Improved Tuna Algorithm Optimization for Gas Turbine Thermal Efficiency Data Cluster Analysis

Abstract

The invention provides a gas turbine thermal efficiency data clustering analysis method optimized by an improved tuna algorithm, which obtains a data set of ambient temperature, compressor pressure ratio and total thermal efficiency of a gas turbine power plant operation record, determines the number of classes in the data set according to the number of data set categories; uses the information entropy reordering maximum and minimum distance product method to initialize the cluster center point of each class; calculates the Euclidean distance from each sample point in the data set to the cluster center point, sorts all Euclidean distances, and divides the cluster center corresponding to the minimum distance into the same cluster; determines the fitness function of TSOFA optimization; uses TSOFA to perform optimization operation on each class cluster, and uses a complex biomimetic model to optimize and obtain a new cluster center; uses the KMC algorithm to re-determine the category of each sample for the new cluster center; calculates the fitness of the new category, and if the fitness is better than the cluster center of the last iteration, replaces it with the new cluster center, otherwise continues to retain the result of the last iteration.

Description

Technical Field

[0001] The invention relates to the technical field of clustering, and in particular to a clustering analysis method for thermal efficiency data of a gas turbine optimized by an improved tuna algorithm. Background Art

[0002] In order to solve the growing global energy demand and save energy from operating gas turbine power plants, it is necessary to analyze the thermal efficiency data of power plants and take corresponding measures for different situations to realize a new green combined heat and power system with efficient energy utilization. Cluster analysis belongs to unsupervised learning. It can divide an overall sample set into several sub-data groups for mining. The goal is to divide the object set into several clusters so that objects with high similarity are classified into the same cluster. K-means clustering (KMC) is an unsupervised clustering method of top-down partition clustering. It has the advantages of strong local search ability, fast convergence speed, high clustering effect on large data sample space, and easy implementation. It has been widely used. However, in the actual gas turbine energy data analysis, the random initialization and the simple search for cluster centers will lead to large errors in the results, and the data cannot be accurately analyzed. K-means++ solves the problem of random initialization to a certain extent, but the optimization path is still too simple and easy to fall into local optimality; the swarm intelligence optimization algorithm, as a random optimization algorithm that imitates the social behavior mechanism of animals, can be unaffected by the initial solution, change the search direction with the help of the wisdom of group cooperation, jump out of the local solution, thereby optimizing the search for cluster centers and improving the defects of KMC. In recent years, it has become a hot research direction to solve the problem of premature maturity by using the position update iteration method combined with biological characteristics of the swarm intelligence optimization algorithm to find the cluster center of the clustering algorithm. Huang He et al. proposed a manta ray algorithm based on random jumping tumbling foraging optimization and K-means joint iteration to cluster analysis movie data sets, but the convergence speed is low; Qin Risheng et al. used weight linear reduction and synchronous learning factor to improve the PSO algorithm to optimize K-means, making the classification results more accurate, and used for daily load curve classification in power grid system planning, but the calculation takes a long time; Shi et al. proposed a hybrid classification method for telecommunications customers in order to mine customers with special behaviors from massive telecommunications language call records, but the data is easily affected by noise points. Tuna Swarm Optimization (TSO) is a global optimization metaheuristic algorithm based on the mutual communication and cooperative foraging of tuna population individuals proposed by Lei et al. in 2021. Inspired by the unique foraging behavior of fish schools, it establishes a complex mathematical model for updating and iteration. Compared with other metaheuristic algorithms, it has certain advantages in convergence, stability and robustness, but there is still room for further optimization. Summary of the invention

[0003] The K-means clustering algorithm is greatly affected by initialization, and the clustering center is randomly generated, which makes the clustering result easily fall into the local optimum and stop iteration. In addition, the simple method of obtaining the clustering center of the cluster easily causes the clustering result to be poor, low precision, poor generalization ability and other problems. The present invention provides a gas turbine thermal efficiency data clustering analysis method optimized by an improved tuna algorithm. The present invention has higher search accuracy, convergence speed and lower clustering cost when facing different test data sets, and has better effect on the clustering of thermal efficiency data of gas turbines in power plants.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows: The improved tuna algorithm optimized gas turbine thermal efficiency data clustering analysis method comprises the following steps: Step 1: Obtain a data set of ambient temperature, compressor pressure ratio, and total thermal efficiency from the operation records of a gas turbine power plant, and determine the number of classes in the data set based on the number of classes in the data set. K ; Step 2: Use the information entropy reordering maximum and minimum distance product method to initialize the cluster center points of each category for the data set obtained in step 1; Step 3: Calculate the Euclidean distance from each sample point in the data set in step 1 to the cluster center point obtained in step 2, sort all Euclidean distances, and divide the cluster center corresponding to the minimum distance into the same cluster; Step 4: Determine the fitness function of TSOFA optimization; Step 5: Use TSOFA to optimize each cluster obtained in step 3, and use a complex biomimetic model to optimize and obtain new cluster centers; Step 6: Use the KMC algorithm to re-determine the category of each sample based on the new cluster center obtained in step 5; Step 7: Calculate the fitness of the new category obtained in step 6. If its fitness is better than the cluster center of the previous iteration, replace it with the new cluster center. Otherwise, keep the result of the previous iteration.

[0005] Furthermore, the elbow method is used in step 1 to determine the number of classes in the dataset. K .

[0006] Furthermore, in step 2, the information entropy reordering maximum and minimum distance product method is used to initialize the cluster center points of each category, specifically including: Step 2.1: Use the maximum and minimum distance product method to select KMC K P pre-selected cluster centers, among which, K P = ωK , ω is a random parameter between [1,2]; Step 2.2: Use the K nearest neighbor method to calculate all remaining data points x Relative to the selected K P Pre-select cluster centers for density estimation, and divide all sample data into K P clusters, other data points in the sample K P Pre-selected cluster center distance estimates R m The order from smallest to largest is as follows:

[0007] in, R m , m =1,… N,N is the number of samples; Step 2.3: Combine the clusters formed by each pre-selected cluster center and calculate the correlation based on the information entropy idea as follows:

[0008] in, m Is with the sample N integers proportional to the size of d Indicates the data dimension, C x is the pre-selected center of the computed cluster, X i For the other points of the preselected cluster; Step 2.4: Sort the calculated correlations of the pre-selected clusters and take the ones with the highest correlations first. K The cluster centers corresponding to the clusters are used as the initial cluster centers after the information entropy reordering maximum and minimum distance product method is improved.

[0009] Furthermore, step 3 is specifically as follows: calculating the Euclidean distance from each sample point to each cluster center, and then classifying the sample points to the cluster center with the nearest distance, so as to achieve distance-based classification.

[0010] Furthermore, the fitness function described in step 4 is expressed as follows:

[0011]

[0012] In the formula, e is the number of population categories, D e Indicates e The distance between the objects of the class population and the center point of the cluster of the class, N eFor the e The number of species in the class.

[0013] Furthermore, in step 5, TSOFA is used to perform an optimization operation, which specifically includes: Step 5.1: Initialize a random population of tuna and set the objective function; Step 5.2: Specify free parameters z , for the control parameters a Perform perturbation-random adjustment strategy; Step 5.3: Calculate the population fitness value and update α 1 , α 2 and p ,in α 1 and α 2 It is the weight coefficient of the spiral foraging phase of the TSOFA algorithm that controls the individual's movement trend toward the best individual and the previous individual. p is the parabolic foraging parameter of TSOFA algorithm; Step 5.4: Determine whether rand < z If it is established, use the random uniform formula to update the position, otherwise determine whether rand <0.5, if not, then use the parabolic foraging method improved by the foraging warning factor to update; if so, then further judge t / t max <rand Is it true? If so, use the perturbation-randomly adjust the control parameters to update the spiral foraging position. Otherwise, use the original spiral foraging strategy. rand is a random vector uniformly distributed in [0,1], t and t max are the current and maximum number of iterations respectively; Step 5.5: Carry out the optimization strategy of chasing in unison to select the individual with the best speed in the school of fish and retain its position; Step 5.6: Determine whether the iteration stop condition is met. If so, output the best individual and its corresponding best fitness. Otherwise, return to step 5.4 to continue the cycle.

[0014] Furthermore, in step 5.2, the control parameters a Perform a perturbation-random adjustment strategy, specifically:

[0015]

[0016] In the formula, amax and a min for a The maximum value of rand is a random number uniformly distributed between [0,1]. λ is the nonlinear perturbation coefficient, λ max and λ min for λ The maximum value of Determine in step 5.4 rand When <0.5, it is no, the parabolic foraging method is updated using the improved foraging warning factor. The specific operations include: Set a warning probability, and then generate a foraging warning factor under this warning probability, and use this foraging warning factor to intervene in the population's search for food source locations to maintain foraging diversity. The original parabolic foraging formula is improved as follows:

[0017] in, X t+i i and X t i For the i Only individuals in t + i The iteration and t The position after iterations; X t best is the current optimal individual; TF is a random number with a value of 1 or -1; p Determined by iteration, the calculation formula is as follows:

[0018] T is the foraging warning factor, and its calculation formula is as follows:

[0019] Among them, the variable r is a random number uniformly distributed between [0,1]; C is a constant with a very small value; P ( r )and sign ( r ) is as follows:

[0020]

[0021] in, Px is a given warning probability.

[0022] Furthermore, the specific operation of the parallel pursuit optimization strategy in step 5.5 is as follows: After each iteration, the better position of the tuna is preserved, while the worse position of the tuna is ignored, so that the individual that obtains more food energy in the population is selected as the best individual, and other individuals pursue this goal, thereby increasing the speed of the entire fish school. The formula for updating the optimal position is as follows:

[0023] in, X t i For the first i Only individuals in The position after iterations, X new is the newly generated position, r new and q is a random number between [0,1]; f ( X ) is the corresponding position X The fitness of .

[0024] Furthermore, the KMC algorithm in step 6 includes the following steps: Step 6.1: Determine the initial cluster center by randomly selecting the sample set C p , where 1≤ p ≤ K , K is the number of target clusters, that is, the number of classes in the data set; Step 6.2: Using the Euclidean distance as the standard, separate the other sample data according to the minimum distance principle and the nearest C p into the same cluster; Step 6.3: Calculate the sample mean of each cluster and update C p ; If there are still sample data transferred to different clusters, C p If the change continues or the change distance does not reach the threshold, return to step 6.2 to continue updating. Otherwise, clustering is completed and the classification clusters and their cluster centers of the sample data set are mined. C p .

[0025] Furthermore, in step 6.3, the cluster centers are calculated C p The formula is as follows:

[0026] In the formula, n p For the p The number of cluster samples, p =1,2... K , C p With sample data x i The Euclidean distance calculation for is as follows:

[0027] In the formula, j =1,2... d, d Indicates the data dimension, x ij For sample data x i exist j vector value in dimension, C pj is the cluster center C p exist j Vector value in dimension.

[0028] The above technical solution has the following advantages or beneficial effects: The present invention proposes a clustering analysis method for gas turbine thermal efficiency data based on improved tuna algorithm optimization, which improves the problems that the traditional K-means clustering (KMC) algorithm is greatly affected by initialization, randomly generates cluster centers, causes the clustering results to easily fall into local optimality and stop iteration, and the simple way of obtaining the cluster centers of the clusters easily causes poor clustering results, low precision, and poor generalization ability, thus being unsuitable for gas turbine energy data analysis.

[0029] The present invention first designs an initialization method for optimizing the KMC cluster center by reordering the information entropy with the maximum and minimum distance product, and improves the influence of data noise on the clustering effect; secondly, in view of the shortcomings of the TSO algorithm, which has a slow convergence speed, is easy to fall into the local optimum, and has limited search accuracy, a TSOFA algorithm is designed. On the basis of the original TSO, a perturbation-random adjustment control parameter strategy is added to enable the algorithm to quickly jump out of the local optimum in the early stage, and better balance the overall search ability in the later stage, and adopt a foraging warning factor to enable the tuna population to adapt to food mutations and quickly jump out of the local optimum; a parallel pursuit optimization strategy is designed, which improves the foraging efficiency of the entire population through comparative competition, improves the global search ability of the algorithm, and speeds up the optimization speed; finally, TSOFA is integrated with the KMC algorithm, and the collaborative iterative optimization method of KMC is optimized to find the cluster center, thereby expanding the optimization range, improving the search accuracy, and obtaining a better clustering effect. Therefore, TSOFA-KMC can achieve effective clustering and has strong applicability.

[0030] When the present invention is applied to cluster analysis of thermal efficiency data sets of gas turbine power plants, it performs better than the tuna optimization algorithm in distinguishing between high-efficiency and low-energy consumption and low-efficiency and high-energy consumption situations of power plants, proving that the present invention helps urban energy coordination decision makers to select appropriate cogeneration systems and achieve the goal of green energy conservation. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 is a flow chart of the improved tuna optimization algorithm involved in the present invention; Figure 3 The clustering fitness convergence curves of TSOFA-KMC of the present invention and various comparison algorithms on various data sets, (a) is the running result of the Vehicle data set, (b) is the running result of the Ionosphere data set, (c) is the running result of the Aggregation data set, (d) is the running result of the Vowel data set, (e) is the running result of the Sonar data set, and (f) is the running result of the WDBC data set; Figure 4 These are the results of running different algorithms on the thermal efficiency dataset of a gas turbine power plant. (a) is the iteration curve of the power plant clustering, and (b) is the three-dimensional clustering effect diagram. DETAILED DESCRIPTION

[0032] The present invention is further described in detail below in conjunction with specific embodiments, which are intended to explain the present invention rather than to limit it.

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. 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 creative work should fall within the scope of protection of the present invention. It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatus.

[0034] The present invention provides a gas turbine thermal efficiency data clustering analysis method optimized by an improved tuna algorithm. Aiming at the problems that the K-means clustering algorithm is greatly affected by initialization, the clustering center is randomly generated, which makes the clustering result easy to fall into the local optimum and stop iteration, and the simple way of obtaining the clustering center of the cluster is easy to cause poor clustering results, low precision and poor generalization ability, thus being unsuitable for gas turbine energy data analysis, a TSOFA-optimized K-means collaborative iterative clustering (TSOFA-KMC) algorithm is proposed. Firstly, a method for initializing KMC cluster centers by reordering information entropy and optimizing the maximum and minimum distance product was designed to improve the influence of data noise on clustering effect. Secondly, a TSOFA algorithm was designed to address the shortcomings of the TSO algorithm, such as slow convergence speed, easy to fall into local optimality and limited search accuracy. On the basis of the original TSO, a perturbation-random adjustment control parameter strategy was added to enable the algorithm to quickly jump out of the local optimal value in the early stage, and better balance the overall search ability in the later stage. The foraging warning factor was used to enable the tuna population to adapt to food mutations and quickly jump out of the local optimal value. A parallel pursuit optimization strategy was designed to improve the foraging efficiency of the entire population through comparative competition, enhance the global search ability of the algorithm, and accelerate the optimization speed. Finally, TSOFA was integrated with the KMC algorithm to collaboratively iteratively optimize the way KMC finds cluster centers, thereby expanding the optimization range, improving the search accuracy, and obtaining better clustering effect. The improvement effects of TSOFA and TSOFA-KMC were tested using multiple standard test functions and data sets, and compared with similar algorithms in literature. The performance indicators show that the proposed TSOFA-KMC algorithm has better convergence speed and search accuracy than other algorithms. When the algorithm is applied to the cluster analysis of the thermal efficiency data set of gas turbine power plants, it can effectively distinguish between the high-efficiency and low-energy consumption and low-efficiency and high-energy consumption of power plants, which helps to select a suitable cogeneration system and achieve the goal of green energy conservation.

[0035] See also Figure 1 The flowchart of the present invention proposes a clustering analysis method for thermal efficiency data of gas turbines based on the improved tuna algorithm optimization. The main work is reflected in the following points: First, the randomness problem of the original KMC initialization cluster center is solved by the information entropy reordering maximum and minimum distance product method, which improves the stability of clustering; secondly, Figure 2, proposed an improved tuna algorithm TSOFA, designed a disturbance-random adjustment control parameter strategy, so that the algorithm can quickly jump out of the local optimal value in the early stage, and better balance the overall search ability in the later stage; designed a foraging warning factor so that the tuna population can adapt to food mutations and quickly jump out of the local optimal value; adopted a parallel pursuit optimization strategy, and improved the foraging efficiency of the entire population by comparing the competitive and superior population individuals, and enhanced the global search ability and optimization speed. Then, through the collaborative iteration of TSOFA and KMC, the iteration complexity and the solution optimization range were increased, the algorithm was prevented from falling into the local optimal value, and the search accuracy was improved, such as Figure 3 Finally, the algorithm of the present invention is applied to the energy situation data set of gas turbine power plant, such as Figure 4 , the factors affecting the thermal efficiency of power plants were analyzed, and good cluster analysis results were achieved.

[0036] The specific steps are as follows: Step 1: Input the data set of ambient temperature, compressor pressure ratio and total thermal efficiency of the gas turbine power plant operation records, and determine the number of classes in the data set according to the number of its categories. K ; Step 2: Use the information entropy reordering maximum and minimum distance product (IRMMP) method to initialize the cluster center of each class for the data set obtained in step 1. Due to the uncertainty of unsupervised learning data features, it is crucial to exclude noise data in cluster analysis. Information entropy can intuitively quantify the amount of information in uncertain phenomenon variables. The larger the entropy, the more changes the system has and the less order it has. Therefore, referring to the information entropy idea to improve the maximum and minimum distance product initialization method, it can reduce the noise data. C p The specific steps of IRMMP are as follows: Step 2.1: Get K P = ωK ,in K is the number of target clustering clusters in the data set, which is usually obtained by the elbow method in experiments. ω is a random parameter between [1,2], selected according to the specific data set; Step 2.2: Use the maximum and minimum distance product method to select KMC K P Pre-selected cluster centers; Step 2.3: Use the K nearest neighbor method to calculate all remaining data points x Relative to the selected K PPre-select cluster centers for density estimation, and divide all sample data into K P clusters, other data points in the sample K P Pre-selected cluster center distance estimates R m The order from smallest to largest is as follows:

[0037] in, R m ,m=1,…N,N is the number of samples; Step 2.4: Then, combine the clusters formed by each pre-selected cluster center and calculate the correlation based on the information entropy idea as follows:

[0038] in, m Is with the sample N integers proportional to the size of d Indicates the data dimension, C x is the pre-selected center of the computed cluster, X i For the other points of the preselected cluster; Step 2.5: Sort the calculated correlations of the pre-selected clusters and take the ones with the highest correlation first. K The cluster centers corresponding to the clusters are used as the initial cluster centers after IRMMP improvement.

[0039] In this way, using IRMMP can solve the problem that the original maximum and minimum distance product does not consider the data density when selecting the initial cluster center, reduce the impact of noise data, and improve the search accuracy.

[0040] Step 3: Calculate the Euclidean distance of each sample point in the data set in step 1 to the cluster center point obtained in step 2, sort all values, and divide the center corresponding to the minimum distance into the same cluster. This method is based on the Euclidean distance between the sample point and each cluster center for classification. By calculating the distance from each sample point to each cluster center, and then classifying the sample point to the cluster center closest to it, distance-based classification is achieved.

[0041] Step 4: Determine the fitness function of TSOFA optimization. The specific operation is as follows: The purpose of optimization is to find the optimal solution. It is necessary to select a suitable fitness function to measure the quality of the solution. It is the only interface between TSOFA and KMC algorithm. Therefore, when solving group optimization problems, if you want to get better optimization results and optimal solutions, you need to construct a fitness function that conforms to the optimal evolutionary direction of the group. KMC uses the distance from the individual to the center and / or the total number of individuals in each category as the fitness function. When the distance or number of points is equal, there will be a problem of insufficient recognition ability affecting the update. Using TSOFA to optimize the cluster center, combined with distance and number, the fitness function used is as follows:

[0042]

[0043] In the formula, e is the number of population categories, D e Indicates e The distance between the objects of the class population and the center point of the cluster of the class, N e For the e The number of populations in the class. It can be seen that the selected fitness depends on the number of cluster samples and the cluster density, which fully reflects the clustering effect and makes the TSOFA optimization result better.

[0044] Step 5: Use TSOFA to optimize each cluster obtained in step 3, and use a complex biomimetic model to optimize and obtain new cluster centers; The principle of the tuna optimization algorithm TSO is as follows: Initialization: TSO generates an initial population in a random and uniform manner throughout the entire search space. The formula is as follows:

[0045] in, X int i It is i The upper and lower bounds of the search space are respectively ub and lb limited , NP is the number of tuna populations, rand is a uniformly distributed random vector in [0,1].

[0046] Form a tight spiral formation to forage. The spiral foraging formula is as follows: when rand <t / t max When , the position of the population is expressed as:

[0047] whenrand ≥ t / t max When , the population position formula is:

[0048]

[0049]

[0050]

[0051] in, t and t max are the current and maximum number of iterations respectively; b is a random number uniformly distributed between [0,1]; X t+1 i For the i Only individuals in t + Position after 1 iteration; X t best and X t rand are the current optimal and random individuals respectively; α 1 and α 2 It is the weight coefficient that controls the individual's tendency to move toward the best individual and the previous individual; a It is a constant that determines the best individual of tuna in the initial stage and the degree of following the previous individual.

[0052] When the population is arranged in a parabolic formation or spontaneously searches for food, the mathematical model is as follows:

[0053]

[0054] in, TF is a random number with value 1 or -1.

[0055] According to the characteristics of thermal efficiency data of gas turbine power plants, the tuna algorithm is improved, a disturbance-random adjustment control parameter strategy is designed, and a foraging warning factor and a go-hand-to-hand pursuit optimization strategy are introduced to obtain a new optimization algorithm TSOFA. The iterative principle is as follows: At the beginning of the algorithm, a random population of tuna is initialized and the objective function is set; then the free parameters are specified. z , for the control parameters a By performing a perturbation-random adjustment strategy and analyzing the position update formula of the TSO parabolic foraging phase, we can see that the global and local search capabilities of the algorithm are controlled by the control parametersa The value has a great impact. However, when TSO establishes a mathematical model, a If the value is fixed, it is easy to fall into the local optimum during the iteration process and affect the global search accuracy. Therefore, a perturbation-random adjustment control parameter strategy is proposed, and the specific implementation formula is:

[0056]

[0057] Among them, a max and a min for a The maximum value of rand is a random number uniformly distributed between [0,1]. λ is the nonlinear perturbation coefficient, λ max and λ min for λ The design of random distribution adjustment can make the control parameters randomly obtain larger or smaller values ​​in the early and late stages of optimization; a concave function that causes disturbance effect is introduced through the above formula, with larger strength in the early stage and smaller strength in the later stage; the comprehensive disturbance-random adjustment control parameter improvement can prevent the algorithm from falling into a local peak in the early stage of iteration, and better improve the global search ability in the later stage, balancing the overall iteration process.

[0058] Then calculate the population fitness value and update α 1 , α 2 and p ,in α 1 and α 2 It is the weight coefficient of the spiral foraging phase of the TSOFA algorithm that controls the individual's movement trend toward the best individual and the previous individual. p is the parabolic foraging parameter of TSOFA algorithm; Determine whether rand < z If it is established, use the random uniform formula to update the position, otherwise determine whether rand <0.5, "No", then use the parabolic foraging method improved by the foraging warning factor to update; if "Yes", further judge t / t max <rand Is it true? If so, use the perturbation-randomly adjust the control parameters to update the spiral foraging position. Otherwise, use the original spiral foraging strategy. randis a random vector uniformly distributed in [0,1], t and t max The current and maximum number of iterations respectively The parabolic foraging method is updated using the improved foraging warning factor. The specific operation is as follows: Tuna needs a lot of food to ensure the energy required for rapid and continuous swimming in the water flow and metabolism. However, during the foraging process, the position of the food source cannot remain unchanged. It will mutate to avoid predator attacks in order to improve the survival rate. Therefore, the position update iteration affects the quality of the entire TSO algorithm. In order to improve the diversity and randomness of tuna foraging, a foraging warning factor is designed to intervene in the original position update during the parabolic foraging stage to achieve the effect of prompting the population to change the foraging direction and reduce the shortcomings of TSO's premature convergence and easy to fall into the local optimum. The implementation steps are as follows: Set a warning probability, and then generate a foraging warning factor under this probability, and use this factor to intervene in the population's search for food source locations to maintain foraging diversity. The original parabolic foraging formula is improved as follows:

[0059] in, T is the foraging warning factor, and its calculation formula is as follows:

[0060] Among them, the variable r is a random number uniformly distributed between [0,1]; C is a constant with a very small value, obtained through a large number of experiments C =0.0001; P ( r )and sign ( r ) is as follows:

[0061]

[0062] in, P x For a given warning probability, take the empirical value P x =0.3. Under the effect of the foraging warning factor, tuna can better adapt to the sudden change of food position when hunting, so that in the parabolic position update iteration link, it can quickly jump out of the local optimal range and accelerate the global decreasing speed.

[0063] The optimization strategy of keeping pace with each other is carried out to select the individuals with the best speed in the school of fish and retain their positions. The specific operation of the optimization strategy of keeping pace with each other is as follows: Tuna can swim quickly in the deep cold sea water of high dimension to find food, and its body temperature must be higher than the temperature of the environment. At this time, the tuna needs to fully call on energy to maintain the high metabolism and rapid digestion of body functions. A good position change can catch more food, thereby ensuring that the tuna generates a lot of energy when it catches food and bursts out at a faster speed. Once the leader in the school of fish has a speed higher than that of other populations, it will drive the speed of the entire school of fish to increase, thereby increasing the foraging amount of the population and making the population stronger. This is the basic idea of ​​the optimization strategy of keeping pace with each other designed by the present invention. This strategy enables the better position of tuna to be preserved after each iteration, while the worse position of tuna will be ignored, so that the individuals in the population that can obtain more food energy (i.e., have the fastest speed) are selected as the best individuals, and other individuals chase with this as the goal, thereby increasing the speed of the entire school of fish. The formula for updating the position by preferential replacement is as follows:

[0064] in, X t i For the first i Only individuals in The position after iterations, X new is the newly generated position, r new and q is a random number between [0,1]; f ( X ) is the corresponding position X The fitness of .

[0065] Determine whether the iteration stop condition is met. If so, output the best individual and its corresponding best fitness. Otherwise, return to the previous step and continue the cycle.

[0066] Step 6: Use KMC to re-determine the category of each sample based on the center obtained in step 5; The KMC algorithm process is as follows: The sample set is randomly selected to determine the initial cluster center C p , where 1≤ p ≤ K , K is the number of target clusters; Using the Euclidean distance as the standard, other sample data are separated from the nearest C p into the same cluster; Calculate the sample mean of each cluster and updateC p ; If there are still sample data transferred to different clusters, C p If the change continues or the change distance does not reach the threshold, return to step 6.2 to continue updating. Otherwise, clustering is completed and the classification clusters and their cluster centers of the sample data set are mined. C p .

[0067] The cluster center C p The formula is as follows:

[0068] In the formula, n p For the p The number of cluster samples, p =1,2... K , C p With sample data x i The Euclidean distance calculation for is as follows:

[0069] In the formula, j =1,2... d, d Indicates the data dimension, x ij For sample data x i exist j vector value in dimension, C pj is the cluster center C p exist j Vector value in dimension.

[0070] Step 7: Calculate the fitness of the new category obtained in step 6. If its fitness is better than the cluster center of the previous iteration, replace it with the new cluster center. Otherwise, keep the result of the previous iteration.

[0071] Figure 3 The clustering fitness convergence curves of TSOFA-KMC of the present invention and various comparison algorithms on each data set, (a) is the running result of the Vehicle data set, (b) is the running result of the Ionosphere data set, (c) is the running result of the Aggregation data set, (d) is the running result of the Vowel data set, (e) is the running result of the Sonar data set, and (f) is the running result of the WDBC data set.

[0072] According to the various iteration curves, it can be analyzed that the convergence accuracy of TSOFA-KMC in different data sets is far better than that of the other five algorithms, and the improved initialization algorithm also has obvious advantages. For the Ionosphere and Vehicle data sets, the algorithm of the present invention is the best in terms of convergence speed and convergence accuracy; for the Sonar and WDBC data sets, by comparing the TSO-KMC convergence curves, it can be seen that the improved TSOFA and KMC collaborative iteration performance is significantly better, proving that the improvement of the swarm intelligence algorithm also performs well after combining with clustering.

[0073] Figure 4 It is the result of running different algorithms on the thermal efficiency data set of gas turbine power plants. It can be seen from the convergence curve (a) that the TSOFA-KMC algorithm proposed in the present invention has a good clustering effect on the energy situation data set of gas turbine power plants. It can be seen from the clustering effect of Figure (b) that the red cluster is in a high-efficiency and low-consumption state. This type of power plant has a high efficiency and can better convert energy into electricity supply; therefore, in areas and times with low heat supply plans, the ambient temperature and compressor pressure of the gas turbine power generation equipment can be adjusted to the range of the red cluster to achieve efficient power generation. The green cluster is in a low-efficiency and high-consumption state, indicating that under higher ambient air temperatures and lower compressor pressure ratios, the total thermal efficiency is lower, the power plant's power generation performance is lower, and more waste is caused. At this time, it is possible to consider using the waste heat of the power plant that is not fully utilized for the combined heat and power system, reusing energy, or when the heat demand is large and other non-fossil energy sources are stable, adjust the production conditions to generate more heat to supplement the shortage of the heating system and prevent excess electricity.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. Improved tuna algorithm optimized gas turbine thermal efficiency data clustering analysis method, characterized in that: The following steps are involved: Step 1: Obtain a data set of ambient temperature, compressor pressure ratio, and total thermal efficiency from the operation records of a gas turbine power plant, and determine the number of classes in the data set based on the number of classes in the data set. K ; Step 2: Use the information entropy reordering maximum and minimum distance product method to initialize the cluster center points of each category for the data set obtained in step 1; Step 3: Calculate the Euclidean distance from each sample point in the data set in step 1 to the cluster center point obtained in step 2, sort all Euclidean distances, and divide the cluster center corresponding to the minimum distance into the same cluster; Step 4: Determine the fitness function of TSOFA optimization; Step 5: Use TSOFA to optimize each cluster obtained in step 3, and use a complex biomimetic model to optimize and obtain new cluster centers; Step 6: Use the KMC algorithm to re-determine the category of each sample based on the new cluster center obtained in step 5; Step 7: Calculate the fitness of the new category obtained in step 6. If its fitness is better than the cluster center of the previous iteration, replace it with the new cluster center. Otherwise, keep the result of the previous iteration.

2. The gas turbine thermal efficiency data clustering analysis method optimized by the improved tuna algorithm according to claim 1 is characterized in that: In step 1, the elbow method is used to determine the number of classes in the data set. K .

3. The gas turbine thermal efficiency data clustering analysis method optimized by improved tuna algorithm according to claim 1 is characterized in that: In step 2, the information entropy reordering maximum and minimum distance product method is used to initialize the cluster center points of each category, including: Step 2.1: Use the maximum and minimum distance product method to select KMC K P pre-selected cluster centers, among which, K P = ωK , ω is a random parameter between [1,2]; Step 2.2: Use the K nearest neighbor method to calculate all remaining data points x Relative to the selected K P Pre-select cluster centers for density estimation, and divide all sample data into K P clusters, other data points in the sample K P Pre-selected cluster center distance estimates R m The order from smallest to largest is as follows: in, R m , m =1,… N,N is the number of samples; Step 2.3: Combine the clusters formed by each pre-selected cluster center and calculate the correlation based on the information entropy idea as follows: in, m Is with the sample N integers proportional to the size of d Indicates the data dimension, C x is the pre-selected center of the computed cluster, X i For the other points of the preselected cluster; Step 2.4: Sort the calculated correlations of the pre-selected clusters and take the ones with the highest correlations first. K The cluster centers corresponding to the clusters are used as the initial cluster centers after the information entropy reordering maximum and minimum distance product method is improved.

4. The gas turbine thermal efficiency data clustering analysis method optimized by the improved tuna algorithm according to claim 1 is characterized in that: Step 3 is as follows: calculate the Euclidean distance from each sample point to each cluster center, and then classify the sample points to the cluster center with the nearest distance to achieve distance-based classification.

5. The improved tuna algorithm optimized gas turbine thermal efficiency data clustering analysis method according to claim 1 is characterized in that: The fitness function described in step 4 is expressed as follows: In the formula, e is the number of population categories, D e Indicates e The distance between the objects of the class population and the center point of the cluster of the class, N e For the e The number of species in the class.

6. The gas turbine thermal efficiency data clustering analysis method optimized by the improved tuna algorithm according to claim 5 is characterized in that: In step 5, TSOFA is used to perform optimization operations, including: Step 5.1: Initialize a random population of tuna and set the objective function; Step 5.2: Specify free parameters z , for the control parameters a Perform perturbation-random adjustment strategy; Step 5.3: Calculate the population fitness value and update α 1, α 2 and p ,in α 1 and α 2 is the weight coefficient of the spiral foraging phase of the TSOFA algorithm that controls the individual's movement trend toward the best individual and the previous individual. p is the parabolic foraging parameter of TSOFA algorithm; Step 5.4: Determine whether rand < z If it is established, use the random uniform formula to update the position, otherwise determine whether rand <0.5, if not, then use the parabolic foraging method improved by the foraging warning factor to update; if so, then further judge t / t max <rand Is it true? If so, use the perturbation-randomly adjust the control parameters to update the spiral foraging position. Otherwise, use the original spiral foraging strategy. rand is a random vector uniformly distributed in [0,1], t and t max are the current and maximum number of iterations respectively; Step 5.5: Carry out the optimization strategy of chasing in unison to select the individual with the best speed in the school of fish and retain its position; Step 5.6: Determine whether the iteration stop condition is met. If so, output the best individual and its corresponding best fitness. Otherwise, return to step 5.4 to continue the cycle.

7. The gas turbine thermal efficiency data clustering analysis method optimized by the improved tuna algorithm according to claim 6 is characterized in that: In step 5.2, the control parameters a Perform a perturbation-random adjustment strategy, specifically: In the formula, a max and a min for a The maximum value of rand is a random number uniformly distributed between [0,1], λ is the nonlinear perturbation coefficient, λ max and λ min for λ The maximum value of Determine in step 5.4 rand When <0.5, it is no, the parabolic foraging method is updated using the improved foraging warning factor. The specific operations include: Set a warning probability, and then generate a foraging warning factor under this warning probability, and use this foraging warning factor to intervene in the population's search for food source locations to maintain foraging diversity. The original parabolic foraging formula is improved as follows: in, X t+i i and X t i For the i Only individuals in t + i The iteration and t The position after iterations; X t best is the current optimal individual; TF is a random number with a value of 1 or -1; p Determined by iteration, the calculation formula is as follows: T is the foraging warning factor, and its calculation formula is as follows: Among them, the variable r is a random number uniformly distributed between [0,1]; C is a constant with a very small value; P ( r )and sign ( r ) is as follows: in, P x is a given warning probability.

8. The improved tuna algorithm optimized gas turbine thermal efficiency data clustering analysis method according to claim 6 is characterized in that: The specific operations of the parallel pursuit optimization strategy in step 5.5 are as follows: After each iteration, the better position of the tuna is preserved, while the worse position of the tuna is ignored, so that the individual that obtains more food energy in the population is selected as the best individual, and other individuals pursue this goal, thereby increasing the speed of the entire fish school. The formula for updating the optimal position is as follows: in, X t i For the first i Only individuals in The position after iterations, X new is the newly generated position, r new and q is a random number between [0,1]; f ( X ) is the corresponding position X The fitness of .

9. The gas turbine thermal efficiency data clustering analysis method optimized by improved tuna algorithm according to claim 1, characterized in that: The KMC algorithm in step 6 includes the following steps: Step 6.1: Determine the initial cluster center by randomly selecting the sample set C p , where 1≤ p ≤ K , K is the number of target clusters, that is, the number of classes in the data set; Step 6.2: Using the Euclidean distance as the standard, separate the other sample data according to the minimum distance principle and the nearest C p into the same cluster; Step 6.3: Calculate the sample mean of each cluster and update C p ; If there are still sample data transferred to different clusters, C p If the change continues or the change distance does not reach the threshold, return to step 6.2 to continue updating. Otherwise, clustering is completed and the classification clusters and their cluster centers of the sample data set are mined. C p .

10. The gas turbine thermal efficiency data clustering analysis method optimized by improved tuna algorithm according to claim 9, characterized in that: Calculate the cluster center in step 6.3 C p The formula is as follows: In the formula, n p For the p The number of cluster samples, p =1,2... K , C p With sample data x i The Euclidean distance calculation for is as follows: In the formula, j =1,2... d,d Indicates the data dimension, x ij For sample data x i exist j vector value in dimension, C pj is the cluster center C p exist j Vector value in dimension.

Images