Optimization Method and System of Injector Based on Improved Orthogonal Test Method

Abstract

The present invention relates to the technical field of electrical digital data processing, and proposes an injector optimization method and system based on an improved orthogonal test method, which combines the advantages of computational fluid dynamics simulation, neural network modeling, and multi-objective optimization algorithms. First, the flow field and acoustic field characteristics of the injector are accurately simulated through a simulation model, providing high-quality data support for optimization. Secondly, an improved orthogonal test method is used for decision variable identification and data augmentation, improving the training effect of the neural network model. Finally, the NSGA-II genetic algorithm is used for multi-objective optimization, enabling the performance of the injector to achieve a balance among multiple objectives. The optimized injector shows significant advantages in terms of velocity distribution, turbulent kinetic energy, and sound power level. After optimization, the flow loss is reduced, the air flow expansion state is improved, the vacuum degree is increased, and at the same time, the flow noise is effectively reduced, enabling the injector to have higher working efficiency and better performance.

Description

Technical Field

[0001] The present invention relates to the technical field of electric digital data processing, and more specifically, to an injector optimization method and system based on an improved orthogonal test method. Background Art

[0002] The statements in this part only provide background technical information related to the present invention, and do not necessarily constitute prior art.

[0003] In modern seawater desalination technology, the multi-effect distillation-thermal vapor compression (MED-TVC) system has been widely used in water-scarce areas due to its high energy efficiency ratio and adaptability. One of the core components of this system, the injector, injects low-pressure fluid by high-pressure working fluid to achieve energy transfer and fluid mixing, and its performance directly affects the overall efficiency and operating cost of the system.

[0004] The existing injector design mainly relies on empirical formulas and experimental methods, and it is often difficult to balance efficiency improvement, structural rationality, and flow stability during the optimization process. The internal flow of the injector involves complex fluid mechanics phenomena, such as shock waves, turbulence, and phase changes, etc. These factors easily lead to energy loss and flow noise, affecting the long-term operation reliability of the equipment. In addition, the geometric structure of the injector has an important impact on its performance, but the existing optimization design methods often lack systematicness and are difficult to precisely adjust multiple geometric parameters to achieve optimal performance.

[0005] Previous research on injector geometric optimization mainly focused on performance maximization, usually using empirical formulas, CFD numerical simulations, or experimental methods for optimization, and then noise reduction was carried out through external equipment during the use of the injector, which increased the cost. At the same time, the noise has a certain impact on the working performance of the injector, affecting the service life of the injector.

[0006] In recent years, the combination of computational fluid dynamics (CFD) and artificial intelligence technology has provided new ideas for injector optimization. There are still many problems in the existing methods: First, the CFD simulation has a large amount of calculation and is difficult to efficiently handle multi-variable and multi-objective optimization problems; second, artificial intelligence models such as neural network models are highly dependent on data quality and may be affected by the distribution of training data, resulting in limited generalization ability; in addition, although the multi-objective optimization algorithm can balance between different objectives, its search efficiency depends on the initial population and fitness evaluation strategy, which may lead to local optimal solutions. Summary of the Invention

[0007] To solve the above problems, the present invention proposes an injector optimization method and system based on an improved orthogonal test method, constructs a model by comprehensively considering flow noise and performance optimization, and combines decision variable identification, an artificial neural network (ANN) surrogate model, and a non-dominated sorting genetic algorithm (NSGA-II) to achieve the optimal design of the injector with the best performance and the minimum flow noise.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] One or more embodiments provide an injector optimization method based on an improved orthogonal test method, including the following steps:

[0010] Construct an injector flow field simulation model and an acoustic field simulation model;

[0011] Based on the constructed models, decision variable identification is carried out by an improved orthogonal test method, with the goal of maximizing the entrainment ratio and minimizing the average sound pressure level, and a target function model is constructed through the trained neural network model;

[0012] Based on the target function model constructed by the trained neural network model, calculate the target function value as the fitness value, and perform multi-objective optimization on the geometric dimensions of the injector to obtain the optimized target dimensions.

[0013] One or more embodiments provide an injector optimization system based on an improved orthogonal test method, including:

[0014] A model construction module configured to construct an injector flow field simulation model and an acoustic field simulation model;

[0015] A target function model construction module configured to, based on the constructed models, carry out decision variable identification by an improved orthogonal test method, with the goal of maximizing the entrainment ratio and minimizing the average sound pressure level, and construct a target function model through the trained neural network model;

[0016] A target optimization module configured to calculate the target function value as the fitness value based on the target function model constructed by the trained neural network model, and perform multi-objective optimization on the geometric dimensions of the injector using the NSGA-II genetic algorithm to obtain the optimized target dimensions.

[0017] Compared with the prior art, the beneficial effects of the present invention are:

[0018] The present invention replaces the traditional CFD optimization calculation with a neural network model, reduces repeated simulation calculations, and improves the optimization efficiency. By adopting a data augmentation technique based on an improved orthogonal test method, the generalization ability of the neural network model is improved, making the optimization results more reliable. Combining with the NSGA-II genetic algorithm enables the entrainment ratio and average sound pressure level of the ejector to be optimized simultaneously, avoiding the limitations brought by single-objective optimization. The optimized ejector shows significant advantages in terms of velocity distribution, turbulent kinetic energy, and sound power level; after optimization, the flow loss is reduced, the air flow expansion state is improved, the vacuum degree is increased, and at the same time, the flow noise is effectively reduced, making the ejector have higher working efficiency and better performance.

[0019] The advantages of the present invention and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention, and the schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute a limitation to the present invention.

[0021] Figure 1 is a flowchart of the ejector optimization method according to Embodiment 1 of the present invention;

[0022] Figure 2 is a schematic diagram of the geometric parameters of the ejector according to Embodiment 1 of the present invention;

[0023] Figure 3 is a schematic diagram of training an artificial neural network model according to Embodiment 1 of the present invention;

[0024] Figure 4 is a comparison diagram of the velocity, turbulent kinetic energy, and sound power level cloud maps of the initial and optimized ejectors according to Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0027] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features in the present invention can be combined with each other. The embodiments will be described in detail below with reference to the drawings.

[0028] Embodiment 1

[0029] In the technical solutions disclosed in one or more embodiments, as Figures 1 to 4 shown, the injector optimization method based on the improved orthogonal test method includes the following steps:

[0030] Step 1: Construct a flow field simulation model and an acoustic field simulation model of the injector;

[0031] Step 2: Based on the constructed models, identify decision variables through the improved orthogonal test method, aiming to maximize the entrainment ratio and minimize the average sound pressure level, and construct an objective function model through the trained neural network model;

[0032] Step 3: Calculate the objective function value as the fitness value based on the objective function model constructed by the trained neural network model, and perform multi-objective optimization on the geometric dimensions of the injector to obtain the optimized target dimensions.

[0033] This method combines the advantages of computational fluid dynamics (CFD) simulation, neural network modeling, and multi-objective optimization algorithms. First, the flow field and acoustic field characteristics of the injector are accurately simulated through the simulation model, providing high-quality data support for optimization. Second, the improved orthogonal test method is used to identify decision variables and enhance data, improving the training effect of the neural network model. Finally, the NSGA-II genetic algorithm is used for multi-objective optimization to balance the performance of the injector among multiple objectives.

[0034] In this embodiment, the traditional CFD optimization calculation is replaced by a neural network model, reducing repeated simulation calculations and improving the optimization efficiency. The data augmentation technology based on the improved orthogonal experiment method is adopted to improve the generalization ability of the neural network model, making the optimization results more reliable. Combining with the NSGA-II genetic algorithm enables the entrainment ratio and average sound pressure level of the ejector to be optimized simultaneously, avoiding the limitations brought by single-objective optimization. The optimized ejector shows significant advantages in terms of velocity distribution, turbulent kinetic energy, and sound power level; after optimization, the flow loss is reduced, the air flow expansion state is improved, the vacuum degree is increased, and at the same time, the flow noise is effectively reduced, enabling the ejector to have working efficiency and better performance.

[0035] For a further technical solution, in step 2, in the improved orthogonal experiment method, multiple offset data corresponding to the current parameter data are obtained by dynamically adjusting the offset amount to expand the parameter data for data augmentation. The augmented data is input into the ejector flow field simulation model and the acoustic field simulation model for orthogonal experiments to obtain the corresponding experimental results, and the experimental results include the entrainment ratio and average sound pressure level of the ejector.

[0036] For a further technical solution, based on the identified decision variables, the data set is augmented as the training set through the improved orthogonal experiment method. Using the parameter values of the decision variables in the training set as the input and the entrainment ratio and average sound pressure level of the ejector model obtained under the corresponding parameter values as the output, the constructed neural network model is trained to obtain the objective function model constructed by the neural network model;

[0037] The experimental results of the orthogonal experiment method in this embodiment include the entrainment ratio and average sound pressure level of the ejector, and these results are used to evaluate the performance of the ejector. By adopting the improved orthogonal experiment method, the design space can be explored more comprehensively, the quality of the training data of the neural network model can be improved, and further the accuracy and stability of the optimization process can be enhanced; the above steps will be specifically described below.

[0038] As Figure 2 shown, the ejector mainly includes eleven main geometric structures, including the nozzle inlet diameter (D nl ), the nozzle converging section length (L nc ), the nozzle throat diameter (D t ), the nozzle diverging section length (L nd ), the nozzle outlet diameter (D no ), the nozzle outlet position (NXP), the pre-mixing chamber length (Lcs), the constant-area mixing chamber diameter (Dmix), the constant-area mixing chamber length (Lmix), the diffuser chamber diameter (Ddiff), and the diffuser chamber length (Ldiff).

[0039] In step 1, to construct the ejector flow field simulation model and the acoustic field simulation model, the following assumptions are first set:

[0040] 1) Assume that the fluid inside the injector is an ideal gas and follows the ideal gas equation;

[0041] 2) The injector wall is adiabatic;

[0042] In Step 1, construct the injector flow field simulation model and the acoustic field simulation model, and the steps are as follows:

[0043] Step 11, Geometry establishment: Use computer-aided engineering software to establish a two-dimensional geometric model of the fluid calculation domain of the injector;

[0044] Specifically, this geometric model defines the structural shape of the injector and contains 11 geometric parameters required for optimization; the purpose of this step is to construct an accurate physical model of the injector.

[0045] Step 12, Mesh generation: Perform mesh generation on the constructed two-dimensional geometric model to discretize the geometric model;

[0046] Furthermore, the mesh is refined near the inner cavity wall of the injector, while ensuring that the average mesh quality of the overall structure is greater than 0.9; by refining the mesh near the inner wall of the injector, the analytical ability of the boundary layer flow is enhanced;

[0047] The purpose of mesh generation is to transform the continuous geometric model into a discrete mesh suitable for numerical calculation, enabling the CFD solver to perform calculations.

[0048] Step 13, CFD simulation: After completing the mesh generation, import it into the CFD solver for numerical calculation to construct the injector flow field simulation model; use the FW-H acoustic analogy method to predict the flow noise and construct the acoustic field simulation model, relying on the meshed flow field data to calculate the acoustic characteristics.

[0049] In the injector flow field simulation model constructed by SST k-ω, the mesh of the calculation domain is used to discretize the Navier-Stokes equations during the simulation, enabling the calculation of fluid parameters (such as velocity, pressure, turbulence characteristics, etc.) in each mesh cell;

[0050] The acoustic solution part uses the FW-H acoustic analogy method to predict the flow noise and relies on the meshed flow field data to calculate the acoustic characteristics. The time step, boundary conditions, and discretization format are solved and iteratively converged based on the mesh cells.

[0051] Among them, FW-H (Ffowcs Williams-Hawkings) is a method based on the acoustic analogy theory for calculating the noise propagation in fluids.

[0052] SST k-ω (Shear Stress Transport k-ω) is a shear stress transport k-ω turbulence model, which combines the advantages of the k-ε model and the k-ω model and is applicable to turbulent simulations near the wall.

[0053] In some embodiments, the ejector flow field simulation model includes a steady-state turbulence model and a transient turbulence model. The steady-state turbulence model uses SST k-ω to obtain the velocity and turbulent kinetic energy distributions, and the transient turbulence model uses SAS scale adaptive simulation.

[0054] A steady-state turbulence model is constructed through the SST k-ω model, and the SST k-ω model is used to predict the internal flow field of the ejector. The specific model is as follows:

[0055] (1);

[0056] (2);

[0057] Among them, k represents the turbulent kinetic energy (Turbulent Kinetic Energy, TKE); ω represents the specific dissipation rate, that is, the ratio of the turbulent dissipation rate; represents the fluid density; represents the time; represents the generation term of the turbulent kinetic energy, characterizing the energy source of the turbulence; and are the generation and dissipation amounts of ω; and are user-defined source terms; and represent the generation and dissipation amounts of k;

[0058] and : represent the spatial coordinate components. The subscripts i and j are used to represent different spatial directions, usually the x and y directions;

[0059] : represent the velocity components. The subscript i corresponds to different spatial directions (for example, corresponds to the velocity in the x direction, corresponds to the velocity in the y direction, corresponds to the velocity in the z direction; describes the flow velocity of the fluid in the corresponding direction. The x direction, y direction, and z direction are three orthogonal directions in the spatial coordinate system;

[0060] : the dynamic viscosity of the fluid, indicating the viscosity of the fluid;

[0061] : Turbulent viscosity, which is used to describe the effect of turbulent motion on momentum transfer;

[0062] and : Are the Prandtl numbers of turbulent kinetic energy and specific dissipation rate respectively, which are used to adjust the diffusion term in the turbulence model;

[0063] Among them, the turbulent viscosity The calculation formula is as follows;

[0064] (3);

[0065] In the formula, S is the order of strain rate, is the low Reynolds number correction coefficient, is a constant.

[0066] represents the correction function, represents the intermediate variable for calculating the correction function The specific formula is as follows:

[0067] (4);

[0068] (5);

[0069] Among them, represents the wall distance, which is used to describe the distance from the fluid point to the wall in the turbulence model;

[0070] In some embodiments, the transient turbulence model in the ejector flow field simulation model is constructed by the SAS model (Scale Adaptive Simulation model), and the specific model is as follows:

[0071] The SAS model introduces an exact transport equation based on the sea surface temperature k-ω model, as shown in Equation (6) and Equation (7). The formula of the SAS model is as follows:

[0072] (6);

[0073] (7);

[0074] Among them, is the additional source term, which is obtained from Equation (8):

[0075] (8);

[0076] Among them, : Represents the model constant, which is used to adjust the relationship between the turbulent viscosity and the turbulent kinetic energy k and the specific dissipation rate ω in the turbulence model. Its value will affect the accuracy of turbulent calculation;

[0077] : Represents a model constant used to adjust the generation term in the turbulent dissipation rate equation;

[0078] : Represents a model constant used to adjust the dissipation term in the turbulent dissipation rate equation;

[0079] : Represents another turbulent Prandtl number used to adjust the diffusion term in the turbulent dissipation rate equation;

[0080] : Represents an additional source term used to account for additional physical effects or corrections in the turbulence model;

[0081] : Represents a set coefficient related to specific modifications or adjustments of the model;

[0082] : von Karman constant, used to describe the characteristics of turbulence near the wall, typically taking a value of about 0.4;

[0083] : Invariant of the velocity gradient, used to describe the shear characteristics of the flow;

[0084] : Represents a characteristic length scale, which may be related to the geometric size of the flow or the integral length scale;

[0085] : Represents a length scale related to the turbulent kinetic energy k, which may be used to adjust the diffusion or dissipation characteristics in the turbulence model;

[0086] : Represents a constant coefficient used to adjust the strength of the source term;

[0087] : Represents the Prandtl number related to the specific dissipation rate ω, used to adjust the diffusion term in the turbulence model;

[0088] : Represents the partial derivative symbol, indicating the partial derivative with respect to the corresponding variable.

[0089] The detailed definitions of other parameters and the values of the constants in the above equations are as follows:

[0090] (9);

[0091] Among them, Represents the strain rate tensor, which is the symmetric part of the velocity gradient tensor and is used to describe the velocity deformation of the fluid; Represents the shear factor; U' represents the magnitude of the velocity gradient, defined as the strain rate tensor The modulus of; U'' represents the modulus of the second-order velocity derivative, related to the curvature or rotational effect of the velocity field;

[0092] represents the damping function, used to adjust the parameters of the model near the wall;

[0093] tanh represents the hyperbolic tangent function, used for smooth transition;

[0094] represents the parameter of the damping function;

[0095] The max and min functions are used to limit the range of, ensuring numerical stability;

[0096] represents the positive diffusion term to prevent numerical instability, taking the larger value of the gradient product and the minimum value. This formula ensures that the diffusion term is non-negative, avoiding negative values or singularities in the calculation;

[0097] represents the component of the velocity vector, indicating the velocity of the fluid in the i-th spatial direction;

[0098] Further, the SAS model can use the standard RANS method in the steady-state region of the constructed geometric model and the LES-like method in the unsteady-state region for dynamic adjustment, thus saving a large amount of computing resources;

[0099] The RANS (Reynolds-Averaged Navier-Stokes) method is based on the Reynolds-averaged Navier-Stokes equations and deals with turbulent flows through time averaging. This method decomposes the instantaneous velocity field into the mean velocity field and the turbulent fluctuation part, and then averages the Navier-Stokes equations to obtain a set of solvable mean equations;

[0100] The LES-like (Large Eddy Simulation) method is a turbulent simulation method between the RANS method and direct numerical simulation (DNS). The LES method retains the large-scale turbulent structures through filtering operations for direct simulation, while the small-scale turbulent structures are modeled through sub-grid models. The LES-like method improves or simplifies the LES method in some aspects to adapt to specific engineering applications.

[0101] The flow noise of the injector is another performance parameter of concern in this study. The flow noise reflects to a certain extent the stability of the fluid dynamics inside the injector. If there is abnormal noise during the operation of the injector, it may be caused by unstable internal flow states, such as vortex, turbulence or cavitation, etc. These phenomena may affect the normal operation and service life of the injector. If the noise generated by the injector during operation is too large, it will have an adverse impact on the surrounding environment and human health. Therefore, pursuing a quieter injector can improve its performance stability and promote the wider application and development of the injector.

[0102] In this embodiment, the noise source and noise monitoring of the injector are realized through a numerical model. The construction process of the numerical model is as follows:

[0103] The detection of the noise index is carried out on a circle with a radius of 1 m centered at the center of the injector. A monitoring point is arranged every 15°, and a total of 13 monitoring points. The noise sound pressure level index is the average value of the 13 points to reflect. The FW-H acoustic analogy method is used in the acoustic model to predict the far-field noise.

[0104] In some embodiments, in step 2, decision variable identification is performed based on the constructed models. By calculating the contributions of various geometric parameters to the entrainment ratio (ER) and the average noise sound pressure level (SPLave), the key decision parameters and their value ranges are finally determined. The decision variable identification is carried out by an improved orthogonal test method, including the following steps:

[0105] Step 21: Obtain the geometric parameters of the injector for analysis, dynamically adjust the offset, and select multiple different level values for each parameter;

[0106] The level value means that there are multiple parameter values of different sizes corresponding to one parameter. Preferably, the offset is dynamically adjusted , and the calculation formula is:

[0107] (10);

[0108] Where represents the intermediate value of the geometric parameter in the orthogonal experimental design, represents the maximum value of the geometric parameter in the orthogonal experimental design, represents the minimum value of the geometric parameter in the orthogonal experimental design; is a known parameter, and are set parameter values; represents the deviation value. As the parameter value increases, the deviation value becomes smaller, thereby optimizing the parameter values of the orthogonal test and being able to more accurately find the influencing parameters;

[0109] Step 22: Randomly combine the horizontal values of the geometric parameters according to the given horizontal values to generate an experimental table and conduct an orthogonal experiment.

[0110] Step 23: Conduct experiments through the constructed injector flow field simulation model and acoustic field simulation model to obtain test results.

[0111] Step 24: Conduct a one-way analysis of variance on the test results to obtain the proportion of the influence values of each geometric parameter on the entrainment ratio (ER) and the average noise sound pressure level (SPLave). Take the parameters with an influence value proportion greater than the set value as decision variables and determine the value range of the decision variables.

[0112] The method for conducting a one-way analysis of variance on the test results to obtain the proportion of the influence values of each geometric parameter on the entrainment ratio (ER) and the average noise sound pressure level (SPLave) includes the following steps:

[0113] Step 241: Take each geometric parameter as an independent group, and set the null hypothesis ( ), and the alternative hypothesis ( ).

[0114] First, collect the simulation data of the entrainment ratio (ER) and the average noise sound pressure level (SPLave) under different geometric parameter settings and organize them into a format suitable for one-way analysis of variance, that is, each geometric parameter group is an independent group, and there are corresponding ER and SPLave values under each group.

[0115] Specifically, in this embodiment, each group contains 11 geometric structures of the injector (such as Figure 2 ), and there are certain differences in the parameters of the geometric structures between groups; for example, the nozzle throat diameter is different between two groups, or the nozzle inlet diameter changes, etc.

[0116] Specifically, establish the null hypothesis ( ), and the alternative hypothesis ( ), where the null hypothesis is that different geometric parameters have no significant effect on ER or SPLave, that is, the means of each group are equal; the alternative hypothesis is that at least one geometric parameter has a significant effect on ER or SPLave, that is, at least two groups have unequal means; significant effect is a relative concept. If there are multiple geometric parameters, after sorting according to the influence size, the corresponding top n geometric parameters can be defined as having a significant effect, where n is a set value.

[0117] Step 242: Calculate the sum of squares within groups (SSW) and the sum of squares between groups (SSB).

[0118] Specifically, SSW measures the difference between data points within each group and the mean of that group, and SSB measures the difference between the means of each group and the overall mean;

[0119] Step 243: Calculate the within-group degrees of freedom (dfW) and the between-group degrees of freedom (dfB);

[0120] To calculate the degrees of freedom, the within-group degrees of freedom (dfW) is the total number of samples minus the number of groups, and the between-group degrees of freedom (dfB) is the number of groups minus one;

[0121] Step 244: Based on the calculated sums of squares and degrees of freedom, calculate the mean square deviations to obtain the within-group mean square deviation MSW and the between-group mean square deviation MSB;

[0122] To calculate the mean square deviation (MS), that is, MSW is equal to SSW divided by dfW, and MSB is equal to SSB divided by dfB;

[0123] Step 244: Divide the between-group mean square deviation MSB by the within-group mean square deviation MSW to obtain the statistic F;

[0124] Step 245: Compare the obtained statistical value with the set critical value to determine whether each group satisfies the null hypothesis, so as to screen out the groups that satisfy the alternative hypothesis, that is, the groups that have an impact on ER or SPLave;

[0125] Specifically, the set critical value can be set to 0.05. Compare the calculated F statistic with the critical value. If the F statistic is greater than the critical value, reject the null hypothesis, indicating that at least one geometric parameter has a significant impact on ER or SPLave;

[0126] Step 246: For the screened groups, extract the data of the same geometric parameters in each group, calculate the mean value corresponding to each geometric parameter, and obtain the proportion of the influence value of each geometric parameter on the entrainment ratio (ER) or the average noise sound pressure level (SPLave);

[0127] Take the first few parameters with larger proportions as decision variables, and determine the approximate value range of each geometric parameter according to the set values that satisfy the conditions of ER or the average noise sound pressure level SPLave.

[0128] Specifically, if multiple groups do not meet the null hypothesis, extract the same parameters, such as extracting the mean value of the nozzle inlet diameter within all groups and extracting the mean value of the nozzle converging section length within all groups. By comparing the mean values, obtain the proportion of the influence value of each parameter; the larger the mean value, the larger the proportion of the influence value.

[0129] If it is found that certain geometric parameters have a significant impact on ER or SPLave, post hoc tests (such as Tukey test) can be further conducted to determine which specific parameters have significant differences, and charts of the means of each group can be drawn to visually show the impact of different geometric parameters on ER and SPLave.

[0130] In the identification of key factors in this embodiment, factors that have a significant impact on the response variable are quickly screened out through F values, and the contribution degree of each factor is clarified through the proportion to guide the priority allocation of resources; while reducing the number of experiments, the optimization direction of process parameters is accurately positioned to improve efficiency or quality; blindly adjusting non-significant factors is avoided to reduce R & D or production costs.

[0131] Traditional data augmentation methods expand data based on existing values according to experience, that is, by manually taking values. In this embodiment, the augmented data of each geometric parameter is dynamically generated through the set deviation value calculation formula, which can adaptively adjust the perturbation amplitude according to the distribution characteristics of the original data, effectively covering the potential feature space boundary while avoiding excessive deviation from real samples; secondly, when selecting key parameters among geometric parameters, combined with the one-way analysis of variance (One-Way ANOVA) method, the independent influence intensity of each geometric parameter on the target performance index can be quantitatively evaluated (such as ranking key parameters through F values), so as to preferentially implement directional enhancement for highly sensitive parameters, making the generated data more focused on the optimization interval of key parameter combinations;

[0132] Specifically, through decision variable identification, the determined decision variables include: the equal-area mixing chamber diameter D mix , the nozzle outlet position NXP, the nozzle expansion section length L nd , the nozzle outlet diameter D no and the pre-mixing chamber length L cs ;

[0133] Furthermore, with the goal of maximizing the entrainment ratio and minimizing the average sound pressure level, the method of constructing the objective function model through the trained neural network model includes the following steps:

[0134] Step S1, obtain the data of the decision variables for the ejector design, expand the training data through the improved orthogonal test method, and construct the data of the geometric parameters including the ejector decision variables and the simulation results of the corresponding entrainment ratio (ER) and average noise sound pressure level (SPLave);

[0135] Specifically, the improved orthogonal test method includes the above steps 21 to 23, which will not be elaborated here;

[0136] Step S2: Use the geometric parameter data in the training set as input, and the entrainment ratio (ER) and average noise sound pressure level (SPLave) as output to train the constructed neural network model;

[0137] Step S3: Calculate the prediction loss according to the prediction output of the neural network model, adjust the parameters of the neural network model, and iteratively train until the prediction accuracy is satisfied to obtain the trained model.

[0138] Step 3: Calculate the objective function value as the fitness value based on the objective function model constructed by the trained neural network model, and use the NSGA-II genetic algorithm to perform multi-objective optimization on the geometric dimensions of the ejector to obtain the optimized target dimensions, including the following steps:

[0139] Step 31: Initialize the population: Randomly generate N initial individuals using Latin Hypercube Sampling (LHS). Each individual is a combination of geometric parameters of the ejector, and set the population size P and the maximum number of iterations G;

[0140] In this embodiment, Latin Hypercube Sampling (LHS) is used to replace the traditional random initialization to improve the coverage of the population and reduce the deviation of the initial population;

[0141] Step 32: Calculate the objective function value based on the trained neural network as the fitness value;

[0142] Step 33: Select the individuals with fitness values greater than the set threshold to form a new parental population;

[0143] Step 34: Perform crossover operation and mutation operation on the individuals in the parental population to obtain the offspring population;

[0144] Step 35: Combine the offspring population and the parental population, and iteratively loop (execute Step 32) until the iteration cut-off condition is satisfied to obtain the final Pareto optimal solution set;

[0145] Step 36: Use the TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) method to fuse the entropy weight method to select the optimal solution among the frontier solutions of the Pareto optimal solution set; including the following steps:

[0146] Step 361: Calculate the objective weights using the entropy weight method;

[0147] In this embodiment, there are two objectives: maximizing the entrainment ratio of the ejector and minimizing the average sound pressure level;

[0148] The objective weight calculation formula is:

[0149] ;

[0150] Among them, is the entropy value of the m-th objective; is the normalized value of individual i on objective m; N is the number of individuals;

[0151] Step 362: Select the optimal solution using the TOPSIS method based on the calculated objective weights;

[0152] 1). Calculate the positive ideal solution A1 and the negative ideal solution A2;

[0153] 2). Calculate the distances from each solution to the positive ideal solution and the negative ideal solution based on the objective weights;

[0154] The distance from each solution to the positive ideal solution is:

[0155] ;

[0156] The distance from each solution to the negative ideal solution is:

[0157] ;

[0158] Among them, represents the fitness value of the i-th individual under the m-th objective in the Pareto optimal solution set;

[0159] 3). Calculate the scoring value of each individual, and select the individual with the highest score as the optimal solution; The calculation of the scoring value is:

[0160] ;

[0161] In the above solution, in a multi-objective optimization problem, the Pareto optimal solution set usually contains multiple possible solutions, each solution having different trade-offs in different objective dimensions. The search efficiency of traditional multi-objective optimization algorithms is limited by the quality of the initial population and the fitness evaluation strategy, which may lead to convergence to local optimal solutions. To improve the global optimality of the solutions and the decision-making quality of the optimization process, the solution of this embodiment uses the TOPSIS method combined with the entropy weight method to screen the optimal solutions. The entropy weight method is used to measure the effective information content of each objective, calculate the weights of each objective according to the information entropy, ensure that subjective weighting is not relied on during the objective trade-off process, but objectively allocate weights according to the degree of data dispersion, and enhance the objectivity of the optimization process. Subsequently, the TOPSIS method is used to calculate the closeness of each Pareto front solution to the ideal solution based on the construction of the ideal solution and the negative ideal solution, and select the optimal solution. The screening of the optimal solution based on the principle of approaching the ideal solution reduces the risk of falling into local optima compared with traditional multi-objective optimization algorithms; avoids the computational cost of directly traversing all solutions, can quickly determine the optimal solution, and improves the convergence speed and solution efficiency of the algorithm.

[0162] In this embodiment, since the traditional CFD simulation has a large amount of calculation, it affects the optimization efficiency. A neural network model is used to construct the objective function model, and the trained neural network is used to quickly predict the entrainment ratio and average sound pressure level of the ejector, thereby replacing the CFD simulation for fitness evaluation. This method reduces the consumption of computing resources, improves the efficiency and accuracy of fitness evaluation, and thus reduces the risk of the algorithm falling into local optimal solutions. The search space coverage rate is improved by an improved orthogonal test method. In the solution, an improved orthogonal test method is used to enhance the parameter data, increasing the data distribution diversity of the decision variables. The data set is expanded by dynamically adjusting the offset, making the search space of the optimization algorithm more uniform, and thus reducing the probability of local optimal solutions. The NSGA-II (Non-dominated Sorting Genetic Algorithm) is adopted in the solution. This algorithm selects individuals based on non-dominated sorting, so that the Pareto front can be explored more evenly during the optimization process, reducing the occurrence of local optimal problems.

[0163] To illustrate the effect of the above method of this embodiment, a comparative test was carried out, and the performance of the initial ejector was compared with that of the ejector optimized by the method of this embodiment.

[0164] By comparing the flow field cloud maps and spectrograms of the ejector before and after optimization, the effectiveness of the optimization method was verified. The optimized ejector shows a more uniform velocity gradient in the flow field distribution, reducing the sound waves generated by the collision, friction and shear between fluid micro-clusters, thereby significantly reducing the flow noise. At the same time, the optimized ejector achieves full expansion under the design conditions, improves the vacuum degree, enhances the entrainment ability of the secondary flow, and further improves the overall performance of the MED-TVC system.

[0165] Spectrum analysis shows that the noise energy of the optimized ejector is significantly reduced within the main frequency bands, verifying the noise control effect. The comparison of the initial and optimized ejectors in terms of velocity, turbulent kinetic energy, and sound power level is shown in Figure 4 .

[0166] As Figure 4 shown in the small figure (a) in , Velocity (m / s): velocity (meters per second), representing the velocity distribution of the fluid. The initial length Lcs of the premixing chamber was 45.94 mm, and the optimized length Lcs’ is 50.82 mm. The initial nozzle exit position NXP was 0, and the optimized position NXP’ is 6.27 mm; the initial radius Rmix was 11.09 mm, and the optimized radius Rmix’ changed to 11.68 mm;

[0167] For the optimized ejector, the length of the premixing chamber increases, and the diameter of the constant-area mixing chamber increases, while the nozzle expansion angle decreases. These optimization measures reduce the shock oscillation in the mixing chamber, lower the energy loss, and thus improve the ejector efficiency.

[0168] In addition, at the nozzle exit of the initial ejector, the primary flow is in an over-expanded state, while the optimized ejector achieves full expansion, improving the vacuum degree and enhancing the entrainment ability of the secondary flow. The velocity profile of the optimized ejector also supports this, with the secondary flow velocity increasing, further enhancing the secondary flow rate and improving the internal energy of the ejector.

[0169] The secondary flow is entrained and flows in layers with the primary flow in the premixing chamber. However, there is a significant velocity difference between the two fluids, resulting in obvious local changes in velocity and density gradients. Therefore, the momentum and energy exchange between fluid micro-clusters become more frequent and intense, accompanied by rapid friction and shear effects, forming the Figure 4 characteristics of the turbulent kinetic energy (Turbulent Kinetic Energy) distribution shown in the small figure (b) in . Specifically, at the interface region where the two fluids interact, the TKE reaches the maximum value. As Figure 4 shown in the small figure (b) in , both the premixing chamber and the diffuser chamber exhibit relatively high TKE. In a cavity flow field, such as an ejector, an increase in TKE usually leads to a higher noise level. The high TKE region of the optimized ejector is significantly reduced, thus lowering the flow noise. Figure 4Sub - figure (c) therein further supports this point, where the distribution of the Acoustic Power Level (APL) closely corresponds to the TKE distribution. In the high - TKE region, the turbulent activity inside the fluid is particularly intense, resulting in an increase in the sound waves generated by the collisions, frictions, and shears between fluid micro - clusters. These sound waves carry energy, which is quantified by the acoustic power level. Therefore, regions with higher TKE are usually associated with higher acoustic power levels, indicating a greater contribution to noise. Therefore, based on the above analysis of these contours, the reasons why the optimized ejector has a lower noise level and better entrainment performance are explained in detail.

[0170] It can be seen that the optimized ejector shows significant advantages in terms of velocity distribution, turbulent kinetic energy, and acoustic power level. After optimization, the flow loss is reduced, the state of air - flow expansion is improved, the vacuum degree is increased, and at the same time, the flow noise is effectively reduced, enabling the ejector to have higher working efficiency and better performance.

[0171] Example 2

[0172] Based on Example 1, in this example, an ejector optimization system based on an improved orthogonal test method is provided, including:

[0173] A model - building module, configured to build an ejector flow - field simulation model and an acoustic - field simulation model;

[0174] A target - function model - building module, configured to identify decision variables through an improved orthogonal test method based on the built models, with the goal of maximizing the entrainment ratio and minimizing the average sound - pressure level, and build a target - function model through a trained neural - network model;

[0175] A target - optimization module, configured to calculate the target - function value as the fitness value based on the target - function model built by the trained neural - network model, and perform multi - objective optimization on the geometric dimensions of the ejector using the NSGA - II genetic algorithm to obtain the optimized target dimensions.

[0176] Furthermore, in the improved orthogonal test method, multiple offset data corresponding to the current parameter data are obtained by dynamically adjusting the offset to expand the parameter data for data augmentation, and the augmented data are input into the ejector flow - field simulation model and the acoustic - field simulation model for orthogonal tests to obtain the corresponding test results, where the test results include the entrainment ratio and the average sound - pressure level of the ejector.

[0177] Furthermore, the target - function model - building module includes a decision - variable identification module, and the decision - variable identification module is configured to identify decision variables through an improved orthogonal test method, including the following steps:

[0178] Obtain the geometric parameters of the ejector for analysis, and select multiple different level values for each parameter;

[0179] Randomly combine the horizontal values of geometric parameters according to the given horizontal values to obtain a generated experimental table and conduct an orthogonal experiment;

[0180] Conduct experiments through the constructed injector flow field simulation model and acoustic field simulation model to obtain test results;

[0181] Conduct a one-way analysis of variance on the test results to obtain the proportion of the influence values of each geometric parameter on the entrainment ratio and the average noise sound pressure level. Take the parameters with an influence value proportion greater than the set value as decision variables and determine the value range of the decision variables.

[0182] It should be noted here that each module in this embodiment corresponds to each step in Embodiment 1, and the specific implementation process is the same, so it will not be repeated here.

[0183] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. The injector optimization method based on the improved orthogonal test method is characterized by: The steps include: Construct the ejector flow field simulation model and the acoustic field simulation model; Based on the constructed models, the decision variables are identified through the improved orthogonal test method, and the objective function model is constructed through the trained neural network model with the goal of maximizing the ejection ratio and minimizing the average sound pressure level. The objective function model constructed based on the trained neural network model calculates the objective function value as the fitness value, performs multi-objective optimization on the geometric dimensions of the injector, and obtains the optimized target dimensions; The decision variable identification is carried out by using the improved orthogonal test method, which includes the following steps: Obtain the geometric parameters of the ejector for analysis, dynamically adjust the offset, and select multiple different level values ​​for each parameter; According to the given horizontal values, the horizontal values ​​of the geometric parameters are randomly combined to obtain the generated experimental table and conduct orthogonal experiments; Experiments were conducted using the constructed ejector flow field simulation model and acoustic field simulation model to obtain test results; A one-way ANOVA was performed on the test results to obtain the influence ratio of each geometric parameter on the induced emission ratio and the average noise sound pressure level. The parameters with an influence ratio greater than the set value were taken as decision variables, and the value range of the decision variables was determined. The method of performing one-way variance analysis on the test results to obtain the influence ratio of each geometric parameter on the induced emission ratio and the average noise sound pressure level includes the following steps: Treat each geometric parameter as an independent group, and set the null hypothesis and alternative hypothesis; the null hypothesis is that different geometric parameters have no significant effect on the entrainment ratio and the average noise sound pressure level, and the alternative hypothesis is that at least one geometric parameter has a significant effect on the entrainment ratio or the average noise sound pressure level; Calculate the within-group sum of squares and between-group sum of squares; Calculate the within-group and between-group degrees of freedom; Based on the calculated sums of squares and degrees of freedom, calculate the mean square error to obtain the within-group mean square error and between-group mean square error; Divide the mean square error between groups by the mean square error within groups to get the statistic F; Compare the obtained statistical value with the set critical value to determine whether each group meets the null hypothesis, so as to screen out the group that meets the alternative hypothesis; For the screened groups, the data of the same geometric parameters in each group are taken out, the mean of each geometric parameter is calculated, and the influence ratio of each geometric parameter on the induced emission ratio or the average noise sound pressure level is obtained.

2. The injector optimization method based on the improved orthogonal test method according to claim 1, characterized in that: The improved orthogonal test method obtains multiple offset data corresponding to the current parameter data by dynamically adjusting the offset, expands the parameter data for data enhancement, and inputs the enhanced data into the ejector flow field simulation model and the sound field simulation model for orthogonal testing to obtain the corresponding test results, which include the ejector's entrainment ratio and the average sound pressure level.

3. The injector optimization method based on the improved orthogonal test method according to claim 1, characterized in that: Based on the identified decision variables, the data set is expanded as the training set through the improved orthogonal test method. The parameter values ​​of the decision variables in the training set are used as input, and the injection ratio and average sound pressure level of the ejector model obtained under the corresponding parameter values ​​are used as output to train the constructed neural network model and obtain the objective function model constructed by the neural network model.

4. The injector optimization method based on the improved orthogonal test method according to claim 1, characterized in that: The objective function model constructed based on the trained neural network model calculates the objective function value as the fitness value, and the NSGA-II genetic algorithm is used to perform multi-objective optimization on the geometric dimensions of the injector to obtain the optimized target dimensions, including the following steps: Latin hypercube sampling is used to randomly generate N initial individuals, each of which is a combination of geometric parameters of the ejector, and the population size P and the maximum number of iterations G are set; Calculate the objective function value based on the trained neural network as the fitness value; Select individuals whose fitness values ​​are greater than the set threshold to form a new parent population; Perform crossover and mutation operations on the individuals of the parent population to obtain the offspring population; The offspring population and the parent population are merged, and the iteration cycle is repeated until the iteration cutoff condition is met to obtain the final Pareto optimal solution set.

5. The injector optimization method based on the improved orthogonal test method according to claim 4, characterized in that: The TOPSIS method is integrated with the entropy weight method to select the optimal solution from the frontier solutions of the Pareto optimal solution set, including the following steps: The entropy weight method is used to calculate the target weight; Based on the calculated objective weights, the TOPSIS method is used to select the optimal solution.

6. The injector optimization system based on the improved orthogonal test method is characterized by: include: A model building module is configured to build an ejector flow field simulation model and an acoustic field simulation model; The objective function model building module is configured to identify decision variables through an improved orthogonal test method based on each constructed model, and to build an objective function model through a trained neural network model with the goal of maximizing the ejection ratio and minimizing the average sound pressure level; A target optimization module is configured to calculate a target function value as a fitness value based on a target function model constructed by the trained neural network model, and to perform multi-objective optimization on the geometric dimensions of the injector using an NSGA-II genetic algorithm to obtain an optimized target dimension; The decision variable identification is carried out by using the improved orthogonal test method, which includes the following steps: Obtain the geometric parameters of the ejector for analysis, dynamically adjust the offset, and select multiple different level values ​​for each parameter; According to the given horizontal values, the horizontal values ​​of the geometric parameters are randomly combined to obtain the generated experimental table and conduct orthogonal experiments; Experiments were conducted using the constructed ejector flow field simulation model and acoustic field simulation model to obtain test results; A one-way ANOVA was performed on the test results to obtain the influence ratio of each geometric parameter on the induced emission ratio and the average noise sound pressure level. The parameters with an influence ratio greater than the set value were taken as decision variables, and the value range of the decision variables was determined. The method of performing one-way variance analysis on the test results to obtain the influence ratio of each geometric parameter on the induced emission ratio and the average noise sound pressure level includes the following steps: Treat each geometric parameter as an independent group, and set the null hypothesis and alternative hypothesis; the null hypothesis is that different geometric parameters have no significant effect on the entrainment ratio and the average noise sound pressure level, and the alternative hypothesis is that at least one geometric parameter has a significant effect on the entrainment ratio or the average noise sound pressure level; Calculate the within-group sum of squares and between-group sum of squares; Calculate the within-group and between-group degrees of freedom; Based on the calculated sums of squares and degrees of freedom, calculate the mean square error to obtain the within-group mean square error and between-group mean square error; Divide the mean square error between groups by the mean square error within groups to get the statistic F; Compare the obtained statistical value with the set critical value to determine whether each group meets the null hypothesis, so as to screen out the group that meets the alternative hypothesis; For the screened groups, the data of the same geometric parameters in each group are taken out, the mean of each geometric parameter is calculated, and the influence ratio of each geometric parameter on the induced emission ratio or the average noise sound pressure level is obtained.

7. The injector optimization system based on the improved orthogonal test method according to claim 6, characterized in that: The improved orthogonal test method obtains multiple offset data corresponding to the current parameter data by dynamically adjusting the offset, expands the parameter data for data enhancement, and inputs the enhanced data into the ejector flow field simulation model and the sound field simulation model for orthogonal testing to obtain the corresponding test results, which include the ejector's entrainment ratio and the average sound pressure level.

8. The injector optimization system based on the improved orthogonal test method according to claim 6, characterized in that: The objective function model building module includes a decision variable identification module, which is configured to identify decision variables through an improved orthogonal test method, including the following steps: The geometric parameters of the ejector are obtained for analysis, and a plurality of different level values ​​are selected for each parameter; According to the given horizontal values, the horizontal values ​​of the geometric parameters are randomly combined to obtain the generated experimental table and conduct orthogonal experiments; Experiments were conducted using the constructed ejector flow field simulation model and acoustic field simulation model to obtain test results; A one-way ANOVA was performed on the test results to obtain the influence ratio of each geometric parameter on the entrainment ratio and the average noise sound pressure level. The parameters with an influence ratio greater than the set value were taken as decision variables, and the value range of the decision variables was determined.

Images