A method for evaluating the effectiveness of high-speed aircraft maneuvering countermeasures under small sample conditions

Through the framework of the high-speed aircraft maneuvering indicator system and the complex blocked neural network model, combined with integrated learning and interval monotonic knowledge, the problem of insufficient data in the maneuvering effectiveness evaluation of high-speed aircraft is solved, and the accuracy and robustness of the evaluation are improved. Achieve rapid and accurate performance evaluation.

CN118798693BActive Publication Date: 2024-11-15TIANMUSHAN LABORATORY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411288367.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-14
Publication Date
2024-11-15
Estimated Expiration
2044-09-14

AI Technical Summary

Technical Problem

The existing high-speed aircraft maneuverability effectiveness evaluation methods have low model fit, poor robustness and generalization capabilities, and strong subjectivity, making it difficult to give effective evaluation results quickly and accurately.

Method used

By establishing a framework for high-speed aircraft maneuverability index system, a complex blocked neural network model is built, and combining the integrated learning framework and expert knowledge of interval monotonicity, the Loss function of the model is optimized to improve the fit and generalization ability of the model.

Benefits of technology

In the case of small sample data, the fit and robustness of the high-speed aircraft maneuverability evaluation model is significantly improved, the accuracy and efficiency of the evaluation are improved, and the problems of low subjectivity and fit of the existing methods are overcome.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118798693B_ABST
    Figure CN118798693B_ABST
Patent Text Reader

Abstract

The invention relates to a method for evaluating the effectiveness of high-speed aircraft maneuver confrontation under a small sample condition, comprising the following steps: S1: analyzing a high-speed aircraft maneuver confrontation process and establishing a high-speed aircraft maneuver confrontation index system framework; S2: performing network structure topology according to the constructed index system framework, establishing a single-block neural network model, and combining to form a final complex block neural network model; S3: establishing a single-block neural network model framework with R MLP models as a base learner and one linear superposition learner as a meta learner; S4: mathematically abstracting "interval monotonicity" expert knowledge to form a regular term of a Loss function of a base learner; and S5: evaluating the effectiveness of high-speed aircraft maneuver confrontation based on the complex block neural network model obtained in S1-S4. The invention combines data and expert knowledge, reduces human subjectivity, and improves the accuracy and robustness of a high-speed aircraft maneuver confrontation effectiveness evaluation network model under a small sample condition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of high-speed aircraft, and in particular relates to a method for evaluating the maneuverability countermeasure effectiveness of a high-speed aircraft under a small sample condition. Background Art

[0002] The framework of the high-speed aircraft maneuverability countermeasure index system involves many factors affecting the maneuverability countermeasure effectiveness. A large amount of data is required to analyze the impact of these factors on the maneuverability countermeasure effectiveness. Due to the large scale of data requirements, the time cost of generating data through a simulation environment is high, which is not conducive to the iteration of analysis methods. In addition, due to economic reasons, there is less real battlefield test data.

[0003] Currently, the existing performance evaluation methods are highly subjective. The existing performance evaluation methods based on neural network models require a large amount of data for model training. When the amount of training data is small, the fit of the performance network model is low, and the robustness and generalization ability of the model are poor, which is prone to overfitting.

[0004] When the data scale is small, establishing a high-accuracy network model of high-speed aircraft maneuverability confrontation effectiveness can replace the simulation environment or the real battlefield environment, and give the results of high-speed aircraft maneuverability confrontation effectiveness evaluation more quickly and accurately, which is conducive to further analyzing the effectiveness of high-speed aircraft maneuverability confrontation strategies. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and to propose a method for evaluating the maneuverability of high-speed aircraft in a small sample situation to achieve fast, accurate and highly robust evaluation.

[0006] The technical solution of the present invention is as follows:

[0007] A method for evaluating the effectiveness of high-speed aircraft maneuvering countermeasures under a small sample condition comprises the following steps:

[0008] Step S1: Analyze the high-speed aircraft maneuvering confrontation process, establish a high-speed aircraft maneuvering confrontation index system framework based on the analysis results, and clarify the index elements of each layer in the high-speed aircraft maneuvering confrontation index system framework;

[0009] Step S2: Perform network structure topology according to the constructed high-speed aircraft maneuverability countermeasure system framework, and form a final complex block neural network model by establishing a combination of single-block neural network models, wherein the input of the complex block neural network model is the leaf node in the high-speed aircraft maneuverability countermeasure system framework, and the output of the complex block neural network model is the high-speed aircraft maneuverability countermeasure effectiveness;

[0010] Step S3: For the single-block neural network model, an integrated learning framework is established with R MLP models as base learners and 1 linear superposition learner as meta-learner. The input of the single-block neural network model is input into m MLP models, a total of m The output results are used as the input of the linear superposition learner, and the output of the single-block neural network model is used as its output for training;

[0011] Step S4: for the base learner in the ensemble learning framework, the “interval monotonicity” expert knowledge is mathematically abstracted to form the regularization term of the Loss function of the base learner;

[0012] Step S5: Evaluate the maneuverability countermeasure performance of high-speed aircraft based on the complex block neural network model obtained in S1-S4.

[0013] Furthermore, the specific process of establishing the high-speed aircraft maneuvering countermeasure system framework in step S1 is as follows:

[0014] S101. Analyze the high-speed aircraft maneuver confrontation process, determine that the value center of the high-speed aircraft maneuver confrontation is the tactical evasion capability, and determine that the indicator that characterizes the effectiveness of the high-speed aircraft maneuver confrontation is the minimum approach distance of the blue side's defense weapons;

[0015] S102, analyzing the high-speed aircraft maneuvering confrontation process, obtaining the indicators that affect the high-speed aircraft maneuvering confrontation effectiveness as maneuvering timing, maneuvering overload, maneuvering direction, and maneuvering duration, and using these four indicators as the second-level indicators of the high-speed aircraft maneuvering confrontation indicator system framework;

[0016] S103. Analyze the high-speed aircraft maneuvering confrontation process, confirm that the maneuvering overload is determined by the high-speed aircraft thrust, the high-speed aircraft dry weight, the high-speed aircraft fuel mass, and the high-speed aircraft fuel consumption rate, and the maneuvering duration is determined by the high-speed aircraft thrust, the high-speed aircraft specific impulse, the high-speed aircraft fuel mass, and the number of high-speed aircraft maneuvers, and use the above corresponding indicators as the third-level indicators of the high-speed aircraft maneuvering confrontation indicator system framework.

[0017] Furthermore, the specific steps of performing network structure topology through the constructed high-speed aircraft maneuverability countermeasure system framework in step S2 include the following contents:

[0018] S201. Establish a single-block neural network model between the third-layer index and the second-layer index in the framework of the high-speed aircraft maneuver confrontation index system, including a maneuver overload single-block neural network model with high-speed aircraft thrust, high-speed aircraft dry weight, high-speed aircraft fuel mass, and high-speed aircraft fuel consumption rate as input and maneuver overload as output, and a maneuver duration single-block neural network model with high-speed aircraft thrust, high-speed aircraft specific impulse, high-speed aircraft fuel mass, and high-speed aircraft maneuver times as input and maneuver duration as output;

[0019] S202, establishing a single-block neural network model between the second-layer index and the first-layer index in the high-speed aircraft maneuvering countermeasure index system framework, including a minimum approach distance single-block neural network model with maneuvering timing, maneuvering overload, maneuvering direction, and maneuvering duration as inputs and a minimum approach distance as output;

[0020] S203. A complex block neural network model of minimum approach distance is established by correspondingly combining the maneuvering overload single-block neural network model and the maneuvering duration single-block neural network model with the maneuvering overload and maneuvering duration inputs in the minimum approach distance single-block neural network model.

[0021] Furthermore, the specific steps of constructing the integrated learning framework in step S3 are:

[0022] S301. Set up training set , test set ,in and For sample and The characteristic vector of and is the corresponding label, and at the same time, the R base learners are determined to be MLP models;

[0023] S302, divide the data size of the training set and the test set based on the ratio of 8:2, and use the MLP-based learner to train the training set in turn, predict the training set, and obtain the predicted training set, and record the predicted training set prediction results of each MLP-based learner as , and the test set is predicted at the same time, and the prediction results are recorded as ;

[0024] S303, repeat the above steps until all MLP-based learners are traversed and the prediction results of all MLP-based learners are and the corresponding labels to form a new training sample set, denoted as D new , the prediction results T of all MLP base learners r Concatenate with the corresponding labels to form a new test set, denoted as T new ;

[0025] S304, D new As the training data of the second-layer meta-learner linear stacking learner, T new Applied to the test set for prediction and obtain the final prediction result.

[0026] Furthermore, in step S4, the specific method of introducing the “interval monotonicity” expert knowledge into the single-block neural network model training Loss function design is:

[0027] S401. According to the network topology result in step S2, the number of input layer nodes, the number of output layer nodes, and the number of hidden layer nodes of different MLP base learners are determined in accordance with the following principles: ,in l is the number of hidden layer nodes, p is the number of input layer nodes, q is the number of nodes in the output layer, a is a constant between 0 and 10;

[0028] S402, determine the "interval monotonicity" expert knowledge of high-speed aircraft maneuvering confrontation, and give interval monotonicity to confirm that the minimum approach distance is non-decreasing with the increase of maneuvering overload, and the minimum approach distance is non-decreasing with the increase of maneuvering time. In the interval, the increase of the minimum approach distance in the random moving direction is non-decreasing. Within the interval, the increase of the minimum approach distance with the random moving direction is non-increasing;

[0029] S403, for the Loss function of the MLP-based learner, a regularization term is added on the basis of the mean square error between the original label data and the fitting data, and a penalty is imposed on the opposite monotonicity in the interval, and the degree of penalty is controlled by changing the coefficient of the regularization term, and the neural network parameters are modified based on the back propagation of the regularization term, so that the MLP-based learner approaches the original monotonicity, wherein the label data is the output of the data set obtained in the simulation environment or the real environment, and the fitting data is the output obtained by fitting the performance network model;

[0030] S404: According to the training set and test set of step S302, by setting the learning rate, back-propagation of the Loss function is performed based on the Adam optimizer and the network parameters are optimized.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The present invention ensures the effectiveness and completeness of the index system framework through the high-speed aircraft maneuverability confrontation index system framework in step S1 and the index elements of each layer in the framework.

[0033] 2. The present invention establishes a block neural network model by performing an efficiency network topology based on the high-speed aircraft maneuverability confrontation index system framework in step S2. When the amount of data used for model training is small, compared with the existing efficiency network model, the framework greatly improves the model's fit.

[0034] 3. The present invention establishes an integrated learning framework in step S3 with R MLP models as base learners and 1 linear superposition learner as meta-learner. Different base learners are used to learn data knowledge from different angles. When the amount of model training data is small, the framework further greatly improves the model's fitting degree on the basis of the block neural network model in step S2. Compared with the existing effective network model, the framework also greatly improves the robustness of the model.

[0035] 4. The present invention introduces the "interval monotonicity" expert knowledge into the Loss function design in step S4, and further corrects and optimizes the model parameters by adding regularization terms. When the amount of model training data is small, the model fitting is further improved. Compared with the existing effective network model, this method also improves the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a framework diagram of the high-speed aircraft maneuvering countermeasure effectiveness evaluation method of the present invention;

[0037] Figure 2 It is a framework diagram of the high-speed aircraft maneuverability countermeasure index system of the present invention;

[0038] Figure 3 It is a network topology diagram of high-speed aircraft maneuverability countermeasure effectiveness of the present invention;

[0039] Figure 4 It is an integrated learning framework diagram for a single-block neural network model according to the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1-4As shown in the figure, a method for evaluating the effectiveness of high-speed aircraft maneuvering confrontation under small sample conditions is mainly aimed at the problem of evaluating the effectiveness of high-speed aircraft maneuvering confrontation. The goal of the method is to improve the problem of strong subjectivity in existing effectiveness evaluation methods. For the situation where the training data obtained during performance evaluation is small, it can improve the fitting of the effectiveness network model, and improve the training robustness and generalization ability of the effectiveness network model.

[0042] The specific steps include:

[0043] S1. Relying on the principle of constructing the index system framework, analyze the high-speed aircraft maneuvering confrontation process, and establish the high-speed aircraft maneuvering confrontation index system framework based on the analysis results, such as Figure 1 As shown in the figure, the indicator elements at each level in the indicator system framework are clarified.

[0044] S1, including the following steps:

[0045] S101. Through the value center theory, the process of high-speed aircraft maneuvering confrontation is analyzed, and the value center of high-speed aircraft maneuvering confrontation is determined to be the tactical evasion capability. The indicator that characterizes the effectiveness of high-speed aircraft maneuvering confrontation is determined to be the minimum approach distance of the blue side's defensive weapons.

[0046] S102. Through the game theory, the process of high-speed aircraft maneuvering confrontation is analyzed. The high-speed aircraft maneuvering confrontation is a confrontation from the perspective of dynamics. The high-speed aircraft avoids the interception of the blue side's defensive weapons through maneuvers. Then, the indicators that affect the effectiveness of high-speed aircraft maneuvering confrontation are obtained, such as maneuvering timing, maneuvering overload, maneuvering direction, and maneuvering duration. Therefore, these four indicators are used as the second-level indicators of the high-speed aircraft maneuvering confrontation indicator system framework.

[0047] S103. The maneuvering overload is determined by the high-speed aircraft thrust, the high-speed aircraft dry weight, the high-speed aircraft fuel mass, and the high-speed aircraft fuel consumption rate; the maneuvering duration is determined by the high-speed aircraft thrust, the high-speed aircraft specific impulse, the high-speed aircraft fuel mass, and the number of high-speed aircraft maneuvers. Therefore, the above corresponding indicators serve as the third-level indicators of the high-speed aircraft maneuvering countermeasure indicator system framework.

[0048] S2. According to the constructed high-speed aircraft maneuverability confrontation index system framework, the network structure topology is constructed and a block neural network model is established, such as Figure 3 As shown, by establishing a combination of single-block neural network models, the final complex block neural network model is formed, wherein the input of the complex block neural network model is the leaf node in the high-speed aircraft maneuverability index system framework, and the output of the complex block neural network model is the high-speed aircraft maneuverability confrontation effectiveness.

[0049] Step S2 has the following sub-steps:

[0050] S201. Establish a single-block neural network model between the third-layer indicators and the second-layer indicators in the framework of the high-speed aircraft maneuvering confrontation indicator system, including a maneuvering overload single-block neural network model with high-speed aircraft thrust, high-speed aircraft dry weight, high-speed aircraft fuel mass, and high-speed aircraft fuel consumption rate as input and maneuvering overload as output; and a maneuvering duration single-block neural network model with high-speed aircraft thrust, high-speed aircraft specific impulse, high-speed aircraft fuel mass, and high-speed aircraft maneuver times as input and maneuvering duration as output.

[0051] S202: Establish a single-block neural network model between the second-layer indicators and the first-layer indicators in the framework of the high-speed aircraft maneuvering countermeasures indicator system, including a minimum approach distance single-block neural network model with maneuvering timing, maneuvering overload, maneuvering direction, and maneuvering duration as inputs and a minimum approach distance as output.

[0052] S203: By correspondingly combining the maneuvering overload single-block neural network model and the maneuvering duration single-block neural network model in S201 with the maneuvering overload and maneuvering duration inputs in the minimum approach distance neural network in S202, a minimum approach distance block neural network framework is established.

[0053] S3. For the single-block neural network model, an integrated learning framework is established with 30 MLP models as base learners and 1 linear superposition learner as meta-learner. That is, the model input of the single-block neural network model is input into 30 MLP models respectively, and a total of 30 output results are obtained. The output results are used as the input of the linear superposition learner, and the output of the single-block neural network model is used as the output for training.

[0054] Step S3 has the following sub-steps:

[0055] S301. Set up training set , test set ,in and For sample and The characteristic vector of and is the corresponding label, m+n=600, that is, 600 data are sampled from the simulation environment, the feature vectors of the three single-block neural network models all have 4 eigenvalues, the label has a label value, and the 30 base learners are determined to be MLP models.

[0056] S302, divide the data size according to the ratio of 8:2 between the training set and the test set, use the MLP base learner to train the training set in turn, and predict the training set, and record the prediction results of the training set of each base learner as , and the test set is predicted at the same time, and the prediction results are recorded as .

[0057] S303, repeat the above steps until all MLP-based learners are traversed and the prediction results of all MLP-based learners are and the corresponding original label y n Splice to form a new training sample set, denoted as D new , the prediction results T of all base learners r and the corresponding original label y m Splice to form a new test set, denoted as T new .

[0058] S304, D new As the training data of the second-layer meta-learner linear stacking learner, T new Applied to the test set for prediction and obtain the final prediction result.

[0059] S4. For the base learner in the integrated learning framework, the expert knowledge of "interval monotonicity" is introduced into the design of the Loss function in the model training process. In ordinary MLP models, the mean square error between the network training output and the label is often used as the Loss function. The "interval monotonicity" is expressed mathematically and added to the Loss function as a regular term.

[0060] Step S4 has the following sub-steps:

[0061] S401. According to the network topology result in step S2, the number of input layer nodes, the number of output layer nodes, and the number of hidden layer nodes of different MLP base learners are determined in accordance with the following principles: ,in l is the number of hidden layer nodes, p is the number of input layer nodes, q is the number of nodes in the output layer, a is a constant between 0 and 10;

[0062] In this embodiment, the number of input nodes of the MLP base learners in the three single-block neural network models is determined to be 4, the number of output nodes is 1, and all have 2 hidden layers, the number of nodes in the first hidden layer is 16, and the number of nodes in the second hidden layer is 4.

[0063] S402, determine the interval monotonicity of the high-speed aircraft maneuvering confrontation, where the minimum approach distance is non-decreasing with the increase of the maneuvering overload, the minimum approach distance is non-decreasing with the increase of the maneuvering time, and when the maneuvering direction is In the interval, the increase of the minimum approach distance in the random moving direction is non-decreasing. Within the interval, the minimum approach distance is non-increasing as the random moving direction increases.

[0064] S403. For the Loss function in the MLP-based learner, a regularization term is added based on the mean square error between the original label data and the fitted data to penalize the opposite monotonicity in the interval. The neural network parameters are modified through back propagation of the regularization term to improve the model accuracy. The specific function is expressed as (1): ,

[0065] Where i: the i-th data, a total of n data, n =600, is the sample label, is the sample fitting value, is a constant coefficient, q=1, m is the number of network inputs with known interval monotonicity, j is the network input dimension, k is the number of non-increasing or non-decreasing intervals, Representation sample No. The input value of the dimension, Indicates The minimum value of the interval, Indicates The maximum value of an interval, if the function is in The intervals are non-increasing, then Takes 1 if the function is in The intervals are non-decreasing, then Take 0, Indicates the network output for dimensional network input in the sample The first term of the Loss function represents the mean square error between the label data and the fitting data. The back propagation of this error will change the network weights, thereby making the network fitting data close to the label data. The second term of the Loss function represents the penalty for violating the monotonicity law within the interval. Represents the degree of penalty. By back-propagating the regularization term to change the network weights, the MLP-based learner approaches the original monotonicity.

[0066] S404, dividing the training set and the test set according to the data set division method in step S3, setting the learning rate, performing back propagation of the Loss function based on the Adam optimizer and optimizing the network parameters.

[0067] Step 5) A complete complex block neural network model for high-speed aircraft maneuvering confrontation is established through steps S1-S4. The complex block neural network model is a high-speed aircraft maneuvering confrontation effectiveness evaluation model. The accuracy of the high-speed aircraft maneuvering confrontation effectiveness evaluation model is determined by its fit on the test set. The specific fit is expressed as (2):

[0068] ,

[0069] in, represents the degree of fit, Represents the mean of the test set error after multiple simulation experiments, represents the mean of the minimum proximity distance in the test set, where the improvement of the complex block neural network model established according to step S2 compared with the ordinary neural network is shown in Table 1;

[0070] According to the integrated learning framework established in step 3), taking the minimum proximity distance single block neural network model as an example, the improvement in fit and robustness compared with the ordinary neural network is shown in Table 2;

[0071] According to step S4, the “interval monotonicity” expert knowledge is introduced into the network model training. Taking the minimum proximity distance neural network as an example, the improvement compared with the ordinary neural network is shown in Table 3.

[0072] Table 1

[0073] Common neural network model Complex Block Neural Network Model The first loss 1.02 0.64 Second loss 0.97 0.66 The third loss 1.18 0.63 The 4th loss 0.89 0.72 The 5th loss 1.08 0.63 Mean loss 1.028 0.656 Variance loss 0.01 0.001 Goodness of fit 86.4% 91.3%

[0074] Table 2

[0075] Common neural network model Ensemble learning framework The first loss 0.9 0.52 Second loss 0.68 0.54 The third loss 0.73 0.51 The 4th loss 0.83 0.53 The 5th loss 0.72 0.5 Mean loss 0.772 0.52 Variance loss 0.00817 0.00025 Goodness of fit 93.71% 95.76%

[0076] Table 3

[0077] Common neural network model Introducing expert knowledge "interval monotonicity" The first loss 0.857 0.684 Second loss 0.846 0.786 The third loss 0.799 0.666 The 4th loss 0.859 0.613 The 5th loss 0.791 0.621 Mean loss 0.8304 0.674 Variance loss 0.0010768 0.0048095 Goodness of fit 93.24% 94.51%

[0078] In summary, the high-speed aircraft confrontation effectiveness evaluation method based on complex block neural network and ensemble learning described in the present invention solves the problem of strong subjectivity of existing evaluation methods by means of data evaluation; constructs the effectiveness network model topology based on the high-speed aircraft maneuver confrontation index system framework, and solves the problem of low model fitting degree when the amount of training data of the effectiveness network model is small; further improves the model fitting degree by establishing an ensemble learning framework, and greatly improves the robustness of the model; and improves the model fitting degree by introducing the "interval monotonicity" expert knowledge into the Loss function design of the network model, and improves the generalization ability of the model, thereby improving the accuracy of the high-speed aircraft maneuver confrontation effectiveness evaluation method.

[0079] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for evaluating the effectiveness of high-speed aircraft maneuvering countermeasures under small sample conditions, characterized in that: The steps include: Step S1: Analyze the high-speed aircraft maneuvering confrontation process, establish a high-speed aircraft maneuvering confrontation index system framework based on the analysis results, and clarify the index elements of each layer in the high-speed aircraft maneuvering confrontation index system framework; Step S2: Perform network structure topology according to the constructed high-speed aircraft maneuverability countermeasure system framework, and form a final complex block neural network model by establishing a combination of single-block neural network models, wherein the input of the complex block neural network model is the leaf node in the high-speed aircraft maneuverability countermeasure system framework, and the output of the complex block neural network model is the high-speed aircraft maneuverability countermeasure effectiveness; Step S3: For the single-block neural network model, an integrated learning framework is established with R MLP models as base learners and 1 linear superposition learner as meta-learner. The input of the single-block neural network model is input into m MLP models, a total of m The output results are used as the input of the linear superposition learner, and the output of the single-block neural network model is used as the output for training; Step S4: for the base learner in the ensemble learning framework, the expert knowledge of "interval monotonicity" is mathematically abstracted to form the regularization term of the Loss function of the base learner; Step S5: Evaluate the maneuverability countermeasure performance of high-speed aircraft based on the complex block neural network model obtained in S1-S4.

2. The method for evaluating the effectiveness of high-speed aircraft maneuvering countermeasures under small sample conditions according to claim 1 is characterized in that: The specific process of establishing the high-speed aircraft maneuvering countermeasure system framework in step S1 is as follows: S101. Analyze the high-speed aircraft maneuver confrontation process, determine that the value center of the high-speed aircraft maneuver confrontation is the tactical evasion capability, and determine that the indicator that characterizes the effectiveness of the high-speed aircraft maneuver confrontation is the minimum approach distance of the blue side's defense weapons; S102, analyzing the high-speed aircraft maneuvering confrontation process, obtaining the indicators that affect the high-speed aircraft maneuvering confrontation effectiveness as maneuvering timing, maneuvering overload, maneuvering direction, and maneuvering duration, and using these four indicators as the second-level indicators of the high-speed aircraft maneuvering confrontation indicator system framework; S103. Analyze the high-speed aircraft maneuvering confrontation process, confirm that the maneuvering overload is determined by the high-speed aircraft thrust, the high-speed aircraft dry weight, the high-speed aircraft fuel mass, and the high-speed aircraft fuel consumption rate, and the maneuvering duration is determined by the high-speed aircraft thrust, the high-speed aircraft specific impulse, the high-speed aircraft fuel mass, and the number of high-speed aircraft maneuvers, and use the above corresponding indicators as the third-level indicators of the high-speed aircraft maneuvering confrontation indicator system framework.

3. The method for evaluating the effectiveness of high-speed aircraft maneuvering countermeasures under small sample conditions according to claim 2 is characterized in that: The specific steps of performing network structure topology by constructing the high-speed aircraft maneuverability countermeasure system framework in step S2 include the following contents: S201. Establish a single-block neural network model between the third-layer index and the second-layer index in the framework of the high-speed aircraft maneuver confrontation index system, including a maneuver overload single-block neural network model with high-speed aircraft thrust, high-speed aircraft dry weight, high-speed aircraft fuel mass, and high-speed aircraft fuel consumption rate as input and maneuver overload as output, and a maneuver duration single-block neural network model with high-speed aircraft thrust, high-speed aircraft specific impulse, high-speed aircraft fuel mass, and high-speed aircraft maneuver times as input and maneuver duration as output; S202, establishing a single-block neural network model between the second-layer index and the first-layer index in the high-speed aircraft maneuvering countermeasure index system framework, including a minimum approach distance single-block neural network model with maneuvering timing, maneuvering overload, maneuvering direction, and maneuvering duration as inputs and a minimum approach distance as output; S203. A complex block neural network model of minimum approach distance is established by correspondingly combining the maneuvering overload single-block neural network model and the maneuvering duration single-block neural network model with the maneuvering overload and maneuvering duration inputs in the minimum approach distance single-block neural network model.

4. The method for evaluating the effectiveness of high-speed aircraft maneuvering countermeasures under small sample conditions according to claim 1 is characterized in that: The specific steps of constructing the integrated learning framework in step S3 are: S301. Set up training set , test set ,in and For sample and The characteristic vector of and is the corresponding label, and at the same time, the R base learners are determined to be MLP models; S302, divide the data size of the training set and the test set based on the ratio of 8:2, and use the MLP-based learner to train the training set in turn, predict the training set, and obtain the predicted training set, and record the predicted training set prediction results of each MLP-based learner as , and the test set is predicted at the same time, and the prediction results are recorded as ; S303, repeat the above steps until all MLP-based learners are traversed and the prediction results of all MLP-based learners are and the corresponding labels to form a new training sample set, denoted as D new , the prediction results T of all MLP base learners r Concatenate with the corresponding labels to form a new test set, denoted as T new ; S304, D new As the training data of the second-layer meta-learner linear stacking learner, T new Applied to the test set for prediction and obtain the final prediction result.

5. The method for evaluating the effectiveness of high-speed aircraft maneuvering countermeasures under small sample conditions according to claim 4 is characterized in that: The specific method of introducing the expert knowledge of "interval monotonicity" into the design of the Loss function for training the single-block neural network model in step S4 is: S401. According to the network topology result in step S2, the number of input layer nodes, the number of output layer nodes, and the number of hidden layer nodes of different MLP base learners are determined in accordance with the following principles: ,in l is the number of hidden layer nodes, p is the number of input layer nodes, q is the number of nodes in the output layer, a is a constant between 0 and 10; S402, determine the "interval monotonicity" expert knowledge of high-speed aircraft maneuvering confrontation, and give interval monotonicity to confirm that the minimum approach distance is non-decreasing with the increase of maneuvering overload, and the minimum approach distance is non-decreasing with the increase of maneuvering time. In the interval, the increase of the minimum approach distance in the random moving direction is non-decreasing. Within the interval, the increase of the minimum approach distance with the random moving direction is non-increasing; S403, for the Loss function of the MLP-based learner, a regularization term is added on the basis of the mean square error between the original label data and the fitting data, and a penalty is imposed on the opposite monotonicity in the interval, and the degree of penalty is controlled by changing the coefficient of the regularization term, and the neural network parameters are modified based on the back propagation of the regularization term, so that the MLP-based learner approaches the original monotonicity, wherein the label data is the output of the data set obtained in the simulation environment or the real environment, and the fitting data is the output obtained by fitting the performance network model; S404: According to the training set and test set of step S302, by setting the learning rate, back-propagation of the Loss function is performed based on the Adam optimizer and the network parameters are optimized.