] S9. Assign the weights and thresholds of the optimal individual to the aging degree prediction model, and use the local search ability of the BP neural network to fine - tune the weights and thresholds;

[0018] S10. Use the trained aging degree prediction model to predict the aging degree of the inter - turn insulation of the dry - type transformer for the input data.

[0019] Optionally, S2 is specifically: Set the number of nodes in the input layer n, the number of nodes in the hidden layer l, the number of nodes in the output layer m, select the training function and training parameters, and initialize the structure of the BP neural network.

[0020] Optionally, S3 is specifically: Define the number of iterations T, the population size N, the upper and lower bounds of the search space ub, lb. Each individual is represented as a vector with dimension dim. Initialize the crayfish group X according to the upper and lower bounds, so that each individual corresponds to a feasible solution:

[0021]

[0022] Among them, the position of each individual is:

[0023] X i,j = lb j +(ub j - lb j )×rand

[0024] In the formula, X i,j represents the position of the individual (i, j), and rand represents a random number between 0 and 1; introduce mirror - reflection learning to generate the opposite solution in the initial stage.

[0025] Optionally, introducing mirror - reflection learning to generate the opposite solution in the initial stage is specifically:

[0026] The boundary control is as follows:

[0027]

[0028] The mutation operator based on mirror reflection learning updates the understanding space:

[0029]

[0030] Calculate the minimum value L in the search space min,j and the maximum value U max,j , λ is the neighborhood elasticity factor, ranging from [0, 1], and mirror reflection learning is used to generate new candidate solutions This gives the search individual a greater chance of jumping out of the local optimum, calculating the new fitness value, and updating it if it is better than the current individual fitness value.

[0031] Optionally, S4 is:

[0032] temp=rand×15+20

[0033] Where temp is the ambient temperature, and rand is a random number between 0 and 1.

[0034] Optionally, during the non-competitive summer stage, introduce the Eagle algorithm to expand the exploration:

[0035]

[0036]

[0037] Where X1(t+1) is the solution of the next iteration of t, t is the current iteration number, T is the maximum iteration number, X best (t) is the best solution before the nth iteration, X M is the average position of the current solution connected at the nth iteration, rand represents a random number between 0 and 1, N is the population size, and dim is the problem dimension size.

[0038] Optionally, the uncontested summer stage in S5 is as follows:

[0039] Calculate cave locations:

[0040]

[0041] Update the location of individual crayfish:

[0042]

[0043] C2=2-(t / T)

[0044] Where, X shade is the cave location, X G represents the optimal position obtained through the number of iterations so far, XL represents the optimal position of the current population, Indicates the updated position. Indicates the position before the update, rand represents a random number between 0 and 1, C2 is a decreasing curve, t is the current number of iterations, and T is the maximum number of iterations.

[0045] Optionally, the competition summer vacation phase in S5 is as follows:

[0046] Calculate cave locations:

[0047]

[0048] Crayfish compete for burrows using the formula:

[0049]

[0050] z=round(rand×(N-1))+1

[0051] Where, X shade is the cave location, X G represents the optimal position obtained through the number of iterations so far, X L represents the optimal position of the current population, Indicates the updated position. Represents the position before the update, z represents a random crayfish individual, rand represents a random number between 0 and 1, N is the population size, and round() represents rounding.

[0052] Optionally, S6 is:

[0053] Calculate food intake, portion size, and food location:

[0054]

[0055] X food =X G

[0056] Where p is food intake, Q is food size, X food is the food location, X G represents the optimal position obtained by the number of iterations so far, temp represents the ambient temperature, μ is the temperature most suitable for crayfish, C1 and σ are parameters that control the amount of crayfish eaten at different temperatures, rand represents a random number between 0 and 1, C3 is the food factor, fitness i is the fitness of the i-th crayfish, fitness food is the fitness of the food location;

[0057] When Q > (C3 + 1) / 2, update the food position and the position of the crayfish individuals:

[0058]

[0059] When Q ≤ (C3 + 1) / 2, directly update the position of the crayfish individuals:

[0060]

[0061] In the formula, represents the updated position, represents the position before update.

[0062] Optionally, the vertical crossover operation is specifically as follows:

[0063]

[0064] In the formula, is the offspring generated after the vertical crossover of the parent X(i) in dimensions d1 and d2. X(i, d1) and X(i, d2) respectively represent the parent X(i) in dimensions d1 and d2. q is a random number between 0 and 1. The offspring individuals generated by the vertical crossover compete with the parent, and the individual with higher fitness is retained.

[0065] Through the above technical solutions, compared with the prior art, the present invention provides a method for evaluating the inter-turn insulation aging degree of dry-type transformers based on an improved crayfish optimization algorithm, which has the following beneficial effects:

[0066] 1. By introducing the mirror reflection learning mechanism, the present invention introduces the "opposite number" of each individual in the population into the search process, effectively expanding the search space, improving the diversity of the population. The new candidate solutions generated by the reflection mechanism are often located at the edge or far area of the search region, thus providing the possibility for the algorithm to jump out of the local optimum, and further enhancing the global search ability;

[0067] 2. By introducing the Tianying algorithm, the present invention can quickly identify the better region at the initial stage of the algorithm, and quickly approach the global optimum solution through competition and foraging strategies. In this stage, the temperature control parameter is used to make the algorithm have strong development ability in the early stage, thus greatly improving the search efficiency;

[0068] 3. The vertical crossover operation of the present invention randomly crosses between different dimensions, combines and recombines the excellent parts of each individual, thus enhancing the local search ability, effectively preventing the population from premature convergence, ensuring that the individuals can jump out of the local optimum trap in the later stage of optimization, and further improving the accuracy and robustness of the overall algorithm;

[0069] 4. The present invention combines multiple improvement strategies such as mirror reflection learning, exploration of the extended Tianying algorithm, and vertical crossover operation, showing stronger robustness when dealing with high-dimensional, multi-extremum, and complex non-linear problems. The optimization result is no longer overly dependent on the initial conditions, and the algorithm can maintain a high search efficiency and accuracy in different types of problems;

[0070] 5. The SAV-COA-BP algorithm proposed by the present invention solves the core pain points in the inter-turn insulation aging assessment of dry-type transformers through five major advantages: global optimization, high-efficiency convergence, high-precision feature extraction, anti-noise design, and engineering practicability, providing highly reliable intelligent decision-making support for the condition-based maintenance of power equipment. Its comprehensive performance is significantly superior to traditional methods and has broad industrial application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to the provided drawings.

[0072] Figure 1 It is a flowchart of the method for evaluating the aging degree of the inter-turn insulation of the dry-type transformer of the present invention;

[0073] Figure 2 It is a schematic diagram of the specular reflection and mirror reflection learning model of the present invention;

[0074] Figure 3 It is a schematic diagram of the influence of temperature on the food intake of crayfish in the embodiment of the present invention;

[0075] Figure 4 It is a schematic diagram of the Tianying algorithm in the embodiment of the present invention;

[0076] Figure 5 It is a schematic diagram of the non-competitive summering stage and the competitive summering stage in the embodiment of the present invention;

[0077] Figure 6 It is a schematic diagram of the foraging stage when the food is too large in the embodiment of the present invention;

[0078] Figure 7 It is a schematic diagram of the principle of vertical crossover operation in the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0079] 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 ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0080] The embodiment of the present invention discloses a method for evaluating the aging degree of inter-turn insulation of dry-type transformers based on an improved crayfish optimization algorithm. Figure 1 As shown, the following steps are included:

[0081] S1. Construct an aging degree prediction model based on BP neural network;

[0082] S2. Determine the topology of the aging degree prediction model;

[0083] S3. Define and initialize the population parameters of the improved crayfish optimization algorithm;

[0084] S4. Define the ambient temperature of the improved crayfish optimization algorithm;

[0085] S5. When the ambient temperature is greater than 30 and the random number is less than 0.5, the algorithm enters the non-competitive heat-avoidance stage and updates the individual position; when the ambient temperature is greater than 30 and the random number is greater than or equal to 0.5, the algorithm enters the competitive heat-avoidance stage and updates the individual position;

[0086] S6. When the ambient temperature is ≤30, the algorithm enters the foraging phase, calculates food intake and food size, and updates the individual position;

[0087] S7, calculate the fitness of all individuals and determine whether the iteration stop condition is met. If so, proceed to S8. If not, return to S4 for the next iteration.

[0088] S8, record the best individual after the iteration, introduce vertical crossover operation, and update the best individual;

[0089] S9, assigning the weight and threshold of the optimal individual to the aging degree prediction model, and using the local search capability of the BP neural network to fine-tune the weight and threshold;

[0090] S10. Use the trained aging degree prediction model to predict the aging degree of the dry-type transformer inter-turn insulation based on the input data.

[0091] The design inspiration of the COA algorithm comes from the behaviors of freshwater crayfish in foraging, summer heat avoidance, and competition. The execution process of this algorithm is divided into three stages: heat avoidance, cave competition, and foraging stage. The heat avoidance stage corresponds to the exploration process of the algorithm, and the cave competition and foraging stages correspond to the exploitation process of the algorithm. The behavior of crayfish is affected by the change of environmental temperature, which guides it to enter different stages. When the environmental temperature is too high, the COA algorithm enters the competition stage or the heat avoidance stage. In the competition stage, other crayfish are also interested in the same cave, and two crayfish will compete for the cave. In the summer heat avoidance stage, there is no other crayfish competing for the cave, so the crayfish will directly enter the cave. When the temperature is suitable for foraging, the COA algorithm enters the foraging stage. At this time, the crayfish will look for food to eat. When the food is large, the crayfish will use its chelipeds to tear the food and eat it. When the food is of medium size, the crayfish will alternately use its second and third walking legs to eat. The feeding behavior of crayfish is also affected by the food intake, which controls the feeding habits of crayfish.

[0092] Further, S2 is specifically as follows: Set the number of input layer nodes n, the number of hidden layer nodes l, the number of output layer nodes m, select the training function and training parameters, and initialize the BP neural network structure.

[0093] Further, S3 is specifically as follows: Define the number of iterations T, the population size N, the upper and lower bounds of the search space ub and lb. Each individual is represented as a vector with dimension dim. Initialize the crayfish population X according to the upper and lower bounds, so that each individual corresponds to a feasible solution:

[0094]

[0095] The position of each individual is:

[0096] X i,j = lb j +(ub j - lb j ) × rand

[0097] In the formula, X i,j represents the position of individual (i, j), and rand represents a random number between 0 and 1; Introduce mirror reflection learning to generate the opposite solution in the initial stage;

[0098] Each crayfish is a 1×dim matrix, and each column matrix represents a solution to a problem. In a set of variables, each variable must be between the upper and lower bounds.

[0099] Further, introducing mirror reflection learning to generate the opposite solution in the initial stage is specifically as follows:

[0100] The boundary control is as follows:

[0101]

[0102] The mutation operator based on mirror reflection learning updates the solution space:

[0103]

[0104] Calculate the minimum value L of the search space min,j and the maximum value U max,j , where λ is the neighborhood elasticity factor with a value range of [0, 1], and new candidate solutions are generated using mirror reflection learning to enable the search individuals to have a greater chance of jumping out of the local optimum. Calculate the new fitness value. If it is better than the fitness value of the current individual, then update. The initial population is adjusted through the mirror reflection learning mechanism to generate opposite solutions to enhance the population diversity, avoid premature convergence, and increase the exploration ability of the search space in the early stage. Corresponding to the initialization population stage, this process adjusts the initial positions and movement modes of the crayfish through the reflection mechanism.

[0105] In the embodiments of the present invention, as Figure 2 shown, the inspiration for mirror reflection learning comes from the mirror reflection of light, that is, light is reflected on an object with a shiny surface; A and B are the incident light and the reflected light respectively. The angle α between A and the normal is called the incident angle, and the angle β between B and the normal is called the reflection angle. According to the law of reflection, α = β;

[0106] Opposite number: Assume that the value range of the real number x is from V L to V U , assume that x ∈ [V L , V U , the opposite number of x can be defined as:

[0107]

[0108] In Figure 2 , O = (x0, 0) is the midpoint of [V L , V U , x = (a, 0) is a random variable on [V L , V U , therefore, the opposite position of x can be obtained through the definition of the opposite number. According to the mirror reflection learning mechanism, assume that the obtained opposite position is (b, 0). According to Figure 2 the following formula can be obtained:

[0109]

[0110] Because the angles between the incident light and the reflected light and the normal are the same, it can be obtained that:

[0111]

[0112] Let \(B0 = \lambda A0\), where \(\lambda\) is a real number greater than 0, we can get:

[0113] \(b=\lambda(x0 - a)+x0\)

[0114] \(=(\lambda + 1)x0-\lambda a\)

[0115] \(=(0.5\lambda + 0.5)\times2x0-\lambda a\)

[0116] \(=(0.5\lambda + 0.5)\times(V\) L +\(V\) U )-\lambda a\)

[0117] From the above formula, we know that when \(\lambda\) changes, \(b\) has different values:

[0118]

[0119] \(b1\) and \(b2\) respectively satisfy When \(\lambda\) is set to 1, it means that mirror reflection learning is learning based on opposition. That is to say, opposition learning is a special case of mirror reflection learning. In the idea of opposition learning, the possible positions of the opposite position \(x\) obtained by mirror reflection learning should be a small neighborhood of. Therefore, in mirror reflection learning, the value of \(\lambda\) is calculated by the following formula:

[0120]

[0121] where \(k1\), \(k2\) are two random numbers uniformly distributed between \([0,1]\), \(R0\) is the neighborhood radius, is the elastic factor. If \(x1>x2\) is satisfied, \(x\) will appear in the right neighborhood of; otherwise, it will appear in the left neighborhood of. Here, if \(R0 = 1\), the maximum value can be obtained, that is, \(x = 2 - x0\), \(x = 2 - x0\) and \(x0\) are symmetric about . From the perspective of symmetry, when \(0>1\), the maximum value obtained is greater than \(x = 2 - x0\). Therefore, in order to balance the search space of the left and right neighborhoods, it is very reasonable to set \(R0\) to \([0,1]\). In addition, is also set to \([0,1]\), that is, all possible values within the neighborhood radius \(R0\) are obtained.

[0122] According to the established mirror reflection learning model, the pair number can be rewritten as:

[0123]

[0124] Furthermore, S4 is specifically:

[0125] temp = rand×15 + 20

[0126] Where temp is the ambient temperature, and rand represents a random number between 0 and 1.

[0127] In the embodiments of the present invention, the change in temperature will cause changes in the behavior of crayfish. When the temperature is higher than 30°C, crayfish will choose a cool place to avoid the heat. At a suitable temperature, crayfish will carry out foraging behavior. The intake of crayfish is affected by temperature. The feeding range of crayfish is between 15°C and 30°C, and 25°C is the best. Therefore, the food intake of crayfish can be approximated as a normal distribution, so that the food intake is affected by temperature. Because crayfish have strong foraging behavior between 20 and 30°C, therefore, COA defines a temperature range of 20 to 35°C. The relationship between the food intake of crayfish and temperature is as Figure 3 shown.

[0128] Furthermore, in the non-competitive heat avoidance stage, the Tianying algorithm is introduced to expand exploration:

[0129]

[0130] Where X1(t + 1) is the solution of the next iteration of t, t is the current iteration number, T is the maximum iteration number, X best (t) is the best solution before the nth iteration, X M is the position average of the current solution connected at the nth iteration, rand represents a random number between 0 and 1, N is the population size, and dim is the problem dimension size.

[0131] In the Tianying algorithm, the next iteration solution X1(t + 1) calculated through the expanded exploration stage serves as an important reference and guiding factor in the current iteration. After being calculated by the formula, X1(t + 1) will be compared with the fitness of each solution in the existing population, so as to guide the update of subsequent individuals. X1(t + 1) can help the population identify more potential areas in the search space and balance the ability of global exploration and local development; during the population update process, if X1(t + 1) shows better fitness than some existing individuals, it will prompt these individuals to move closer to this solution, thus guiding the entire population to quickly gather towards a better solution. This method of guiding the update by using the next candidate solution obtained in the current iteration enables the algorithm to enhance the efficiency of local mining and fine optimization while retaining the global search ability.

[0132] In the embodiments of the present invention, as Figure 4As shown in the figure, the Sky Eagle algorithm simulates the behavior of the Sky Eagle during hunting, showing every step of the hunting action. Therefore, the optimization process of the proposed AO algorithm is divided into four methods: in the divergent search space, it uses contour flight and short descent attack to search; in the convergent search space, it uses low flight and slow descent attack to search; the algorithm models the Sky Eagle hunting behavior as a mathematical optimization example, which determines the best solution under specific constraints;

[0133] This embodiment of the present invention uses the first phase of the Skyhawk algorithm, the extended exploration phase, to optimize the crayfish algorithm. During the extended exploration phase, the Skyhawk identifies prey areas and selects the optimal hunting area by soaring vertically at high altitudes. During this phase, the Skyhawk determines the size of the search space from high altitudes and explores the location of prey. Figure 4 The high-altitude soaring behavior of the eagle under vertical bending is shown. During the COA process, when the temperature is above 30℃, the crayfish enters the summer vacation stage. Here, the first stage of the eagle optimization algorithm (AO) is introduced to update the position by randomly selecting different behavioral strategies (foraging or competing).

[0134] Furthermore, the uncontested summer stage in S5 is as follows:

[0135] When the temperature is greater than 30°C, Figure 5 As shown, the temperature is too high. At this time, crayfish will choose to enter the cave to avoid the heat. Calculate the location of the cave:

[0136]

[0137] When rand>0.5, it means that there are no other crayfish competing for the cave, and the crayfish will directly enter the cave to avoid the heat. The position of the individual crayfish is updated:

[0138]

[0139] C2=2-(t / T)

[0140] Where, X shade is the cave location, X G represents the optimal position obtained through the number of iterations so far, X L represents the optimal position of the current population, Indicates the updated position. represents the position before the update, rand represents a random number between 0 and 1, C2 is a decreasing curve, t is the current iteration number, and T is the maximum iteration number. During the summer retreat phase, the crayfish's goal is to approach the cave, which represents the optimal solution. At this time, the crayfish will approach the cave, which brings the individual closer to the optimal solution and enhances the development ability of COA, making the algorithm converge faster.

[0141] Furthermore, the competition summer stage in S5 is specifically as follows:

[0142] Calculate cave locations:

[0143]

[0144] When rand>0.5, it means that other crayfish are also interested in the cave. At this time, they will compete for the cave. The crayfish compete for the cave according to the formula:

[0145]

[0146] z=round(rand×(N-1))+1

[0147] Where, X shade is the cave location, X G represents the optimal position obtained through the number of iterations so far, X L represents the optimal position of the current population, Indicates the updated position. represents the position before the update, z represents a random crayfish, rand represents a random number between 0 and 1, N represents the population size, and round() represents rounding to the nearest integer. During the competition phase, crayfish compete with each other and adjust their positions based on the positions of other crayfish. This adjustment expands the COA's search range and enhances the algorithm's exploration capabilities.

[0148] Furthermore, S6 is specifically:

[0149] When the temperature is ≤30℃, it is suitable for crayfish to forage. Calculate food intake, food size and food location:

[0150]

[0151] X food =X G

[0152] Where p is food intake, Q is food size, X food is the food location, X G represents the optimal position obtained by the number of iterations so far, temp represents the ambient temperature, μ is the temperature most suitable for crayfish, C1 and σ are parameters that control the amount of crayfish eaten at different temperatures, rand represents a random number between 0 and 1, C3 is the food factor, fitness i is the fitness of the i-th crayfish, fitness food is the fitness of the food location;

[0153] When Q>(C3+1) / 2, it means the food is too big, such asFigure 6 As shown, the crayfish will use its claws to tear the food and alternately eat with its second and third walking feet, updating the food position and the crayfish's individual position:

[0154]

[0155] In the embodiment of the present invention, in order to simulate the alternating process, a combination of sine function and cosine function is used to simulate the alternating process, and the food obtained by the crayfish is also related to the food intake;

[0156] When Q ≤ (C3 + 1) / 2, directly update the crayfish's individual position:

[0157]

[0158] In the formula, represents the updated position, represents the position before update.

[0159] In the foraging stage, the crayfish uses different feeding methods according to the size of its food. The food X food represents the optimal solution. When the size of the food is suitable for the crayfish to eat, the crayfish will approach the food. When Q is too large, it indicates that there is a significant difference between the crayfish and the optimal solution. Therefore, X food should be reduced to be closer to the food. And the randomness of the crayfish food intake enhancement algorithm is controlled. Through the foraging stage, the algorithm will gradually approach the optimal solution, enhancing the developability of the algorithm and making it have good convergence ability.

[0160] Furthermore, the vertical crossover operation is specifically as follows:

[0161]

[0162] In the formula, is the offspring generated after the vertical crossover of the parent X(i) in the d1 dimension and the d2 dimension. X(i, d1) and X(i, d2) respectively represent the parent X(i) in the d1 dimension and the d2 dimension. q is a random number between 0 and 1. The offspring individuals generated by the vertical crossover compete with the parent, and the individual with higher fitness is retained.

[0163] The vertical crossover is an arithmetic crossover that operates on all individuals between two different dimensions. During the entire iteration process, the parent population searched by the vertical crossover comes from the population of the dominant solutions of the horizontal crossover, which can prevent the population from falling into local optima. The principle of the vertical crossover operation is as Figure 7As shown, where r and 1 - r are the proportionality coefficients of the selected parental individuals. To avoid the loss and damage of the elite individuals in the previous generation caused by the crossover operation, after the pairwise crossover of individuals, only one offspring is generated for each vertical crossover operation, and the elite retention strategy is adopted to store the elite individuals in this iteration. Elite individuals are particles with excellent fitness values during the operation of the algorithm. By storing these particles, the convergence speed of the algorithm can be effectively improved, providing an opportunity to jump out of the local optimum for the stagnant dimension without destroying another dimension that may be the global optimum. This crossover operation randomly selects two dimensions for crossover, aiming to enhance the local search ability. Corresponding to the foraging stage, this part of the code adjusts the position through a mechanism similar to the genetic algorithm, enabling the crayfish to better adapt to the search environment.

[0164] In an embodiment of the present invention, a BP neural network based on an improved crayfish optimization algorithm is used to evaluate the inter-turn insulation aging degree of a dry-type transformer. Specifically:

[0165] Partial discharge tests are respectively carried out on Nomex insulating papers with different aging degrees, and the aging degree is divided into 6 levels, respectively representing different aging degrees. Since the discharge waveforms corresponding to the same pattern are similar, for the same discharge pattern, the statistical characteristic quantities of the partial discharge signal are similar. The statistical characteristic quantities describe the distribution characteristics of partial discharge from the perspective of probability, and the characteristics of the probability distribution function can be quantitatively described by the central moments of each order, including the first moment (mean μ), the second moment (standard deviation σ), the third moment (skewness Sk), the fourth moment (kurtosis Ku), the cross-correlation coefficient (cc), and the quality factor (Q f ) etc. In addition, the characteristics of the partial discharge signal can also be characterized by the cross-correlation coefficient and the phase asymmetry degree. The above statistical characteristic quantities are obtained by dividing the partial discharge signal into positive and negative half-cycles under the power frequency cycle. Since the partial discharge patterns in the embodiments of the present invention are relatively simple, only the following eleven statistical characteristic parameters are extracted from the partial discharge of the specimens in the four aging stages during electrical aging as the parameters for reliability evaluation: namely, the positive half-wave skewness (Sk+), the positive half-wave kurtosis (Ku+), the negative half-wave skewness (Sk-), the negative half-wave kurtosis (Ku-), the cross-correlation coefficient (cc), the discharge quantity factor (QF), the cross-correlation coefficient (CC), the total discharge quantity (ΣQ), the maximum discharge quantity (Qmax), the average discharge quantity (Qavg), the total discharge times (N), and the inception discharge voltage (PDIV).

[0166] Taking the aging data as the data samples of the regression model, the partial discharge statistical characteristic quantities collected at different electrical aging stages are obtained according to the calculation method of partial discharge statistical characteristic quantities, which can be used as the sample data corresponding to the six modes. Some of the training sample data of partial discharge statistical characteristic quantities are shown in Table 1. In the table, S represents the insulation aging state of the specimen, which corresponds one-to-one with the partial discharge modes measured during six aging stages.

[0167] Table 1 Training Sample Data of Partial Discharge Statistical Characteristic Quantities

[0168]

[0169]

[0170] The comparison of the evaluation results obtained by the SAV-COA-BP algorithm in the embodiments of the present invention with other algorithms is shown in Table 2:

[0171] Table 2 Comparison of Evaluation Results

[0172] Algorithm Convergence iteration times Root mean square error Mean absolute percentage error Traditional BP 1250 0.19727 5.52161％ COA - BP 680 0.17859 4.21219％ SAV - COA - BP 325 0.02643 1.01780％

[0173] It can be seen that the number of iterative calculations, root mean square error (RMSE), and mean absolute percentage error (MAPE) of the SAV-COA-BP algorithm are better than those of the traditional BP neural network and the BP neural network optimized by COA.

[0174] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other.

[0175] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm, characterized in that The following steps are involved: S1. Construct an aging degree prediction model based on BP neural network; S2. Determine the topology of the aging degree prediction model; S3. Define and initialize the population parameters of the improved crayfish optimization algorithm; S4. Define the ambient temperature of the improved crayfish optimization algorithm; S5. When the ambient temperature is greater than 30 and the random number is less than 0.5, the algorithm enters the non-competitive heat-avoidance stage and updates the individual position; when the ambient temperature is greater than 30 and the random number is greater than or equal to 0.5, the algorithm enters the competitive heat-avoidance stage and updates the individual position; S6. When the ambient temperature is ≤30, the algorithm enters the foraging phase, calculates food intake and food size, and updates the individual position; S7, calculate the fitness of all individuals and determine whether the iteration stop condition is met. If so, proceed to S8. If not, return to S4 for the next iteration. S8, record the best individual after the iteration, introduce vertical crossover operation, and update the best individual; S9, assigning the weight and threshold of the optimal individual to the aging degree prediction model, and using the local search capability of the BP neural network to fine-tune the weight and threshold; S10. Use the trained aging degree prediction model to predict the aging degree of the dry-type transformer inter-turn insulation based on the input data.

2. The method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, characterized in that S2 specifically includes: setting the number of input layer nodes n, the number of hidden layer nodes l, the number of output layer nodes m, selecting the training function and training parameters, and initializing the BP neural network structure.

3. A method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, characterized in that, S3 specifically defines the number of iterations T, the population size N, the upper and lower bounds ub and lb of the search space, each individual is represented as a vector of dimension dim, and the crayfish population X is initialized according to the upper and lower bounds so that each individual corresponds to a feasible solution: The position of each individual is: X i,j = lb j + (ub j - lb j ) × rand where X i,j represents the position of the individual (i,j), and rand represents a random number between 0 and 1; mirror reflection learning is introduced to generate the opposite solution in the initial stage.

4. A method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 3, characterized in that The introduction of mirror reflection learning to generate opposing solutions in the initial stage is as follows: Border controls are as follows: The mutation operator based on mirror reflection learning updates the understanding space: Calculate the minimum value L of the search space min,j and the maximum value U max,j , where λ is the neighborhood elasticity factor with a value range of [0, 1], and a new candidate solution is generated using mirror reflection learning to enable the search individual to have a greater chance of jumping out of the local optimum, calculate the new fitness value, and update it if it is better than the fitness value of the current individual.

5. A method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, characterized in that, S4 is specifically: temp=rand×15+20 Where temp is the ambient temperature, and rand is a random number between 0 and 1.

6. The method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, characterized in that In the non-competitive summer stage, the Sky Eagle algorithm is introduced for extended exploration: Where, X1(t + 1) is the solution of the next iteration of t, t is the current iteration number, T is the maximum iteration number, X best (t) is the best solution before the nth iteration, X M is the position average of the current solution connected at the nth iteration, rand represents a random number between 0 and 1, N is the population size, and dim is the problem dimension size.

7. A method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, characterized in that The specific contention-free summer stage in S5 is: Calculate cave locations: Update the location of individual crayfish: C2=2-(t / T) Wherein, X shade is the cave location, X G represents the optimal location obtained so far through the number of iterations, X L represents the optimal location of the current population, represents the updated location, represents the location before update, rand represents a random number between 0 and 1, C2 is a decreasing curve, t is the current iteration number, and T is the maximum iteration number.

8. A method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, characterized in that, The specific competition summer vacation stage in S5 is: Calculate cave locations: Crayfish compete for burrows using the formula: z=round(rand×(N-1))+1 Wherein, X shade is the cave location, X G represents the optimal location obtained so far through the number of iterations, X L represents the optimal location of the current population, represents the updated location, represents the location before update, z represents a random crayfish individual, rand represents a random number between 0 and 1, N is the population size, and round() represents rounding to the nearest integer.

9. The method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, wherein S6 specifically: Calculate food intake, portion size, and food location: X food = X G Where p is the food intake, Q is the food size, and X food is the food position, and X G represents the optimal position obtained so far through the number of iterations, temp represents the environmental temperature, μ is the temperature most suitable for crayfish, C1 and σ are parameters for controlling the food intake of crayfish at different temperatures, rand represents a random number between 0 and 1, C3 is the food factor, and fitness i is the fitness of the i-th crayfish, and fitness food is the fitness of the food position; When Q>(C3+1) / 2, update the food position and the individual crayfish position: When Q≤(C3+1) / 2, directly update the individual position of the crayfish: In the formula, represents the updated position, represents the position before update.

10. A method for evaluating the aging degree of the inter-turn insulation of a dry-type transformer based on an improved crayfish optimization algorithm according to claim 1, characterized in that, The vertical cross operation is as follows: In the formula, is the offspring generated after the vertical crossover of the parent X(i) in the d1 dimension and the d2 dimension. X(i, d1) and X(i, d2) respectively represent the parent X(i) in the d1 dimension and the d2 dimension. q is a random number between 0 and 1. The offspring individuals generated by the vertical crossover compete with the parent, and the individual with higher fitness is retained.

Images