Multifunctional radar behavior prediction modeling method based on plan behavior theory

Abstract

The invention discloses a multifunctional radar behavior prediction modeling method based on a plan behavior theory, and the method comprises the steps: constructing a mapping relation between a human plan behavior theory and multifunctional radar plan behaviors, constructing a multifunctional radar plan behavior prediction theoretical structure model, and carrying out the mathematical modeling of each analysis module; secondly, establishing a multifunctional radar simulation scene, acquiring multifunctional radar parameters and behavior data, constructing an example data set, preprocessing the example data set, constructing a feature sample for model training, and verifying the accuracy of the model; and finally, constructing a multifunctional radar plan behavior prediction system based on a plan behavior theory based on the behavior prediction model with qualified verification accuracy, and inputting radar behavior data to complete radar behavior prediction. According to the method, mathematical modeling of multifunctional radar behavior prediction is effectively realized, and the problems that a behavior prediction basic support theory is lacked and a behavior prediction method system is not established are solved.

Description

Technical Field

[0001] The invention belongs to the technical field of multifunctional radar behavior prediction modeling, and in particular relates to a multifunctional radar behavior prediction modeling method based on planned behavior theory. Background Art

[0002] Multi-Function Radar (MFR) behavior prediction is a key part of MFR cognitive technology. This technology can help accurately grasp the target situation. Through this technology, we can quantitatively understand its type and operating status, providing a key basis for us to formulate efficient response strategies and decisions. Behavior prediction of multi-function radar is a relatively cutting-edge direction.

[0003] At present, there are two main types of behavior prediction methods for multi-function radars: one is direct prediction based on recognition results, and the other is evolutionary prediction based on reasoning results. Direct prediction based on recognition results means that prediction is achieved by directly modeling and learning the conditional probability relationship between historical radar behavior sequences and future radar behavior sequences; evolutionary prediction based on reasoning results means that the radar behavior sequence is controlled by a certain optimization problem, and based on the optimization problem objective function obtained by reasoning, the optimization problem is solved again to predict future radar behavior.

[0004] The paper “OU Jian, CHEN Yongguang, ZHAO Feng, et al. Novel approach for the cognition and prediction of multifunction radar behaviors based on predictive state representations. Sensors, 2017, 17(3): 632” studies the prediction of multifunction radar signals based on predictive state representations for the prediction of multifunction radar behaviors, and verifies the method using the Mercury radar data as an example. However, the behavior of the Mercury radar is relatively simple, and the signal hierarchical model in the method has limitations. The simple representation of behavior is not enough to describe the behavior of other multifunction radars. The theory of planned behavior is one of the most important theories in the field of intentional behavior research. Its main theory is to use behavioral attitudes, subjective norms, and perceived behavioral control to reflect behavioral intentions. The theory of planned behavior analyzes the influence mechanism of the three dimensions in the framework of the theory of planned behavior model on individual behavioral intentions by processing the collected information, and thus predicts and explains the individual's final actual behavior with individual behavioral intentions as the medium.

[0005] Existing research rarely focuses on the prediction of multi-function radar behavior, and most of them are constructed using traditional hierarchical models, which makes it difficult to make advance deployment of the opponent's multi-function radar behavior intentions. Therefore, it is of great value to study a multi-function radar behavior prediction modeling method based on the theory of planned behavior. Summary of the invention

[0006] To solve the above technical problems, the present invention provides a multi-function radar behavior prediction modeling method based on the planned behavior theory, which can provide rule guidance for the multi-function radar behavior prediction and achieve high-accuracy prediction of the multi-function radar behavior.

[0007] The technical solution adopted by the present invention is: a multifunctional radar behavior prediction modeling method based on the planned behavior theory, and the specific steps are as follows:

[0008] S1. Construct a multifunctional radar planned behavior prediction model based on the planned behavior theory;

[0009] S2. Based on step S1, a multifunctional radar planned behavior prediction system based on the planned behavior theory is constructed on a high-performance platform;

[0010] The multifunctional radar planning behavior prediction system comprises: a data loading module, a data initialization module, a multifunctional radar behavior prediction module, and a result output module.

[0011] The multifunctional radar behavior prediction module is the multifunctional radar planned behavior prediction model constructed in step S1; the multifunctional radar behavior prediction module includes: pulse-level behavior intention prediction, feature-level behavior intention prediction, and tactical-level behavior intention prediction.

[0012] S3, the data loading module loads the radar behavior data into the data initialization module, and after the data initialization, inputs it into the multi-function radar behavior prediction module to perform radar behavior prediction, and outputs the result through the result output module;

[0013] Among them, before the multifunctional radar planned behavior prediction model performs radar behavior prediction, model training is first performed to verify the prediction accuracy. After the accuracy is qualified, it is used in the multifunctional radar planned behavior prediction system to perform radar behavior prediction.

[0014] Furthermore, the step S1 is specifically as follows:

[0015] S11. Construct the mapping relationship between the theory of human planned behavior and the theory of multifunctional radar planned behavior;

[0016] (1) Data collection;

[0017] In the theory of human planned behavior, data is collected to analyze the components of an individual's attitude toward a specific behavior, subjective norms, and intuitive behavioral control, and to calculate the corresponding strength. The collection methods include questionnaires and interviews; the collected data include individual characteristics, group characteristics, and external environment characteristics.

[0018] Correspondingly, in the multifunctional radar behavior prediction, the data collection method in the theory of human planned behavior is extended to the data collection tool by analogy, and the data collection results in the theory of human planned behavior are extended to the multifunctional radar behavior prediction by analogy. Among them, in the multifunctional radar behavior prediction, the data collection tool includes: the detection equipment in the radar passive detection; the data results collected by the multifunctional radar behavior prediction include: the multifunctional radar signal parameters intercepted by the detection equipment, and the analysis results of the radar passive detection system on the multifunctional radar working environment information.

[0019] (2) behavioral attitude;

[0020] In the theory of human planned behavior, behavioral attitude reflects the influence of an individual's own characteristics on his or her behavior, that is, the individual's evaluation of the degree to which he or she likes or dislikes performing a specific behavior. The individual's own characteristics include: his or her own values ​​and age.

[0021] Corresponding to the multi-function radar behavior prediction, the radar behavior attitude reflects the impact of the radar's own characteristics on the radar behavior. Among them, the radar's own characteristics include: multi-function radar behavior state transition probability characteristics, waveform parameter characteristics, antenna scanning mode characteristics; the radar behavior includes: radar behavior state, working mode, signal parameters. According to the multi-function radar's own characteristics ξ, the radar's behavior attitude intensity A(σ′) to the behavior σ′ is recorded as f A (σ′,ξ).

[0022] (3) Subjective norms;

[0023] In the theory of human planned behavior, subjective norms reflect the influence of the individual's social group on the individual's behavior, that is, the external pressure and expectations perceived by the individual when deciding whether to perform a specific behavior.

[0024] Correspondingly, in the prediction of multifunctional radar behavior, the influencing factors of radar subjective norms are the radar specifications of the multifunctional radar radiation source, including: multifunctional radar behavior state, working mode transfer rules, and multifunctional radar waveform beam selection rules. They can be compared to the external expectations perceived by individuals when performing specific behaviors in the theory of human planned behavior. That is, the radar specifications in the knowledge base of the multifunctional radar radiation source mechanism can be regarded as the expectations of experts on the behavior that the radar should perform.

[0025] Assume that there is a radar rule ψ in the knowledge base of the multifunctional radar radiation source mechanism, and the subjective norm intensity SN(σ′) of the radar on the behavior σ′ is recorded as fSN (σ′|ψ).

[0026] (4) perceived behavioral control;

[0027] In the theory of human planned behavior, perceived behavioral control reflects the ability and experience of an individual, as well as the influence of the external environment on individual behavior, that is, the degree to which it is easy or difficult for an individual to perform a certain behavior in a certain way. Among them, perceived behavioral control in the theory of human planned behavior includes: internal factors and external factors, that is, one's own ability and the external environment.

[0028] Corresponding to the multi-function radar behavior prediction, radar perceptual behavior control refers to the difficulty faced by the radar when executing a specific behavior, which also includes internal factors and external factors, also known as radar planned behavior constraint control. Among them, the internal factors of radar perceptual behavior control refer to certain internal cost criteria and constraints that must be followed when the multi-function radar is working, including: the time resource constraints faced by the radar's internal resources and task scheduling, which can be compared to the individual's ability in the theory of human planned behavior; external factors refer to the working environment of the multi-function radar, ε PBC The impact of the characteristics of our own and the enemy's other equipment on the multi-function radar's planned behavior can be compared to the external resource conditions in the theory of human planned behavior.

[0029] According to the radar history and current behavior σ 1:T , the radar's internal resource constraintsζ and the radar's working environmentε PBC The perceived behavioral control intensity PBC(σ′) of the multifunctional radar on the behavior σ′ is denoted as f PBC (σ 1:T ,σ′|ζ,ε PBC ).

[0030] (5) behavioral intention;

[0031] In human behavior prediction, behavioral intention refers to the possibility of an individual performing a specific behavior, that is, the individual's judgment on the subjective probability of taking a specific behavior.

[0032] Correspondingly, in the multi-function radar behavior prediction, radar behavior intention refers to the future behavior that the multi-function radar intends to take according to its own task scheduling strategy, waveform parameter characteristics, beam scanning characteristics, following the multi-function radar behavior rules and the radar's own time and energy resource constraints, under the influence of its own and the opponent's equipment and target status.

[0033] (6) Actual behavior;

[0034] In human behavior prediction, actual behavior refers to the actual behavior that an individual ultimately performs under the dual guidance of behavioral intention and perceived behavioral control.

[0035] Correspondingly, in the multi-function radar behavior prediction, the actual behavior of the radar refers to the behavior sequence actually executed by the multi-function radar in each scheduling interval, that is, the actual behavior that the radar finally executes under the dual guidance of behavioral intention and perceived behavioral control.

[0036] The difference between radar actual behavior and radar behavior intention is that radar behavior intention refers to the sequence of behaviors that the radar selects for the next scheduling interval after each scheduling interval ends, and the event has not yet occurred; while radar actual behavior refers to the behavior that the radar has executed and has occurred in the current scheduling interval. Among them, if the multi-function radar does not change after selecting the task to be performed in the next scheduling interval, it is believed that the radar behavior intention can reflect the actual behavior of the radar that will occur in the future.

[0037] S12, based on the mapping relationship in step S11, constructing a theoretical structural model of multi-function radar plan behavior prediction;

[0038] The multifunctional radar plan behavior prediction theoretical structure model comprises: a radar plan behavior attitude analysis module, a radar plan behavior specification analysis module, a radar plan behavior constraint control analysis module, and a radar plan behavior intention analysis module.

[0039] Among them, the analysis of the multi-function radar behavior intensity is divided into three levels, including: pulse level, feature level, and tactical level. Pulse level behavior is the basic expression, including: carrier frequency, pulse repetition rate, and pulse width; feature level behavior is the characteristic expression of pulse level behavior, including: carrier frequency modulation, repetition rate modulation, and pulse width modulation; tactical level behavior is the top-level expression, including: behavior status, working mode, and radar function.

[0040] S13, based on step S12, according to the improved expected utility maximization formula, mathematical modeling is performed on the multifunctional radar plan behavior prediction theoretical structure model and its various analysis modules to obtain the multifunctional radar plan behavior prediction model;

[0041] S131, performing radar plan behavior attitude analysis based on the radar plan behavior attitude analysis module to obtain the behavior attitude strength;

[0042] Firstly, the historical and current radar behaviors are input into the radar plan behavior attitude analysis module to analyze the individual characteristics of the multi-function radar.

[0043] The individual characteristic analysis of the multifunctional radar includes: pulse level behavior analysis, feature level behavior analysis, and tactical level behavior analysis, as follows:

[0044] (1) Pulse-level behavior analysis;

[0045] The pulse level behavior intensity analysis is the characteristic parameters of any pulse level, including: carrier frequency RF, pulse repetition interval PRI, pulse width PW.

[0046] Set the multi-function radar pulse level behavior parameter to be predicted as x pulee , the multi-function radar's planned behavior attitude strength for the L+1th pulse parameter The expression is as follows:

[0047]

[0048] in, represents the L+1th pulse parameter, represents the predicted value of the L+1th pulse parameter, and α represents the coefficient. Formula (1) maps the radar's planned behavior attitude intensity for the L+1th pulse parameter to between 0 and 1. near When it approaches 1, and When the deviation is large, it approaches 0.

[0049] (2) Feature-level behavior analysis;

[0050] The characteristic level behavior intensity analysis is the characteristic category corresponding to the pulse level parameters, including: radar PRI modulation type, RF modulation type, and PW modulation type.

[0051] Set the characteristic level behavior parameter of the multi-function radar to be predicted as x feature , the multi-function radar selects the feature-level behavior x for the next moment feature The planned behavior attitude strength A(x feature ) is expressed as follows:

[0052]

[0053] Among them, x feature Represents a specific PRI modulation type, and M represents all possible corresponding feature-level behavior types stored in the multifunction radar waveform library, which is obtained by behavioral recognition of radar waveforms. Formula (2) means that the behavioral attitude intensity of the multifunction radar to the modulation type existing in its waveform library is 1, and the behavioral attitude intensity to other modulation types is 0.

[0054] (3) Tactical-level behavioral analysis;

[0055] Tactical level behavior intensity analysis is the characteristics of any tactical level, including: radar behavior status, radar working mode, and radar function.

[0056] Set the tactical level behavior parameter of the multi-function radar to be predicted as x tactical , the tactical behavior of the radar has Markov properties, and the multi-function radar behavior state transition probability matrix is ​​obtained by calculation

[0057] Among them, p ij =p(x tactical+1 =s j |x tactical =s i ), that is, the multi-function radar is composed of tactical level behaviors i Towards tactical level behavior j The probability of switching.

[0058] According to the calculated multi-function radar tactical level behavior state transition probability and behavior state at time t s(t) = s i , the radar executes event s(t+1)=s at time t+1 j The intensity of behavioral attitude A(s j )=p ij .

[0059] After the individual characteristics of the multifunctional radar are analyzed, according to the multifunctional radar's own characteristics ξ, the radar's behavioral attitude intensity towards the behavior σ′ is recorded as A(σ′)=f A (σ′,ξ).

[0060] S132, performing radar plan behavior specification analysis based on the radar plan behavior specification analysis module to obtain the behavior specification strength;

[0061] First, the historical and current radar behaviors, historical and current working environments ε SN Input the radar plan behavior specification analysis module, that is, input it into the multi-function radar radiation source mechanism knowledge base for matching, and obtain the associated mechanism knowledge ψ'.

[0062] The analysis of behavioral norms is divided into two levels: pulse level and feature level, and tactical level, as follows:

[0063] (1) Pulse-level and feature-level planning behavior specifications;

[0064] First calculate the degree of difference As the basis for the planned behavior specification strength SN(σ′) of the multifunction radar for the pulse-level or characteristic-level behavior σ′, where the radar behavior σ′ corresponds to the parameter Θ σ′ , is the radar in the current scene ε SN Parameters required for the behavior state s(t).

[0065] Radar behavior parameter Θ σ′ Required parameters of radar The degree of difference between them is calculated by Euclidean distance, which is expressed as follows:

[0066]

[0067] According to the degree of difference between the parameters, the planned behavior specification strength SN(σ′) of the multi-function radar for pulse-level or characteristic-level behavior is expressed as The expression is as follows:

[0068]

[0069] Among them, k s Represents the coefficient.

[0070] (2) Tactical-level planning behavior norms;

[0071] Assume that the tactical level behavior of multi-function radar in the radar mechanism knowledge base has a switching relationship ψ trans , the inversion result of the tactical level behavior at the current moment is μ t , the possible tactical level behavior at the next moment is The current working environment of the radar is ε SN .

[0072] The tactical level behavior inversion result μ t Radar working environment SN Input into the radar knowledge base for matching, and according to the transfer rules of tactical level behavior, obtain the associated mechanism knowledge, that is, ψ' = ψ trans . Tactical level behavior switching relationship Degree of compliance with radar code of conduct Expressed as The expression is as follows:

[0073]

[0074] After the multi-function radar behavior norm analysis, based on the associated mechanism knowledge ψ' in the multi-function radar radiation source mechanism knowledge base, the radar's subjective norm intensity SN(σ') for behavior σ' is recorded as f SN (σ′|ψ′).

[0075] S133, performing radar plan behavior constraint control analysis based on the radar plan behavior constraint control analysis module to obtain behavior constraint control strength;

[0076] Radar plan behavior constraint control input includes: multi-function radar system resource constraint limit ζ, multi-function radar working environment characteristics ε PBC And the radar history and the current radar behavior σ 1:T .

[0077] Among them, the resource constraints of the multi-function radar system include: time resource constraints, energy resource constraints, computer resource constraints; multi-function radar working environment characteristics PBC Including: characteristics of our other equipment, tactical characteristics of the opponent's targets, and movement characteristics of the opponent's targets.

[0078] According to the input, the planned behavior constraint control strength PBC(σ′) of the multifunctional radar for possible behavior σ′ is expressed as f PBC (σ′|ζ,ε PBC ,σ 1:T ), the specific expression is as follows:

[0079]

[0080] Among them, f ζ (σ′) represents the radar plan behavior constraint control generated by internal factors, represents the radar planned behavior constraint control generated by external factors. g(.) represents the planned behavior constraint control calculation function, and the input is f ζ (σ′), The final calculation result expression of the multi-function radar planned behavior constraint control strength is as follows:

[0081]

[0082] Among them, w ζ and Represent weights respectively.

[0083] S134, calculating the utility function according to the improved formula for maximizing expected utility and constructing a radar plan behavior intention analysis module, and inputting the behavior attitude strength, behavior norm strength, and behavior constraint control strength obtained in steps S131-S133 into the radar plan behavior intention analysis module;

[0084] According to the improved expected utility maximization formula, the mathematical representation of radar planning behavior is constructed, and the expected result of the improved expected utility maximization behavior is σ * The expression is as follows:

[0085]

[0086] Where X represents a finite set of all possible outcomes of the behavior, i.e., X = x1, x2, ..., x N , N represents the number of behaviors; U(·) represents the utility function defined on the set of possible behaviors of an individual Σ, and the input P σ represents the probability distribution of the impact of the behavior, u σ (x i ) indicates that the execution behavior σ causes the impact x i The utility generated when ρ (x i ) indicates that when action σ is executed, the result is x i probability.

[0087] uσ (·) The specific expression is as follows:

[0088]

[0089] Where f(·) represents the functional expression, A(σ), SN(σ), and PBC(σ) represent the behavioral attitude, behavioral norms, and constraint control of the individual when performing behavior σ, respectively.

[0090] According to formula (9), the utility function expression is as follows:

[0091] U(P σ )=w A f A (σ)+w SN f SN (σ)+w PBC f PBC (σ) (10)

[0092] Among them, w A 、w SN 、w PBC They represent the weights of behavioral attitude, behavioral norms and behavioral constraint control respectively.

[0093] Individuals perform actions i The subjective probability The judgment expression is as follows:

[0094]

[0095] Among them, λ∈0,+∞) represents the rationality parameter. When λ=0, individuals ignore the expected utility of different strategies and have equal probability of executing each behavior; when λ→+∞, the probability of individuals executing the behavior that maximizes the expected utility is 1, and the probability of executing other behaviors is 0.

[0096] Using U(P σ ) and formula (11) to calculate the subjective judgment of the probability of executing different behaviors at the next moment, that is, the radar plan behavior intention Individual behavior σ i The intensity of behavioral intention And the individual's final behavior is determined by the behavioral intention and behavioral constraint control, so the individual's final behavior prediction result B is expressed as follows:

[0097] B=Rul(BI,PC) (12)

[0098] Where BI=[B1,B2,…,B N ] T represents the intensity of an individual’s behavioral intention to perform each behavior, [·] Trepresents the transposition operation; PC = [PBC(σ1), PBC(σ2),…, PBC(σ N )] represents the behavioral constraint control of each behavior; Rule(·) represents the mapping relationship between behavioral intention, perceived behavioral control and individual final behavior, which is achieved by training machine learning models, including support vector machines and neural networks.

[0099] Furthermore, in step S3, the multifunctional radar planned behavior prediction model performs model training to verify the prediction accuracy, as follows:

[0100] A1. Establish a multi-function radar simulation scenario, obtain multi-function radar parameters and behavior data, build a sample data set and perform preprocessing, and build feature samples;

[0101] Firstly, a multifunctional radar simulation scenario is established to generate n sets of pulse data according to the radar tactical-level behaviors executed in various scenarios.

[0102] The pulse characteristic parameters include: pulse carrier frequency RF and its modulation type, pulse width PW and its modulation type, pulse repetition interval PRI and its modulation type. In addition to the pulse characteristic parameters, according to the desired behavior intentions at different levels, tactical level behavior settings can also be performed, including: radar behavior status, radar working mode, and radar function.

[0103] Then, according to the behavioral intentions at different levels, feature samples are constructed through preprocessing.

[0104] The feature sample is defined as D = (X, y), where X and y represent corresponding sequences and labels that meet different requirements.

[0105] For pulse-level behavior, X is l represents the sequence length, and y represents the predicted parameter true value label;

[0106] For feature-level behavior, X is l represents the sequence length, and y represents the predicted corresponding feature-level behavior label;

[0107] For tactical-level behavior, X is n t It represents the number of corresponding parameter types, and y represents the desired tactical-level behavior label.

[0108] Finally, for pulse-level behavior, feature-level behavior, and tactical-level behavior, based on the constructed feature samples, different error levels k Err_leel , a total of data_num groups of multi-function radar data sets data_set are generated.

[0109] Among them, the sequence length is data_length; kErr_leel , data_num and data_length are generated according to actual conditions.

[0110] A2, based on the feature samples obtained in step A1, input the multi-function radar plan behavior prediction model for training to verify the prediction accuracy;

[0111] A21. Divide all sample data into training sets and test sets according to actual conditions. Send the training set to the GRU neural network for training. Dynamically adjust the learning rate and batch size according to the accuracy of the training set until the model loss value converges, completing the behavior prediction model training.

[0112] Since there are three analysis modules: behavior attitude, behavior norms, and behavior constraint control, the loss function expression of the behavior prediction model is as follows:

[0113] l ss=l ss d +l ss s +l ss p (13)

[0114] in, represents the loss function of the behavior attitude analysis module, y=[y1,…,y i ]、 Represent the actual value and the predicted value vector of the model output, respectively. represents the dynamic error weight; represents the loss function of the behavior specification analysis module, Δy n Represents y p The difference operation of represents the square of the vector 2-norm, N data represents the number of groups of sequences in the dataset, α s and λ s Represents a constant factor, and introduces L1 regularization to control excessive growth of differences; represents the loss function of the behavior constraint control analysis module, represents the prediction parameter set, represents the target parameter set, β and η represent constant factors.

[0115] A22, sending the test set data generated in step A21 to the trained model to determine the accuracy;

[0116] Prediction accuracy

[0117] Among them, N test represents the number of samples in the test set, represents the intention value of the network output at the tth moment, y tRepresents the true value of the network output at the tth moment, T test_ot represents the total number of network outputs; δ(·) represents the impulse function.

[0118] Represents a binary judgment impulse function for the intended value and the real value, which is used to distinguish right from wrong; for impulse-level behavior:

[0119]

[0120] Among them, threshold pulse Indicates the correct range threshold of the pulse parameter. The conditions are set according to the actual situation. The result is 1, which proves that the intention value and the true value match and are correct; when The result is 0, which proves that the intended value and the actual value do not match and are wrong.

[0121] For feature-level and tactical-level behaviors:

[0122]

[0123] when The result is 1, which proves that the intention value and the true value match and are correct; when The result is 0, which proves that the intended value and the true value do not match and are wrong;

[0124] After the prediction accuracy of the behavior prediction model passes the test, ac p Greater than the correct threshold T acc When the model is verified to be valid, the model data is then exported for use in the multi-function radar planning behavior prediction system to perform radar behavior prediction.

[0125] Among them, T acc Set the value according to the actual situation.

[0126] Furthermore, in step S3, the data loading module uses a program to read the radar behavior data, inputs the data initialization module to initialize the data, and finally imports the read and initialized data into the multi-functional radar behavior prediction module to output the prediction result.

[0127] The data initialization includes: dividing the data into short and long ones, that is, padding the pulse sequence whose data length is less than data_length with zeros; dividing the pulse sequence whose data length is greater than data_length with the length of data_length.

[0128] Beneficial effects of the present invention: The method of the present invention constructs a mapping relationship between human planned behavior theory and multifunctional radar planned behavior, constructs a multifunctional radar planned behavior prediction theory structural model and mathematically models its various analysis modules to obtain a multifunctional radar planned behavior prediction model, then establishes a multifunctional radar simulation scene, obtains multifunctional radar parameters and behavior data, constructs an example data set and performs preprocessing, constructs feature samples for model training, verifies the model accuracy, and finally constructs a multifunctional radar planned behavior prediction system based on the planned behavior theory based on the behavior prediction model with qualified verification accuracy, and inputs radar behavior data to complete radar behavior prediction. The method of the present invention uses the planned behavior theory to mathematically model the multifunctional radar behavior prediction, provides rule guidance for the multifunctional radar behavior prediction, provides a specific method for high-accuracy multifunctional radar behavior prediction, and verifies the feasibility and effectiveness through the application of multifunctional radar pulse repetition interval prediction, solving the problem of lack of basic supporting theory for behavior prediction and failure to establish a behavior prediction method system. BRIEF DESCRIPTION OF THE DRAWINGS

[0129] Figure 1 The present invention is a flowchart of a multi-function radar behavior prediction modeling method based on the theory of planned behavior.

[0130] Figure 2 It is a structural diagram of a multifunctional radar planned behavior prediction system based on the planned behavior theory in an embodiment of the present invention.

[0131] Figure 3 Schematic diagram of the mapping relationship between human planned behavior and multi-function radar planned behavior in an embodiment of the present invention.

[0132] Figure 4 Schematic diagram of the theoretical structural model of multi-function radar planning behavior prediction in an embodiment of the present invention.

[0133] Figure 5 Schematic diagram of the behavior hierarchy of the multi-function radar in an embodiment of the present invention.

[0134] Figure 6 Schematic diagram of the meaning of radar planning behavior attitude in an embodiment of the present invention.

[0135] Figure 7 Schematic diagram of a radar planning behavior attitude analysis module in an embodiment of the present invention.

[0136] Figure 8 It is a schematic diagram of the specific meaning of the radar planning behavior specification in an embodiment of the present invention.

[0137] Fig. 9 Schematic diagram of a radar planning behavior specification analysis module in an embodiment of the present invention.

[0138] Fig.10It is a schematic diagram of the specific meaning of the radar plan behavior constraint control in an embodiment of the present invention.

[0139] Fig.11 Schematic diagram of a radar planning behavior constraint control analysis module in an embodiment of the present invention.

[0140] Fig.12 Schematic diagram of a radar planning behavior intention analysis module in an embodiment of the present invention.

[0141] Fig.13 A schematic diagram of constructing a loss function based on the theory of planned behavior in an embodiment of the present invention.

[0142] Fig.14 Schematic diagram of pulse level behavior prediction for different modulation types in an embodiment of the present invention.

[0143] Fig.15 Schematic diagram of the accuracy of pulse level behavior of different modulation types at different error levels in an embodiment of the present invention.

[0144] Fig.16 Schematic diagram of feature-level behavior prediction accuracy in an embodiment of the present invention.

[0145] Fig.17 Schematic diagram of tactical level behavior prediction accuracy in an embodiment of the present invention. DETAILED DESCRIPTION

[0146] The method of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0147] like Figure 1 As shown, a flow chart of a multi-function radar behavior prediction modeling method based on the planned behavior theory of the present invention, the specific steps are as follows:

[0148] S1. Construct a multifunctional radar planned behavior prediction model based on the planned behavior theory;

[0149] S2. Based on step S1, a multifunctional radar planned behavior prediction system based on the planned behavior theory is constructed on a high-performance platform;

[0150] like Figure 2 As shown, the multifunctional radar planned behavior prediction system includes: a data loading module, a data initialization module, a multifunctional radar behavior prediction module, and a result output module.

[0151] The multifunctional radar behavior prediction module is the multifunctional radar planned behavior prediction model constructed in step S1; the multifunctional radar behavior prediction module includes: pulse-level behavior intention prediction, feature-level behavior intention prediction, and tactical-level behavior intention prediction.

[0152] S3, the data loading module loads the radar behavior data into the data initialization module, and after the data initialization, inputs it into the multi-function radar behavior prediction module to perform radar behavior prediction, and outputs the result through the result output module;

[0153] Among them, before the multifunctional radar planned behavior prediction model performs radar behavior prediction, model training is first performed to verify the prediction accuracy. After the accuracy is qualified, it is used in the multifunctional radar planned behavior prediction system to perform radar behavior prediction.

[0154] In this embodiment, the step S1 is specifically as follows:

[0155] S11. Construct the mapping relationship between the theory of human planned behavior and the theory of multifunctional radar planned behavior, such as Figure 3 As shown;

[0156] (1) Data collection;

[0157] In the theory of human planned behavior, the collected data is used to analyze the individual's attitude towards a specific behavior, subjective norms, and the components of intuitive behavioral control, and calculate the corresponding strength. The collection methods (tools) include: questionnaires, interviews; the collected data include: individual characteristics, group characteristics, and external environment characteristics.

[0158] Correspondingly, in the multifunctional radar behavior prediction, the data collection method in the theory of human planned behavior is extended to the data collection tool by analogy, and the data collection results in the theory of human planned behavior are extended to the multifunctional radar behavior prediction by analogy. Among them, in the multifunctional radar behavior prediction, the data collection tool includes: radar detection equipment; the data results collected by the multifunctional radar behavior prediction include: multifunctional radar signal parameters intercepted by the detection equipment, and the analysis results of the radar passive detection system on the multifunctional radar working environment information.

[0159] (2) behavioral attitude;

[0160] In the theory of human planned behavior, behavioral attitude reflects the influence of an individual's own characteristics on his or her behavior, that is, the individual's evaluation of the degree to which he or she likes or dislikes performing a specific behavior. The individual's own characteristics include: his or her own values ​​and age.

[0161] Correspondingly, in the multi-function radar behavior prediction, the radar behavior attitude reflects the impact of the radar's own characteristics on the radar behavior.

[0162] Among them, the radar characteristics include: multi-function radar behavior state transition probability characteristics, waveform parameter characteristics, antenna scanning mode characteristics; the radar behavior includes: radar behavior state, working mode, signal parameters. According to the multi-function radar characteristics ξ, the radar's behavior attitude intensity A(σ′) to the behavior σ′ is recorded as f A(σ′,ξ).

[0163] (3) Subjective norms;

[0164] In the theory of human planned behavior, subjective norms reflect the influence of the individual's social group on the individual's behavior, that is, the external pressure and expectations perceived by the individual when deciding whether to perform a specific behavior.

[0165] In the prediction of multifunctional radar behavior, the influencing factors of radar subjective norms are the radar specifications of the multifunctional radar radiation source, including: multifunctional radar behavior state, working mode transfer rules, and multifunctional radar waveform beam selection rules. They can be compared to the external expectations perceived by individuals when performing specific behaviors in the theory of human planned behavior. That is, the radar specifications in the knowledge base of the multifunctional radar radiation source mechanism can be regarded as the expectations of experts on the behavior that the radar should perform.

[0166] Assume that there is a radar rule ψ in the knowledge base of the multifunctional radar radiation source mechanism, and the subjective norm intensity SN(σ′) of the radar on the behavior σ′ is recorded as f SN (σ′|ψ).

[0167] (4) perceived behavioral control;

[0168] In the theory of human planned behavior, perceived behavioral control reflects an individual's ability and experience, as well as the influence of the external environment on the individual's behavior, that is, the degree to which it is easy or difficult for an individual to perform a particular behavior in a certain way.

[0169] Among them, perceived behavioral control in the theory of human planned behavior includes: internal factors and external factors, namely, one's own abilities and the external environment.

[0170] In the prediction of multi-function radar behavior, radar perceived behavior control refers to the difficulty faced by the radar when executing a specific behavior, which also includes internal and external factors, and is also called radar planned behavior constraint control.

[0171] Among them, the internal factors of radar perception behavior control refer to the internal cost criteria and constraints that the multifunctional radar must follow when working, including: the time resource constraints faced by the radar's internal resources and task scheduling, which can be compared to the individual capabilities in the theory of human planned behavior; the external factors refer to the working environment of the multifunctional radar. PBC The impact of the characteristics of our own and the enemy's other equipment on the multi-function radar's planned behavior can be compared to the external resource conditions in the theory of human planned behavior.

[0172] According to the radar history and current behavior σ 1:T , the radar's internal resource constraintsζ and the radar's working environmentε PBC The perceived behavioral control intensity PBC(ε′) of the multifunctional radar on the behavior σ′ is denoted as f PBC(σ 1:T ,σ′|ζ,ε PBC ).

[0173] (5) behavioral intention;

[0174] In human behavior prediction, behavioral intention refers to the possibility of an individual performing a specific behavior, that is, the individual's judgment on the subjective probability of taking a specific behavior.

[0175] In the multi-function radar behavior prediction, radar behavior intention refers to the future behavior that the multi-function radar intends to take according to its own task scheduling strategy, waveform parameter characteristics, beam scanning characteristics, following the multi-function radar behavior rules and the radar's own time and energy resource constraints, under the influence of its own and the opponent's equipment and target status.

[0176] (6) Actual behavior;

[0177] In human behavior prediction, actual behavior refers to the actual behavior that an individual ultimately performs under the dual guidance of behavioral intention and perceived behavioral control.

[0178] In the multi-function radar behavior prediction, the actual radar behavior refers to the behavior sequence actually executed by the multi-function radar in each scheduling interval, that is, the actual behavior finally executed by the radar under the dual guidance of behavioral intention and perceived behavioral control.

[0179] The difference between the actual radar behavior and the radar behavior intention is that the radar behavior intention refers to the sequence of behaviors that the radar selects for the next scheduling interval after each scheduling interval, and the event has not yet occurred; while the actual radar behavior refers to the behavior that the radar has executed and occurred in the current scheduling interval.

[0180] Among them, if the multi-function radar does not change after selecting the next scheduling interval to perform the task, it is believed that the radar behavior intention can reflect the actual behavior of the radar that will occur in the future.

[0181] S12, based on the mapping relationship of step S11, construct a theoretical structural model of multi-function radar plan behavior prediction, such as Figure 4 As shown;

[0182] The multifunctional radar plan behavior prediction theoretical structure model comprises: a radar plan behavior attitude analysis module, a radar plan behavior specification analysis module, a radar plan behavior constraint control analysis module, and a radar plan behavior intention analysis module.

[0183] Among them, Figure 5As shown in the figure, the analysis of the multi-function radar behavior intensity is divided into three levels, including: pulse level, feature level, and tactical level. Pulse level behavior is the basic expression, including: carrier frequency, pulse repetition rate, and pulse width; feature level behavior is the characteristic expression of pulse level behavior, including: carrier frequency modulation, repetition rate modulation, and pulse width modulation; tactical level behavior is the top-level expression, including: behavior status, working mode, and radar function.

[0184] S13, based on step S12, according to the improved expected utility maximization formula (Formula (8)), mathematical modeling is performed on the multifunctional radar plan behavior prediction theoretical structure model and its various analysis modules to obtain a multifunctional radar plan behavior prediction model;

[0185] S131、The specific meaning of radar plan behavior attitude is as follows Figure 6 As shown, based on the radar plan behavior attitude analysis module, the radar plan behavior attitude analysis is performed to obtain the behavior attitude strength;

[0186] Radar Program Behavior Attitude Analysis Module Figure 7 As shown, the historical and current radar behaviors are first input into the radar plan behavior attitude analysis module to perform individual characteristic analysis of the multi-function radar.

[0187] The individual characteristic analysis of the multifunctional radar includes: pulse level behavior analysis, feature level behavior analysis, and tactical level behavior analysis, as follows:

[0188] (1) Pulse-level behavior analysis;

[0189] The pulse level behavior intensity analysis is the characteristic parameters of any pulse level, including: carrier frequency (RF), pulse repetition interval (PRI), pulse width (PW).

[0190] Set the multi-function radar pulse level behavior parameter to be predicted as x pulse , the multi-function radar's planned behavior attitude strength for the L+1th pulse parameter The expression is as follows:

[0191]

[0192] in, represents the L+1th pulse parameter, represents the predicted value of the L+1th pulse parameter, and α represents the coefficient. Formula (1) maps the radar's planned behavior attitude intensity for the L+1th pulse parameter to between 0 and 1. near When it approaches 1, and When the deviation is large, it approaches 0.

[0193] (2) Feature-level behavior analysis;

[0194] The characteristic level behavior intensity analysis is the characteristic category corresponding to the pulse level parameters, including: radar PRI modulation type, RF modulation type, and PW modulation type.

[0195] Set the characteristic level behavior parameter of the multi-function radar to be predicted as x feature , the multi-function radar selects the feature-level behavior x for the next moment feature The planned behavior attitude strength A(x feature ) is expressed as follows:

[0196]

[0197] Among them, x feature represents a specific PRI modulation type, and M represents all possible corresponding feature-level behavior types stored in the multifunction radar waveform library, which can be obtained by behavioral identification of the radar waveform. Formula (2) means that the behavior attitude intensity of the multifunction radar to the modulation type existing in its waveform library is 1, and the behavior attitude intensity to other modulation types is 0.

[0198] (3) Tactical-level behavioral analysis;

[0199] Tactical level behavior intensity analysis is the characteristics of any tactical level, including: radar behavior status, radar working mode, and radar function.

[0200] Set the tactical level behavior parameter of the multi-function radar to be predicted as x tactical , the tactical behavior of the radar has Markov properties, and the multi-function radar behavior state transition probability matrix is ​​obtained by calculation

[0201] Among them, p ij =p(x tactical+1 =s j |x tactical =s i ), that is, the multi-function radar is composed of tactical level behaviors i Towards tactical level behavior j The probability of switching.

[0202] According to the calculated multi-function radar tactical level behavior state transition probability and behavior state at time t s(t) = s i , the radar executes event s(t+1)=s at time t+1 j The intensity of behavioral attitude A(s j )=p ij .

[0203] After the individual characteristics of the multifunctional radar are analyzed, according to the multifunctional radar's own characteristics ξ, the radar's behavioral attitude intensity towards the behavior σ′ is recorded as A(σ′)=fA (σ′,ξ).

[0204] S132、Specific meaning of radar program behavior code is as follows Figure 8 As shown, based on the radar plan behavior specification analysis module, the radar plan behavior specification analysis is performed to obtain the behavior specification strength;

[0205] Radar plan behavior specification analysis module such as Fig. 9 As shown, firstly, the historical and current radar behaviors, historical and current working environments ε SN Input the radar plan behavior specification analysis module, that is, input it into the multi-function radar radiation source mechanism knowledge base for matching, and obtain the associated mechanism knowledge ψ'.

[0206] The analysis of behavioral norms is divided into two levels: pulse level and feature level, and tactical level, as follows:

[0207] (1) Specification of planned behaviors at the pulse level and feature level (calculation of parameter difference degree);

[0208] First calculate the degree of difference As the basis for the planned behavior specification strength SN(σ′) of the multifunction radar for the pulse-level or characteristic-level behavior σ′, where the radar behavior σ′ corresponds to the parameter Θ σ′ , is the radar in the current scene ε SN Parameters required for the behavior state s(t).

[0209] Radar behavior parameter Θ σ′ Required parameters of radar The degree of difference between them is calculated by Euclidean distance, which is expressed as follows:

[0210]

[0211] According to the degree of difference between the parameters, the planned behavior specification strength SN(σ′) of the multi-function radar for pulse-level or characteristic-level behavior is expressed as The expression is as follows:

[0212]

[0213] Among them, k s represents the coefficient. Since the behavior parameter Θ σ′ With the required parameters It may obey a certain probability distribution, and the degree of difference between the two can be represented by a more complex mathematical model; It can also be built through machine learning models such as neural networks.

[0214] (2) Tactical-level planning behavior norms (tactical-level behavior transfer rules);

[0215] Assume that the tactical level behavior of multi-function radar in the radar mechanism knowledge base has a switching relationship ψ trans , the inversion result of the tactical level behavior at the current moment is μ t , the possible tactical level behavior at the next moment is The current working environment of the radar is ε SN .

[0216] The tactical level behavior inversion result μ t Radar working environment SN Input into the radar knowledge base for matching, and according to the transfer rules of tactical level behavior, obtain the associated mechanism knowledge, that is, ψ' = ψ trans . Tactical level behavior switching relationship Degree of compliance with radar code of conduct Expressed as The expression is as follows:

[0217]

[0218] in, It can also be constructed through machine learning models such as support vector machines and neural networks for complex representation.

[0219] After the multi-function radar behavior norm analysis, based on the associated mechanism knowledge ψ' in the multi-function radar radiation source mechanism knowledge base, the radar's subjective norm intensity SN(σ') for behavior σ' is recorded as f sN (σ′|ψ′).

[0220] S133、The specific meaning of radar plan behavior constraint control is as follows Fig.10 As shown, radar plan behavior constraint control analysis is performed based on the radar plan behavior constraint control analysis module to obtain the behavior constraint control strength;

[0221] Radar plan behavior constraint control analysis module such as Fig.11 As shown in the figure, the radar planning behavior constraint control input includes: multi-function radar system resource constraint limit ζ, multi-function radar working environment characteristics ε PBC And the radar history and the current radar behavior σ 1:T .

[0222] Among them, the resource constraints of the multi-function radar system include: time resource constraints, energy resource constraints, computer resource constraints; multi-function radar working environment characteristics PBC Including: characteristics of our other equipment, tactical characteristics of the opponent's targets, and movement characteristics of the opponent's targets.

[0223] According to the input, the planned behavior constraint control strength PBC(σ′) of the multifunctional radar for possible behavior σ′ is expressed as fPBc (σ′|ζ,ε PBC ,σ 1:T ), the specific expression is as follows:

[0224]

[0225] Among them, f ζ (σ′) represents the radar plan behavior constraint control generated by internal factors, represents the radar planned behavior constraint control generated by external factors. g(.) represents the planned behavior constraint control calculation function, and the input is f ζ (σ′), The final calculation result expression of the multi-function radar planned behavior constraint control strength is as follows:

[0226]

[0227] Among them, w ζ and Represent weights respectively.

[0228] S134, calculating the utility function according to the improved formula for maximizing expected utility and constructing a radar plan behavior intention analysis module, and inputting the behavior attitude strength, behavior norm strength, and behavior constraint control strength obtained in steps S131-S133 into the radar plan behavior intention analysis module;

[0229] Radar plan behavior intention analysis module such as Fig.12 As shown, according to the improved expected utility maximization formula, the mathematical representation of radar planning behavior is constructed, and the expected result of the improved expected utility maximization behavior is σ * The expression is as follows:

[0230]

[0231] Where X represents a finite set of all possible outcomes of the behavior, i.e., X = x1, x2, ..., x N , N represents the number of behaviors; U(·) represents the utility function defined on the set ∑ of possible behaviors of an individual, and the input P σ represents the probability distribution of the impact of the behavior, u σ (x i ) indicates that the execution behavior σ causes the impact x i The utility generated when σ (x i ) indicates that when action σ is executed, the result is x i probability.

[0232] u σ (·) The specific expression is as follows:

[0233] u σ (x i )=f(A(σ),SN(σ),PBC(σ),x i ) (9)

[0234] Where f(·) represents the functional expression, A(σ), SN(σ), and PBC(σ) represent the behavioral attitude, behavioral norms, and constraint control of the individual when performing behavior σ, respectively.

[0235] According to formula (9), the utility function expression is as follows:

[0236] U(P σ )=w A f A (σ)+w SN f SN (σ)+w PBC f PBC (σ)(10)

[0237] Among them, w A 、w SN 、w PBC They represent the weights of behavioral attitude, behavioral norms, and behavioral constraints. It should be noted that the behavioral utility U(P σ ) and behavioral attitudes, behavioral norms, and behavioral constraint controls can be represented by a more complex model.

[0238] Individuals perform actions i The subjective probability The judgment expression is as follows:

[0239]

[0240] Among them, λ∈0,+∞) represents the rationality parameter. When λ=0, individuals ignore the expected utility of different strategies and have equal probability of executing each behavior; when λ→+∞, the probability of individuals executing the behavior that maximizes the expected utility is 1, and the probability of executing other behaviors is 0.

[0241] Using U(P σ ) and formula (11) to calculate the subjective judgment of the probability of executing different behaviors at the next moment, that is, the radar plan behavior intention Individual behavior σ i The intensity of behavioral intention Since the final behavior of an individual is determined by the behavioral intention and behavioral constraint control, the final behavior prediction result B of the individual is expressed as follows:

[0242] B=Rul(BI,PC) (12)

[0243] Where BI=[B1,B2,…,B N ] T represents the intensity of an individual’s behavioral intention to perform each behavior, [·] T represents the transposition operation; PC = [PBC(σ1), PBC(σ2),…, PBC(σ N )] represents the behavioral constraint control of each behavior; Rule(·) represents the mapping relationship between behavioral intention, perceived behavioral control and individual final behavior, which is achieved by training machine learning models, including support vector machines and neural networks.

[0244] In this embodiment, in step S3, the multifunctional radar planned behavior prediction model performs model training to verify the prediction accuracy, as follows:

[0245] A1. Establish a multi-function radar simulation scenario, obtain multi-function radar parameters and behavior data, build a sample data set and perform preprocessing, and build feature samples;

[0246] First, a multi-function radar simulation scenario is established, and n sets of pulse data are generated according to the radar tactical level behaviors executed in various scenarios. Among them, the pulse characteristic parameters are set including: pulse carrier frequency RF and its modulation type, pulse width PW and its modulation type, pulse repetition interval PRI and its modulation type. The parameter settings required for the multi-function radar pulse level and tactical level are shown in Table 1. The specific values ​​of the parameters need to be determined according to the actual data set.

[0247] Table 1

[0248]

[0249] In addition to the pulse characteristic parameters, the tactical level behavior of the multi-function radar can also be set according to the desired behavior intentions of the multi-function radar at different levels, as shown in Table 2, including radar behavior status, radar working mode, and radar function.

[0250] Table 2

[0251]

[0252] Then, according to the behavioral intentions at different levels, feature samples are constructed through preprocessing.

[0253] The feature sample is defined as D = (X, y), where X and y represent corresponding sequences and labels that meet different requirements.

[0254] For pulse-level behavior, X is l represents the sequence length, y represents the predicted parameter true value label, and x pulse Indicates the corresponding parameter type, including: PRI, RF, PW;

[0255] For feature-level behavior, X is l represents the sequence length, y represents the predicted corresponding feature-level behavior label, and x pulse Indicates the corresponding parameter type, including: PRI, RF, PW;

[0256] For tactical-level behavior, X is ple_n t Indicates the number of corresponding parameter types, including: PRI, RF, PW, and y indicates the desired tactical-level behavior label.

[0257] Finally, for pulse-level behavior, feature-level behavior, and tactical-level behavior, based on the constructed feature samples, different error levels k Err_leel , a total of data_num groups of multi-function radar data sets data_set are generated.

[0258] Among them, the sequence length l is data_length; k Err_leel , data_num and data_length are generated according to actual conditions.

[0259] In this embodiment, multifunctional radar PRI sequence prediction, PRI modulation type prediction and working mode prediction are performed, which respectively represent the prediction contents at the pulse level, feature level and tactical level.

[0260] In this embodiment, the multi-function radar PRI sequence prediction and PRI modulation type prediction are performed, and the multi-function radar PRI parameters and modulation type settings are specifically shown in Table 3. The error level k Err_leel =[0,0.05,0.1,0.15,0.2], a total of data_num_PRI=5000 sets of multi-function radar PRI feature sample data were generated, among which The sequence length data_length_PRI of each set of sample data is 300.

[0261] Table 3

[0262]

[0263] In this embodiment, the multi-function radar working mode prediction is performed, and the multi-function radar working mode parameter settings are specifically shown in Table 4, and a total of data_num_WorkMode=3600 sets of multi-function radar working mode sample data are generated, where X=[P RF ,P PI ,P P ], the sequence length of each set of sample data is data_length_WorkMode=300.

[0264] Table 4

[0265] Working Mode RF / GHz PRI / us PW / us VS 8-12 3.3-10 1-3 RWS 8-12 6-10 1-3 TAS 8-12 30-200 0.1-25 STT 8-12 10-250 0.1-20

[0266] A2, based on the feature samples obtained in step A1, input the multi-function radar plan behavior prediction model for training to verify the prediction accuracy;

[0267] A21. Divide all sample data into training sets and test sets according to actual conditions. Send the training set to the GRU neural network for training. Dynamically adjust the learning rate and batch size according to the accuracy of the training set until the model loss value reaches the requirement, completing the behavior prediction model training.

[0268] In this embodiment, the sample data is divided into 8:2 ratios, and the loss function based on the theory of planned behavior is constructed as follows: Fig.13 As shown in the figure, since there are three analysis modules: behavior attitude, behavior norms, and behavior constraint control, the loss function expression of the behavior prediction model is as follows:

[0269] l ss=l ss d +l ss s +l ss p (13)

[0270] in, represents the loss function of the behavior attitude analysis module, Represent the actual value and the predicted value vector of the model output, respectively. Represents the dynamic error weight. The larger the error, the lower the weight. represents the loss function of the behavior specification analysis module, Δy n Represents y p The difference operation of represents the square of the vector 2-norm, N data represents the number of groups of sequences in the dataset, α s and λ s represents a constant factor, and L1 regularization is introduced later to control the excessive growth of differences; represents the loss function of the behavior constraint control analysis module, represents the prediction parameter set, represents the target parameter set, β and η represent constant factors. The log term is used to reduce the impact of outliers, thereby reducing sensitivity to extreme values.

[0271] A22, sending the test set data generated in step A21 to the trained model to determine the accuracy;

[0272] Prediction accuracy

[0273] Among them, Ntest represents the number of samples in the test set, represents the intention value of the network output at the tth moment, y t Represents the true value of the network output at the tth moment, T test_o t represents the total number of network outputs; δ(·) represents the impulse function.

[0274] Represents a binary judgment impulse function for the intended value and the real value, which is used to distinguish right from wrong; for impulse-level behavior:

[0275]

[0276] Among them, threshold pulse Indicates the correct range threshold of the pulse parameter, and the condition is set according to the actual situation; in this embodiment, threshold pulse =0.05. The result is 1, which proves that the intention value and the true value match and are correct; when The result is 0, which proves that the intended value and the actual value do not match and are wrong.

[0277] For feature-level and tactical-level behaviors:

[0278]

[0279] when The result is 1, which proves that the intention value and the true value match and are correct; when The result is 0, which proves that the intended value and the actual value do not match and are wrong.

[0280] After the prediction accuracy of the behavior prediction model passes the test, ac p Greater than the set correct threshold T acc When , the model is verified to be valid. In the test of this embodiment, the threshold T acc Take 85%. Then derive the model data and use it in the multifunctional radar planning behavior prediction system to predict radar behavior.

[0281] In this embodiment, step S3 is specifically as follows:

[0282] The data loading module uses a program to read radar behavior data and inputs the data initialization module to initialize the data.

[0283] The data initialization includes: dividing the data into short and long lengths for subsequent identification and reading. That is, in this embodiment, zero padding is performed for pulse sequences whose data length is less than data_length_PRI or data_length_WorkMode; and the pulse sequences whose data length is greater than data_length_PRI or data_length_WorkMode are divided into data_length_PRI or data_length_WorkMode for subsequent import into the model for prediction.

[0284] Finally, the read and initialized data is imported into the multi-function radar behavior prediction module and the prediction results are output.

[0285] like Fig.14 As shown, the pulse level behavior prediction under different PRI modulation types in this embodiment includes: constant PRI, group PRI, sliding PRI, and staggered PRI. Fig.14 The middle blue area indicates the interval where the PRI predicted values ​​are judged to be correct. It can be seen that when the measured values ​​do not fluctuate much, the predicted values ​​are all in a correct interval.

[0286] This embodiment performs PRI sequence prediction for a multifunctional radar. The PRI sequence prediction accuracy obtained under different PRI modulation types is as follows: Fig.15 As shown in the figure, the accuracy of pulse-level behavior of different modulation types at different error levels. It can be seen from the figure that in the test, the PRI value prediction accuracy of the three modulation types of constant, group variable, and sliding variable are all correctly predicted at different error levels. The prediction accuracy of the staggered modulation type decreases significantly as the error level increases, but it is still at a relatively high level. According to the test results, it can be determined that the multifunctional radar behavior prediction model described in this embodiment is effective and feasible for the prediction of the multifunctional radar PRI sequence, which verifies the correctness and effectiveness of the present invention for the prediction of pulse-level behavior under different circumstances.

[0287] This embodiment further verifies the prediction accuracy of the multifunctional radar behavior prediction model for feature-level behavior and tactical-level behavior. The verification results are as follows: Fig.16 , 17 As shown. It can be seen that in the face of PRI modulation type recognition in feature-level behavior and working mode recognition in tactical-level behavior, the method of the present invention shows good performance, with an accuracy rate greater than 85%, which verifies the correctness and effectiveness of the method of the present invention for predicting feature-level and tactical-level behaviors in different situations.

[0288] In summary, the planned behavior prediction theoretical model proposed by the method of the present invention has the ability to be extended to other levels of radar behavior prediction, and can be used to implement a multi-function radar behavior prediction algorithm based on the planned behavior theory through deep learning through other deep learning network models and loss functions. The method of the present invention uses the planned behavior theory to mathematically model the multi-function radar behavior prediction, provides rule guidance for the multi-function radar behavior prediction, provides a specific method for high-accuracy multi-function radar behavior prediction, and verifies the feasibility and effectiveness through the application of multi-function radar pulse repetition interval prediction, solving the problem of lack of basic support theory for behavior prediction and the lack of a behavior prediction method system.

[0289] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A multi-function radar behavior prediction modeling method based on the theory of planned behavior, the specific steps are as follows: S1. Construct a multifunctional radar planned behavior prediction model based on the planned behavior theory; S2. Based on step S1, a multifunctional radar planned behavior prediction system based on the planned behavior theory is constructed on a high-performance platform; The multifunctional radar planning behavior prediction system comprises: Data loading module, data initialization module, multi-function radar behavior prediction module, result output module; The multifunctional radar behavior prediction module is the multifunctional radar planned behavior prediction model constructed in step S1; the multifunctional radar behavior prediction module includes: pulse-level behavior intention prediction, feature-level behavior intention prediction, and tactical-level behavior intention prediction; S3, the data loading module loads the radar behavior data into the data initialization module, and after the data initialization, inputs it into the multi-function radar behavior prediction module to perform radar behavior prediction, and outputs the result through the result output module; Among them, before the multifunctional radar planned behavior prediction model performs radar behavior prediction, model training is first performed to verify the prediction accuracy. After the accuracy is qualified, it is used in the multifunctional radar planned behavior prediction system to perform radar behavior prediction.

2. A multi-function radar behavior prediction modeling method based on the theory of planned behavior according to claim 1, It is characterized in that The step S1 is specifically as follows: S11. Construct the mapping relationship between the theory of human planned behavior and the theory of multifunctional radar planned behavior; (1) Data collection; In the theory of human planned behavior, the collected data is used to analyze the components of an individual's behavioral attitude, subjective norms, and intuitive behavioral control towards a specific behavior, and calculate the corresponding strength. The collection methods include: questionnaires and interviews; the collected data include: individual characteristics, group characteristics, and external environment characteristics; Correspondingly, in the multifunctional radar behavior prediction, the data collection method in the human planned behavior theory is extended to the data collection tool by analogy, and the data collection results in the human planned behavior theory are extended to the multifunctional radar behavior prediction by analogy, wherein, in the multifunctional radar behavior prediction, the data collection tool includes: the detection equipment in the radar passive detection; the data results collected by the multifunctional radar behavior prediction include: the multifunctional radar signal parameters intercepted by the detection equipment, and the analysis results of the radar passive detection system on the multifunctional radar working environment information; (2) behavioral attitude; In the theory of human planned behavior, behavioral attitude reflects the influence of an individual's own characteristics on his or her behavior, that is, the individual's evaluation of the degree to which he or she likes or dislikes performing a specific behavior, wherein the individual's own characteristics include: his or her own values, age; Correspondingly, in the multi-function radar behavior prediction, the radar behavior attitude reflects the influence of the radar's own characteristics on the radar behavior, wherein the radar's own characteristics include: multi-function radar behavior state transition probability characteristics, waveform parameter characteristics, antenna scanning mode characteristics; the radar behavior includes: radar behavior state, working mode, signal parameters; according to the multi-function radar's own characteristics ξ, the radar's behavior attitude intensity A(σ′) to the behavior σ′ is recorded as f A (σ′,ξ); (3) Subjective norms; In the theory of human planned behavior, subjective norms reflect the influence of the social group to which an individual belongs on the individual's behavior, that is, the external pressures and expectations perceived by the individual when deciding whether to perform a particular behavior; In the prediction of multifunctional radar behavior, the influencing factors of radar subjective norms are the radar norms of the multifunctional radar radiation source, including: multifunctional radar behavior state, working mode transfer rules, and multifunctional radar waveform beam selection rules, which can be compared to the external expectations perceived by individuals when performing specific behaviors in the theory of human planned behavior. That is, the radar norms in the multifunctional radar radiation source mechanism knowledge base can be regarded as the expectations of experts on the behavior that the radar should perform; Assume that there is a radar rule ψ in the knowledge base of the multifunctional radar radiation source mechanism, and the subjective norm intensity SN(σ′) of the radar on the behavior σ′ is recorded as f SN (σ′|ψ); (4) perceived behavioral control; In the theory of human planned behavior, perceived behavioral control reflects the ability and experience of an individual, as well as the influence of the external environment on the individual's behavior, that is, the degree to which it is easy or difficult for an individual to perform a particular behavior in a certain way; Among them, the perceived behavioral control in the theory of human planned behavior includes: internal factors and external factors, namely, one's own ability and external environment; In the multi-function radar behavior prediction, radar perceived behavior control refers to the difficulty faced by the radar when executing a specific behavior, which also includes internal and external factors, and is also called radar planned behavior constraint control; Among them, the internal factors of radar perception behavior control refer to the internal cost criteria and constraints that the multifunctional radar must follow when working, including: the time resource constraints faced by the radar's internal resources and task scheduling, which can be compared to the individual capabilities in the theory of human planned behavior; the external factors refer to the working environment of the multifunctional radar. PBC The influence of the characteristics of other equipment of the enemy and the enemy on the planned behavior of the multifunctional radar can be compared to the external resource conditions in the theory of human planned behavior; According to the radar history and current behavior σ 1:T , the radar's internal resource constraintsζ and the radar's working environmentε PBC The perceived behavioral control intensity PBC(ε′) of the multifunctional radar on the behavior σ′ is denoted as f PBC (σ 1:T ,σ′|ζ,ε PBc ); (5) behavioral intention; In human behavior prediction, behavioral intention refers to the possibility of an individual to perform a specific behavior, that is, the individual's judgment on the subjective probability of taking a specific behavior; In the multi-function radar behavior prediction, radar behavior intention refers to the future behavior that the multi-function radar intends to take according to its own task scheduling strategy, waveform parameter characteristics, beam scanning characteristics, and the multi-function radar behavior rules and the radar's own time and energy resource constraints, under the influence of its own and the opponent's equipment and target status; (6) Actual behavior; In human behavior prediction, actual behavior refers to the actual behavior that individuals ultimately perform under the dual guidance of behavioral intention and perceived behavioral control; In the multi-function radar behavior prediction, the actual behavior of the radar refers to the behavior sequence actually executed by the multi-function radar in each scheduling interval, that is, the actual behavior finally executed by the radar under the dual guidance of behavioral intention and perceived behavioral control; The difference between radar actual behavior and radar behavior intention is that radar behavior intention refers to the sequence of behaviors that the radar selects for the next scheduling interval after each scheduling interval ends, and the event has not yet occurred; while radar actual behavior refers to the behavior that the radar has executed and occurred in the current scheduling interval; Among them, if the multifunctional radar does not change after selecting the next scheduling interval to perform the task, it is considered that the radar behavior intention can reflect the actual behavior that the radar will take place in the future; S12, based on the mapping relationship in step S11, constructing a theoretical structural model of multi-function radar plan behavior prediction; The multifunctional radar plan behavior prediction theoretical structure model includes: a radar plan behavior attitude analysis module, a radar plan behavior specification analysis module, a radar plan behavior constraint control analysis module, and a radar plan behavior intention analysis module; Among them, the analysis of the multi-function radar behavior intensity is divided into three levels, including: pulse level, feature level, and tactical level; pulse level behavior is the basic expression, including: carrier frequency, pulse repetition rate, and pulse width; feature level behavior is the characteristic expression of pulse level behavior, including: carrier frequency modulation, repetition rate modulation, and pulse width modulation; tactical level behavior is the top-level expression, including: behavior status, working mode, and radar function; S13, based on step S12, according to the improved expected utility maximization formula, mathematical modeling is performed on the multifunctional radar plan behavior prediction theoretical structure model and its various analysis modules to obtain the multifunctional radar plan behavior prediction model; S131, performing radar plan behavior attitude analysis based on the radar plan behavior attitude analysis module to obtain the behavior attitude strength; Firstly, the historical and current radar behaviors are input into the radar plan behavior attitude analysis module to analyze the individual characteristics of the multi-function radar; The individual characteristic analysis of the multifunctional radar includes: pulse level behavior analysis, feature level behavior analysis, and tactical level behavior analysis, as follows: (1) Pulse-level behavior analysis; The pulse level behavior intensity analysis is the characteristic parameters of any pulse level, including: carrier frequency RF, pulse repetition interval PRI, pulse width PW; Set the multi-function radar pulse level behavior parameter to be predicted as x pulse , the multi-function radar's planned behavior attitude strength for the L+1th pulse parameter The expression is as follows: in, represents the L+1th pulse parameter, represents the predicted value of the L+1th pulse parameter, and α represents the coefficient; Formula (1) maps the radar's planned behavior attitude intensity for the L+1th pulse parameter to between 0 and 1. near When it approaches 1, and When the deviation is large, it approaches 0; (2) Feature-level behavior analysis; The characteristic level behavior intensity analysis is the characteristic category corresponding to the pulse level parameters, including: radar PRI modulation type, RF modulation type, PW modulation type; Set the characteristic level behavior parameter of the multi-function radar to be predicted as x feature , the multi-function radar selects the feature-level behavior x for the next moment feature The planned behavior attitude strength A(x feature ) is expressed as follows: Among them, x feature represents a specific PRI modulation type, M represents all possible corresponding feature-level behavior types stored in the multifunctional radar waveform library, which is obtained by behavioral recognition of radar waveforms; Formula (2) means that the behavioral attitude intensity of the multifunctional radar to the modulation type existing in its waveform library is 1, and the behavioral attitude intensity to other modulation types is 0; (3) Tactical-level behavioral analysis; Tactical level behavior intensity analysis is the characteristics of any tactical level, including: radar behavior status, radar working mode, radar function; Set the tactical level behavior parameter of the multi-function radar to be predicted as x tactical , the tactical behavior of the radar has Markov properties, and the multi-function radar behavior state transition probability matrix is ​​obtained by calculation Among them, p ij =p(x tactical+1 =s j |x tactical =s i ) is a multi-function radar with tactical level behavior i Tactical level behavior j The probability of switching; According to the calculated multi-function radar tactical level behavior state transition probability and behavior state at time t s(t) = s i , the radar executes event s(t+1)=s at time t+1 j The intensity of behavioral attitude A(s j )=p ij ; After the individual characteristics of the multifunctional radar are analyzed, according to the multifunctional radar's own characteristics ξ, the radar's behavioral attitude intensity towards the behavior σ′ is recorded as A(σ′)=f A (σ′,ξ); S132, performing radar plan behavior specification analysis based on the radar plan behavior specification analysis module to obtain the behavior specification strength; First, the historical and current radar behaviors, historical and current working environments ε SN Input the radar plan behavior specification analysis module, that is, input it into the multi-function radar radiation source mechanism knowledge base for matching, and obtain the associated mechanism knowledge ψ'; The analysis of behavioral norms is divided into two levels: pulse level and feature level, and tactical level, as follows: (1) Pulse-level and feature-level planning behavior specifications; First calculate the degree of difference As the basis for the planned behavior specification strength SN(σ′) of the multifunction radar for the pulse-level or characteristic-level behavior σ′, where the radar behavior σ′ corresponds to the parameter Θ σ′ , is the radar in the current scene ε SN Required parameters under behavioral state s(t); Radar behavior parameter Θ σ′ Required parameters of radar The degree of difference between them is calculated by Euclidean distance, which is expressed as follows: According to the degree of difference between the parameters, the planned behavior specification strength SN(σ′) of the multi-function radar for pulse-level or characteristic-level behavior is expressed as The expression is as follows: Among them, k s represents the coefficient; (2) Tactical-level planning behavior norms; Assume that the tactical level behavior of multi-function radar in the radar mechanism knowledge base has a switching relationship ψ trans , the inversion result of the tactical level behavior at the current moment is μ t , the possible tactical level behavior at the next moment is The current working environment of the radar is ε SN ; The tactical level behavior inversion result μ t Radar working environment SN Input into the radar knowledge base for matching, and according to the transfer rules of tactical level behavior, obtain the associated mechanism knowledge, that is, ψ' = ψ trans ; Tactical level behavior switching relationship The degree of compliance with radar behavior standards Expressed as The expression is as follows: After the multi-function radar behavior norm analysis, based on the associated mechanism knowledge ψ' in the multi-function radar radiation source mechanism knowledge base, the radar's subjective norm intensity SN(σ') for behavior σ' is recorded as f SN (σ′|ψ′); S133, performing radar plan behavior constraint control analysis based on the radar plan behavior constraint control analysis module to obtain behavior constraint control strength; Radar plan behavior constraint control input includes: multi-function radar system resource constraint limit ζ, multi-function radar working environment characteristics ε PBC And the radar history and the current radar behavior σ 1:T ; Among them, the resource constraints of the multi-function radar system include: time resource constraints, energy resource constraints, computer resource constraints; multi-function radar working environment characteristics PBC Including: characteristics of our other equipment, tactical characteristics of the opponent's targets, and movement characteristics of the opponent's targets; According to the input, the planned behavior constraint control strength PBC(σ′) of the multifunctional radar for possible behavior σ′ is expressed as f PBC (σ′|ζ,ε PBC ,σ 1:T ), the specific expression is as follows: Among them, f ζ (σ′) represents the radar plan behavior constraint control generated by internal factors, represents the radar planned behavior constraint control generated by external factors; g(.) represents the planned behavior constraint control calculation function, and the input is f ζ (σ′), The final calculation result expression of the multi-function radar planned behavior constraint control strength is as follows: Among them, w ζ and Respectively represent weights; S134, calculating the utility function according to the improved formula for maximizing expected utility and constructing a radar plan behavior intention analysis module, and inputting the behavior attitude strength, behavior norm strength, and behavior constraint control strength obtained in steps S131-S133 into the radar plan behavior intention analysis module; According to the improved expected utility maximization formula, the mathematical representation of radar planning behavior is constructed, and the expected result of the improved expected utility maximization behavior is σ * The expression is as follows: Where X represents a finite set of all possible outcomes of the behavior, i.e. X = x 1 ,x 2 ,...,x N , N represents the number of behaviors; U(·) represents the utility function defined on the set ∑ of possible behaviors of an individual, and the input P σ represents the probability distribution of the impact of the behavior, u σ (x i ) indicates that the execution behavior σ causes the impact x i The utility generated when σ (x i ) indicates that when the action σ is executed, the result is x i The probability of u σ (·) The specific expression is as follows: you σ (x i )=f(A(σ),SN(σ),PBC(σ),x i ) (9) Among them, f(·) represents the function expression, A(σ), SN(σ), and PBC(σ) represent the behavioral attitude, behavioral norms, and constraint control of the individual when performing behavior σ, respectively; According to formula (9), the utility function expression is as follows: U(P σ )=w A f A (s)+w SN f SN (s)+w PBC f PBC (s) (10) Among them, w A 、w SN 、w PBC Respectively represent the weights of behavioral attitude, behavioral norms and behavioral constraint control; Individuals perform actions i The subjective probability The judgment expression is as follows: Among them, λ∈0,+∞) represents the rationality parameter; when λ=0, the individual ignores the expected utility of different strategies and the probability of executing each behavior is equal; when λ→+∞, the probability of the individual executing the behavior that maximizes the expected utility is 1, and the probability of executing other behaviors is 0; Using U(P σ ) and formula (11) to calculate the subjective judgment of the probability of executing different behaviors at the next moment, that is, the radar plan behavior intention Individual behavior σ i The intensity of behavioral intention And the individual's final behavior is determined by the behavioral intention and behavioral constraint control, so the individual's final behavior prediction result B is expressed as follows: B=Rul(BI,PC) (12) Where BI = [B 1 ,B 2 ,...,B N ] T represents the intensity of an individual’s behavioral intention to perform each behavior, [·] T Represents transposition operation; PC = [PBC (σ 1 ),PBC(σ 2 ),…,PBC(σ N )] represents the behavioral constraint control of each behavior; Rule(·) represents the mapping relationship between behavioral intention, perceived behavioral control and individual final behavior, which is achieved by training machine learning models, including support vector machines and neural networks.

3. A multi-function radar behavior prediction modeling method based on the theory of planned behavior according to claim 1, It is characterized in that In step S3, the multifunctional radar planned behavior prediction model performs model training to verify the prediction accuracy, as follows: A1. Establish a multi-function radar simulation scenario, obtain multi-function radar parameters and behavior data, build a sample data set and perform preprocessing, and build feature samples; First, a multi-function radar simulation scenario is established to generate n sets of pulse data according to the radar tactical-level behaviors executed in various scenarios; The pulse characteristic parameters include: pulse carrier frequency RF and its modulation type, pulse width PW and its modulation type, pulse repetition interval PRI and its modulation type; and in addition to the pulse characteristic parameters, according to the desired behavior intentions at different levels, tactical level behavior settings can also be performed, including: radar behavior status, radar working mode, radar function; Then, according to the behavioral intentions at different levels, feature samples are constructed through preprocessing; Define the feature sample as D = (X, y), where X and y represent the corresponding sequence and label that meet different requirements; For pulse-level behavior, X is l represents the sequence length, and y represents the predicted parameter true value label; For feature-level behavior, X is l represents the sequence length, and y represents the predicted corresponding feature-level behavior label; For tactical-level behavior, X is n t represents the number of corresponding parameter types, and y represents the desired tactical-level behavior label; Finally, for pulse-level behavior, feature-level behavior, and tactical-level behavior, based on the constructed feature samples, different error levels k Err_leel , a total of data_num groups of multi-function radar data sets data_set are generated; Among them, the sequence length is data_length; k Err_leel , data_num and data_length are generated according to the actual situation; A2, based on the feature samples obtained in step A1, input the multi-function radar plan behavior prediction model for training to verify the prediction accuracy; A21. Divide all sample data into training sets and test sets according to actual conditions. Send the training set to the GRU neural network for training. Dynamically adjust the learning rate and batch size according to the accuracy of the training set until the model loss value converges, completing the behavior prediction model training. Since there are three analysis modules: behavior attitude, behavior norms, and behavior constraint control, the loss function expression of the behavior prediction model is as follows: l ss=l ss d +l ss s +l ss p (13) in, represents the loss function of the behavior attitude analysis module, y = [y 1 ,…,y i ]、 Represent the actual value and the predicted value vector of the model output, respectively. represents the dynamic error weight; represents the loss function of the behavior specification analysis module, Δy n Represents y p The difference operation of represents the square of the vector 2-norm, N data represents the number of groups of sequences in the dataset, α s and λ s Represents a constant factor, and introduces L1 regularization to control excessive growth of differences; represents the loss function of the behavior constraint control analysis module, represents the prediction parameter set, represents the target parameter set, β and η represent constant factors; A22, sending the test set data generated in step A21 to the trained model to determine the accuracy; Prediction accuracy Among them, N test represents the number of samples in the test set, represents the intention value of the network output at the tth moment, y t Represents the true value of the network output at the tth moment, T test_ot represents the total number of network outputs; δ(·) represents the impulse function; Represents a binary judgment impulse function for the intended value and the real value, which is used to distinguish right from wrong; for impulse-level behavior: Among them, threshold pulse Indicates the correct range threshold of the pulse parameter. The condition is set according to the actual situation. The result is 1, which proves that the intention value and the true value match and are correct; when The result is 0, which proves that the intended value and the true value do not match and are wrong; For feature-level and tactical-level behaviors: when The result is 1, which proves that the intention value and the true value match and are correct; when The result is 0, which proves that the intended value and the true value do not match and are wrong; After the prediction accuracy of the behavior prediction model passes the test, ac p Greater than the correct threshold T acc When the model is verified to be valid, the model data is then exported for use in the multifunctional radar planning behavior prediction system to perform radar behavior prediction; Among them, T acc Set the value according to the actual situation.

4. A multi-function radar behavior prediction modeling method based on the theory of planned behavior according to claim 1, It is characterized in that In step S3, the data loading module uses a program to read the radar behavior data, inputs the data initialization module to initialize the data, and finally imports the read and initialized data into the multi-function radar behavior prediction module to output the prediction result; The data initialization includes: dividing the data into short and long ones; that is, padding the pulse sequence whose data length is less than data_length with zeros; and dividing the pulse sequence whose data length is greater than data_length into short and long ones, and the dividing length is data_length.

Images