GIS isolation switch fault diagnosis method

Abstract

The invention discloses a GIS isolation switch fault diagnosis method. According to the method, a voltage signal and a current signal of the GIS isolation switch are obtained and then noise reduction processing is carried out by using a wavelet filtering algorithm; modal function components of the current signals in different states are extracted through EMD to serve as frequency domain features, and the frequency domain features and time domain features are fused through a PCA algorithm to obtain a fused feature set; dividing the fusion feature set into a training sample set and a test sample set; optimizing kernel parameters and penalty factors of the support vector machine model by improving an original whale optimization algorithm, and training and testing the support vector machine model according to the training sample set and the test sample set to obtain a GIS isolation switch fault diagnosis model; inputting the fusion feature set of the GIS disconnecting switch into a GIS disconnecting switch fault diagnosis model, and identifying a GIS disconnecting switch fault; the high-efficiency fault diagnosis of the high-voltage isolation switch is realized by adopting the finite characteristic quantity when the fault samples are few.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for power grid equipment, and in particular to a fault diagnosis method for GIS disconnect switches. Background Technology

[0002] Intelligent operation and maintenance of power grid equipment is a crucial component of power grid intelligence. GIS disconnect switches, as the most widely used type of switchgear, operate under high current and high voltage conditions for extended periods and are susceptible to various mechanical failures due to outdoor environments and climate. Surveys indicate that mechanical failures caused by incomplete opening and closing, reduced strength of the transmission mechanism, failure of the stationary contact spring clip, and jamming of connecting rods, shafts, and contacts account for a significant proportion. Therefore, it is necessary to conduct research on real-time online status monitoring of GIS disconnect switches, implement intelligent early warning systems for mechanical failures, promptly detect and diagnose faults, and prevent further escalation of accidents and substantial losses.

[0003] On the one hand, traditional GIS disconnector fault diagnosis methods are inadequate and have low accuracy, easily leading to false alarms or even misdiagnosis. On the other hand, using artificial intelligence algorithms to diagnose artificially simulated mechanical faults in high-voltage disconnectors can result in unsimulated mechanical faults appearing in actual applications, leading to misdiagnosis or missed diagnosis of these unsimulated faults. Current research on the mechanical condition detection of GIS disconnectors largely relies on the disconnector's motor current signal, operating torque, angle-time data obtained from attitude sensors, and vibration signals during the opening and closing processes for condition detection and fault identification. Voltage and current signals contain rich mechanical condition information of GIS disconnectors, making the search for suitable fault feature extraction methods crucial for GIS disconnector fault diagnosis. Commonly used signal analysis methods include Fourier transform, wavelet transform, and empirical mode decomposition. However, because the current signal of a GIS disconnector is non-stationary, these methods struggle to obtain its localized feature information. The Whale Optimization Algorithm (WOA) is a novel heuristic optimization algorithm inspired by three distinct foraging behaviors of humpback whales: random wandering, encircling food, and using a spiral bubble net to capture it. The Whale Optimization Algorithm's optimization accuracy remains largely consistent with the increasing dimensionality of the problem, and it involves minimal parameter setting issues. Its superior mechanism design, novel approach, and ease of operation make it a promising tool for scientific research and development.

[0004] Traditional feature extraction methods, based solely on the time or frequency domain, struggle to effectively extract all signal information. Wavelet decomposition, on the other hand, is difficult to define a definite dimension for signals under different conditions, hindering the implementation of fault diagnosis algorithms in GIS disconnector online monitoring systems. Furthermore, due to inherent limitations of the whale optimization algorithm, it often suffers from excessively long optimization times, susceptibility to local optima, and poor robustness during iteration. Summary of the Invention

[0005] The purpose of this invention is to provide a fault diagnosis method for GIS disconnect switches, which solves the problems that traditional feature extraction methods are difficult to effectively extract all the information of the signal, wavelet decomposition is not conducive to the implementation of fault diagnosis algorithms in GIS disconnect switch online monitoring systems, and due to the limitations of the whale optimization algorithm itself, it often suffers from problems such as excessively long optimization time, easy getting trapped in local optima, and poor robustness during the iteration process.

[0006] This invention provides a method for diagnosing faults in GIS disconnect switches, the method comprising:

[0007] Obtain the voltage and current signals of the GIS disconnector;

[0008] The voltage and current signals are denoised using a wavelet filtering algorithm.

[0009] The modal function components of the current signal under different states are extracted by EMD as frequency domain features. Among them, the energy entropy, approximate entropy, sample entropy, fuzzy entropy and permutation entropy of the first 5 components are calculated as 5 types of information entropy as feature entropy, forming a feature vector; the current signal and voltage signal are extracted as time domain features.

[0010] The frequency domain features and time domain features are fused using the PCA algorithm to obtain a fused feature set;

[0011] The fused feature set is divided into a training sample set and a test sample set;

[0012] The kernel parameters and penalty factor of the support vector machine model were optimized by improving the original whale optimization algorithm, resulting in an optimized support vector machine model.

[0013] The support vector machine model is trained and tested based on the training sample set and the test sample set to obtain the GIS disconnect switch fault diagnosis model;

[0014] The fused feature set of the GIS disconnect switch is input into the GIS disconnect switch fault diagnosis model to identify GIS disconnect switch faults.

[0015] Furthermore, the kernel parameters and penalty factor of the support vector machine model are optimized by improving the whale optimization algorithm, resulting in an optimized support vector machine model, including:

[0016] An enhanced global search mechanism, a quasi-adversarial learning mechanism, a bounce boundary handling mechanism, and a mirror evolution mechanism are introduced into the original whale optimization algorithm to search for the optimal solution in a multi-dimensional search space to obtain the optimal fitness. The original whale optimization algorithm includes three stages: random search foraging behavior, shrinking and surrounding prey behavior, and spiral bubble net predation behavior.

[0017] The penalty factor and kernel parameters corresponding to the optimal fitness are used as the optimal parameters of the support vector machine model.

[0018] Furthermore, the enhanced global search mechanism includes:

[0019] The judgment conditions for random search foraging behavior and spiral bubble net predation behavior in the original whale optimization algorithm have been adjusted:

[0020] The spiral bubble net predation behavior is executed when the random number p < 0.5 and |A| < 1;

[0021] When the random number p < 0.5 and |A| ≥ 1, the behavior of shrinking and surrounding the prey is executed;

[0022] When the random number p ≥ 0.5, random search foraging behavior is executed;

[0023] Where A is the position update coefficient of the individual whale during the search process of the original whale optimization algorithm, which is used to control the movement step size and direction of the individual whale in the search space; the random number p is used to determine the behavior pattern of the individual whale during the search process.

[0024] Furthermore, the quasi-adversarial learning mechanism includes:

[0025] When performing random foraging behavior, a new position is generated based on the whale's current position and its reverse position relative to the midpoint of the dimension, thus diversifying the whale's position.

[0026] Furthermore, the rebound boundary processing includes:

[0027] A collision bounce mechanism is used. When an individual whale's position exceeds the boundary of the search space, the portion that exceeds the boundary will bounce back into the boundary. The length of the bounce in each dimension is equal to the length that exceeds the boundary.

[0028] Furthermore, the mirror evolution mechanism includes:

[0029] Generate two whale populations, S1 and S2, and set the size of both populations to N.

[0030] The whale population S1 is initialized using the Chebyshev chaotic map;

[0031] The whale population S2 is initialized using a Logistic chaotic mapping.

[0032] Whale population S1 evolves using an improved whale algorithm, while whale population S2 evolves according to the strategy of the original whale algorithm.

[0033] After whale populations S1 and S2 have been updated through the different evolutionary strategies and positions described above, N / 3 whale individuals with the worst fitness in whale population S1 are replaced by random whale individuals from whale population S2. If the fitness of the whale individual to be exchanged in whale population S1 is worse than that of the whale individual selected for exchange in whale population S2, then the exchange is performed; otherwise, the original position of the whale individual with the worst fitness in whale population S1 is retained and no exchange is performed.

[0034] Compare the fitness values ​​of the best whale individuals in whale population S1 and whale population S2. If the fitness value of the best whale individual in whale population S1 is worse than that of the best whale individual in whale population S2, then the best whale individuals in the two whale populations are swapped. Then, the next iteration is executed, and the two whale populations continue to perform a new round of interactive evolution.

[0035] After the required number of iterations is reached, output the best whale individual in population S1.

[0036] This invention offers the following advantages: The GIS disconnector fault diagnosis method of this invention uses five types of information entropy—energy entropy, approximate entropy, sample entropy, fuzzy entropy, and permutation entropy—from the first five components of EMD decomposition as feature entropies to form a feature vector. Simultaneously, it extracts the time-domain features of voltage and current signals and fuses these time-frequency domain features for diagnosis. This invention proposes an improved whale optimization algorithm, using the penalty factor and kernel parameters corresponding to the optimal fitness as the optimal parameters of the support vector machine model. This enables efficient fault diagnosis of high-voltage disconnectors using a limited number of features when there are few fault samples. It solves the problem that the fault model trained by the original support vector machine algorithm cannot accurately divide the space, thus improving the accuracy of the fault diagnosis model. Attached Figure Description

[0037] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0038] Figure 1 A flowchart of the GIS disconnector fault diagnosis method provided by the present invention;

[0039] Figure 2 Here is an example flowchart of the overall fault diagnosis method for GIS disconnect switches provided by the present invention;

[0040] Figure 3 A detailed example flowchart of the GIS disconnector fault diagnosis method provided by the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The technical solutions provided by various embodiments of this invention will be described in detail below with reference to the accompanying drawings.

[0042] Please see Figures 1 to 3 The present invention provides a method for diagnosing faults in GIS disconnect switches, the method comprising:

[0043] S101, acquire the voltage and current signals of the GIS disconnect switch.

[0044] S102, the voltage and current signals are denoised using a wavelet filtering algorithm.

[0045] Specifically, wavelet filtering algorithm is used to denoise the voltage and current signals in order to remove interference signals that may be generated during the opening and closing of the GIS disconnector.

[0046] S103 extracts the modal function components of the current signal under different states as frequency domain features through EMD. Among them, the energy entropy, approximate entropy, sample entropy, fuzzy entropy and permutation entropy of the first 5 components are calculated as 5 types of information entropy as feature entropy to form a feature vector; the current signal and voltage signal are extracted as time domain features.

[0047] Specifically, EMD stands for Empirical Mode Decomposition.

[0048] S104, the frequency domain features and time domain features are fused using the PCA algorithm to obtain a fused feature set.

[0049] Specifically, the PCA algorithm is the Principal Component Analysis (PCA) algorithm.

[0050] S105, the fused feature set is divided into a training sample set and a test sample set.

[0051] S106. The kernel parameters and penalty factor of the support vector machine model are optimized by improving the original whale optimization algorithm to obtain the optimized support vector machine model.

[0052] The original whale optimization algorithm includes three stages: random search for food, shrinking and surrounding prey, and spiral bubble net hunting.

[0053] Random foraging behavior: When whales engage in random foraging behavior, the current position of an individual whale is updated by randomly selecting another individual whale and moving towards it. At this time, the whale moves with a larger step size, making the global search of the algorithm more thorough. The mathematical formula for this stage is:

[0054]

[0055]

[0056] In equation (1), A, D, and C are parameter vectors used to characterize random foraging behavior, and X rand The position vector of a randomly selected whale individual is X(t). The position vector of the t-th generation whale individual is X(t). The coefficient vector is updated by equation (2):

[0057]

[0058]

[0059] In equation (2), a and r are the variable parameters in the whale optimization algorithm. The value of a decreases from 2 to 0 as the number of iterations increases, while r is updated in the range of 0 to 1.

[0060] Shrinking and Encircling Prey Behavior: Once a whale individual finds food, the algorithm enters the shrinking and encircling phase. The whale individual performs a local search and focuses its attack on the prey. At this point, the optimal whale has found food, and its position approximates the prey's location. The current whale individual then moves towards the globally optimal whale according to formula (3). The mathematical formula for this phase is:

[0061]

[0062]

[0063] Among them, X best The location of the target prey is the location of the globally optimal individual whale.

[0064] Spiral bubble net hunting behavior: When a whale performs spiral bubble net hunting behavior, the distance between the current whale and the current best whale is calculated using formula (4). Then, a spiral function is constructed based on their positions, i.e., the whale moves in a spiral from the current whale as the starting point to the current best whale as the ending point. The mathematical formula is as follows:

[0065]

[0066]

[0067] In equation (4), the distance between the current individual and the global best individual is D', the constant b in the logarithmic spiral formula is set to 1, and the parameter l∈[0,1] is a uniformly distributed random number.

[0068] This invention introduces an enhanced global search mechanism, a quasi-adversarial learning mechanism, a bounce boundary handling mechanism, and a mirror evolution mechanism into the original whale optimization algorithm. It searches for the optimal solution in a multi-dimensional search space to obtain the optimal fitness. The penalty factor and kernel parameters corresponding to the optimal fitness are used as the optimal parameters of the support vector machine model.

[0069] This invention adjusts the criteria for judging the random search and foraging behavior and spiral bubble net predation behavior of whale populations in the improved algorithm. It introduces the historical optimal position of individuals during the random search of the whale algorithm, broadening the search range in the early stages of iteration and ensuring the convergence of the algorithm in the later stages. A quasi-adversarial learning mechanism is introduced after the whale individual position is updated, increasing population diversity. The unified boundary handling method for delayed sets is improved, mitigating the problem of reduced population diversity caused by large-scale convergence after a large number of individuals exceeding the boundary are uniformly processed. The basic whale algorithm cannot effectively balance its global development and local exploration capabilities. Although it has excellent solution performance, its solution stability is sometimes weak, and it is prone to getting trapped in local optima. To further improve the optimization performance and refine the optimization mechanism, a mirror population strategy is introduced, establishing an information exchange channel between populations with two different evolutionary strategies, thereby balancing and regulating the algorithm's global development and local exploration capabilities.

[0070] I. Enhancing the Global Search Mechanism: In the whale optimization algorithm, the global search performed by individual whales during the optimization process is insufficient. The value range of A is (-a, a). When the number of iterations... When the number of iterations increases, a gradually decreases from 2 to 1. At this time, the whale algorithm may execute one of three strategies: random search, surrounding the target, and bubble net predation. Random search will only be executed when |A|≥1 and P<0.5.

[0071] And when the number of iterations When a < 1, |A| < 1, and the algorithm stops performing random search behavior after the mid-term, resulting in weak global search ability, easy getting trapped in local optima, and unstable solution. Therefore, in the improved algorithm, the judgment conditions for random search foraging behavior and spiral bubble net predation behavior in the original whale optimization algorithm are adjusted:

[0072] The spiral bubble net predation behavior is executed when the random number p < 0.5 and |A| < 1;

[0073] When the random number p < 0.5 and |A| ≥ 1, the behavior of shrinking and surrounding the prey is executed;

[0074] When the random number p ≥ 0.5, random search foraging behavior is executed;

[0075] Where A is the position update coefficient of the individual whale during the search process of the original whale optimization algorithm, which is used to control the movement step size and direction of the individual whale in the search space; the random number p is used to determine the behavior pattern of the individual whale during the search process.

[0076] This increases the probability of the algorithm performing a random search, allowing for a more thorough global search of individual whales and improving the algorithm's solution capability. The improved formula for updating the prey encirclement behavior is:

[0077]

[0078]

[0079] X temp This is the improved position vector of a randomly selected individual whale.

[0080] Introducing the individual's historical best position Among them when hour:

[0081]

[0082] otherwise:

[0083]

[0084] With the iteration of the algorithm, Depend on Towards the global optimum Closer proximity. In the early stages of iteration, it broadened the search range and avoided blind searching; in the later stages of iteration, it ensured the convergence of the algorithm.

[0085] II. Quasi-adversarial learning mechanism:

[0086] The basic idea of ​​the random number inverse learning mechanism is to generate a new position for a whale based on its current position and its reverse position relative to the midpoint of a dimension during random foraging behavior. This diversifies the whale's positions and enhances the algorithm's global performance. The adversarial learning formulas for each dimension when updating the whale's position using the inverse learning strategy are as follows:

[0087]

[0088] newX j To update the position of individual whales using a reverse learning strategy; The minimum value in the j-dimensional location space. Let be the maximum value in the j-dimensional location space. However, reverse learning only considers the opposite position of a certain position, and the selection of points is too fixed and lacks flexibility. In order to better increase the diversity of the population and ensure the actual improvement of global search ability, an adversarial learning strategy is adopted to optimize the individual positions after each iteration.

[0089]

[0090] Equation (9) represents the newly generated quasi-adversarial point between the midpoint and the reverse point in the j-th dimension search space. After obtaining the quasi-adversarial point, the fitness value of qonewX is compared with that of X, and the better one is selected as the new individual position point. qonewX is the position of the whale individual updated using an adversarial learning strategy.

[0091] III. Handling of the rebound boundary:

[0092] The Whale Optimization Algorithm performs boundary checks at the beginning of each iteration, detecting and handling boundary violations in various dimensions of the entire population after the previous iteration. If an individual exceeds the boundary in a certain dimension, the position value of that dimension is set as the boundary value. However, this uniform handling after each iteration is detrimental to the algorithm's optimization. Specifically, during each iteration, when a whale performs a random search, it uses any random individual from the current generation's population. If this random individual is a whale that has already been updated and exceeded the boundary in this round, it will cause a chain reaction in the whale population, causing more whales to exceed the boundary. Using the original boundary checks method, a large number of individuals in the population will concentrate at the boundary, damaging population diversity and disrupting the original evolutionary results, leading to an overall decrease in the Whale Optimization Algorithm's solution capability. Therefore, during the iteration process, after each whale's position is updated, its new position should be detected and processed promptly. This chapter implements boundary violation checks immediately after an individual's position is updated and changes the original method of setting the boundary value immediately upon exceeding the boundary. A collision bounce mechanism is adopted. When an individual whale's position exceeds the boundary of the search space, the portion exceeding the boundary is bounced back into the boundary. The bounce length in each dimension is equal to the length exceeding the boundary. This improvement enhances the activity of the population and improves the stability of the algorithm.

[0093] IV. Mirror Evolution Mechanism: The mirror evolution mechanism is based on the improved whale optimization algorithm proposed in the above three improvements, and includes the following steps:

[0094] Generate two whale populations, S1 and S2, and set the size of both populations to N.

[0095] The whale population S1 is initialized using the Chebyshev chaotic map;

[0096] X t+1 =cos(a·arccosX) t (10)

[0097] Where a = 4, X0 ∈ [-1, 1].

[0098] The whale population S2 is initialized using a Logistic chaotic mapping.

[0099] X′ t+1 =μX′ t (1-X′ t (11)

[0100] Where μ = 4 and X′0 = 0.5.

[0101] Whale population S1 evolves using an improved whale algorithm, while whale population S2 evolves according to the original whale algorithm strategy. After whale populations S1 and S2 have completed their evolutionary strategies and position updates, random whale individuals from whale population S2 replace the N / 3 whale individuals with the worst fitness in whale population S1. If the fitness of the whale individual to be swapped in whale population S1 is worse than that of the selected whale individual to be swapped in whale population S2, then the swap is performed; otherwise, the original position of the whale individual with the worst fitness in whale population S1 is retained without swapping.

[0102] Compare the fitness values ​​of the best whale individuals in whale populations S1 and S2. If the fitness value of the best whale individual in whale population S1 is worse than that of the best whale individual in whale population S2, then the best whale individuals in the two whale populations are swapped. Then, the next iteration is executed, and the two whale populations continue to perform a new round of interactive evolution. After the required number of iterations is reached, the best whale individual in population S1 is output.

[0103] S107, The support vector machine model is trained and tested according to the training sample set and the test sample set to obtain the GIS disconnect switch fault diagnosis model.

[0104] S108, input the fused feature set of the GIS disconnect switch into the GIS disconnect switch fault diagnosis model to identify GIS disconnect switch faults.

[0105] As shown in the above embodiments, this method involves extracting current signal components as frequency domain features using EMD, extracting current and voltage signals as time domain features, fusing these time and frequency domain features, and employing an improved whale optimization algorithm to optimize the kernel parameters of the support vector machine. The fused feature set is then used as model input for GIS disconnector fault diagnosis and classification. This method achieves fusion feature extraction of voltage and current signals from GIS disconnectors under different states, improving the convergence accuracy and speed of the fault diagnosis model.

[0106] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention.

Claims

1. A method for diagnosing faults in GIS disconnect switches, characterized in that, The method includes: Obtain the voltage and current signals of the GIS disconnector; The voltage and current signals are denoised using a wavelet filtering algorithm. The modal function components of the current signal under different states are extracted by EMD as frequency domain features. Among them, the energy entropy, approximate entropy, sample entropy, fuzzy entropy and permutation entropy of the first 5 components are calculated as 5 types of information entropy as feature entropy, forming a feature vector; the current signal and voltage signal are extracted as time domain features. The frequency domain features and time domain features are fused using the PCA algorithm to obtain a fused feature set; The fused feature set is divided into a training sample set and a test sample set; The kernel parameters and penalty factor of the support vector machine model were optimized by improving the original whale optimization algorithm, resulting in an optimized support vector machine model. The support vector machine model is trained and tested based on the training sample set and the test sample set to obtain the GIS disconnect switch fault diagnosis model; The fused feature set of the GIS disconnect switch is input into the GIS disconnect switch fault diagnosis model to identify GIS disconnect switch faults.

2. The GIS disconnector fault diagnosis method as described in claim 1, characterized in that, The kernel parameters and penalty factor of the support vector machine model were optimized by improving the whale optimization algorithm, resulting in an optimized support vector machine model, including: An enhanced global search mechanism, a quasi-adversarial learning mechanism, a bounce boundary handling mechanism, and a mirror evolution mechanism are introduced into the original whale optimization algorithm to search for the optimal solution in a multi-dimensional search space to obtain the optimal fitness. The original whale optimization algorithm includes three stages: random search foraging behavior, shrinking and surrounding prey behavior, and spiral bubble net predation behavior. The penalty factor and kernel parameters corresponding to the optimal fitness are used as the optimal parameters of the support vector machine model.

3. The GIS disconnector fault diagnosis method as described in claim 2, characterized in that, The enhanced global search mechanism includes: The judgment conditions for random search foraging behavior and spiral bubble net predation behavior in the original whale optimization algorithm have been adjusted: The spiral bubble net predation behavior is executed when the random number p < 0.5 and |A| < 1; When the random number p < 0.5 and |A| ≥ 1, the behavior of shrinking and surrounding the prey is executed; When the random number p ≥ 0.5, random search foraging behavior is executed; Where A is the position update coefficient of the individual whale during the search process of the original whale optimization algorithm, which is used to control the movement step size and direction of the individual whale in the search space; the random number p is used to determine the behavior pattern of the individual whale during the search process.

4. The GIS disconnector fault diagnosis method as described in claim 3, characterized in that, The quasi-adversarial learning mechanism includes: When performing random foraging behavior, a new position is generated based on the whale's current position and its reverse position relative to the midpoint of the dimension, thus diversifying the whale's position.

5. The GIS disconnector fault diagnosis method as described in claim 4, characterized in that, The rebound boundary processing includes: A collision bounce mechanism is used. When an individual whale's position exceeds the boundary of the search space, the portion that exceeds the boundary will bounce back into the boundary. The length of the bounce in each dimension is equal to the length that exceeds the boundary.

6. The GIS disconnector fault diagnosis method as described in claim 5, characterized in that, The mirror evolution mechanism includes: Generate two whale populations, S1 and S2, and set the size of both populations to N. The whale population S1 is initialized using the Chebyshev chaotic map; The whale population S2 is initialized using a Logistic chaotic mapping. Whale population S1 evolves using an improved whale algorithm, while whale population S2 evolves according to the strategy of the original whale algorithm. After whale populations S1 and S2 have been updated through the different evolutionary strategies and positions described above, N / 3 whale individuals with the worst fitness in whale population S1 are replaced by random whale individuals from whale population S2. If the fitness of the whale individual to be exchanged in whale population S1 is worse than that of the whale individual selected for exchange in whale population S2, then the exchange is performed; otherwise, the original position of the whale individual with the worst fitness in whale population S1 is retained and no exchange is performed. Compare the fitness values ​​of the best whale individuals in whale population S1 and whale population S2. If the fitness value of the best whale individual in whale population S1 is worse than that of the best whale individual in whale population S2, then the best whale individuals in the two whale populations are swapped. Then, the next iteration is executed, and the two whale populations continue to perform a new round of interactive evolution. After the required number of iterations is reached, output the best whale individual in population S1.

Images