Control parameter collaborative optimization method based on hybrid heuristic optimization

Abstract

The invention discloses a control parameter collaborative optimization method based on hybrid heuristic optimization, which comprises the following steps of: performing collaborative initialization on Ns basic optimization algorithms through a hybrid heuristic optimization thread, and performing parallel solution on a target optimization problem based on each algorithm native iteration mechanism; generating feature space data containing algorithm superior communities, and extracting a current optimal solution of each algorithm; constructing a cross-algorithm winning community interaction mechanism, importing the feature space data into a winning community feature library, and outputting an objective function optimization coefficient of the algorithm; and carrying out stability verification and robustness analysis on all objective function optimization coefficients through a multi-objective fusion decision unit, and determining a globally optimal solution. According to the method, multiple heuristic algorithms are combined, random exchange of the individuals of the superior community is achieved, the influence of the defect of a single algorithm on a solving result is avoided to the maximum extent, and fast initialization, parallel solving and iteration of multi-algorithm collaboration can be achieved.

Description

Technical Field

[0001] The present invention relates to a method for collaborative optimization of control parameters based on hybrid heuristic optimization, and realizes the optimization of control parameters through computer technology. Background Art

[0002] The problem of optimizing and solving control parameters in an MPC controller is actually the solution of a constrained optimization problem. When both the objective function and the constraint conditions are linear functions, the optimization problem is a linear programming problem. When the objective function J and the constraint conditions are not all linear functions, the optimization problem is a non-linear programming problem. Most of the models involved in the optimal operation of an air-conditioning system are non-linear models, and the objective function established based on the non-linear model is also a non-linear function. Therefore, the solution of control parameters is the solution of a non-linear constrained optimization problem. The commonly used methods for solving non-linear constrained optimization problems can be roughly divided into two categories: traditional optimization methods and intelligent optimization methods. The specific optimization algorithms, advantages and disadvantages of each method are shown in Table 1.

[0003] Table 1 Methods for Solving Non-linear Constrained Optimization Problems

[0004]

[0005] As can be seen from Table 1, traditional optimization methods have problems such as low operation efficiency and being prone to falling into local optima. At the same time, traditional optimization methods are difficult to handle problems such as non-convex feasible regions, having multiple local extrema, discontinuous feasible regions, and all or part of the variables being discrete or integer. Due to its powerful global search ability and advantages in dealing with non-linear problems and complex non-convex optimization problems, intelligent optimization methods have currently been widely applied in actual production processes and achieved good results.

[0006] Among intelligent optimization methods, heuristic algorithms are relatively widely used in current research and applications. Heuristic algorithms are a class of problem-solving methods based on experience and intuition, and guide the optimization and decision-making processes through heuristic rules, strategies, and search techniques. Common heuristic algorithms include genetic algorithms, ant colony algorithms, simulated annealing algorithms, particle swarm algorithms, etc. However, there are always some inherent disadvantages when applying a single heuristic algorithm. For example, the genetic algorithm has poor local search ability and is prone to premature convergence; the ant colony algorithm has a long search time and the algorithm results are sensitive to parameters. The invention patent with the publication number of CN116300430B discloses an MPC control parameter optimization method and its application in a parallel platform. By establishing an evaluation function for MPC control parameter optimization and using a real-number-coded differential evolution algorithm to perform meta-heuristic search in the parameter space, parameter optimization is completed, and the model predictive controller can obtain better control effects. However, this method is still prone to falling into local search traps. Summary of the Invention

[0007] The object of the present invention is to overcome the deficiencies of the prior art and provide a collaborative optimization method for control parameters based on hybrid heuristic optimization, which combines computer technology to avoid the influence of the defects of each algorithm on the solution results to the greatest extent and improve the probability of finding the global optimal solution. The core of this method lies in randomly exchanging individuals in the winning communities of different algorithms for optimization and solution.

[0008] The object of the present invention is achieved by the following technical solutions:

[0009] A collaborative optimization method for control parameters based on hybrid heuristic optimization, the method comprising:

[0010] Collaboratively initialize Ns basic optimization algorithms through a hybrid heuristic optimization thread, and perform parallel solution of the target optimization problem based on the native iteration mechanism of each algorithm;

[0011] When any algorithm satisfies the first convergence condition or the initial iteration number iter1 reaches the Maxiter1 threshold, generate the characteristic space data containing the winning community of the algorithm, and extract the current optimal solution of each algorithm;

[0012] Construct an interactive mechanism for cross-algorithm winning communities, import the characteristic space data into the winning community feature library, and perform feature sampling and cross-population embedding operations from different winning communities based on the population diversity characteristic coefficient;

[0013] When a single algorithm satisfies the second convergence condition or the total iteration number iter2 reaches the Maxiter2 threshold, output the optimization coefficient of the objective function of the algorithm;

[0014] Perform stability verification and robustness analysis on all objective function optimization coefficients through a multi-objective fusion decision unit to determine the global optimal solution.

[0015] As a preferred method, the hybrid heuristic optimization thread is constructed by injecting a population characteristic migration mechanism and a parameter adaptive adjustment strategy into a benchmark optimization thread, and the benchmark optimization thread includes the native operation operators of genetic algorithm, particle swarm optimization algorithm and firefly swarm optimization algorithm;

[0016] The cross-algorithm winning community interactive mechanism includes:

[0017] Use a feature vector fusion unit to perform multi-dimensional feature splicing on the characteristic space data to generate a winning community feature vector with cross-algorithm relevance;

[0018] Calculate the population diversity characteristic coefficient based on the winning community feature vector, and determine the sampling ratio and feature stratification interval of cross-population embedding according to this coefficient;

[0019] Among them, the construction method of the parameter adaptive adjustment strategy includes:

[0020] Monitoring the change rate of the objective function value of each algorithm within a continuous preset iteration window;

[0021] When the change rate is lower than the dynamic adjustment threshold, trigger the parameter perturbation module to perform non-linear adjustment on the current algorithm core parameters:

[0022] Perform an exponential amplification operation on the mutation probability of the genetic algorithm;

[0023] Perform a piecewise decay operation on the inertia weight of the particle swarm algorithm;

[0024] Perform an adaptive recalibration operation on the luciferin coefficient of the firefly algorithm.

[0025] As a preferred method, the specific manifestations of the native operation operators of the benchmark optimization thread including the genetic algorithm, particle swarm optimization algorithm, and firefly swarm optimization algorithm are:

[0026] The genetic algorithm adopts a chromosome coding mechanism, and its operation operators include a gene recombination unit based on the roulette wheel selection operator, a population evolution unit of the single-point crossover operator, and a gene perturbation unit of the basic bit mutation operator;

[0027] The particle swarm optimization algorithm adopts a velocity-position update model, and its operation operators include an individual optimal position tracking unit and a global optimal position fusion unit;

[0028] The firefly swarm optimization algorithm adopts a brightness attraction mechanism, and its operation operators include a relative brightness calculation unit and a dynamic distance response unit.

[0029] As a preferred method, the initialization process of the benchmark optimization thread includes:

[0030] Genetic algorithm dimension: Set a gene coding parameter space including population size, chromosome length, crossover probability threshold, and mutation probability threshold, and generate an initial population that satisfies a uniform distribution within the feasible domain of the control parameters;

[0031] Particle swarm algorithm dimension: Define a kinematic parameter space including the number of particles, learning factor matrix, and inertia weight curve, and generate a position matrix with random initial velocities within the boundary of the feasible domain of the control parameters;

[0032] Firefly algorithm dimension: Construct a bio-inspired parameter space including the number of fireflies, attribute dimensions, and search space topological structure, and generate an initial position tensor that satisfies a Gaussian distribution within a predefined multi-dimensional feasible domain.

[0033] As a preferred method, the iterative evolution process of the native operation operators includes:

[0034] The genetic algorithm executes a three-stage evolutionary process of gene selection - crossover - mutation, and drives the iterative update of the population through the fitness function evaluation unit;

[0035] The particle swarm algorithm executes a two-stage motion process of velocity vector update - position vector correction, and realizes information interaction through the individual historical optimal solution memory unit and the global optimal solution sharing unit;

[0036] The firefly algorithm executes a three-stage bionic process of brightness calculation - position attraction - parameter update, and realizes the evolution of swarm intelligence through the relative distance calculation unit and the attraction intensity adjustment unit.

[0037] As a preferred method, the construction method of the superior community feature library includes:

[0038] Establish multi-dimensional feature labels for each superior community, including algorithm identifiers, iteration stage feature codes, target function value change gradients, and population diversity indices;

[0039] Adopt a feature coding mechanism to convert the multi-dimensional feature labels into traceable feature vectors, and construct a superior community feature map with spatio-temporal correlation.

[0040] As a preferred method, the feature sampling and cross-population embedding operations specifically include:

[0041] Determine the feature extraction amount of each superior community based on the dynamic sampling ratio calculation module, and the dynamic sampling ratio is positively correlated with the population diversity index;

[0042] Adopt a hierarchical feature sampling strategy, divide the superior communities into multiple feature sub-spaces according to the fitness distribution, and perform adaptive sampling based on probability density within each sub-space;

[0043] Inject the extracted feature samples into the target population through the random position insertion algorithm, and the random position insertion algorithm includes a gene sequence recombination verification unit and a population capacity balance unit.

[0044] As a preferred method, the parameter perturbation module includes:

[0045] The genetic algorithm mutation enhancement unit, which adopts an exponential mutation probability function;

[0046] The particle swarm inertia weight adjustment unit, which adopts a piecewise linear attenuation function;

[0047] The firefly luciferin adjustment unit, which adopts an adaptive update formula.

[0048] As a preferred method, the determination method of the second convergence condition includes:

[0049] Construct a joint decision-making model for dual convergence criteria. When the total number of iterations iter2 > Miniter2 and the change in the objective function value Δf ≤ ε, trigger the first convergence flag;

[0050] When the continuous iteration times when the algorithm falls into local optimum exceed the preset threshold, trigger the second convergence flag;

[0051] The joint decision-making model for dual convergence criteria uses a weighted voting mechanism to comprehensively judge the termination time of the algorithm.

[0052] As a preferred method, the stability verification and robustness analysis include:

[0053] Establish an evaluation index for the fluctuation degree of the solution space, calculate the standard deviation σ and coefficient of variation CV of each candidate optimal solution in multiple independent runs; construct a multi-dimensional robustness evaluation matrix, including parameter sensitivity analysis, boundary condition testing and noise interference experiments;

[0054] Use the analytic hierarchy process to determine the weight coefficients of each evaluation index, and determine the final optimal solution through weighted comprehensive evaluation.

[0055] As a preferred method, the method further includes an optimization process traceability mechanism:

[0056] Construct an algorithm evolution log database to record multi-dimensional evolution data including population gene distribution, fitness change curve, and parameter adjustment trajectory in real time;

[0057] Design a visual analysis interface to convert the multi-dimensional evolution data into a visual analysis report including a three-dimensional convergence surface, a parameter sensitivity heat map, and a population diversity radar map.

[0058] A control parameter collaborative optimization system based on hybrid heuristic optimization, including:

[0059] A hybrid heuristic optimization engine module, a winning community interaction module, a parameter adaptive adjustment module, a multi-objective decision-making module, and an algorithm basic library module. Among them: the hybrid heuristic optimization engine module is respectively connected to the algorithm basic library module and the winning community interaction module, and transmits algorithm native iteration data and winning community feature space data respectively; the parameter adaptive adjustment module is bidirectionally connected to the hybrid heuristic optimization engine module, and transmits the target function change rate data in real time and feedbacks parameter adjustment instructions; the multi-objective decision-making module receives the optimization coefficient data of all algorithm channels and performs stability verification and robustness analysis;

[0060] The hybrid heuristic optimization engine module includes:

[0061] A collaborative initialization unit: connected to the algorithm basic library module, initialize the parameter space of Ns basic optimization algorithms, and establish parallel solution channels;

[0062] Native Iterative Execution Unit: Built-in genetic algorithm operators, particle swarm optimization operators, and firefly swarm optimization operators to drive each algorithm to perform iterative operations according to the native mechanism;

[0063] Convergence Monitoring Unit: Connected to each algorithm channel, and judges the iteration termination condition through the convergence condition discrimination module and the Maxiter counter module;

[0064] Feature Extraction Unit: Generates feature space data containing the superior communities of the algorithms, and extracts the current optimal solution vectors of each channel; The superior community interaction module includes:

[0065] Feature Vector Fusion Unit: Receives feature space data from each algorithm channel, and performs multi-dimensional feature splicing to generate cross-algorithm associated feature vectors;

[0066] Diversity Analysis Unit: Determines the feature sampling ratio and stratification interval based on the population diversity feature coefficient calculation module;

[0067] Cross-Population Embedding Unit: Connects to the hybrid heuristic optimization engine module, and performs dynamic migration and population recombination operations on the superior feature vectors;

[0068] The parameter adaptive adjustment module includes:

[0069] Iterative Window Monitoring Unit: Statistics the change rate of the objective function values of each algorithm through a sliding time window;

[0070] Perturbation Trigger Unit: When the change rate is lower than the dynamic adjustment threshold, activates the parameter non-linear adjustment channel;

[0071] Algorithm Regulator Group: Includes a genetic algorithm mutation probability amplifier, a particle swarm inertia weight attenuator, and a firefly luciferin calibrator;

[0072] The multi-objective decision-making module includes:

[0073] Stability Verification Unit: Performs robustness tests on each optimization coefficient through the Monte Carlo simulation module;

[0074] Decision Fusion Unit: Adopts the fuzzy comprehensive evaluation method to screen the global optimal solution from the Pareto front solution set;

[0075] Result Output Interface: Generates an analysis report containing the optimized parameter configuration scheme and the objective function response surface;

[0076] The algorithm basic library module stores:

[0077] Genetic Algorithm Core Operator Group: Includes a selection operator library, a crossover operator library, a mutation operator library, and a fitness function library;

[0078] Particle swarm optimization component set: including a velocity update calculator, a position updater, and a neighborhood topology structure template;

[0079] Firefly algorithm function package: covering a luciferin update module, a movement probability calculator, and an attraction calculation unit;

[0080] The feature migration operations performed by the cross-population embedding unit include: calculating the similarity of the feature space based on the Mahalanobis distance; adopting a population recombination strategy using the quantum rotation gate mechanism; and a feature stratified sampling method based on information entropy weight;

[0081] The parameter non-linear adjustment channel includes: an exponential amplification function for the mutation probability of the genetic algorithm; a piecewise decay strategy for the inertia weight of the particle swarm; and an adaptive calibration model for the luciferin coefficient of the firefly.

[0082] An electronic device, the electronic device includes:

[0083] At least one processor; and, a memory communicatively connected to the at least one processor; wherein,

[0084] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the previously described collaborative optimization method for control parameters based on hybrid heuristic optimization.

[0085] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the previously described collaborative optimization method for control parameters based on hybrid heuristic optimization.

[0086] The present invention has at least the following beneficial effects: The core of the present invention is to combine multiple heuristic algorithms and implement random exchange of individuals in the winning communities, which maximally avoids the influence of the defects of a single algorithm on the solution result, thereby significantly increasing the probability of finding the global optimal solution. With the help of computer technology, it is possible to achieve rapid initialization, parallel solution, and iteration of multi-algorithm collaboration. By extracting the features of the winning communities, it is ensured that the exchanged individuals have high value. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to disclose the technical details of the embodiments of the present invention, the drawings involved in the embodiments will be briefly introduced next. It should be emphasized that these drawings only present several embodiments of the present invention and should not be regarded as defining the scope of the invention. For those of ordinary skill in the art, other related drawings can still be derived based on these drawings without creative labor.

[0088] Figure 1 It is a flowchart for optimizing and solving by a hybrid heuristic algorithm;

[0089] Figure 2 It is the calculation flow chart of the genetic algorithm;

[0090] Figure 3 It is the calculation flow chart of the particle swarm optimization algorithm;

[0091] Figure 4 It is the calculation flow chart of the glowworm swarm optimization algorithm. Specific implementation manners

[0092] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following description.

[0093] In the following content, the embodiments of the present disclosure will be described in detail with the aid of the accompanying drawings. However, it should be clear that the present disclosure is not limited to the specific forms shown here. On the contrary, it should be understood to cover various variations, equivalent forms and / or alternative solutions of the embodiments of the present disclosure. In the process of elaborating the drawings, the same reference numerals will be used to label similar components.

[0094] In various embodiments of the present disclosure, expressions such as "first", "second", "the first" or "the second" are used to modify different components, rather than indicating order and / or importance, nor imposing restrictions on the corresponding components. For example, the first user device and the second user device respectively represent different user devices, although they both belong to the category of user devices. Similarly, the first component can be named the second component, and the second component can also be named the first component, which does not change their essential attributes within the scope of the present disclosure.

[0095] In this disclosure, terms are used to elaborate specific embodiments and do not constitute a limitation to the present disclosure. In this context, the use of the singular form also covers the plural form unless otherwise clearly expressed in the text. In the process of elaboration, it should be understood that terms such as "comprising" or "having" are intended to indicate the existence of features, quantities, steps, operations, structural components, parts or combinations thereof, and do not preclude the possibility or addition of one or more other features, quantities, steps, operations, structural components, parts or combinations thereof in advance.

[0096] It should be clear that although the following description provides detailed specific details to help comprehensively understand the example embodiments. However, professionals in the field should know that even without these specific details, the example embodiments can still be implemented. For example, the system can be shown in the form of a block diagram to avoid excessive details interfering with the clarity of the example. In some other cases, in order to maintain the clarity of the example, unnecessary details of those well-known processes, structures and technologies may be omitted.

[0097] A method for collaborative optimization of control parameters based on hybrid heuristic optimization, the method comprising:

[0098] Collaboratively initialize Ns basic optimization algorithms through a hybrid heuristic optimization thread, and perform parallel solution of the target optimization problem based on the native iteration mechanism of each algorithm;

[0099] When any algorithm meets the first convergence condition or the initial iteration number iter1 reaches the Maxiter1 threshold, generate feature space data including the winning communities of the algorithms, and extract the current optimal solutions of each algorithm;

[0100] Construct an interaction mechanism for cross-algorithm winning communities, import the feature space data into a winning community feature library pre-set in a computer, and perform feature sampling and cross-population embedding operations from different winning communities based on the population diversity feature coefficient;

[0101] When a single algorithm meets the second convergence condition or the total iteration number iter2 reaches or is greater than the Maxiter2 threshold, output the optimization coefficient of the objective function of the algorithm;

[0102] Perform stability verification and robustness analysis on all objective function optimization coefficients through a multi-objective fusion decision unit to determine the global optimal solution.

[0103] The hybrid heuristic optimization thread uses computer multi-threading technology to execute different optimization algorithms in parallel in the same program. This can make full use of the multi-core processing power of the computer and improve the execution efficiency of the algorithms. Through collaborative initialization, the parameter space of the algorithms can be quickly initialized, laying a foundation for parallel solution. Through hybrid heuristic optimization and parallel solution, the optimization efficiency and accuracy can be effectively improved, and the calculation cost can be reduced. At the same time, the collaborative optimization method can integrate the advantages of different algorithms and improve the overall optimization performance.

[0104] In a preferred embodiment, the hybrid heuristic optimization thread is constructed by injecting a population feature migration mechanism (or cross-algorithm winning community interaction mechanism) and a parameter adaptive adjustment strategy into a benchmark optimization thread, and the benchmark optimization thread includes the native operation operators of genetic algorithms, particle swarm optimization algorithms, and firefly swarm optimization algorithms; the population feature migration mechanism is an optimization algorithm strategy that extracts features or information from the population of one algorithm (source algorithm) and transfers it to another algorithm (target algorithm) to enhance the search ability and optimization performance of the target algorithm.

[0105] The cross-algorithm winning community interaction mechanism includes:

[0106] Use a feature vector fusion unit to perform multi-dimensional feature splicing on the feature space data to generate a winning community feature vector with cross-algorithm relevance;

[0107] Calculate the population diversity characteristic coefficient based on the eigenvector of the winning community, and determine the sampling ratio and characteristic stratification interval for cross-population embedding according to this coefficient;

[0108] The cross-algorithm winning community interaction mechanism refers to the information exchange and feature sharing among the winning populations of multiple algorithms to promote the collaboration of the entire optimization process and the improvement of the optimization effect, including population feature migration. Through the interaction of the winning populations, the collaborative optimization between algorithms is enhanced, the limitations of a single algorithm are avoided, and the overall optimization performance is improved.

[0109] Among them, the construction method of the parameter adaptive adjustment strategy includes:

[0110] Monitor the change rate of the objective function value of each algorithm within a continuous preset iteration window;

[0111] When the change rate is lower than the dynamic adjustment threshold, trigger the parameter perturbation module to perform non-linear adjustment on the core parameters of the current algorithm:

[0112] Perform an exponential amplification operation on the mutation probability of the genetic algorithm;

[0113] Perform a piecewise decay operation on the inertia weight of the particle swarm algorithm;

[0114] Perform an adaptive recalibration operation on the luciferin coefficient of the firefly algorithm.

[0115] As Figure 1 shown, there are Ns kinds of heuristic algorithms for mixing. Initialize all algorithms at the same time and solve the optimization problem according to the solution process of each algorithm itself. When any one of the algorithms converges or the internal calculation iteration times iter1 of the algorithm exceeds the maximum iteration limit Maxiter1, a winning community is generated, and the optimal solution is calculated at the same time. Extract the winning communities of different algorithms into the winning community library, randomly select individuals from different winning communities and insert them into the random populations of other random algorithms for exchange, and then continue to optimize. When the total iteration times iter2 of a single algorithm is greater than the total minimum iteration times Miniter2 and the change value of the optimal solution of the objective function is less than or equal to the threshold or reaches the maximum iteration times Maxiter2, output the optimal solution of the algorithm. Finally, compare the optimal solutions of all algorithms to obtain the final optimal solution.

[0116] By injecting a mechanism into the benchmark optimization thread, the migration of population characteristics between different algorithms and the dynamic adjustment of parameters are realized. This mechanism can better utilize the advantages of different algorithms and improve the optimization effect. The population characteristic migration and parameter self-adaptive adjustment strategy can improve the adaptability and robustness of the algorithm, enabling the algorithm to better adapt to different optimization problems and environmental changes. At the same time, this strategy can also prevent the algorithm from falling into local optimal solutions and improve the global optimization ability.

[0117] In a preferred embodiment, the native operation operators of the genetic algorithm, particle swarm optimization algorithm, and firefly swarm optimization algorithm are included in the benchmark optimization thread, and the specific manifestations are as follows:

[0118] The genetic algorithm adopts a chromosome coding mechanism, and its operation operators include a gene recombination unit based on the roulette wheel selection operator, a population evolution unit of the single-point crossover operator, and a gene perturbation unit of the basic bit mutation operator;

[0119] The particle swarm optimization algorithm adopts a velocity-position update model, and its operation operators include an individual optimal position tracking unit and a global optimal position fusion unit;

[0120] The firefly swarm optimization algorithm adopts a brightness attraction mechanism, and its operation operators include a relative brightness calculation unit and a dynamic distance response unit.

[0121] The benchmark optimization thread includes the native operation operators of the genetic algorithm, particle swarm optimization algorithm, and firefly swarm optimization algorithm. Among them, the genetic algorithm adopts a chromosome coding mechanism and realizes evolution through operations such as gene recombination and population evolution. The particle swarm optimization algorithm adopts a model of velocity and position update and realizes optimization through the interaction of individual and global optimal positions. The firefly swarm optimization algorithm adopts a brightness attraction mechanism and realizes optimization by calculating relative brightness and dynamic distance response. By combining the advantages of different algorithms, this method can search the solution space more comprehensively and improve the accuracy and efficiency of optimization. At the same time, the use of the native operation operators of each algorithm ensures the stability and reliability of the algorithm.

[0122] In a preferred embodiment,

[0123] The initialization process of the benchmark optimization thread includes:

[0124] Genetic algorithm dimension: Set a gene coding parameter space including population size, chromosome length, crossover probability threshold, and mutation probability threshold, and generate an initial population that satisfies a uniform distribution within the feasible domain of control parameters;

[0125] Particle swarm algorithm dimension: Define a kinematic parameter space including the number of particles, learning factor matrix, and inertia weight curve, and generate a position matrix with random initial velocities within the boundary of the feasible domain of control parameters;

[0126] Firefly algorithm dimension: Construct a biologically inspired parameter space including the number of fireflies, attribute dimensions, and search space topology, and generate an initial position tensor that satisfies the Gaussian distribution within a predefined multi-dimensional feasible region.

[0127] In a preferred embodiment,

[0128] The iterative evolution process of the native operation operator includes:

[0129] The genetic algorithm executes a three-stage evolution process of gene selection - crossover - mutation, and drives the population to iteratively update through the fitness function evaluation unit;

[0130] The particle swarm algorithm executes a two-stage motion process of velocity vector update - position vector correction, and realizes information interaction through the individual historical best solution memory unit and the global best solution sharing unit;

[0131] The firefly algorithm executes a three-stage bionic process of brightness calculation - position attraction - parameter update, and realizes swarm intelligence evolution through the relative distance calculation unit and the attraction intensity adjustment unit.

[0132] The iterative evolution process of the native operation operator includes the three stages of gene selection - crossover - mutation of the genetic algorithm, the two stages of velocity vector update - position vector correction of the particle swarm algorithm, and the three stages of brightness calculation - position attraction - parameter update of the firefly algorithm. These processes drive the population to iteratively update through the fitness function evaluation unit, realizing the self-optimization and evolution of the algorithm. Through clear evolution processes and operation operators, the optimization process of the algorithm becomes clearer and more controllable. At the same time, through the iterative update of the population, the adaptability and robustness of the algorithm are enhanced.

[0133] In a preferred embodiment,

[0134] The method for constructing the excellent community feature library includes:

[0135] Establish multi-dimensional feature labels for each excellent community, including algorithm identifiers, iteration stage feature codes, target function value change gradients, and population diversity indices;

[0136] The multi-dimensional feature tags are converted into traceable feature vectors by using a feature encoding mechanism, and a superior community feature map with spatio-temporal correlation is constructed. The construction method of the superior community feature library includes establishing multi-dimensional feature tags for each superior community, and these tags contain information such as algorithm identifiers, iteration stage feature codes, change gradients of objective function values, and population diversity indices. Then, the feature encoding mechanism is used to convert these multi-dimensional feature tags into traceable feature vectors, and a superior community feature map with spatio-temporal correlation is constructed. By constructing the superior community feature library, this method can record and trace the feature information of each superior community, providing a basis for subsequent cross-population embedding operations. At the same time, the feature map with spatio-temporal correlation can also better reflect the evolution and interaction process of the superior community.

[0137] In a preferred embodiment, the feature sampling and cross-population embedding operations specifically include:

[0138] Based on the dynamic sampling ratio calculation module, determine the feature extraction amount of each superior community, and the dynamic sampling ratio is positively correlated with the population diversity index;

[0139] Adopt a hierarchical feature sampling strategy, divide the superior community into multiple feature sub-spaces according to the fitness distribution, and perform adaptive sampling based on probability density within each sub-space;

[0140] Inject the extracted feature samples into the target population through a random position insertion algorithm, and the random position insertion algorithm includes a gene sequence recombination verification unit and a population capacity balance unit. The feature sampling and cross-population embedding operations include determining the feature extraction amount of each superior community based on the dynamic sampling ratio calculation module, dividing the superior community into multiple feature sub-spaces according to the fitness distribution by using a hierarchical feature sampling strategy, and performing adaptive sampling based on probability density within each sub-space. Then, inject the extracted feature samples into the target population through the random position insertion algorithm, realizing the migration of features and the recombination of populations. The dynamic sampling and hierarchical feature sampling strategies realize the effective extraction and utilization of the features of the superior community. At the same time, injecting the feature samples into the target population through the random position insertion algorithm promotes the improvement of population diversity and intelligence. This method also accelerates the convergence speed and accuracy of the optimization process.

[0141] In a preferred embodiment,

[0142] The parameter perturbation module includes:

[0143] A genetic algorithm mutation enhancement unit, adopting an exponential mutation probability function;

[0144] A particle swarm inertia weight adjustment unit, adopting a piecewise linear attenuation function;

[0145] A firefly luciferin adjustment unit, adopting an adaptive update formula.

[0146] In a preferred embodiment,

[0147] The method for determining the second convergence condition includes:

[0148] Construct a joint decision-making model for dual convergence criteria. When the total number of iterations iter2 > Miniter2 and the change in the objective function value Δf ≤ ε, trigger the first convergence flag; the first convergence flag ensures timely termination when sufficient accuracy is achieved, avoiding resource waste. The first convergence flag can be used to initiate feature migration.

[0149] When the continuous iteration count of the algorithm falling into the local optimum exceeds a preset threshold, trigger the second convergence flag; the second convergence flag serves as a fallback condition to prevent the algorithm from being unable to terminate due to complex problems. The second convergence flag can be used to trigger the termination process.

[0150] The joint decision-making model for dual convergence criteria uses a weighted voting mechanism to comprehensively judge the algorithm termination timing. The weight of the first convergence flag can be set to 0.4, and the weight of the second convergence flag can be set to 0.6. By combining the two termination conditions, the algorithm can be terminated when the conditions are met, avoiding wasting time on unnecessary iterations.

[0151] In a preferred embodiment,

[0152] The stability verification and robustness analysis include:

[0153] Establish an evaluation index for the fluctuation degree of the solution space, calculate the standard deviation σ and coefficient of variation CV of each candidate optimal solution in multiple independent runs; construct a multi-dimensional robustness evaluation matrix, including parameter sensitivity analysis, boundary condition testing, and noise interference experiments;

[0154] Use the analytic hierarchy process to determine the weight coefficients of each evaluation index, and determine the final optimal solution through weighted comprehensive evaluation.

[0155] In a preferred embodiment,

[0156] The method further includes an optimization process traceability mechanism:

[0157] Construct an algorithm evolution log database to record multi-dimensional evolution data including population gene distribution, fitness change curve, and parameter adjustment trajectory in real time;

[0158] Design a visual analysis interface to convert the multi-dimensional evolution data into a visual analysis report including a three-dimensional convergence surface, a parameter sensitivity heat map, and a population diversity radar map.

[0159] A control parameter collaborative optimization system based on hybrid heuristic optimization includes:

[0160] A hybrid heuristic optimization engine module, an elite community interaction module, a parameter self-adaptive adjustment module, a multi-objective decision-making module, and an algorithm basic library module, where: The hybrid heuristic optimization engine module is respectively connected to the algorithm basic library module and the elite community interaction module, and transmits algorithm native iteration data and elite community feature space data respectively; The parameter self-adaptive adjustment module is bidirectionally connected to the hybrid heuristic optimization engine module, and transmits the objective function change rate data in real time and feeds back parameter adjustment instructions; The multi-objective decision-making module receives the optimization coefficient data of all algorithm channels and performs stability verification and robustness analysis;

[0161] The hybrid heuristic optimization engine module includes:

[0162] A collaborative initialization unit: connected to the algorithm basic library module, initializes the parameter space of Ns basic optimization algorithms, and establishes parallel solution channels;

[0163] A native iteration execution unit: built-in genetic algorithm operators, particle swarm optimization operators, and firefly swarm optimization operators, driving each algorithm to perform iterative operations according to the native mechanism;

[0164] A convergence monitoring unit: connected to each algorithm channel, judging the iteration termination condition through a convergence condition discrimination module and a Maxiter counter module;

[0165] A feature extraction unit: generates feature space data containing the elite community of the algorithm, and extracts the current optimal solution vector of each channel;

[0166] The elite community interaction module includes:

[0167] A feature vector fusion unit: receives the feature space data from each algorithm channel, and performs multi-dimensional feature splicing to generate a cross-algorithm associated feature vector;

[0168] A diversity analysis unit: determines the feature sampling ratio and stratification interval based on the population diversity feature coefficient calculation module;

[0169] A cross-population embedding unit: connected to the hybrid heuristic optimization engine module, performs dynamic migration and population recombination operations on the elite feature vectors;

[0170] The parameter self-adaptive adjustment module includes:

[0171] An iteration window monitoring unit: statistically calculates the change rate of the objective function value of each algorithm through a sliding time window;

[0172] A perturbation trigger unit: when the change rate is lower than the dynamic adjustment threshold, activates the parameter non-linear adjustment channel;

[0173] An algorithm regulator group: includes a genetic algorithm mutation probability amplifier, a particle swarm inertia weight attenuator, and a firefly luciferin calibrator;

[0174] The multi-objective decision-making module includes:

[0175] A stability verification unit: performing a robustness test on each optimization coefficient through the Monte Carlo simulation module;

[0176] A decision fusion unit: screening for the global optimal solution of the Pareto front solution set using the fuzzy comprehensive evaluation method;

[0177] A result output interface: generating an analysis report containing the optimized parameter configuration scheme and the response surface of the objective function;

[0178] The algorithm basic library module stores:

[0179] A genetic algorithm core operator group: including a selection operator library, a crossover operator library, a mutation operator library, and a fitness function library;

[0180] A particle swarm optimization component set: including a velocity update calculator, a position updater, and a neighborhood topology structure template;

[0181] A firefly algorithm function package: covering a luciferin update module, a movement probability calculator, and an attraction calculation unit;

[0182] The feature migration operations performed by the cross-population embedding unit include: calculating the similarity of the feature space based on the Mahalanobis distance; a population recombination strategy using the quantum rotation gate mechanism; a feature stratified sampling method based on the information entropy weight;

[0183] The parameter non-linear adjustment channel includes: an exponential amplification function for the mutation probability of the genetic algorithm; a piecewise decay strategy for the inertia weight of the particle swarm; an adaptive calibration model for the firefly luciferin coefficient.

[0184] This system is based on a hybrid heuristic optimization method, integrating multiple basic optimization algorithms, and establishing a parallel solution channel through a collaborative initialization unit. In the native iterative execution unit, genetic algorithm operators, particle swarm optimization operators, and firefly swarm optimization operators are built-in to drive each algorithm to perform iterative operations according to its native mechanism. The convergence monitoring unit is responsible for monitoring the iterative process of each algorithm, and judging the iteration termination condition through the convergence condition discrimination module and the Maxiter counter module. The feature extraction unit generates feature space data containing the superior communities of the algorithms, and extracts the current optimal solution vectors of each channel. These feature space data are transmitted to the superior community interaction module, where the feature vector fusion unit performs multi-dimensional feature splicing to generate cross-algorithm correlation feature vectors. The diversity analysis unit determines the feature sampling ratio and stratification interval based on the population diversity feature coefficient calculation module, while the cross-population embedding unit performs dynamic migration and population recombination operations on the superior feature vectors to enhance the search ability of the algorithm and avoid premature convergence. The parameter adaptive adjustment module calculates the change rate of the objective function values of each algorithm through the iteration window monitoring unit. When the change rate is lower than the dynamic adjustment threshold, the perturbation trigger unit activates the parameter non-linear adjustment channel. The algorithm regulator group includes regulators for different algorithms, such as the genetic algorithm mutation probability amplifier, the particle swarm inertia weight attenuator, and the firefly luciferin calibrator, which are used to adjust the algorithm parameters in real time to optimize the search process. The multi-objective decision-making module receives the optimization coefficient data of all algorithm channels, and the stability verification unit performs a robustness test on each optimization coefficient through the Monte Carlo simulation module. The decision fusion unit uses the fuzzy comprehensive evaluation method to screen the global optimal solution from the Pareto front solution set, and finally generates an analysis report containing the optimized parameter configuration scheme and the objective function response surface through the result output interface.

[0185] The present invention selects the genetic algorithm, the particle swarm optimization algorithm, and the firefly swarm optimization algorithm as the basic algorithms to establish a hybrid heuristic optimization solution algorithm. The basic principles and calculation processes of each basic algorithm are described as follows.

[0186] (1) Genetic Algorithm

[0187] The genetic algorithm (Genetic Algorithm, GA) is an adaptive global optimization algorithm that simulates the inheritance and evolution of organisms in the natural environment. Its basic idea follows the principle of "natural selection, survival of the fittest" in biological evolution theory. The algorithm mainly represents individuals through coding, evaluates the quality of individuals using the fitness function, and uses evolutionary operations such as selection, crossover, and mutation to achieve optimization search.

[0188] The calculation flow chart of the genetic algorithm is as Figure 2 shown.

[0189] (2) Particle Swarm Optimization Algorithm

[0190] The Particle Swarm Optimization (PSO) is a swarm intelligence optimization algorithm proposed based on the foraging behavior rules of bird flocks. In the algorithm, the search space of the optimization problem is analogized to the flight space of birds, and the optimal solution to be found is analogized to the food to be searched. Each bird is abstracted as a massless particle, and the particle contains two feature vectors: position and velocity. During the optimization process, the position and velocity of each particle are first randomly initialized, and then the position and velocity are updated iteratively.

[0191] The calculation flow chart of the Particle Swarm Optimization algorithm is as Figure 3 shown.

[0192] (3) Glowworm Swarm Optimization Algorithm

[0193] The Glowworm Swarm Optimization Algorithm (GSOA) was proposed by Krishnanand et al. based on the principle of glowworm luminescence. In the basic Glowworm Swarm Optimization Algorithm, each glowworm is distributed within the definition space of the objective function. Glowworms carry their own luciferin and have their own field of vision. The field of vision is called the regional decision range, and its size is affected by the number of surrounding glowworms. When the density of surrounding glowworms is low, the regional decision radius will increase to facilitate the search for more glowworms; conversely, the decision radius will decrease. Eventually, the glowworms will gather around the glowworm with the highest fitness value, thereby determining the optimal value of the objective function.

[0194] The calculation flow chart of the Glowworm Optimization Algorithm is as Figure 4 shown.

[0195] In one embodiment, a method for collaborative optimization of control parameters of a subway station air conditioning system based on hybrid heuristic optimization (method for collaborative optimization of control parameters of a subway station air conditioning system) collaboratively initializes multiple basic optimization algorithms (including genetic algorithms, particle swarm optimization algorithms, and firefly algorithms) through a hybrid heuristic optimization thread, and performs parallel solutions for key control parameters of the subway station air conditioning system such as outlet air temperature, wind speed, and humidity.

[0196] When any algorithm meets the first convergence condition or the initial iteration number iter1 reaches the set threshold Maxiter1, characteristic space data containing the winning community is generated, and the optimal settings of the outlet air temperature and wind speed in the current winning solution are extracted. The combination of these parameters directly affects the energy consumption and passenger comfort of the air conditioning system.

[0197] Next, construct an interaction mechanism for the winning communities across algorithms. Use the feature vector fusion unit to perform multi-dimensional feature splicing on the feature space data to form a winning community feature vector with cross-algorithm relevance. In this process, integrate the advantages of each algorithm in optimizing the supply air temperature and wind speed control, enhance the complementarity of the overall parameters, and use this feature vector to calculate the population diversity characteristic coefficient, so as to determine the sampling ratio and feature stratification interval for cross-population embedding, and thus more effectively adjust the relevant control parameters.

[0198] When a certain algorithm meets the second convergence condition or the total number of iterations iter2 reaches another set threshold Maxiter2, output the optimized coefficients of the objective function of this algorithm, such as the energy consumption ratio and the comfort index, which clearly reflect the optimal combination strategy of the supply air temperature and wind speed through these coefficients. Subsequently, use the multi-objective fusion decision-making unit to perform stability verification and robustness analysis on all the optimized coefficients of the objective function to ensure that the key parameters such as the supply air temperature, wind speed, and humidity can maximize the comfort of passengers and the energy efficiency of the system in the actual operation environment of the subway station.

[0199] To enhance the combination effect of the parameters, a population feature migration mechanism and a parameter adaptive adjustment strategy are injected into the method, and the change rates of the objective function values of each algorithm regarding the supply air temperature, wind speed, and humidity within the specified iteration window are monitored in real time. When the change rate is lower than the dynamic adjustment threshold, trigger the parameter perturbation module to perform non-linear adjustment on the core parameters, including dynamically optimizing the mutation probability of the genetic algorithm, segmentally decaying the inertia weight of the particle swarm optimization algorithm, and adaptively recalibrating the luciferin coefficient of the firefly algorithm, to ensure the optimal adjustment of the supply air temperature and wind speed during the iteration process.

[0200] Through the collaborative action of each link, the overall solution deepens the integration and optimization of the control parameters in the subway station air conditioning system to achieve the goals of reducing energy consumption, improving passenger comfort, and strengthening system intelligent management.

[0201] In another embodiment, we provide another solution.

[0202] A method for collaborative optimization of subway station air conditioning system parameters based on hybrid heuristic optimization (method for collaborative optimization of subway station air conditioning system parameters), the method comprising:

[0203] Perform collaborative initialization on the genetic algorithm, particle swarm optimization algorithm, and firefly swarm optimization algorithm through the hybrid heuristic optimization thread, and perform parallel solution on the multi-objective optimization problem of energy efficiency - comfort of the air conditioning system based on the native iteration mechanism of each algorithm, where the optimization parameters include the supply air temperature set value, fan frequency, chilled water valve opening, and fresh air ratio;

[0204] When any algorithm meets the first convergence condition or the initial iteration number iter1 reaches the Maxiter1 threshold, characteristic space data including temperature field distribution characteristics, equipment energy consumption characteristics, and passenger density characteristics is generated, and the current optimal equipment control parameter combination of each algorithm is extracted;

[0205] Construct a cross-algorithm superior community interaction mechanism, import the characteristic space data into the superior community characteristic library containing historical operating condition characteristics, calculate the population diversity characteristic coefficient based on the heat load fluctuation coefficient and the passenger density change rate, and accordingly perform the sampling of air supply parameters and the cross-population embedding operation between different superior communities;

[0206] When a single algorithm meets the second convergence condition or the total iteration number iter2 reaches the Maxiter2 threshold, the equipment energy efficiency optimization coefficient and the temperature uniformity index of the algorithm are output;

[0207] Through the multi-objective fusion decision-making unit, the stability verification of all optimization coefficients under dynamic load conditions is carried out, and the global optimal solution is determined based on the energy consumption fluctuation tolerance and the temperature deviation threshold. The real-time meteorological parameters and passenger flow density monitoring data are introduced in the verification process;

[0208] The hybrid heuristic optimization thread is constructed by injecting an equipment parameter migration mechanism and an environmental parameter adaptive strategy, where:

[0209] The parameter migration mechanism realizes the migration and recombination of the fan frequency parameter matrix between different algorithms. The environmental parameter adaptive strategy includes:

[0210] Monitor the change rate of the COP value of each algorithm within 3 consecutive iteration windows. When the change rate is lower than the dynamic adjustment threshold, trigger:

[0211] Perform an exponential amplification operation on the mutation probability of the supply air temperature of the genetic algorithm, and the amplification coefficient is positively correlated with the real-time passenger flow density;

[0212] Perform a piecewise attenuation operation on the inertia weight of the particle swarm algorithm, and the attenuation rate is linked to the change rate of outdoor temperature and humidity;

[0213] Perform an adaptive recalibration operation on the luciferin coefficient of the firefly algorithm, and the calibration reference value takes the optimal value of the same period in the current period of history.

[0214] The present invention also provides a method for collaborative optimization of subway station air-conditioning refrigeration control parameters based on hybrid heuristic optimization, that is, a method for collaborative optimization of subway station air-conditioning refrigeration control parameters. The method includes the following steps:

[0215] Initialization stage: Parameterize the key control parameters of the subway station air-conditioning refrigeration system, including the following parameters:

[0216] Temperature setpoint (Tset )

[0217] Humidity set point (H set )

[0218] Supply air volume (Q air )

[0219] Chiller operating frequency (f chiller )

[0220] Compressor operating pressure (P compressor )

[0221] Hybrid heuristic optimization thread construction: Based on genetic algorithm, particle swarm optimization algorithm (PSO) and firefly swarm optimization algorithm (FA), initialize the optimization population and encode the control parameters of the air-conditioning system as optimization variables. Set the initial population size and parameter encoding range, and establish a population characteristic migration mechanism and a parameter adaptive adjustment strategy.

[0222] Parallel solution and collaborative optimization:

[0223] Each optimization algorithm performs parallel solution on the control parameters of the air-conditioning system based on its native iteration mechanism (such as gene recombination of genetic algorithm, speed and position update of PSO, and brightness attraction mechanism of FA).

[0224] When any algorithm meets the first convergence condition (such as the change rate of the objective function value is lower than the threshold or the number of iterations reaches the preset value), extract the current optimal control parameter configuration of this algorithm, that is, extract the winning population characteristic data of this algorithm, and generate a multi-dimensional characteristic vector containing the current optimal solution of the air-conditioning system (such as the optimal temperature set point and the optimal supply air volume).

[0225] Cross-algorithm winning community interaction and embedding (cross-algorithm winning community interaction mechanism):

[0226] Import the winning community characteristic data of each algorithm into the winning community characteristic library, use the feature vector fusion unit to perform multi-dimensional feature splicing on the feature space data, generate a winning community characteristic vector with cross-algorithm relevance, calculate the population diversity characteristic coefficient based on the winning community characteristic vector, and dynamically determine the sampling ratio and feature stratification interval of cross-population embedding according to this coefficient.

[0227] Dynamically adjust the feature sampling ratio according to the population diversity coefficient, and extract representative control parameter configurations from the winning community. Use the random position insertion algorithm to inject the extracted control parameter configurations into the target algorithm population to achieve cross-algorithm knowledge transfer and the improvement of population diversity.

[0228] Parameter Adaptive Adjustment Strategy: Monitor the change rate of the objective function value of each algorithm within a continuous preset iteration window. When the change rate is lower than the dynamic adjustment threshold, trigger the parameter perturbation module to perform non-linear adjustment on the current core parameters of the algorithm.

[0229] Optimization Termination and Global Optimal Solution Confirmation:

[0230] When a single algorithm reaches Maxiter2 in the total number of iterations iter2 or the change amount of the objective function value Δf ≤ ε, where ε is the convergence threshold, output the optimal control parameter configuration of the algorithm.

[0231] Through the multi-objective fusion decision-making unit, perform stability verification and robustness analysis on the optimal control parameter configurations of each algorithm (such as standard deviation < 0.15), and finally determine the globally optimal combination of subway station air-conditioning control parameters to ensure the efficient operation of the system and the comfortable experience of passengers.

[0232] In another embodiment, another method for collaborative optimization of temperature control parameters of a subway station air-conditioning system based on hybrid heuristic optimization is provided, that is, a method for collaborative optimization of air-conditioning refrigeration control parameters.

[0233] Controlled Parameters: Platform Hall Ambient Temperature (Target Parameter: 26 ± 1°C, Constraint Range: 24 - 28°C), Air-Conditioning Supply Air Volume (Target Parameter: Dynamically Adjusted, Constraint Range: 2000 - 5000 m 3 / h)

[0234] Hybrid Optimization Thread Initialization and Parameter Modeling

[0235] Benchmark Optimization Thread Construction

[0236] Genetic Algorithm (GA) Dimension

[0237] Control Parameters: Air-Conditioning Start / Stop Threshold, Supply Air Volume Adjustment Step

[0238] Initialization: Population Size = 50, Chromosome Encoding Length = 10 (5-bit Temperature Parameter + 5-bit Air Volume Parameter), Crossover Probability = 0.8, Mutation Probability = 0.05

[0239] Particle Swarm Optimization (PSO) Dimension

[0240] Control Parameters: Temperature PID Coefficients (Kp, Ki, Kd), Supply Air Volume Change Rate

[0241] Initialization: Number of Particles = 30, Learning Factor c1 = 1.5 (Temperature Parameter Weight), c2 = 1.2 (Air Volume Parameter Weight), Inertia Weight ω = 0.9

[0242] Firefly Algorithm (FA) Dimension

[0243] Control parameters: temperature setpoint offset, dynamic response time of air supply volume

[0244] Initialization: number of fireflies = 40, luciferin decay coefficient = 0.4, attraction coefficient β = 0.3

[0245] Comprehensive optimization objective:

[0246]

[0247] Treal: actual temperature, Tset: set temperature (26°C)

[0248] Qactual: actual air supply volume, Qmax: maximum air supply volume (5000m 3 / h)

[0249] Epower: real-time power of air conditioning system (kW)

[0250] Weight coefficients: α = 0.6 (temperature priority), β = 0.3 (air volume efficiency), γ = 0.1 (energy consumption)

[0251] Interaction and parameter optimization of superior communities

[0252] Cross-algorithm feature migration

[0253] Extraction of GA superior community: select individuals with the top 10% fitness values (temperature fluctuation < 0.5°C and optimal air volume adjustment step)

[0254] Extraction of PSO superior community: select the global optimal particle (PID parameter combination with temperature overshoot < 5%)

[0255] Extraction of FA superior community: screen the firefly with the highest brightness value (air supply response time < 30 seconds)

[0256] Feature embedding and parameter adjustment

[0257] GA→PSO migration: insert the gene sequence of the air supply volume adjustment step of GA into the PSO particle velocity update formula.

[0258] FA→GA migration: introduce the temperature offset parameter of FA as a mutation operator into the GA chromosome.

[0259] Parameter adaptive adjustment strategy

[0260] Trigger condition: temperature fluctuation > 0.8°C for 5 consecutive iterations or air supply volume exceeding the limit

[0261] Genetic algorithm: the mutation probability is adjusted according to exponential growth

[0262] where t is the number of consecutive ineffective iterations;

[0263] Particle Swarm Optimization: Piecewise Decay of Weights

[0264] Firefly Algorithm: Dynamic Re - calibration of Luciferin Coefficient

[0265] Convergence Criterion and Generation of Global Optimal Solution

[0266] Convergence Conditions:

[0267] The First Convergence Flag: The total number of iterations > 50 and the change in the objective function value Δf ≤ 0.01

[0268] The Second Convergence Flag: The air supply volume has not exceeded the constraint range for 10 consecutive iterations and the standard deviation of temperature < 0.3°C

[0269] Verification of Global Optimal Solution:

[0270] Stability Analysis: Select 3 groups of candidate solutions (temperature - air volume parameter combinations) and conduct 72 - hour on - site measurements during the evening rush hour (passenger flow of 2000 people / h) at a certain subway station:

[0271]

[0272] After robustness testing, the requirements for temperature and air supply volume are met.

[0273] Although the preferred embodiments of the present invention have been described in detail, those skilled in the art can still make further changes and modifications to these embodiments after grasping the basic creative concepts. Therefore, the appended claims are intended to cover these preferred embodiments and all changes and modifications that fall within the scope of the present invention. The above - mentioned are only the preferred embodiments of the present invention and are not used to limit its scope. It should be noted that any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A control parameter collaborative optimization method based on hybrid heuristic optimization, characterized in that: The method comprises: The Ns basic optimization algorithms are collaboratively initialized through hybrid heuristic optimization threads, and the target optimization problem is solved in parallel based on the native iteration mechanism of each algorithm; When any algorithm meets the first convergence condition or the initial number of iterations iter1 reaches the Maxiter1 threshold, feature space data containing the winning community of the algorithm is generated, and the current optimal solution of each algorithm is extracted; Constructing a cross-algorithm winning community interaction mechanism, importing the feature space data into the winning community feature library, and performing feature sampling and cross-population embedding operations from different winning communities based on the population diversity feature coefficient; When a single algorithm meets the second convergence condition or the total number of iterations iter2 reaches the Maxiter2 threshold, the objective function optimization coefficient of the algorithm is output; The stability verification and robustness analysis of all objective function optimization coefficients are carried out through the multi-objective fusion decision-making unit to determine the global optimal solution.

2. The control parameter collaborative optimization method based on hybrid heuristic optimization according to claim 1 is characterized in that: The hybrid heuristic optimization thread is constructed by injecting a population feature migration mechanism and a parameter adaptive adjustment strategy into a benchmark optimization thread, wherein the benchmark optimization thread includes native operation operators of a genetic algorithm, a particle swarm optimization algorithm, and a firefly swarm optimization algorithm; The cross-algorithm winning community interaction mechanism includes: Using a feature vector fusion unit to perform multi-dimensional feature splicing on the feature space data to generate a superior community feature vector with cross-algorithm correlation; Calculating the population diversity characteristic coefficient based on the superior community characteristic vector, and determining the sampling ratio and characteristic stratification interval of cross-population embedding according to the coefficient; The method for constructing the parameter adaptive adjustment strategy includes: Monitor the rate of change of the objective function value of each algorithm within a continuous preset iteration window; When the change rate is lower than the dynamic adjustment threshold, the parameter perturbation module is triggered to perform nonlinear adjustment on the core parameters of the current algorithm: an exponential amplification operation of the mutation probability is performed on the genetic algorithm; Perform piecewise decay operation of inertia weight on particle swarm algorithm; Performs an adaptive recalibration of the fluorescein coefficients for the firefly algorithm.

3. The control parameter collaborative optimization method based on hybrid heuristic optimization according to claim 1 is characterized in that: The benchmark optimization thread includes native operation operators of genetic algorithm, particle swarm optimization algorithm and firefly swarm optimization algorithm, which are specifically manifested as follows: The genetic algorithm adopts the chromosome encoding mechanism, and its operation operators include the gene recombination unit based on the roulette selection operator, the population evolution unit based on the single-point crossover operator, and the gene perturbation unit based on the basic position mutation operator; The particle swarm optimization algorithm adopts a speed-position update model, and its operation operators include an individual optimal position tracking unit and a global optimal position fusion unit; The firefly swarm optimization algorithm adopts a brightness attraction mechanism, and its operation operator includes a relative brightness calculation unit and a dynamic distance response unit.

4. The control parameter collaborative optimization method based on hybrid heuristic optimization according to claim 3 is characterized by: The iterative evolution process of the native operation operator includes: The genetic algorithm executes a three-stage evolutionary process of gene selection-crossover-mutation, and drives the iterative update of the population through the fitness function evaluation unit; The particle swarm algorithm executes a two-stage motion process of velocity vector update-position vector correction, and realizes information interaction through the individual historical optimal solution memory unit and the global optimal solution sharing unit; The firefly algorithm executes a three-stage bionic process of brightness calculation-position attraction-parameter update, and realizes swarm intelligent evolution through relative distance calculation unit and attraction intensity adjustment unit.

5. The control parameter collaborative optimization method based on hybrid heuristic optimization according to any one of claims 1 to 4, characterized in that: The method for constructing the winning community feature library includes: Establish a multi-dimensional feature label for each winning community, including algorithm identifier, iteration stage feature code, objective function value change gradient and population diversity index; A feature encoding mechanism is used to convert the multidimensional feature labels into traceable feature vectors, and a feature map of superior communities with temporal and spatial correlation is constructed.

6. The control parameter collaborative optimization method based on hybrid heuristic optimization according to claim 1 is characterized in that: The feature sampling and cross-population embedding operations specifically include: Determine the feature extraction amount of each winning community based on a dynamic sampling ratio calculation module, wherein the dynamic sampling ratio is positively correlated with the population diversity index; A stratified feature sampling strategy is adopted to divide the winning community into multiple feature subspaces according to the fitness distribution, and adaptive sampling based on probability density is performed in each subspace; The extracted feature samples are injected into the target population through a random position insertion algorithm, wherein the random position insertion algorithm comprises a gene sequence recombination verification unit and a population capacity balancing unit.

7. The control parameter collaborative optimization method based on hybrid heuristic optimization according to claim 2 is characterized in that: The parameter perturbation module comprises: The genetic algorithm mutation enhancement unit adopts an exponential mutation probability function; The particle swarm inertia weight adjustment unit adopts a piecewise linear attenuation function; Firefly luciferin regulation unit, using an adaptive update formula.

8. The control parameter collaborative optimization method based on hybrid heuristic optimization according to claim 1 is characterized in that: The determination method of the second convergence condition includes: Construct a dual convergence criterion joint decision model, and trigger the first convergence flag when the total number of iterations iter2>Miniter2 and the change in the objective function value Δf≤ε; The second convergence flag is triggered when the number of continuous iterations of the algorithm falling into the local optimum exceeds a preset threshold; The dual convergence criterion joint decision-making model adopts a weighted voting mechanism to comprehensively judge the termination timing of the algorithm.

9. A control parameter collaborative optimization system based on hybrid heuristic optimization, characterized in that: include: Hybrid heuristic optimization engine module, winning community interaction module, parameter adaptive adjustment module, multi-objective decision module and algorithm basic library module, wherein: the hybrid heuristic optimization engine module is connected to the algorithm basic library module and the winning community interaction module respectively, and transmits the algorithm native iteration data and the winning community feature space data respectively; the parameter adaptive adjustment module is bidirectionally connected to the hybrid heuristic optimization engine module, transmits the objective function change rate data in real time and feeds back the parameter adjustment instructions; the multi-objective decision module receives the optimization coefficient data of all algorithm channels, and performs stability verification and robustness analysis; The hybrid heuristic optimization engine module includes: Collaborative initialization unit: connects the algorithm basic library module, initializes the parameter space of Ns basic optimization algorithms, and establishes a parallel solution channel; Native iterative execution unit: Built-in genetic algorithm operator, particle swarm optimization operator and firefly swarm optimization operator, driving each algorithm to perform iterative operations according to the native mechanism; Convergence monitoring unit: connected to each algorithm channel, judging the iteration termination condition through the convergence condition judgment module and the Maxiter counter module; Feature extraction unit: generates feature space data containing the winning community of the algorithm and extracts the current optimal solution vector of each channel; the winning community interaction module includes: Feature vector fusion unit: receives feature space data from each algorithm channel, performs multi-dimensional feature splicing to generate cross-algorithm correlation feature vectors; Diversity analysis unit: Determine the characteristic sampling ratio and stratification interval based on the population diversity characteristic coefficient calculation module; Cross-population embedding unit: connects to the hybrid heuristic optimization engine module to perform dynamic migration of winning feature vectors and population reorganization operations; The parameter adaptive adjustment module includes: Iteration window monitoring unit: statistics the change rate of each algorithm's objective function value through a sliding time window; Disturbance trigger unit: When the rate of change is lower than the dynamic adjustment threshold, the parameter nonlinear adjustment channel is activated; Algorithm regulator group: including genetic algorithm mutation probability amplifier, particle swarm inertia weight attenuator, and firefly luciferin calibrator; The multi-objective decision-making module comprises: Stability verification unit: The robustness test of each optimization coefficient is carried out through the Monte Carlo simulation module; Decision fusion unit: Use fuzzy comprehensive evaluation method to screen the global optimal solution of Pareto frontier solution set; Result output interface: Generates an analysis report including the optimization parameter configuration scheme and the objective function response surface; The algorithm base library module stores: Genetic algorithm core operator group: including selection operator library, crossover operator library, mutation operator library and fitness function library; Particle swarm optimization component set: including velocity update calculator, position updater and neighborhood topology template; Firefly algorithm function package: including luciferin update module, movement probability calculator and attraction calculation unit; The feature migration operation performed by the cross-population embedding unit includes: feature space similarity calculation based on Mahalanobis distance; population recombination strategy using quantum rotating door mechanism; feature stratified sampling method based on information entropy weight; The parameter nonlinear adjustment channel includes: an exponential amplification function of the genetic algorithm mutation probability; a segmented attenuation strategy of the particle group inertia weight; and an adaptive calibration model of the firefly luciferin coefficient.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the control parameter collaborative optimization method based on hybrid heuristic optimization as described in any one of claims 1 to 8.

Images