A wind tunnel control driving parameter intelligent generation method based on an XGBoost machine learning method

Abstract

The application belongs to the technical field of aerospace vehicle ground simulation test and machine learning, and discloses a wind tunnel control driving parameter intelligent generation method based on an XGBoost machine learning method, which comprises the following steps: data screening and processing, input / output variables participating in modeling are screened out from wind tunnel historical test operation data, and control driving parameters after flow field stabilization are extracted to form a data set; feature analysis, input / output variables of the model are determined according to the control relationship between the wind tunnel flow field adjusting mechanism position and the flow field pressure parameter, and combined with the correlation analysis atlas; model construction, the data set is divided into a training set and a test set, XGBoost algorithm is adopted to establish each driving parameter prediction model, and parameter optimization, training, testing and verification are iteratively carried out; parameter generation and verification, based on the trained XGBoost parameter prediction model, input the current test condition, and intelligently generate the control driving parameter combination.

Description

Technical Field

[0001] This application belongs to the field of ground simulation testing and machine learning technology for aerospace vehicles, and particularly relates to an intelligent generation method for wind tunnel control start-up parameters based on the XGBoost machine learning method. Background Technology

[0002] A certain transient transonic wind tunnel possesses high-fidelity geometric simulation and high Reynolds number testing capabilities, playing a crucial role in the aerodynamic simulation testing of my country's spacecraft. As a transient transonic wind tunnel, it is characterized by multiple controlled objects, strong coupling of flow field parameters, and rapid instantaneous / steady-state switching.

[0003] To effectively shorten the test time and improve the accuracy of flow field control, it is required to accurately give the control start-up parameters for each test, including the initial preset values ​​of the positions of the main pressure regulating valve (hereinafter referred to as main regulator), main exhaust valve (hereinafter referred to as main exhaust), stagnation chamber flow valve / pressure regulating valve (hereinafter referred to as stagnation flow / stagnation regulator), grid finger mechanism and other flow field adjustment mechanisms, as well as the pressure parameters of the main ejector (hereinafter referred to as main ejector) and stagnation chamber ejector (hereinafter referred to as auxiliary ejector).

[0004] Precisely setting the initial preset value of the flow field adjustment mechanism position will enable the wind tunnel transient pressurization process to quickly enter the steady-state adjustment stage of the flow field, thereby significantly shortening the steady-state adjustment process. At the same time, precise setting of flow field pressure parameters will ensure that the flow field closed-loop control is in the best adjustment condition, laying the foundation for high-precision flow field control.

[0005] Typically, the setting of start-up control parameters is determined by personnel based on prior knowledge and manual querying of historical run data. This experience-based parameter setting method often has the following problems: 1. Each operating condition requires a long time to prepare parameters for the first start-up; 2. Differences in the experience of personnel may lead to significant deviations in control parameters; 3. Manually querying historical run data may be limited by the coverage of existing experience cases and cannot fully meet the changing needs of new operating conditions. The quality of the given start-up control parameters directly affects the length of the blowing time and even the success or failure of the blowing. For wind tunnels with huge energy consumption, establishing test flow field conditions more quickly and shortening the blowing time is of great practical significance for energy conservation. Summary of the Invention

[0006] The purpose of this application is to overcome the problems of the prior art by disclosing an intelligent generation method for wind tunnel control start-up parameters based on the XGBoost machine learning method, which is used to automatically provide accurate control start-up parameters.

[0007] The objective of this application is achieved through the following technical solution:

[0008] A method for intelligently generating wind tunnel control start-up parameters based on the XGBoost machine learning approach, the method comprising:

[0009] S1: Data screening and processing, including: screening input / output variables involved in modeling from historical wind tunnel test operation data, and extracting control start-up parameters after the flow field stabilizes to form a dataset;

[0010] S2: Feature analysis, including: determining the input / output variables of the model based on the control relationship between the location of the wind tunnel flow field adjustment mechanism and the flow field pressure parameters, combined with the correlation analysis spectrum;

[0011] S3: Model building, including: dividing the dataset obtained in step S1 into training set and test set, using the XGBoost algorithm to build prediction models for each driving parameter, and iteratively carrying out parameter tuning, training, testing and validation;

[0012] S4: Parameter generation and verification, including: intelligently generating control driving parameter combinations based on the trained XGBoost parameter prediction model and the input of the current test conditions.

[0013] According to a preferred embodiment, the data selected in step S1 includes:

[0014] Time series data: main load pressure Assisted pressure Tonic displacement , stationary displacement Grid finger displacement , standing displacement Main row displacement Angle of attack α Total pressure static pressure Reference Mach number Gas source pressure ,

[0015] And, nominal Mach number Atmospheric pressure test section Type and model of incoming flow blockage at α=0° Data, incoming flow congestion It is the sum of the blockage degree including the model, support mechanism and corresponding auxiliary mechanism.

[0016] According to a preferred embodiment, in step S1, the process of forming the dataset includes:

[0017] In total pressure and reference Mach number After stabilization, extract the values ​​of the positions of each flow field regulating mechanism and the flow field pressure parameters at the start time of α from the time series data, and compare them with the incoming flow blockage degree. Atmospheric pressure nominal Mach number Create a dataset.

[0018] According to a preferred embodiment, the control relationship between the position of the wind tunnel flow field adjustment mechanism and the flow field pressure parameters is as follows:

[0019] Main pressure By the dominant shift Closed-loop regulation; total pressure Displacement by main row Closed-loop control, and subject to main traction pressure. Impact; Assisted pressure Displacement by stationary adjustment Closed-loop control; static pressure Displacement by gate finger and stagnant displacement Controlled individually or jointly, and subject to auxiliary pressure. Influence;

[0020] Meanwhile, the degree of inflow blockage For static pressure This has an impact on total pressure. and static pressure Decide to refer to Mach number Size.

[0021] According to a preferred embodiment, step S3 includes: establishing an XGBoost output regression task, establishing a prediction model for each control parameter that needs to be intelligently generated, and performing hyperparameter tuning for each.

[0022] According to a preferred embodiment, in step S3, the hyperparameters are tuned using a Bayesian optimization method, with the root mean square error (RMSE) of the dataset as the optimization criterion. The hyperparameters of each independent XGBoost model are tuned, and the optimal parameter combination is used to train the model on the training set.

[0023] According to a preferred embodiment, step S3 further includes: evaluating the performance of each model using a test set, wherein the performance evaluation process uses the RMSE of the training set and the test set as the evaluation metric.

[0024] According to a preferred embodiment, in step S4, the known quantity in the current test conditions is the total pressure. nominal Mach number Gas source pressure Atmospheric pressure Flow blockage test section ;

[0025] The combination of control parameters to be intelligently generated is: main descent pressure. Assisted pressure Tonic displacement , stationary displacement Grid finger displacement , standing displacement Main row displacement .

[0026] According to a preferred embodiment, step S4 further includes parameter verification, which compares and verifies the intelligently generated driving parameters, the manually given driving parameters, and the closed-loop steady-state values ​​in the time series data based on the absolute error Δ, thereby verifying the feasibility of the model's intelligent parameters.

[0027] The aforementioned main solution and its various further alternative solutions can be freely combined to form multiple solutions, all of which are solutions that can be adopted and are claimed in this application. Those skilled in the art, after understanding the solution of this application, will realize that there are many combinations based on the prior art and common general knowledge, all of which are technical solutions to be protected in this application, and will not be exhaustively listed here.

[0028] The beneficial effects of this application are:

[0029] This application analyzes historical wind tunnel test data and constructs multiple initial control parameter generation models. By inputting test condition data, it automatically generates each initial control parameter. Compared to existing technologies, the initial control parameters for wind tunnels are transformed from being given manually based on experience to being intelligently generated. This increases the success rate of the first wind run, reduces the wind tunnel control system adjustment time, minimizes human error from manually given parameters, significantly improves wind tunnel testing efficiency and accuracy, and enhances the digitalization level of wind tunnels. It has significant engineering application value in the aerospace field. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall process of the intelligent generation method for wind tunnel control start-up parameters in this application;

[0031] Figure 2 This is a schematic diagram showing the control relationship between the position of the wind tunnel flow field adjustment mechanism and the flow field pressure parameters.

[0032] Figure 3 This is a correlation analysis graph of the dataset. Detailed Implementation

[0033] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features can be combined with each other. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0034] Example

[0035] Taking a transient transonic wind tunnel as an example, such as Figure 1 As shown, a method for intelligently generating wind tunnel control driving parameters based on XGBoost machine learning includes the following steps S1-S4.

[0036] Step S1: Data filtering and processing.

[0037] Step S11, Data Filtering. Input / output variables that can be used for modeling are filtered from the experimental outlines, experimental task sheets, control data, and other sensor data from previous years. The filtered data fields are shown in Table 1. Used to determine and The correction relationship; The incoming flow blockage degree is the sum of the blockage degrees of the model, support mechanism and other auxiliary mechanisms when the model α=0°, and is a specific numerical value. The current atmospheric pressure value is collected by sensors before each driving session; and The test conditions are specified on the task sheet, and the control data consists of the operational data of the position of the flow field adjustment mechanism and the flow field pressure parameters. It is a time-series data covering the entire process from the start to the end of the test.

[0038] Table 1. Data selected for modeling

[0039]

[0040] Step S12, Data Processing. Extract the location of the flow field regulation mechanism and the flow field pressure parameter values ​​at the stable flow field moment for each train from the time series data, and compare them with... , , Together, they serve as a dataset for machine learning. The method for extracting numerical values ​​from time-series data is as follows: starting with α (…). and Using the time after stabilization as a marker, continuously select from the time series data. n We take the average of the points; in this embodiment, we take... n =20. Specifically, extract The value of time The final dataset is shown in Table 2.

[0041] Table 2. Dataset example table formed after data extraction according to the method of this application.

[0042]

[0043] Step S2, Feature Analysis

[0044] Based on the control relationship between the location of the wind tunnel flow field adjustment mechanism and the flow field pressure parameters, and combined with the correlation analysis spectrum, the model input / output variables are determined. The specific steps are as follows:

[0045] Step S21, Parameter Mechanism Analysis. Analyze the control relationship between the position of the wind tunnel adjustment mechanism and the flow field pressure parameters, such as... Figure 2 As shown, that is: Depend on Closed-loop regulation; Depend on Closed-loop control, and subject to Influence; Depend on Closed-loop control; Depend on and Controlled individually or jointly, and subject to Impact. Additionally... It will also affect To have an impact and Decide Size.

[0046] Step S22, parameter correlation analysis. For example... Figure 3 As shown, and , , , High correlation; and , , , , High correlation; and , , , , , High correlation; and , , , , , High correlation; and , , , , High correlation; and , , , , , High correlation; and , , , The correlation is high.

[0047] Step S23, Model Input / Output Selection. Based on the parameter mechanism in Step S21 and the correlation analysis results in Step S22, input variables that have a significant impact on the target parameters are selected. The known conditions are... , , , , , In addition, the stabilizing current, stabilizing adjustment, and auxiliary traction of the stabilizing pump system need to be opened, closed, and set together; according to the Mach number control strategy, It is also related to the activation status of the venting system; while the main exhaust, as the exhaust channel for the entire wind tunnel, is directly connected to the outside atmosphere. Also with The relevant information is shown in Table 3 for the features and target selection of each prediction model in this embodiment.

[0048] Table 3 Feature and target selection used in training the XGBoost prediction model

[0049]

[0050] Step S3, Model Building

[0051] Using XGBoost to implement a multi-output regression task, the model undergoes iterative parameter tuning, training, testing, and validation. The specific steps are as follows:

[0052] Step S31: Divide the dataset into a training set and a test set in an 8:2 ratio.

[0053] Step S32: Use the XGBRegressor class to build multiple independent XGBoost models, each model is responsible for predicting a target parameter.

[0054] Step S33: Perform hyperparameter tuning using Bayesian optimization. Using the root mean square error (RMSE) of the dataset as the optimization criterion, perform hyperparameter tuning for each independent XGBoost model, tuning the following key parameters: n_estimators, learning_rate, max_depth, subsample, colsample_bytree, gamma, min_child_weight, reg_alpha, and reg_lambda. Then, train the model on the training set using the optimal parameter combinations.

[0055] Step S34: Evaluate the performance of each model using the test set, employing the RMSE of the training and test sets as the evaluation metric.

[0056] Step S4: Intelligent Parameter Generation and Verification

[0057] Based on the trained XGBoost model, the experimental conditions are input, and the control parameter combination is output. The specific steps are as follows:

[0058] Step S41, according to each of and Correct the relationship, and add the information on the task sheet. Revised to .

[0059] Step S42, the order of predicting start-up parameters is shown in Table 4. Each time the system starts up, the test conditions (i.e.:) are input. , , , , This allows for the intelligent generation of control driving parameter combinations (i.e.: , , , , , , ).

[0060] Table 4 Input-output relationship for parameter prediction using the trained XGBoost prediction model

[0061]

[0062] Step S43, parameter verification: Based on the absolute error Δ, the intelligently generated start-up parameters, the manually given start-up parameters, and the closed-loop steady-state values ​​in the time series data are compared and verified. The results are shown in Tables 5, 6, and 7. Analysis shows that the intelligently generated control parameter combination is closer to the overall closed-loop steady-state value and can be used for start-up parameters in transient wind tunnel tests.

[0063] Table 5 Comparison of intelligent driving parameters, manual driving parameters, and closed-loop steady-state values ​​extracted from time-series data

[0064]

[0065] Table 6 Comparison of intelligent driving parameters, manual driving parameters, and closed-loop steady-state values ​​extracted from time-series data

[0066]

[0067] Table 7 Comparison of intelligent driving parameters, manual driving parameters, and closed-loop steady-state values ​​extracted from time-series data

[0068]

[0069] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for intelligent generation of wind tunnel control driving parameters based on XGBoost machine learning, characterized in that, The intelligent generation method for wind tunnel control start-up parameters includes: S1: Data screening and processing, including: screening input / output variables involved in modeling from historical wind tunnel test operation data, and extracting control start-up parameters after the flow field stabilizes to form a dataset; S2: Feature analysis, including: determining the input / output variables of the model based on the control relationship between the location of the wind tunnel flow field adjustment mechanism and the flow field pressure parameters, combined with the correlation analysis spectrum; Step S2 includes: Step S21, parameter mechanism analysis, the control relationship between the position of the wind tunnel flow field adjustment mechanism and the flow field pressure parameters is as follows: Main pressure By the dominant shift Closed-loop regulation; total pressure Displacement by main row Closed-loop control, and subject to main traction pressure. Impact; Assisted pressure Displacement by stationary adjustment Closed-loop control; static pressure Displacement by gate finger and stagnant displacement Controlled individually or jointly, and subject to auxiliary pressure. Influence; Meanwhile, the degree of inflow blockage For static pressure This has an impact on total pressure. and static pressure Decide to refer to Mach number Size; Step S22, Parameter Correlation Analysis: With nominal Mach number , , , High correlation; and , , , , High correlation; and , , Gas source pressure , , High correlation; and , , , , , High correlation; and , , , , High correlation; and , , , , , High correlation; and , , , High correlation; Step S23, Model input and output selection: Based on the parameter mechanism in step S21 and the correlation analysis results in step S22, the input variables that have a greater impact on the target parameters are selected. S3: Model building, including: dividing the dataset obtained in step S1 into training set and test set, using the XGBoost algorithm to build prediction models for each driving parameter, and iteratively carrying out parameter tuning, training, testing and validation; S4: Parameter generation and verification, including: intelligently generating control driving parameter combinations based on the trained XGBoost parameter prediction model and the input of the current test conditions.

2. The intelligent generation method for wind tunnel control start-up parameters as described in claim 1, characterized in that, The data filtered in step S1 includes: Time series data: main load pressure Assisted pressure Tonic displacement , stationary displacement Grid finger displacement , standing displacement Main row displacement Angle of attack α Total pressure static pressure Reference Mach number Gas source pressure , And, nominal Mach number Atmospheric pressure test section Type and model of incoming flow blockage at α=0° Data, incoming flow congestion It is the sum of the blockage degree including the model, support mechanism and corresponding auxiliary mechanism.

3. The intelligent generation method for wind tunnel control start-up parameters as described in claim 2, characterized in that, In step S1, the process of forming the dataset includes: In total pressure and reference Mach number After stabilization, extract the values ​​of the positions of each flow field regulating mechanism and the flow field pressure parameters at the start time of α from the time series data, and compare them with the incoming flow blockage degree. Atmospheric pressure nominal Mach number Create a dataset.

4. The intelligent generation method for wind tunnel control start-up parameters as described in claim 1, characterized in that, Step S3 includes: establishing an XGBoost output regression task, building a prediction model for each control parameter that needs to be intelligently generated, and performing hyperparameter tuning for each.

5. The intelligent generation method for wind tunnel control start-up parameters as described in claim 4, characterized in that, In step S3, the Bayesian optimization method is used to perform hyperparameter tuning. The root mean square error (RMSE) of the dataset is used as the optimization criterion to perform hyperparameter tuning for each independent XGBoost model, and the optimal parameter combination is used to train the model on the training set.

6. The intelligent generation method for wind tunnel control start-up parameters as described in claim 5, characterized in that, Step S3 also includes: evaluating the performance of each model using a test set, and the performance evaluation process uses the RMSE of the training set and the test set as the evaluation metric.

7. The intelligent generation method for wind tunnel control start-up parameters as described in claim 1, characterized in that, In step S4, the known quantity under the current test conditions is the total pressure. nominal Mach number Gas source pressure Atmospheric pressure Flow blockage test section ; The combination of control parameters to be intelligently generated is: main descent pressure. Assisted pressure Tonic displacement , stationary displacement Grid finger displacement , standing displacement Main row displacement .

8. The intelligent generation method for wind tunnel control start-up parameters as described in claim 7, characterized in that, Step S4 also includes parameter verification. Based on the absolute error Δ, the intelligently generated driving parameters, the manually given driving parameters, and the closed-loop steady-state values ​​in the time series data are compared and verified to verify the feasibility of the intelligent parameters of the model.

Images