Radiation source working mode identification method for incomplete information

Abstract

The invention relates to a radiation source working mode identification method for incomplete information, and belongs to the technical field of radiation source working mode identification. Comprising the following steps: determining a rule basic style, selecting characteristic parameters to describe a radiation source working mode, and constructing a radiation source working mode identification expression form of disjunction confidence rules; modeling a confidence rule base, and determining premise feature attribute parameters and identification result confidence; identifying the working mode of the radiation source, determining the individual matching degree of the feature attribute input relative to the feature attribute reference value, determining the activation weight of the input information to the confidence rule, and integrating the activation rule to obtain the final output of the confidence rule base. The method has a very good effect on a radiation source working mode presenting incomplete information, and particularly has more advantages under a small sample condition.

Description

Technical Field

[0001] The invention relates to a radiation source working mode recognition method facing incomplete information, and belongs to the technical field of radiation source working mode recognition. Background Art

[0002] By acquiring electromagnetic signals, we can analyze who is using the spectrum, when the spectrum is used, and whether there are identifiable features or working modes. This is the premise for situational awareness, task management, and interference resource management. By intercepting and identifying signals, mining signal rules based on their characteristics and determining the working status of radiation sources, it is the basis for threat estimation, countermeasure selection, and other links. Its level directly determines the effectiveness of the confrontation.

[0003] With the widespread application of technologies such as phased arrays and digital arrays, radiation sources can flexibly realize the control capabilities of beam pointing, waveform agility, and random modulation, and at the same time have multiple functions such as performing search, tracking, and guidance. In order to achieve different detection capabilities, radiation sources generally adopt multiple working modes, such as various multi-function radars, including Tracking-While-Scan (TWS), Track and Search (TAS), Single Target Track, and other working modes. Due to the diversification of radiation source working modes, the functional boundaries are blurred and the signal styles overlap during the use of radiation sources, making it more difficult to identify the radiation source working mode. The radiation source identification process depends on the correct measurement of the radiation source parameters. Although the frequency measurement accuracy is usually high enough, due to various reasons, errors will still occur in the measurement or calculation of parameters such as pulse width and pulse repetition interval. In addition, due to the influence of factors such as the environment and sensors, some parameters of the radiation source are missing, resulting in incomplete information on the radiation source. For example, since the measuring device cannot always keep a wide open range for all frequencies within its operating frequency band, some parameters of the radiation source may be lost, making the characteristic parameters of the radiation source incomplete. Existing methods are still insufficient in this regard, and it is necessary to study new methods to identify the working mode of the radiation source under incomplete conditions. Summary of the invention

[0004] In view of the shortcomings of the above-mentioned prior art, the present invention proposes a method for identifying the working mode of a radiation source with incomplete information. The method proposed in the present invention has a good effect on the working mode of a radiation source presenting incomplete information, especially under small sample conditions.

[0005] The invention provides a radiation source working mode recognition method for incomplete information, which is characterized in that it comprises the following steps:

[0006] Step 1) determine the basic style of the rule, select characteristic parameters to describe the working mode of the radiation source, and construct the radiation source working mode recognition expression form of the disjunctive confidence rule;

[0007] Step 2) Build a confidence rule base to determine the premise feature attribute parameters and the confidence of the recognition result;

[0008] Step 3) Identify the working mode of the radiation source, determine the individual matching degree of the feature attribute input relative to the feature attribute reference value, determine the activation weight of the input information to the confidence rule, integrate the activation rules, and obtain the final output of the confidence rule base.

[0009] Preferably, the specific steps of step 1) are:

[0010] (1) The pulse sequence of a radiation source in a certain working mode is composed of multiple coherent processing intervals (CPI). In the same CPI cycle, its related characteristic parameters remain stable. Four parameters, PW, PRI, PGI and PN, are selected to describe the working mode. The radiation source characteristics are formally described by PGW = {PW, PRI, PGI, PN}, where PW represents the pulse width, PRI represents the pulse repetition interval, PGI represents the pulse group interval, and PN represents the number of pulses in the pulse group. The radiation source working mode is modeled by PGW to reflect the joint change law of the radiation source working mode parameters;

[0011] The pulse groups are divided according to the characteristic parameters of the working mode of the radiation source. In a pulse group, the characteristic parameters of the working mode of the radiation source remain stable. That is, when the radiation source is in a certain working mode, it corresponds to a set of data sets composed of PW, PRI, PGI, and PN. The input training data set can be expressed by the matrix X The formal description is:

[0012]

[0013] Among them, L represents the total number of input training data sets, and M represents the number of attributes of each training data. Here, the attributes of each training data are mainly composed of PW, PRI, PGI, and PN.

[0014] (2) Construct the radiation source working mode recognition expression form of the disjunctive confidence rule. On the basis of the IF-THEN rule, the premise feature attribute weight, rule weight and the confidence of the conclusion part are integrated to establish the disjunctive paradigm confidence rule R k ,Right now

[0015]

[0016] Among them, R k represents the kth rule, represents the mth feature reference value in the kth rule, D j (j=1,2,L,N) represents the jth recognition result, Indicates that the input in the kth rule is relative to D j The confidence level, θ k represents the weight of the kth rule, represents the weight of the i-th attribute in the k-th rule; the above IF-THEN rules are represented as disjunctive confidence rules, and multiple disjunctive confidence rules constitute a disjunctive confidence rule base, denoted as DBRB. DBRB processes the input information, synthesizes the rules activated by the input information, and then obtains the recognition result of the radiation source working mode;

[0017] Preferably, the specific steps of step 2) are:

[0018] Building DBRB requires the following steps:

[0019] (1) Determine the DBRB premise feature attribute A based on the actual work scenario requirements m ;

[0020] (2)x m (m=1,2,…,T k ) represents the mth premise attribute, T k Indicates the number of feature attributes, according to the input information x m Determine the number of prerequisite feature attribute reference values ​​and prerequisite attribute reference values Any confidence rule needs to contain at least one premise feature attribute;

[0021] (3) For the recognition result D j Construct the corresponding confidence pass It reflects whether the k-th confidence rule is complete. If it meets the following requirements, it means that the k-th confidence rule is complete. Otherwise, it means that the k-th rule is incomplete.

[0022]

[0023] (4) Check whether all the reference values ​​of the premise characteristic attributes have been included in the DBRB.

[0024] Preferably, the specific steps of step 3) are:

[0025] 1) Determine whether the input information is complete. If the information is complete, directly determine the feature attribute input x m (m=1,2,…,T k ) relative to the characteristic attribute reference value A m,j The individual matching degree α m,j(m=1,2,…,T k ,j=1,2,…,J m ), J m Indicates the number of reference values ​​of the mth attribute. Assuming that the working mode of the radiation source is identified, the characteristic attribute input x m Reference value A for feature attribute m,j The individual matching degree α m,j Calculated by the following formula:

[0026]

[0027] m represents the feature attribute number, j represents the reference value number, α m,j (m=1,2,…,T k ,j=1,2,…,J m ) represents the input component x m (m=1,2,…,T k ) for the characteristic attribute reference value A m,j The individual matching degree of J m Indicates the number of reference values ​​of the mth feature attribute.

[0028] 2) If the attribute value is incomplete in the input information, the confidence distribution of the attribute is completely unknown. At this time, the individual matching degree of the attribute is 0, indicating that the incomplete attribute value cannot be matched with any attribute reference value. Determine the weight θ of each confidence rule k For information with incomplete attribute values, it is necessary to further calculate the rule attenuation factor and adjust the rule weight according to the incomplete attribute weight. Suppose the input data x i Some attribute values ​​are missing, that is, the radiation source data obtained is incomplete. By defining the rule attenuation factor δ m To quantitatively describe the impact of incomplete information on working mode recognition;

[0029]

[0030] in, is the attribute weight of missing data;

[0031] 3) Determine the activation weight ω of the input information x for the kth confidence rule k , reflects the degree to which the kth rule is activated. Since DBRB takes “∨” as the combination relationship between the premise feature attribute reference values, the individual matching degree Take the cumulative calculation method and then perform normalization:

[0032]

[0033] 4) Integrate the activation rules. kWhen ≠0, it means that the kth rule is activated. The rules in DBRB can be integrated using the evidence reasoning algorithm. The steps are as follows:

[0034] (1) The confidence level of the conclusion part Perform the basic probability mass (BPA) conversion, that is, activate the weight ω k Confidence in the conclusion Perform fusion processing:

[0035]

[0036] in, Indicates relative to the evaluation result D j BPA, represents the BPA relative to the set D, that is, corresponding to the evaluation result D not set to any j BPA, is associated with the activation weight of the kth rule, It is related to the incompleteness of the evaluation result of the kth rule.

[0037] (2) Combine K rules to obtain the recognition result D j The confidence level β j .

[0038] Through the evidence reasoning parsing algorithm, all the rules in the confidence rule base can be combined, and the final output S(x) of the confidence rule base is:

[0039]

[0040] in, Indicates that the result obtained after rule combination is relative to the evaluation result D j confidence level.

[0041] The present invention provides a radiation source working mode recognition method for incomplete information, and the beneficial effects are as follows:

[0042] (1) Reduce the size of the confidence rule base for identifying the working mode of the radiation source to avoid the combinatorial explosion problem. When the radiation source has M characteristic attributes, the number of reference values ​​of the mth characteristic of the radiation source is l m , using the Conjunctive Belief Rule-Base (CBRB) based on the conjunctive normal form, the size of the recognition database is Since the size of the recognition database will grow exponentially with the number of features, the size of the recognition database can be reduced by changing the connection relationship between the preconditions and using the disjunctive belief rule-base (DBRB) based on the disjunctive paradigm. If there are 8 feature attributes and each feature attribute has 5 reference values, the number of rules of CBRB will reach 390,625, while the number of rules of DBRB is only 5 to complete the construction of the rule base. Under the condition of having the same reference value, using DBRB to build the rule base can effectively avoid the combinatorial explosion problem.

[0043] (2) By adopting the disjunctive connection relationship, it is possible to effectively process incomplete radiation source information, so that the system can output the radiation source working mode recognition results.

[0044] (3) In the form of a confidence rule base, a nonlinear model can be established between the input radiation source information and the output working mode information.

[0045] In summary, the method proposed in the present invention has a good effect on the working mode of the radiation source presenting incomplete information, especially under the condition of small sample. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is the linear combination diagram of DBRB prerequisites;

[0048] Figure 2 The modeling process diagram for the confidence rule base;

[0049] Figure 3 This is the working mode diagram of the radiation source;

[0050] Figure 4 This is the evolution curve of the optimization algorithm. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] A radiation source working mode recognition method for incomplete information in this embodiment includes the following steps:

[0053] 1) Determine the basic style of the rule

[0054] In the identification of radiation source working mode, a certain radiation source working mode can be identified based on the data set composed of pulse width, pulse repetition interval, pulse group interval, and the number of pulses in the pulse group. Assuming that a piece of data in the data set is {PW=10us,PRI=57us,PGI=52us,PN=60,MODE}, the premise part of the confidence rule base can be described as

[0055] X k :X1 is PW=10, X2 is PRI=57, X3 is PGI=52, X4 is PN=60,

[0056] Assuming that the working mode of the radiation source includes four results, namely MODE = {RWS, VS, TWS, STT}, the corresponding evaluation result can be expressed as D = {RWS (D1), VS (D2), TWS (D3), STT (D4)}.

[0057] Assume X k Determining that it belongs to the working mode RWS means that the confidence of the mode RWS is 1, and the confidence of other modes VS / TWS / STT is 0, which can be expressed as β = {1, 0, 0, 0}. The recognition result and the recognition confidence can be combined to express, that is, {(D, β)} = {(D1, 1), (D2, 0), (D3, 0), (D4, 0)}. A complete confidence rule using the CBRB conjunction normal form can be expressed as:

[0058] R k :if(X1 is 10)∧(X2 is 57)∧(X3 is 52)∧(X4 is 60),then{(D1,1),(D2,0),(D3,0),(D4,0)};

[0059] In order to effectively reduce the number of rules in the database, the DBRB disjunctive normal form can be adopted to avoid the "combinatorial explosion" problem. Its confidence rule structure can be expressed as:

[0060] R k :if(X1 is 10)∨(X2 is 57)∨(X3 is 52)∨(X4 is 60),then{(D1,1),(D2,0),(D3,0),(D4,0)};

[0061] 2) Size of disjunctive confidence rules

[0062] Once the number of premise attributes and their reference values ​​is determined, the size of the DBRB-based confidence rule base can be determined. m | represents the number of reference values ​​of the mth feature, then the minimum size of the disjunctive confidence rule base DBRB can be expressed as

[0063]

[0064] That is, the minimum size of the disjunctive confidence rule base is the maximum number of reference values ​​in all features. M |=P,

[0065] min(size DBRB )=P

[0066] For the disjunctive confidence rule base, the number of rules is not fixed, but has a certain range. The number of rules can be larger than the minimum scale, and new rules can be added until the reference values ​​of all features are arranged and combined to obtain a new rule base. At this time, the rule base has the maximum scale, which is expressed as

[0067]

[0068] At this point, the number of disjunctive confidence rules will be equal to the number of conjunctive confidence rules. When the radiation source has M features, each feature has |A m | reference value, the size K of the disjunctive confidence rule base is and between.

[0069] The minimum size of the disjunctive confidence rule base is only equal to the number of feature reference values ​​|A m | is related to the number of features M, so the disjunctive rule can effectively reduce the size of the rule base and is more suitable for situations with multiple features.

[0070] Assume that the radiation source feature is composed of four premise attribute features, represented as X1~X4 respectively. If the reference value A1~A3 of each radiation source feature is set to 3, then the number of CBRB confidence rule bases is 3. 4 =81, while if it is a DBRB confidence rule base, its scale is between the minimum scale of 3 and the maximum scale of 81. Therefore, by constructing confidence rules through DBRB, the number of rule bases has a greater advantage, especially for situations with a large number of attribute features and a large number of feature reference values, which can more effectively reduce the rule storage space.

[0071] 3) Modeling process of the confidence rule base

[0072] The method for determining the working mode of the radiation source requires collecting the waveform parameters of the radiation during the working process of the radiation source, performing corresponding pulse sorting processing, and then using the sorted pulse data for long-term correlation to further extract the radiation source signal parameter information. Then, the working mode of the radiation source is determined by the confidence rule reasoning method. Due to the influence of external factors or information processing methods, the extracted radiation source signal parameters may be incomplete or even missing. In this case, the DBRB reasoning method can still give the radiation source working mode recognition result.

[0073] Step 1: Determine the radiation source characteristics A m (m=1,2,…,M), corresponding to PW, PRI, PGI, and PN characteristic information respectively.

[0074] Step 2: Determine the characteristics x of each radiation source m Set the corresponding reference value

[0075] Step 3: Using the radiation source characteristic reference value Construct confidence rules and set the rule weight θ of the kth confidence rule k , for the recognition result D j (j=1,2,L,N) Set the recognition confidence Repeat this step until all parameters are set for K rules.

[0076] Step 4: Verify the confidence rules to check whether all reference values ​​of all radiation source characteristics are covered If the rules cover all reference values Then go to step 5, otherwise determine the missing reference value and go to step 3.

[0077] Step 5: Complete the construction of the radiation source working mode confidence rule base.

[0078] The confidence rule base can effectively utilize various types of information and establish a nonlinear model between input feature parameters and output working mode recognition results in the recognition of radiation source working modes. Compared with the traditional IF-THEN rules, the DBRB confidence rule base provides a knowledge expression and pattern recognition method that is closer to reality.

[0079] 4) Optimization of confidence rule base model parameters

[0080] When the reference value of the premise attribute of the radiation source is unknown, it is necessary to use the premise attribute reference value as the estimated parameter of DBRB for learning and training, so as to obtain better recognition results. By establishing a premise attribute reference value optimization model, the premise attribute reference value is used as the parameter to be estimated, including the initial weight of the rule in DBRB and the confidence of each evaluation level in the conclusion part, and using indicators such as mean square error as the objective function, the differential evolution algorithm is used as the search engine to solve the optimization algorithm.

[0081] 5) Cross-validation

[0082] The proposed model is evaluated using the cross-validation method. The radiation source data set is divided into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the model prediction effect. There are many ways to divide the training set and the test set. If the data used in a training set is also used in the test set, the so-called "cross" is formed. Cross-validation is suitable for training and optimizing models under small sample conditions. K-fold cross-validation is a commonly used method. First, the data set is divided into k subsets of equal size and mutually exclusive, that is, k-1 parts are randomly selected as training sets each time, and the remaining subset is used as the test set. In this way, k rounds of validation can be performed, k models can be obtained, and the corresponding errors can be obtained each time. The k errors are averaged to obtain the cross-validation error, and the performance of the model is evaluated by the error. Assuming that a model has multiple parameters, different parameters will correspond to different cross-validation errors. The parameters with the smallest error can be selected as the parameters of the model. The advantage of k-fold cross-validation is that all data can be used for training and testing, effectively avoiding the occurrence of overfitting problems.

[0083] 6) Experimental results and analysis under incomplete conditions

[0084] Using small sample incomplete radiation source feature data as the test data set, the corresponding recognition accuracy is tested and compared with the corresponding algorithm. In order to apply the traditional algorithm to the recognition of incomplete data, the missing attribute value can be replaced by the average value of the corresponding attribute, and then the recognition is performed to obtain the recognition result of the corresponding method, and finally the different methods are compared and analyzed.

[0085] Experiments show that when the radiation source attributes are complete, the recognition accuracy of the DBRB algorithm is 90%, while when the radiation source attributes are missing, the recognition accuracy based on the DBRB algorithm is 86.67%, a decrease of 3.7%. When the SVM algorithm is used, the recognition accuracy can only reach 75.33% when the radiation source is missing.

[0086] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and all of them belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A radiation source working mode recognition method for incomplete information, characterized in that The following steps are involved: Step 1) determine the basic style of the rule, select characteristic parameters to describe the working mode of the radiation source, and construct the radiation source working mode recognition expression form of the disjunctive confidence rule; Step 2) Build a confidence rule base to determine the premise feature attribute parameters and the confidence of the recognition result; Step 3) Identify the working mode of the radiation source, determine the individual matching degree of the feature attribute input relative to the feature attribute reference value, determine the activation weight of the input information to the confidence rule, integrate the activation rules, and obtain the final output of the confidence rule base.

2. A method for identifying radiation source working modes based on incomplete information according to claim 1, characterized in that The specific steps of step 1) are: (1) The pulse sequence of a radiation source in a certain working mode is composed of multiple coherent processing intervals (CPI). In the same CPI cycle, its related characteristic parameters remain stable. Four parameters, PW, PRI, PGI and PN, are selected to describe the working mode. The radiation source characteristics are formally described by PGW = {PW, PRI, PGI, PN}, where PW represents the pulse width, PRI represents the pulse repetition interval, PGI represents the pulse group interval, and PN represents the number of pulses in the pulse group. The radiation source working mode is modeled by PGW to reflect the joint change law of the radiation source working mode parameters; The pulse groups are divided according to the characteristic parameters of the working mode of the radiation source. In a pulse group, the characteristic parameters of the working mode of the radiation source remain stable. That is, when the radiation source is in a certain working mode, it corresponds to a set of data sets composed of PW, PRI, PGI, and PN. The input training data set can be expressed by the matrix X The formal description is: in, L Represents the total number of input training data sets, M Indicates the number of attributes of each training data. Here, the attributes of each training data are mainly composed of PW, PRI, PGI, and PN; (2) Construct the radiation source working mode recognition expression form of the disjunctive confidence rule. On the basis of the IF-THEN rule, the premise feature attribute weight, rule weight and the confidence of the conclusion part are integrated to establish the disjunctive paradigm confidence rule R k ,Right now Among them, R k represents the kth rule, represents the mth feature reference value in the kth rule, D j (j=1,2,L,N) represents the jth recognition result, Indicates that the input in the kth rule is relative to D j The confidence level, θ k represents the weight of the kth rule, Represents the weight of the i-th attribute in the k-th rule; the above IF-THEN rules are expressed as disjunctive confidence rules, and multiple disjunctive confidence rules constitute a disjunctive confidence rule base, denoted as DBRB. DBRB processes the input information, synthesizes the rules activated by the input information, and then obtains the radiation source working mode recognition result.

3. A method for identifying radiation source working modes based on incomplete information according to claim 1, characterized in that The specific steps of step 2) are: Building DBRB requires the following steps: (1) Determine the DBRB premise feature attribute A based on the actual work scenario requirements m ; (2)x m (m=1,2,…,T k ) represents the mth premise attribute, T k Indicates the number of feature attributes, according to the input information x m Determine the number of prerequisite feature attribute reference values ​​and prerequisite attribute reference values Any confidence rule needs to contain at least one premise feature attribute; (3) For the recognition result D j Construct the corresponding confidence pass It reflects whether the k-th confidence rule is complete. If it meets the following requirements, it means that the k-th confidence rule is complete. Otherwise, it means that the k-th rule is incomplete. (4) Check whether all the reference values ​​of the premise characteristic attributes have been included in the DBRB.

4. A method for identifying radiation source working modes based on incomplete information according to claim 1, characterized in that The specific steps of step 3) are: 1) Determine whether the input information is complete. If the information is complete, directly determine the feature attribute input x m (m=1,2,…,T k ) relative to the characteristic attribute reference value A m,j The individual matching degree α m,j (m=1,2,…,T k ,j=1,2,…,J m ), J m Indicates the number of reference values ​​of the mth attribute. Assuming that the working mode of the radiation source is identified, the characteristic attribute input x m Reference value A for feature attribute m,j The individual matching degree α m,j Calculated by the following formula: m represents the feature attribute number, j represents the reference value number, α m,j (m=1,2,…,T k ,j=1,2,…,J m ) represents the input component x m (m=1,2,…,T k ) for the characteristic attribute reference value A m,j The individual matching degree of J m Indicates the number of reference values ​​of the mth feature attribute; 2) If the attribute value is incomplete in the input information, the confidence distribution of the attribute is completely unknown. At this time, the individual matching degree of the attribute is 0, indicating that the incomplete attribute value cannot be matched with any attribute reference value. Determine the weight θ of each confidence rule k For information with incomplete attribute values, it is necessary to further calculate the rule attenuation factor and adjust the rule weight according to the incomplete attribute weight. Suppose the input data x i Some attribute values ​​are missing, that is, the radiation source data obtained is incomplete. By defining the rule attenuation factor δ m To quantitatively describe the impact of incomplete information on working mode recognition; in, is the attribute weight of missing data; 3) Determine the activation weight ω of the input information x for the kth confidence rule k , reflects the degree to which the kth rule is activated. Since DBRB takes "∨" as the combination relationship between the reference values ​​of the premise feature attributes, the individual matching degree Take the cumulative calculation method and then perform normalization: 4) Integrate the activation rules. k When ≠0, it means that the kth rule is activated. The rules in DBRB can be integrated using the evidence reasoning algorithm. The steps are as follows: (1) The confidence level of the conclusion part Perform the basic probability mass (BPA) conversion, that is, activate the weight ω k Confidence in the conclusion Perform fusion processing: in, Indicates relative to the evaluation result D j BPA, represents the BPA relative to the set D, that is, corresponding to the evaluation result D not set to any j BPA, is associated with the activation weight of the kth rule, It is related to the incompleteness of the evaluation result of the kth rule; (2) Combine K rules to obtain the recognition result D j The confidence level β j ; Through the evidence reasoning parsing algorithm, all the rules in the confidence rule base can be combined, and the final output S(x) of the confidence rule base is: in, Indicates that the result obtained after rule combination is relative to the evaluation result D j confidence level.

Images