Hybrid variable optimization design method and system for power device cooling system and storage medium

Abstract

The invention discloses a hybrid variable optimization design method and system for a power device cooling system and a storage medium, and the method comprises the steps: completely maintaining the essential characteristics of different types of variables through a differential processing mode of mapping ordered type variables to a one-dimensional hidden space and mapping disordered type variables to a high-dimensional hidden space, and achieving the optimal design of the hybrid variables of the power device cooling system. According to the method, the model can accurately describe a numerical value-category mixed variable coupling action mechanism, the fitting precision and generalization ability of the model are remarkably improved, the comprehensive influence of different variable combinations on the acoustic performance of the cooling system is accurately reflected, and a reliable model basis is provided for subsequent optimization. According to the method, the total sound pressure level response serves as a core optimization design target, and efficient optimization design of the power device cooling system containing the mixed variables is achieved through the whole process of parameterized acoustic analysis, hidden space Gaussian process model training and iterative optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power device cooling system optimization, in particular to a power device cooling system mixed variable optimization design method, system and storage medium. BACKGROUND

[0002] As a key auxiliary system of core power equipment such as engines and motors, the design rationality of the power device cooling system directly affects the noise control level of the power device. In actual engineering scenarios, the design of the cooling system needs to consider both numerical design variables (such as air duct structure size, fan configuration, etc.) and categorical design variables (such as cooling medium type, fan speed level, air duct layout scheme, etc.). The coupling effect of the two types of variables makes the system optimization design challenging.

[0003] Traditional cooling system optimization methods mostly focus on the optimization of a single type of variable, such as using gradient descent, genetic algorithm, etc. optimization strategies for numerical variables only, or enumerating and screening categorical variables. However, such methods do not fully consider the mixed characteristics of numerical and categorical variables, resulting in limitations in the optimization process. Ignoring the differences between variable types can lead to insufficient accuracy of the optimization model, making it difficult to accurately reflect the comprehensive influence of different variable combinations on system performance. Moreover, traditional methods such as enumeration are prone to problems such as rapid increase in computational load, low optimization efficiency, and difficulty in finding global optimal solutions when dealing with multiple categories and multiple numerical variable combinations. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application proposes a power device cooling system mixed variable optimization design method.

[0005] To achieve the above-mentioned purpose, the technical solution adopted by the present application is as follows: A power device cooling system mixed variable optimization design method, comprising the following steps: S1, generating an input matrix containing initial training samples , with a size of ; S2, combining the input matrix formed by initial samples corresponding total sound pressure level response to form a training sample set ; S3, mapping the label value of the ordered categorical design variable to a one-dimensional ordered coordinate point in a one-dimensional hidden space; S4, mapping the label value of the unordered categorical design variable to a multi-dimensional unordered coordinate point in a high-dimensional hidden space; S5, training a hidden space Gaussian process model according to the training sample set ; S6, iterating the numerical design variables and the categorical design variables to obtain the input variables after iteration And obtain the input variables after iteration. Corresponding total sound pressure level response S7, Update sample count ,Will Merged into the training sample set ,judge Is it greater than the preset threshold N? If yes, proceed to step S8; otherwise, return to step S3. In S8, the training sample set... The numerical and categorical design variables corresponding to the minimum total sound pressure level response are used as the optimization design scheme for the power unit cooling system.

[0006] Further, step S3 specifically includes: for an ordered categorical design variable with L class values, sorting the label values ​​of the ordered categorical design variable from smallest to largest to obtain a one-dimensional variable sequence. ; Introducing inclusion An ordered sequence of coordinate points in a one-dimensional latent space To map one-to-one one-dimensional variable sequences Parametric values ​​of ordered categorical design variables in the interval; values ​​of one-dimensional latent space coordinates within the interval Inside; and , The values ​​of the remaining one-dimensional latent space coordinates satisfy .

[0007] Further, step S4 specifically includes: for an unordered categorical design variable with M category values, the sequence of label values ​​of the M unordered categorical design variables is represented as an unordered variable sequence. ; Introducing inclusion indivual A multidimensional unordered sequence of coordinate points in hidden space One-to-one mapping sequence The parameterized values ​​of unordered categorical design variables in the model, where the non-zero coordinate component values ​​at each coordinate point are... All within the range Inside.

[0008] Furthermore, in step S5, the hyperparameters of the latent space Gaussian process model that need to be determined through training include: the regression constant term. Process variance The feature scale coefficients implicit in the correlation function, and the values ​​of the one-dimensional latent space coordinates corresponding to the categories of ordered categorical design variables. The non-zero coordinate component values ​​of the high-dimensional latent space coordinate points corresponding to the category values ​​of unordered categorical design variables. .

[0009] Further, the hyperparameters can be obtained by maximizing the log-likelihood function, and the covariance matrix corresponding to the maximized log-likelihood function is obtained ; the log-likelihood function is: ; is the covariance matrix of , is the number of samples, and the process variance , the characteristic scale coefficient, and the value of the one-dimensional latent space coordinate point , the value of the non-zero coordinate component of the high-dimensional latent space coordinate point are all implicitly contained in the covariance matrix ; the regression constant term is calculated by the following formula: is a vector with all elements being 1.

[0010] Further, the element in the th row and the th column of the covariance matrix is: ; is the correlation function between the th sample response and the th sample response , which contains the characteristic scale coefficient, the value of the one-dimensional latent space coordinate point , and the value of the non-zero coordinate component of the high-dimensional latent space coordinate point .

[0011] Further, the formula of the correlation function is as follows: ; is the distance component of the th design variable. For numerical design variables, , ; for ordered categorical design variables, , ; and for unordered categorical design variables, , ; is the characteristic scale coefficient corresponding to the th design variable; is the total number of ordered categorical design variables; is the numerical design variable value of the th design variable of the th sample;​ the i-th sample the numerical design variable value of the j-th design variable of the i-th sample the ordinal of the ordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the ordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the ordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the ordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the ordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the multi-dimensional unordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the multi-dimensional unordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the multi-dimensional unordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the multi-dimensional unordered coordinate point sequence corresponding to the j-th design variable of the i-th sample the ordinal of the multi-dimensional unordered coordinate point sequence corresponding to the j-th design variable of the i-th sample a unit vector with the i-th element being 1 and the rest elements being 0 a unit vector with the i-th element being 1 and the rest elements being 0 a unit vector with the i-th element being 1 and the rest elements being 0 a unit vector with the i-th element being 1 and the rest elements being 0

[0012] Further, in step S7, the input variable corresponding to the maximum of the improved expected acquisition function is obtained by maximizing the improved expected acquisition function .

[0013] the posterior predictive mean of the latent space Gaussian process model corresponding to the input variable ; the minimum value of the response in the training sample set , the cumulative distribution function of the standard normal distribution, the probability density function of the standard normal distribution the posterior predictive standard deviation of the latent space Gaussian process model corresponding to the input variable .

[0014] The application also provides a power device cooling system mixed variable optimization design system, comprising: a matrix generation module, used for generating an input matrix containing i-th initial training samples , the size of the input matrix is ; a training sample generation module, used for combining the input matrix formed by the i-th initial sample to form a total sound pressure level response corresponding to the input matrix, and forming a training sample set The ordered variable mapping module maps the labeled values ​​of ordered categorical design variables to one-dimensional ordered coordinate points in a one-dimensional latent space; the unordered variable mapping module maps the labeled values ​​of unordered categorical design variables to multi-dimensional unordered coordinate points in a high-dimensional latent space; the training module is used to map the values ​​of unordered categorical design variables to multi-dimensional unordered coordinate points in a high-dimensional latent space based on the training sample set. Training a latent space Gaussian process model; an iteration module for iterating over numerical and categorical design variables to obtain the iterated input variables. And obtain the input variables after iteration. Corresponding total sound pressure level response The loop module is used to update the number of samples. ,Will Merged into the training sample set ,judge Whether it is greater than the preset threshold N; Optimize the design scheme output module, used to process the training sample set The numerical and categorical design variables corresponding to the minimum total sound pressure level response are used as the optimization design scheme for the power unit cooling system.

[0015] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements a hybrid variable optimization design method for a power unit cooling system.

[0016] The present invention has the following beneficial effects: By mapping ordered categorical variables to a one-dimensional latent space and unordered categorical variables to a high-dimensional latent space, this invention fully preserves the essential characteristics of different types of variables, avoiding feature loss or unreasonable associations caused by variable parameterization in traditional methods. This enables the model to accurately characterize the coupling mechanism between the two types of variables, accurately reflecting the comprehensive impact of different variable combinations on the acoustic performance of the cooling system, significantly improving the model's fitting accuracy and generalization ability, and providing a reliable model foundation for subsequent optimization. A latent space Gaussian process model is used to train the training sample set, combined with efficient iterative optimization of the mixed variables. By iteratively updating the sample set and dynamically determining the termination condition, redundant computation is effectively reduced, avoiding the problems of surging computational load and long optimization cycles in traditional methods. Simultaneously, the Gaussian process model possesses powerful nonlinear fitting and prediction capabilities, enabling it to quickly focus on the optimal variable combination region during iteration, improving the search efficiency for the global optimum, and ensuring that ideal optimization results are obtained efficiently within a preset threshold. This invention uses the total sound pressure level response as the core optimization objective, achieving accurate mapping between mixed variables and acoustic performance through a complete process of parameterized model acoustic analysis, latent space model training, and iterative optimization.

[0017] In addition to the above described objects, features and advantages, the present application has other objects, features and advantages. The present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated by reference in their entirety. In the drawings: Figure 1 is a flow chart of the method of the present application. DETAILED DESCRIPTION

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the present application and should not be construed as limiting the present application.

[0020] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0021] It should be noted that all directional indications, such as up, down, left, right, front, back, etc., in the embodiments of the present application are only used to explain the relative positional relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly.

[0022] In addition, the descriptions of "first", "second", etc. in the present application are only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the fact that the technical solutions can be realized by those skilled in the art. When the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the scope of protection claimed by the present application.

[0023] Please refer to Figure 1 In a preferred embodiment of the present application, a hybrid variable optimization design method of a power device cooling system is provided, which comprises steps S1, S2, S3, S4, S5, S6, S7 and S8.

[0024] S1, generating an input matrix containing initial training samples , The size of the input matrix is . The total number of types of numerical design variables. The total number of categories for categorical design variables.

[0025] S2, combined The input matrix formed by the initial samples Corresponding total sound pressure level response To form a training sample set Typical total sound pressure level response Available Vector representation, forming the total sound pressure level response vector Total sound pressure level response vector Depend on Each sample corresponds to The total sound pressure level response is formed by the combination of these factors. A vector, or a matrix with only one column. That is, the input matrix. The total sound pressure level response corresponding to the first sample in the dataset is the total sound pressure level response. The first element in the vector. This is understandable. The total sound pressure level corresponding to each initial sample can be obtained experimentally, or by constructing a system that simultaneously contains... Numerical design variables and The power unit cooling system parametric analysis model with categorical design variables is obtained through simulation. The simulation software can be ANSYS Fluent, Siemens Simcenter, or MSC Actran.

[0026] S3 maps the labeled values ​​of ordered categorical design variables to one-dimensional ordered coordinate points in a one-dimensional latent space.

[0027] S4 maps the labeled values ​​of unordered categorical design variables to multidimensional unordered coordinate points in a high-dimensional latent space.

[0028] S5, based on the training sample set Train a Gaussian process model in the latent space.

[0029] S6 iterates through the numerical and categorical design variables to obtain the iterated input variables. And obtain the input variables after iteration. Corresponding total sound pressure level response . It can be obtained through experiments or by constructing a parametric analysis model of the power unit cooling system and conducting simulation.

[0030] S7, Update sample count ,Will Merging into the training sample set , determining whether it is greater than a preset threshold N: if yes, go to step S8, if no, return to step S3. , the number of rows in the size of increases by 1, and the size increases by 1 row at the bottom based on the original size, such as The size of the original is 2x5, and after updating and merging , the size becomes 3x5, and each parameter of is supplemented to the matrix The last row is increased by 1, and the last element is increased by 1, and the added element is .

[0031] S8, the numerical design variable and the category design variable corresponding to the minimum total sound pressure level response in the training sample set are taken as the optimization design scheme of the cooling system of the power device.

[0032] The power device cooling system mixed variable optimization design method provided by the application, through the differential processing mode of mapping ordered category type variable to one-dimensional hidden space and mapping unordered category type variable to high-dimensional hidden space, the essential characteristics of different types of variables are completely preserved, avoiding the feature loss or unreasonable association problem caused by variable parameterization in traditional methods. Make the model be able to accurately depict the coupling mechanism of the two types of variables, accurately reflect the comprehensive influence of different variable combinations on the acoustic performance of the cooling system, and significantly improve the fitting precision and generalization ability of the model, providing a reliable model basis for subsequent optimization. The training sample set is trained by using the Gaussian process model in the hidden space, combined with efficient iterative optimization of mixed variables, the sample set is updated by iteration and the termination condition is dynamically judged, which effectively reduces the redundant calculation amount and avoids the problem of rapid increase of calculation amount and long optimization period in traditional methods. At the same time, the Gaussian process model has strong nonlinear fitting and prediction ability, which can quickly focus on the optimal variable combination area in the iteration process, improve the search efficiency of the global optimal solution, and ensure that the ideal optimization result is obtained efficiently within the preset threshold. The total sound pressure level response is taken as the core optimization target, and through the whole process of parameterized model acoustic analysis, hidden space model training and iterative optimization, the accurate mapping of mixed variables and acoustic performance is realized. A scientific parameterized mapping scheme is designed for ordered and unordered category type variables, so that the method can adapt to different types and different quantities of mixed variable combination scenes, and is suitable for the cooling system design of various engines, motors and other power devices.

[0033] In some embodiments of the present application, step S3 specifically comprises: for an ordered categorical design variable with L category values, sorting the label values of the ordered categorical design variable from small to large to obtain a one-dimensional variable sequence ; introducing an ordered coordinate point sequence comprising one-dimensional latent space coordinate points (ordered coordinate points in one-dimensional latent space) to one-to-one correspondingly map the parameterized values of the ordered categorical design variable in the one-dimensional variable sequence ; the value of the one-dimensional latent space coordinate point is within the interval ; and , the values of the remaining one-dimensional latent space coordinate points satisfy . It can be understood that each ordered categorical design variable needs to be processed by step S3 to form a one-dimensional variable sequence and an ordered coordinate point sequence , and the values of L for different ordered categorical design variables can be different or the same.

[0034] By sorting the ordered categorical design variables, constructing a one-dimensional latent space coordinate point sequence for one-to-one correspondence mapping, the hierarchical relationship of the ordered variables is ensured not to be lost in the parameterization process. The values of the one-dimensional latent space coordinate points are limited in the interval [0, 1], and the first and last coordinates are fixed as 0 and 1, and the intermediate coordinates are strictly increasing, which not only realizes the standardized expression of the ordered variables, but also accurately preserves the ordered correlation characteristics between variables. Avoiding the problem of distortion of ordered information in traditional parameterization methods, the subsequent model training can accurately capture the gradient influence of ordered variables on the acoustic performance of the cooling system, providing a reliable ordered variable input basis for the optimization model.

[0035] In some embodiments of the present application, step S4 specifically comprises: for an unordered categorical design variable with M category values, the sequence formed by the label values of the M unordered categorical design variables is denoted as an unordered variable sequence ; since there is no order difference between the unordered categorical design variables, the unordered variable sequence can be sorted randomly when determining the arrangement order, and the order is fixed after sorting is completed, and the subsequent order does not change. Introducing a multi-dimensional unordered coordinate point sequence comprising M -dimensional latent space coordinate points (multi-dimensional unordered coordinate points in high-dimensional latent space) to one-to-one correspondingly map the parameterized values of the unordered categorical design variable in the sequence , wherein the non-zero coordinate component values in each coordinate point are within the interval . Since there is no mutual influence between the unordered categorical design variables, each The non-zero coordinate components of points in the implicit space are not in the same dimension, and there is no basis for comparison or influence between them. It is understandable that each unordered categorical design variable requires step S4 to form an unordered variable sequence. and multidimensional unordered coordinate point sequences And different unordered categorical design variables The values ​​can be different or the same.

[0036] For unordered categorical design variables, a sequence of M M-dimensional latent space coordinate points is used for mapping. By differentiating the non-zero coordinates across different dimensions, effective distinction between different categories of variables is achieved while avoiding the introduction of unreasonable numerical correlations. The non-zero coordinate component values ​​are limited to the [0,1] interval, ensuring the consistency and standardization of variable parameterization. This step fully preserves the essential characteristics of unordered variables, enabling the model to accurately identify the independent influence of different categories of variables on the total sound pressure level response, thus improving the scientific rigor and accuracy of unordered variable handling in mixed-variable optimization.

[0037] In some embodiments of the present invention, in step S5, the hyperparameters of the latent space Gaussian process model to be determined through training include: a regression constant term. Process variance The feature scale coefficients implicit in the correlation function, and the values ​​of the one-dimensional latent space coordinates corresponding to the categories of ordered categorical design variables. The non-zero coordinate component values ​​of the high-dimensional latent space coordinate points corresponding to the category values ​​of unordered categorical design variables. .

[0038] The types of hyperparameters required for the latent space Gaussian process model have been clarified, covering regression constants, process variance, feature scale coefficients, and latent space coordinates corresponding to each category variable, forming a complete hyperparameter system. This makes the model training objective clearer and enables it to comprehensively capture the mixed characteristics of numerical and categorical variables and their combined effects on the acoustic response.

[0039] In some embodiments of the present invention, the hyperparameters can be obtained in the following manner: By maximizing the log-likelihood function, we obtain the covariance matrix corresponding to the maximization of the log-likelihood function. The log-likelihood function is: ; for The covariance matrix, The process variance is the number of samples. Feature scaling coefficients and the values ​​of coordinates in the one-dimensional latent space. Non-zero coordinate component values ​​of coordinate points in high-dimensional hidden space All are implicit in the covariance matrix middle. Let be the log-likelihood function. This is the total sound pressure level response vector.

[0040] The optimized hyperparameters can be obtained by using an optimization algorithm that maximizes the log-likelihood function. This optimization algorithm can be a particle swarm optimization algorithm or a genetic algorithm, among others.

[0041] The regression constant term It is calculated using the following formula: ; For all elements to be 1 A vector, where T represents the transpose of the vector. This is the total sound pressure level response vector.

[0042] By maximizing the log-likelihood function to obtain hyperparameter values, the correlation information between the total sound pressure level response and the covariance matrix in the training sample set is fully utilized. Taking into account the number of samples, the dimension of the response vector, and the characteristics of the covariance matrix, the hyperparameters can be adaptively adjusted to fit the data distribution pattern, providing a core guarantee for the model's prediction accuracy.

[0043] The The Line number The elements of the column are: ; For the first Response of a sample With the Response of a sample The correlation function between them includes the feature scale coefficient and the values ​​of the coordinates of the one-dimensional latent space points. Non-zero coordinate component values ​​of coordinate points in high-dimensional hidden space .

[0044] The definition of the elements of the covariance matrix clearly defines the correlation between sample responses and implicitly includes key parameters such as process variance, feature scaling coefficients, and latent space coordinates, achieving organic integration of multiple parameters. This enables the model to fully explore the potential correlations in the sample data, improves the model's ability to characterize the mapping relationship between mixed variables and acoustic responses, and provides a reliable model foundation for subsequent iterative optimization.

[0045] In some embodiments of the present invention, the formula for the relevant function is as follows: ; For the first Distance components of each design variable, For numerical design variables , ; For ordered categorical design variables , ; For unordered categorical design variables , ; For the first Characteristic scaling coefficients of each design variable; For ordered categorical types, set the total number of categories. The total number of types of unordered categorical design variables. For the i-th sample Numerical design variable values ​​for each design variable; For the first The first sample Numerical design variable values ​​for each design variable; For the i-th sample The ordinal number of the ordered coordinate point sequence corresponding to each design variable. For the first in the ordered sequence of coordinate points The values ​​of each coordinate point; For the first The first sample The ordinal number of the ordered coordinate point sequence corresponding to each design variable. For the first in an ordered sequence of coordinate points The values ​​of each coordinate point; For the i-th sample The ordinal number of the multidimensional unordered coordinate point sequence corresponding to each design variable. For the multidimensional unordered coordinate point sequence, the first... Non-zero coordinate component values ​​of each coordinate point; For the first The first sample The ordinal number of the multidimensional unordered coordinate point sequence corresponding to each design variable. For the multidimensional unordered coordinate point sequence, the first... Non-zero coordinate component values ​​of each coordinate point; Indicates the first A unit vector with 1 element and 0 elements; Indicates the first A unit vector with one element being 1 and the rest being 0.

[0046] for and ,when When, it corresponds to numerical design variables; , corresponding to ordinal categorical design variables; , corresponding to unordered categorical design variables. The input matrix is arranged in this order, i.e. the input matrix is arranged in this order, i.e. the input matrix is arranged in this order, i.e. the input matrix is arranged in this order, i.e. the input matrix is arranged in this order, i.e. the input matrix is arranged in this order, i.e. the input matrix is arranged in this order, i.e. the input matrix is arranged in this order, i.e. the input matrix

[0047] The correlation function integrates the distance components of the numerical, ordinal categorical and unordered categorical design variables in the form of product, and realizes the accurate quantification of the influence weight of different types of variables. The differentiated distance component calculation logic is defined for the three types of variables, the numerical variable is based on the absolute value difference, the ordinal categorical variable is based on the implicit space coordinate difference, and the unordered categorical variable is based on the high-dimensional implicit space vector two norm, which ensures the pertinence and rationality of the correlation description of each type of variable. The feature scale coefficient is introduced in the function, which can adaptively adjust the influence degree of different design variables on the overall sound pressure level response, and improves the adaptation ability of the model to the variable coupling relationship.

[0048] In some embodiments of the present application, in step S7, the input variable is obtained by maximizing the improved expected acquisition function. The improved expected acquisition function is: ; The optimization algorithm can be a hybrid variable particle swarm optimization algorithm and a hybrid variable genetic algorithm.

[0049] is the posterior predictive mean of the implicit space Gaussian process model corresponding to the input variable ; is the minimum value of the response in the training sample set , i.e. the minimum value of all elements in the overall sound pressure level response vector y. is the cumulative distribution function of the standard normal distribution, is the probability density function of the standard normal distribution; is the posterior predictive standard deviation of the implicit space Gaussian process model corresponding to the input variable . ​

[0050] Specifically, and It can be calculated using the following formula: for The related function vector, its first... Each element is a related function. ; .Will Input variables in replace It can be calculated.

[0051] An improved expected acquisition function is combined with the posterior predicted mean and standard deviation of a latent space Gaussian process model. An iterative criterion is constructed using the cumulative distribution function and probability density function of a standard normal distribution. This function targets the minimum total sound pressure level response in the training sample set, guiding variable iteration towards regions with superior acoustic performance. Simultaneously, the standard deviation term considers the potential optimization value of unexplored regions, effectively preventing iteration from getting trapped in local optima. The iterative mechanism eliminates the need for additional discretization of categorical variables, allowing direct co-optimization with numerical variables, adapting to the characteristics of mixed variables, and significantly improving iteration efficiency. By maximizing this acquisition function to obtain the optimized input variables, each iteration ensures targeted supplementation of high-quality samples, accelerating the convergence speed of the training sample set and enabling the optimization process to efficiently approach the global optimum within a preset threshold.

[0052] This invention also provides a hybrid variable optimization design system for a power unit cooling system, used to implement a hybrid variable optimization design method for a power unit cooling system, comprising: a matrix generation module for generating a matrix containing... Input matrix of initial training samples , The size is Training sample generation module, used to combine The input matrix formed by the initial samples Corresponding total sound pressure level response To form a training sample set The ordered variable mapping module maps the labeled values ​​of ordered categorical design variables to one-dimensional ordered coordinate points in a one-dimensional latent space; the unordered variable mapping module maps the labeled values ​​of unordered categorical design variables to multi-dimensional unordered coordinate points in a high-dimensional latent space; the training module is used to map the values ​​of unordered categorical design variables to multi-dimensional unordered coordinate points in a high-dimensional latent space based on the training sample set. Training a latent space Gaussian process model; an iteration module for iterating over numerical and categorical design variables to obtain the iterated input variables. And obtain the input variables after iteration. Corresponding total sound pressure level response The loop module is used to update the number of samples. ,Will Merged into the training sample set ,judge Whether it is greater than the preset threshold N; Optimize the design scheme output module, used to process the training sample set The numerical and categorical design variables corresponding to the minimum total sound pressure level response are used as the optimization design scheme for the power unit cooling system.

[0053] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements a hybrid variable optimization design method for a power unit cooling system.

[0054] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A hybrid variable optimization design method for a power unit cooling system, characterized in that, Includes the following steps: S1, generates a sequence containing Input matrix of initial training samples , The size is , The total number of types of numerical design variables. The total number of categories for categorical design variables; S2, combined The input matrix formed by the initial samples Corresponding total sound pressure level response To form a training sample set ; S3 maps the labeled values ​​of ordered categorical design variables to ordered coordinate points in a one-dimensional latent space; S4 maps the labeled values ​​of unordered categorical design variables to unordered coordinate points in a high-dimensional latent space; S5, based on the training sample set Training a latent space Gaussian process model; S6 iterates through the numerical and categorical design variables to obtain the iterated input variables. And obtain the input variables after iteration. Corresponding total sound pressure level response ; S7, Update sample count ,Will Merged into the training sample set ,judge Is it greater than the preset threshold N? If yes, proceed to step S8; otherwise, return to step S3. S8, training sample set The numerical and categorical design variables corresponding to the minimum total sound pressure level response are used as the optimization design scheme for the power unit cooling system.

2. The hybrid variable optimization design method for a power unit cooling system according to claim 1, characterized in that, Step S3 specifically includes: For an ordered categorical design variable with L class values, sort the class values ​​of the ordered categorical design variable from smallest to largest to obtain a one-dimensional variable sequence. ; Introducing include An ordered sequence of coordinate points in a one-dimensional latent space To map one-to-one one-dimensional variable sequences Parametric values ​​of ordered categorical design variables in the interval; values ​​of one-dimensional latent space coordinates within the interval Inside; and , The values ​​of the remaining one-dimensional latent space coordinates satisfy .

3. The hybrid variable optimization design method for a power unit cooling system according to claim 2, characterized in that, Step S4 specifically includes: For an unordered categorical design variable with M class values, the sequence of label values ​​of the M unordered categorical design variables is represented as the unordered variable sequence. ; Introducing include indivual A multidimensional unordered sequence of coordinate points in hidden space One-to-one mapping sequence The parameterized values ​​of unordered categorical design variables in the model, where the non-zero coordinate component values ​​at each coordinate point are... All within the range Inside.

4. The hybrid variable optimization design method for a power unit cooling system according to claim 3, characterized in that, In step S5, the hyperparameters of the latent space Gaussian process model that need to be determined through training include: the regression constant term. Process variance The feature scale coefficients implicit in the correlation function, and the values ​​of the one-dimensional latent space coordinates corresponding to the categories of ordered categorical design variables. The non-zero coordinate component values ​​of the high-dimensional latent space coordinate points corresponding to the category values ​​of unordered categorical design variables. .

5. The hybrid variable optimization design method for a power unit cooling system according to claim 4, characterized in that, The hyperparameters can be obtained in the following way: By maximizing the log-likelihood function, we obtain the covariance matrix corresponding to the maximization of the log-likelihood function. ; The log-likelihood function is: ； for The covariance matrix, The process variance is the number of samples. Feature scaling coefficients and the values ​​of coordinates in the one-dimensional latent space. Non-zero coordinate component values ​​of coordinate points in high-dimensional hidden space All are implicit in the covariance matrix middle; The regression constant term It is calculated using the following formula: ； For all elements to be 1 vector.

6. The hybrid variable optimization design method for a power unit cooling system according to claim 5, characterized in that, The covariance matrix The Line number The elements of the column are: ； For the first Response of a sample With the Response of a sample The correlation function between them includes the feature scale coefficient and the values ​​of the coordinates of the one-dimensional latent space points. Non-zero coordinate component values ​​of coordinate points in high-dimensional hidden space .

7. The hybrid variable optimization design method for a power unit cooling system according to claim 6, characterized in that, The formula for the relevant function is as follows: ； For the first Distance components of each design variable, For numerical design variables , ; For ordered categorical design variables , ; For unordered categorical design variables , ; For the first Characteristic scaling coefficients of each design variable; The total number of categories is set for ordered categorical types; For the i-th sample, the th Numerical design variable values ​​for each design variable; For the first The first sample Numerical design variable values ​​for each design variable; For the i-th sample, the th The ordinal number in the ordered coordinate point sequence corresponding to each design variable; For the first The first sample The ordinal number in the ordered coordinate point sequence corresponding to each design variable; For the i-th sample, the th The ordinal number of the multidimensional unordered coordinate point sequence corresponding to each design variable; For the first The first sample The ordinal number of the multidimensional unordered coordinate point sequence corresponding to each design variable; Indicates the first A unit vector with 1 element and 0 elements; Indicates the first A unit vector with one element being 1 and the rest being 0.

8. The hybrid variable optimization design method for a power unit cooling system according to claim 1, characterized in that, In step S7, the input variables when the improved desired acquisition function is maximized are obtained by maximizing the improved desired acquisition function. ; The improved desired acquisition function is as follows: ； Input variables The posterior prediction mean of the corresponding latent space Gaussian process model; For training sample set The minimum value of the response. The cumulative distribution function of the standard normal distribution. is the probability density function of the standard normal distribution; Input variables The posterior prediction standard deviation of the corresponding latent space Gaussian process model.

9. A hybrid variable optimization design system for a power unit cooling system, used to implement the hybrid variable optimization design method for a power unit cooling system according to any one of claims 1 to 8, characterized in that, include: The matrix generation module is used to generate matrices containing... Input matrix of initial training samples , The size is ; Training sample generation module, used to combine The input matrix formed by the initial samples Corresponding total sound pressure level response To form a training sample set ; The ordered variable mapping module is used to map the labeled values ​​of ordered categorical design variables to one-dimensional ordered coordinate points in a one-dimensional latent space; The unordered variable mapping module is used to map the labeled values ​​of unordered categorical design variables to multidimensional unordered coordinate points in a high-dimensional latent space; The training module is used to train the training sample set. Training a latent space Gaussian process model; The iteration module is used to iterate over numerical and categorical design variables to obtain the iterated input variables. And obtain the input variables after iteration. Corresponding total sound pressure level response ; The loop module is used to update the number of samples. ,Will Merged into the training sample set ,judge Is it greater than the preset threshold N? The optimized design output module is used to process the training sample set. The numerical and categorical design variables corresponding to the minimum total sound pressure level response are used as the optimization design scheme for the power unit cooling system.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the hybrid variable optimization design method for the power unit cooling system as described in any one of claims 1 to 8.

Images