Radar radiation source threat level evaluation method and system based on XGBoost

Abstract

The invention discloses a radar radiation source threat level evaluation method and system based on XGBoost, and the method comprises the steps: constructing a radar radiation source unified description word data set, carrying out the preprocessing of the data set, and dividing the data set into a training set and a test set; extracting radar radiation source features, and calculating each feature parameter value; constructing a radar radiation source threat level evaluation model based on XGBoost; training and optimizing an XGBoost model, and storing optimal parameters; and performing threat level evaluation on the radar radiation source in the test set by using the trained model. The system comprises a data set generation module, a preprocessing module, a feature extraction module, a model construction module, a model updating module and a threat level evaluation module. Compared with a traditional evaluation method, the radar radiation source hidden relation characteristics can be mined, the artificial subjective influence is reduced, and the reliability of a radar radiation source threat evaluation result is improved.

Description

Technical Field

[0001] The present invention relates to the field of electronic countermeasure technology, and in particular to an XGBoost-based radar radiation source threat level assessment method and system. Background Art

[0002] Situation assessment and situation prediction of the electromagnetic environment are extremely important tactical goals. As an important component of situation assessment and a direct reference for situation prediction, the accuracy and rationality of radar radiation source threat level assessment will directly affect the judgment of the battle situation and decision-making deployment.

[0003] The threat level assessment of radar emitters is mainly aimed at the multi-dimensional information obtained after signal recognition, including the characteristic parameters of each emitter after de-overlapping, the platform and the current working status, etc., and fully utilizes the limited resources of the party to implement soft and hard electromagnetic countermeasures against the emitters with the greatest threat. As the electromagnetic environment becomes increasingly complex, various intentional and unintentional interference pulses may be mixed into the actual signal to play a camouflage effect, which puts higher requirements on the threat level assessment algorithm. The accuracy and rationality of traditional threat assessment methods can no longer meet the needs of the increasingly complex electromagnetic environment.

[0004] Many scholars in China have conducted research on the threat level assessment algorithm for radiation sources. Reference 1 (Niu Hai, Jiang Ning. Threat level judgment method based on multi-attribute decision making [J]. Aerospace Electronic Countermeasures, 2001 (04): 35-38.) The decision matrix is ​​formed by coordinating the scheme set and the attribute set to obtain target-related information, and on this basis, the threat level assessment is realized through the membership function and weight value. Reference 2 (Tong Youtang, Wang Jianming. Air Target Fuzzy Pattern Recognition Threat-Judgment Model [J]. Systems Engineering and Electronic Technology (English Edition), 2003, 14 (1): 41-46.) A fuzzy recognition algorithm model under sea-to-air platform is proposed to realize the threat level assessment of air targets. Reference 3 (Chen Jiahui, Ke Hongfa. A new method for evaluating equipment combat effectiveness combining grey correlation and set pair analysis [J]. Ordnance Automation, 2022, 41 (06): 85-90.) A threat level assessment method based on fuzzy comprehensive evaluation and pairwise comparison method is proposed to further improve the assessment accuracy. However, these methods are all based on traditional rule evaluation and rely heavily on manual experience.

[0005] With the rapid development of machine learning and deep learning in recent years, there have been great breakthroughs in their practical applications. How to overcome the shortcomings of traditional radar emitter threat assessment methods through machine learning and deep learning is an urgent problem to be solved. The present invention uses machine learning methods to deeply mine the high-dimensional hidden features of radiation source information, and adaptively learns and adjusts, which reduces the influence of subjective factors during assessment and can effectively improve the accuracy of radiation source threat level assessment. Summary of the invention

[0006] The purpose of the present invention is to provide a radar emitter threat level assessment method and system based on XGBoost to improve the accuracy of radar emitter threat level assessment.

[0007] The technical solution to achieve the purpose of the present invention is: a radar emitter threat level assessment method based on XGBoost, the steps are as follows:

[0008] Step 1: obtain a unified description word data set of radar emitters after radar emitter sorting and identification, remove abnormal samples, and mark the threat level;

[0009] Step 2: Preprocess the data set and divide it into training set and test set according to the proportion;

[0010] Step 3: Extract radar radiation source features. The extracted features include radar position parameters, performance parameters and state parameters, and calculate the characteristic value of each parameter.

[0011] Step 4: Build a radar emitter threat level assessment model based on XGBoost, and obtain the loss function of the assessment model, as well as the first-order derivative and second-order derivative of the loss function;

[0012] Step 5: Using training set samples, using an iterative method to obtain the optimal leaf weight through the loss function, using a derivation method to obtain the objective function through the loss function, and substituting the optimal leaf weight into the objective function to determine the optimal objective function; updating the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function to obtain the latest objective function, thereby establishing a radar emitter threat level assessment model based on the latest objective function;

[0013] Step 6: Use the trained radar emitter threat level assessment model to assess the threat level of the radar emitters in the test set.

[0014] A radar emitter threat level assessment system based on XGBoost, the system is used to implement the radar emitter threat level assessment method based on XGBoost, the system includes a data set generation module, a preprocessing module, a feature extraction module, a model construction module, a model update module and a threat level assessment module, wherein:

[0015] The data set generation module obtains the unified description word data set of radar emitters after radar emitters are sorted and identified, removes abnormal samples, and marks the threat level;

[0016] The preprocessing module preprocesses the data set and divides it into training set and test set according to the proportion;

[0017] Feature extraction module: extracts radar radiation source features. The extracted features include radar position parameters, performance parameters and state parameters, and calculates the characteristic values ​​of each parameter.

[0018] The model building module builds a radar emitter threat level assessment model based on XGBoost, obtains the loss function of the assessment model, and the first-order derivative and second-order derivative of the loss function;

[0019] The model updating module uses the training set samples to obtain the optimal leaf weight through the loss function using an iterative method, obtains the objective function through the loss function using a derivation method, and substitutes the optimal leaf weight into the objective function to determine the optimal objective function; updates the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function to obtain the latest objective function, thereby establishing a radar emitter threat level assessment model based on the latest objective function;

[0020] The threat level assessment module uses the trained radar emitter threat level assessment model to assess the threat level of the radar emitters in the test set.

[0021] Compared with the prior art, the present invention has the following significant advantages:

[0022] (1) The present invention uses the machine learning model XGBoost to mine the hidden features of the radiation source, combines it with artificial experience features, and comprehensively evaluates the threat level of the radiation source. Compared with the traditional method, it reduces the influence of subjective factors during the evaluation and improves the accuracy of the threat level evaluation of the radar radiation source;

[0023] (2) The present invention uses the XGBoost model to form various complex decision surfaces in the pattern space, and adaptively learns and adjusts to mine the hidden relationship features of the radiation source, which can adapt to the complex and changeable electromagnetic environment and solve the problem of poor adaptability of traditional methods to complex electromagnetic environment scenes;

[0024] (3) The XGBoost model of the present invention can be applied to complex scenarios with multiple targets and multiple factors, which is consistent with the threat assessment of radiation sources in complex electromagnetic environments, and improves the reliability of the threat assessment results of radar radiation sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1It is a flow chart of the radar radiation source threat level assessment based on XGBoost of the present invention.

[0026] Figure 2 It is a curve chart showing the transformation of the loss function value of the XGBoost model with the number of training rounds in an embodiment of the present invention.

[0027] Figure 3 It is a curve chart showing how the accuracy of threat level assessment of the XBGoost model changes with the number of training rounds in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms in the specification and claims of this application and the above-mentioned figure descriptions: "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.

[0029] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0030] Combination Figure 1 The radar radiation source threat level assessment method based on XGBoost of the present invention comprises the following steps:

[0031] Step 1: obtain a unified description word data set of radar emitters after radar emitter sorting and identification, remove abnormal samples, and mark the threat level;

[0032] Step 2: Preprocess the data set and divide it into training set and test set according to the proportion;

[0033] Step 3: Extract radar radiation source features. The extracted features include radar position parameters, performance parameters and state parameters, and calculate the characteristic value of each parameter.

[0034] Step 4: Build a radar emitter threat level assessment model based on XGBoost, and obtain the loss function of the assessment model, as well as the first-order derivative and second-order derivative of the loss function;

[0035] Step 5: Using training set samples, using an iterative method to obtain the optimal leaf weight through the loss function, using a derivation method to obtain the objective function through the loss function, and substituting the optimal leaf weight into the objective function to determine the optimal objective function; updating the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function to obtain the latest objective function, thereby establishing a radar emitter threat level assessment model based on the latest objective function;

[0036] Step 6: Use the trained radar emitter threat level assessment model to assess the threat level of the radar emitters in the test set.

[0037] As a specific example, the step 1 is as follows:

[0038] Generate different types of radar signals, and obtain a unified description of radar radiation source information after sorting and identification;

[0039] The accuracy of the radar radiation source information description words after sorting and identification is confirmed by expert experience. The radar radiation source information description words include the radar radiation source's azimuth, altitude, carrier frequency, carrier frequency type pulse width, pulse width type, repetition rate, repetition rate type, data rate, scanning mode, scanning cycle, and distance information;

[0040] Abnormal samples are removed and the threat level is marked.

[0041] As a specific example, step 2 is as follows:

[0042] The data set obtained after step 1 is preprocessed to remove samples with missing values ​​and outliers, and then divided into a training set S and a test set T in a ratio of 7:3.

[0043] As a specific example, in step 3, the radar position parameters include distance, azimuth, and altitude; the performance parameters include carrier frequency, carrier frequency type, pulse width, pulse width type, repetition rate, and repetition rate type; and the status parameters include data rate, scanning mode, and platform type.

[0044] As a specific example, the calculation of the radar position parameter characteristic value is as follows:

[0045] Distance feature v d for:

[0046]

[0047] Orientation eigenvalue v doa for:

[0048]

[0049] Height eigenvalue v h :

[0050]

[0051] Where d is the distance from the target radiation source, l 1 , l 2 h are the distances corresponding to the maximum threat level and the minimum threat level of the radiation source respectively; 1 、h 2 are the minimum threat altitude and maximum threat altitude of the radiation source respectively, and h is the altitude of the target radiation source.

[0052] As a specific example, the calculation of the characteristic value of the performance parameter is as follows:

[0053] The carrier frequency, pulse width, and repetition rate are normalized using the sigmoid function to obtain the carrier frequency characteristic values ​​v RF , pulse width characteristic value v PW , repetition frequency eigenvalue v PRF ;

[0054] The carrier frequency type, pulse width type, and repetition rate type are all processed using one-hot coding to obtain the carrier frequency type characteristic value v RFT , pulse width type characteristic value v PWT , repetition type characteristic value v PRFT .

[0055] As a specific example, the calculation of the characteristic value of the working state parameter is as follows:

[0056] Scanning mode characteristic value

[0057] The data rate is normalized using the sigmoid function to obtain the data rate eigenvalue v rate ;

[0058] The platform type is processed using one-hot encoding to obtain the platform type feature value v type .

[0059] As a specific example, in step 4, a radar radiation source threat level assessment model based on XGBoost is constructed, and the model feature x i x is formed by concatenating the eigenvalues ​​of each parameter i =[v d ,v doa ,v h ,v RF ,v PW ,v PRF ,v RFT ,v PWT ,v PRFT ,v scan ,v rate ,v type ];

[0060] Based on the XGBoost method, an evaluation model is defined. The evaluation model is a strong evaluation model composed of K basic models. The threat assessment value of the i-th sample is

[0061]

[0062] in is the hypothesis space of the base model, f(·) is the regression tree, represents the feature of sample i, d represents the feature dimension; q represents the leaf node, q(x i ) means mapping sample i to a leaf node. is the weight value of the leaf node q, and T represents the number of leaf nodes in the decision tree;

[0063] The objective function of the evaluation model is:

[0064]

[0065] in, is a regular term used to limit the complexity of XGBoost. γ and λ are both hyperparameters. j represents the weight of a leaf node, K represents the number of trees, and f k represents the kth tree, N is the number of samples; represents the threat assessment value, y i is the tag value, is the model training loss function.

[0066] As a specific example, in step 5, the parameters related to the accumulated values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function are updated to obtain the latest objective function, which is as follows:

[0067] The second-order Tate expansion is used for the threat level assessment model loss function:

[0068]

[0069] Where L is the objective function, Represent the first and second order derivatives of the loss function, respectively. is the threat assessment value obtained by the t-1th round of iteration, because It is obtained after t-1 rounds of iterations. At the tth round of iteration, it is regarded as a constant value of the t-1th round of iteration;

[0070] Assume Q j ={j|q(x i )=j} is the set of all feature data in the jth leaf node, then the objective function after t rounds of iterations is Simplified to:

[0071]

[0072] Thus, the optimal leaf weight is calculated And the final objective function

[0073]

[0074] in, are the accumulated values ​​of the first-order derivative and the second-order derivative on the j-th leaf node, respectively, and λ and γ are constants.

[0075] The present invention also provides a radar emitter threat level assessment system based on XGBoost, which is used to implement the radar emitter threat level assessment method based on XGBoost. The system includes a data set generation module, a preprocessing module, a feature extraction module, a model construction module, a model update module and a threat level assessment module, wherein:

[0076] The data set generation module obtains the unified description word data set of radar emitters after radar emitters are sorted and identified, removes abnormal samples, and marks the threat level;

[0077] The preprocessing module preprocesses the data set and divides it into training set and test set according to the proportion;

[0078] Feature extraction module: extracts radar radiation source features. The extracted features include radar position parameters, performance parameters and state parameters, and calculates the characteristic values ​​of each parameter.

[0079] The model building module builds a radar emitter threat level assessment model based on XGBoost, obtains the loss function of the assessment model, and the first-order derivative and second-order derivative of the loss function;

[0080] The model updating module uses the training set samples to obtain the optimal leaf weight through the loss function using an iterative method, obtains the objective function through the loss function using a derivation method, and substitutes the optimal leaf weight into the objective function to determine the optimal objective function; updates the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function to obtain the latest objective function, thereby establishing a radar emitter threat level assessment model based on the latest objective function;

[0081] The threat level assessment module uses the trained radar emitter threat level assessment model to assess the threat level of the radar emitters in the test set.

[0082] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0083] Example

[0084] This embodiment provides a radar emitter threat level assessment method based on XGBoost, the steps are as follows:

[0085] 1. Build a sample dataset

[0086] Specifically, 20 different types of radar signals are generated, and after sorting and identification, a unified description word of radar radiation source information is obtained, with a total of 21,300 samples. Then, the accuracy of the radar radiation source information description word after sorting and identification is confirmed through expert experience, abnormal samples are removed, and the threat level is marked. The radar radiation source information description word mainly includes the radar radiation source's azimuth, altitude, carrier frequency, carrier frequency type pulse width, pulse width type, repetition rate, repetition rate type, data rate, scanning mode, scanning cycle, distance and other information.

[0087] 2. Preprocess the above data set, remove samples with missing values ​​and outliers, and divide them into training set S and test set T in a ratio of 7:3.

[0088] 3. Perform feature extraction on the above training set S to obtain the target features of the radar radiation source. The present invention is suitable for processing complex scenes with multiple targets and multiple factors, and is consistent with the threat assessment of radiation sources in complex electromagnetic environments. The radiation source itself has multiple types of parameters, each type corresponds to multiple specific parameters. Its threat assessment requires the joint assessment of multiple types of parameters. The assessment factors are numerous and complex, and there is a certain correlation between the assessment parameters. In this step, the feature extraction is performed by calculating and processing the relevant dimensional parameters in the radar radiation source information description word and unifying the dimensions.

[0089] 1) The present invention comprehensively considers the attributes of each radar parameter and the feasibility of obtaining parameters in actual scenarios, and selects radar position parameters, performance parameters and state parameters as XGBoost model feature inputs. Specifically, the position parameters of the radar radiation source are considered: azimuth, distance, height; performance parameters: carrier frequency, carrier frequency type, pulse width, pulse width type, repetition rate, repetition rate type; state parameters: data rate, scanning mode, platform type.

[0090] 2) Calculation of position parameter eigenvalues

[0091] ① Radiation source distance assessment weight eigenvalue

[0092] The distance of the radiation source can directly reflect the relative relationship between the two parties. The evaluation method in the present invention considers whether the distance of the radiation source terminal is within our combat radius and whether our various deployments are within the effective attack range of the other party. The distance of the radiation source is one of the most influential evaluation factors in threat assessment. The closer the distance, the higher the threat level corresponding to the radiation source. Let the distance of the radiation source be d, and assume that the distance corresponding to the maximum threat level and the distance corresponding to the minimum threat level are l respectively. 1 , l 2 , radiation source distance characteristic parameter evaluation value v d for:

[0093]

[0094] ② Calculation of characteristic values ​​for radiation source orientation assessment

[0095] There are three main types of radiation source azimuth conditions: slow change, instantaneous change, and constant. Slow change in azimuth indicates that the enemy target is conducting reconnaissance and search tasks, which are non-targeted range tasks and pose a relatively small threat; instantaneous change in azimuth usually corresponds to the enemy's targeted search and attack on us, with clear tactical goals and intentions, and the threat is relatively large compared to slow change; constant azimuth indicates that the radiation source is in a relatively fixed state, most of which are fixedly arranged reconnaissance radars, and the threat level is the lowest. According to the above method, the radiation source azimuth characteristic parameter evaluation value v doa for:

[0096]

[0097] ③ Calculation of characteristic values ​​for radiation source height assessment

[0098] The threat level of the radiation source height is similar to that of the radiation source distance. Its threat level is inversely proportional to the height. Suppose the minimum threat height and maximum threat height of the radiation source are h 1 and h 2 , the current radiation source height is h, and the radiation source height parameter characteristic evaluation value v is obtained h :

[0099] 2) Calculation of characteristic values ​​of performance parameters

[0100] In the present invention, carrier frequency, carrier frequency type, pulse width, pulse width type, repetition rate, and repetition rate type are selected as radiation source performance characteristics; the carrier frequency, pulse width, and repetition rate are normalized using a sigmoid function to obtain characteristic values ​​v RF , v PW 、v PRF , where the sigmoid function is: The carrier frequency type, pulse width type, and repetition rate type are all processed using one-hot coding to obtain the characteristic values ​​of v RFT , vPWT 、v PRFT .

[0101] 3) Calculation of radiation source state parameter characteristics

[0102] ① Calculation of characteristic values ​​of scanning mode parameters

[0103] The radar scanning mode of the evaluation method of the present invention includes periodic scanning, non-periodic scanning and long-term resident scanning. The non-periodic scanning mode indicates that the enemy unit has not locked or tracked our specific target, which belongs to the conventional reconnaissance or search mode, and the threat level is considered to be low at this time; the periodic scanning mode indicates that the enemy radar has found our specific target and is ready to start a specific action, and the threat level is higher than that of the non-periodic scanning; the long-term resident scanning indicates that the radar has no special action mode and the threat level is the lowest.

[0104] According to the above rules, calculate the scanning mode eigenvalue v scan

[0105]

[0106] ② Calculation of data rate and platform type characteristic values

[0107] The data rate is normalized using the sigmoid function to obtain the eigenvalue v rate ; The platform type is processed using one-hot encoding to obtain the characteristic value v type .

[0108] 4) Through the above steps, the feature extraction of the data set is completed, and the radar radiation source feature S of the training set is obtained. i ,y i )}, i=1,2,...,N, N is the number of samples, x i is the feature of sample i, which is composed of the above-mentioned feature parameter values, x i =[v d ,v doa ,v h ,v RF ,v PW ,v PRF ,v RFT ,v PWT ,v PRFT ,v scan ,v rate ,v type ],y i is the label value, that is, the radar radiation source threat level corresponding to the sample.

[0109] 4. In an example of this embodiment, an evaluation model is defined using the XGBoost method, which is a strong classifier model composed of K basic models. Assuming that K trees are trained, the final evaluation value for the i-th sample is:

[0110]

[0111] in is the hypothesis space of the basic model; q represents each

[0112] The construction of the basic model, represents the characteristics of sample i, q(x i ) means mapping sample i to a leaf node, T means the number of leaf nodes in the decision tree or regression tree; d means the feature dimension, is the weight value of the leaf node;

[0113] The loss function is:

[0114]

[0115] in, is a regular term to prevent overfitting and is used to limit the complexity of XGBoost. T is the number of leaf nodes. γ and λ are both hyperparameters. w j Indicates the weight of a leaf node. represents the threat assessment value, y i is the tag value, is a differentiable threat assessment loss function.

[0116] 5. Model training and optimization

[0117] Get the loss function of the XBGoost model, as well as the first-order and second-order derivatives of the loss function;

[0118] In one example of this embodiment, a second-order Tate expansion is used for the loss function:

[0119]

[0120] in, Represent the first and second order derivatives of the loss function respectively, since It is obtained after t-1 rounds of iteration. In the tth round of iteration, it can be regarded as a constant value of the t-1th round of iteration of the label data set.

[0121] Assume Q j ={j|q(x i )=j} is the set of all feature data in the jth leaf node, then:

[0122]

[0123] To obtain the optimal leaf weights Taking the derivative of the above equation and setting it equal to zero, we get:

[0124]

[0125] in, is the cumulative value of the first-order derivative and the second-order derivative on the jth leaf node, λ and γ are constants, Represents the optimal objective function at the tth iteration, and the optimal weight is substituted into the objective function to obtain it.

[0126] The optimal leaf weight is obtained through the loss function using the iterative method, the objective function is obtained through the loss function using the derivation method, and the optimal leaf weight is substituted into the objective function to determine the optimal objective function, thereby obtaining the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node; the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function are updated to obtain the latest objective function; and an XGBoost threat level assessment model based on the latest objective function is established.

[0127] In this example, the AUC curve is used to represent the model performance, where AUC is defined as the area under the ROC curve and the coordinate axis, and is a value between 0 and 1. The larger the value, the better the model evaluation effect. In this embodiment, the AUC value is obtained through each round of iteration to determine whether it has increased or decreased. When the AUC no longer changes (or decreases), the iterative calculation ends. Specifically, when the AUC no longer increases, it means that continued iterative training no longer contributes to the evaluation effect of the XGBoost model, and the model parameters are saved at this time. Figure 2 It is a curve chart showing the transformation of the loss function value of the XGBoost model with the number of training rounds in an embodiment of the present invention.

[0128] 6. Use the trained XGBoost model to assess the threat level of radar emitters in the test set.

[0129] Specifically, the test set extracts radar emitter features according to the feature extraction process in the above steps, inputs them into the trained XBGoost model, obtains the threat level of the radar emitter, and calculates the evaluation accuracy based on the label value. Figure 3 It is a curve showing how the accuracy of threat level assessment of the XBGoost model in an embodiment of the present invention changes with the number of training rounds.

[0130] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

[0131] It should be understood that in order to simplify the present invention and help those skilled in the art understand the various aspects of the present invention, in the above description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes described in a single embodiment or described with reference to a single figure. However, the present invention should not be interpreted as the features included in the exemplary embodiments are all necessary technical features of the patent claims.

Claims

1. A radar emitter threat level assessment method based on XGBoost, It is characterized in that Here are the steps: Step 1: obtain a unified description word data set of radar emitters after radar emitter sorting and identification, remove abnormal samples, and mark the threat level; Step 2: Preprocess the data set and divide it into training set and test set according to the proportion; Step 3: Extract radar radiation source features. The extracted features include radar position parameters, performance parameters and state parameters, and calculate the characteristic value of each parameter. Step 4: Build a radar emitter threat level assessment model based on XGBoost, and obtain the loss function of the assessment model, as well as the first-order derivative and second-order derivative of the loss function; Step 5: Using training set samples, using an iterative method to obtain the optimal leaf weight through the loss function, using a derivation method to obtain the objective function through the loss function, and substituting the optimal leaf weight into the objective function to determine the optimal objective function; updating the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function to obtain the latest objective function, thereby establishing a radar emitter threat level assessment model based on the latest objective function; Step 6: Use the trained radar emitter threat level assessment model to assess the threat level of the radar emitters in the test set.

2. According to the XGBoost-based radar emitter threat level assessment method of claim 1, It is characterized in that The step 1 is specifically as follows: Generate different types of radar signals, and obtain a unified description of radar radiation source information after sorting and identification; The accuracy of the radar radiation source information description words after sorting and identification is confirmed by expert experience. The radar radiation source information description words include the radar radiation source's azimuth, altitude, carrier frequency, carrier frequency type pulse width, pulse width type, repetition rate, repetition rate type, data rate, scanning mode, scanning cycle, and distance information; Abnormal samples are removed and the threat level is marked.

3. According to the XGBoost-based radar emitter threat level assessment method of claim 1, It is characterized in that The step 2 is specifically as follows: The data set obtained after step 1 is preprocessed to remove samples with missing values ​​and outliers, and then divided into a training set S and a test set T in a ratio of 7:

3.

4. According to the XGBoost-based radar emitter threat level assessment method of claim 1, It is characterized in that In step 3, the radar position parameters include distance, azimuth, and altitude; the performance parameters include carrier frequency, carrier frequency type, pulse width, pulse width type, repetition rate, and repetition rate type; and the status parameters include data rate, scanning mode, and platform type.

5. According to the XGBoost-based radar emitter threat level assessment method of claim 4, It is characterized in that The calculation of radar position parameter eigenvalues ​​is as follows: Distance feature v d for: Orientation eigenvalue v doa for: Height eigenvalue v h : Where d is the distance from the target radiation source, l 1 , l 2 h are the distances corresponding to the maximum threat level and the minimum threat level of the radiation source respectively; 1 、h 2 are the minimum threat altitude and maximum threat altitude of the radiation source respectively, and h is the altitude of the target radiation source.

6. According to claim 5, the radar emitter threat level assessment method based on XGBoost, It is characterized in that The calculation of the characteristic value of the performance parameter is as follows: The carrier frequency, pulse width, and repetition rate are normalized using the sigmoid function to obtain the carrier frequency characteristic values ​​v RF , pulse width characteristic value v PW , repetition frequency eigenvalue v PRF ; The carrier frequency type, pulse width type, and repetition rate type are all processed using one-hot coding to obtain the carrier frequency type characteristic value v RFT , pulse width type characteristic value v PWT , repetition type characteristic value v PRFT .

7. According to claim 6, the radar emitter threat level assessment method based on XGBoost, It is characterized in that The calculation of the characteristic value of the working state parameter is as follows: Scanning mode characteristic value The data rate is normalized using the sigmoid function to obtain the data rate eigenvalue v rate ; The platform type is processed using one-hot encoding to obtain the platform type feature value v type .

8. According to the XGBoost-based radar emitter threat level assessment method of claim 7, It is characterized in that In step 4, a radar radiation source threat level assessment model based on XGBoost is constructed, and the model feature x i x is formed by concatenating the eigenvalues ​​of each parameter i =[v d ,v doa ,v h ,v RF ,v PW ,v PRF ,v RFT ,v PWT ,v PRFT ,v scan ,v rate ,v type ]; Based on the XGBoost method, an evaluation model is defined. The evaluation model is a strong evaluation model composed of K basic models. The threat assessment value of the i-th sample is in is the hypothesis space of the base model, f(·) is the regression tree, represents the characteristics of sample i, and d represents the feature dimension; q represents a leaf node, q(x i ) means mapping sample i to a leaf node. is the weight value of the leaf node q, and T represents the number of leaf nodes in the decision tree; The objective function of the evaluation model is: in, is a regular term used to limit the complexity of XGBoost. γ and λ are both hyperparameters. j represents the weight of a leaf node, K represents the number of trees, and f k represents the kth tree, N is the number of samples; represents the threat assessment value, y i is the tag value, is the model training loss function.

9. According to claim 8, the radar emitter threat level assessment method based on XGBoost, It is characterized in that In step 5, the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function are updated to obtain the latest objective function, as follows: The second-order Tate expansion is used for the threat level assessment model loss function: Where L is the objective function, Represent the first and second order derivatives of the loss function, respectively. is the threat assessment value obtained by the t-1th round of iteration, because It is obtained after t-1 rounds of iterations. At the tth round of iteration, it is regarded as a constant value of the t-1th round of iteration; Assume Q j ={j|q(x i )=j} is the set of all feature data in the jth leaf node, then the objective function after t rounds of iterations is Simplified to: Thus, the optimal leaf weight is calculated And the final objective function obj (t)* : in, are the accumulated values ​​of the first-order derivative and the second-order derivative on the j-th leaf node, respectively, and λ and γ are constants.

10. A radar emitter threat level assessment system based on XGBoost, It is characterized in that The system is used to implement the radar radiation source threat level assessment method based on XGBoost according to any one of claims 1 to 9, and the system includes a data set generation module, a preprocessing module, a feature extraction module, a model construction module, a model update module and a threat level assessment module, wherein: The data set generation module obtains the unified description word data set of radar emitters after radar emitters are sorted and identified, removes abnormal samples, and marks the threat level; The preprocessing module preprocesses the data set and divides it into training set and test set according to the proportion; Feature extraction module: extracts radar radiation source features. The extracted features include radar position parameters, performance parameters and state parameters, and calculates the characteristic values ​​of each parameter. The model building module builds a radar emitter threat level assessment model based on XGBoost, obtains the loss function of the assessment model, and the first-order derivative and second-order derivative of the loss function; The model updating module uses the training set samples to obtain the optimal leaf weight through the loss function using an iterative method, obtains the objective function through the loss function using a derivation method, and substitutes the optimal leaf weight into the objective function to determine the optimal objective function; updates the parameters related to the cumulative values ​​of the first-order derivative and the second-order derivative at the leaf node in the optimal objective function to obtain the latest objective function, thereby establishing a radar emitter threat level assessment model based on the latest objective function; The threat level assessment module uses the trained radar emitter threat level assessment model to assess the threat level of the radar emitters in the test set.

Images