Radiation source target threat level evaluation method and device, electronic equipment and storage medium

Abstract

The invention provides a radiation source target threat level evaluation method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining threat element data of a radiation source; processing the threat element data by adopting a pre-constructed threat level evaluation model to obtain a threat level evaluation reasoning result of the radiation source; outputting a threat level evaluation reasoning result of the radiation source; according to the method, model averaging is carried out on a network structure converted from a plurality of searched local optimal sequences based on a Stacking restart strategy, and then constrained adjustment is carried out on a current optimal node sequence through a model averaging result, so that a new node sequence jumps out of local optimum and retains advantages and characteristics of the new node sequence at the same time, the purpose of restart is more effectively achieved, and the method is suitable for large-scale popularization and application. Therefore, the accuracy of the Bayesian network threat level evaluation model is improved.

Description

Technical Field

[0001] The present application belongs to the technical field of radiation source target threat assessment, and in particular, relates to a radiation source target threat level assessment method, device, electronic device and storage medium. Background Art

[0002] In complex regional environments, the electromagnetic spectrum becomes the key link connecting multiple action areas and the basis of information transmission and command and control. Controlling electromagnetic power means being able to control the electromagnetic environment, ensuring that communications, radar and navigation systems are not interfered with, and effectively interfering with or destroying related systems that are unfavorable to oneself. The distributed electromagnetic application mode is an efficient and flexible strategy in the context of complex electromagnetic environments. It disperses electromagnetic countermeasure resources on various nodes on a networked and informationized platform to achieve multi-electromagnetic signal transmission and reception, as well as distributed coordination of electromagnetic interference and anti-interference. Reliable resource allocation strategies can help achieve precise and flexible distributed electromagnetic applications, thus laying a solid foundation for the fight for electromagnetic control. A wide variety of multi-threat-level radars and communication signals are widely present in regional environments. With the rise of unmanned equipment, multiple unmanned platforms carrying electronic countermeasures are also distributed in all corners of the region. In addition, the rapid changes in regional situations also make it impossible for the current designated resource allocation method to respond to emergencies in a timely manner. Therefore, autonomous electromagnetic resource allocation is a necessary measure to cope with complex and changing regional environments. An accurate and effective radiation source target threat level assessment model can help decision makers quickly identify key targets and make reasonable and effective resource allocation decisions, thereby improving the effectiveness of electromagnetic countermeasures. How to accurately and quickly establish an accurate radiation source target threat level assessment model to meet the real-time and accuracy requirements of electromagnetic resource allocation in complex electromagnetic environments is an urgent problem that needs to be solved. Over the years, domestic and foreign scholars have conducted a lot of research and exploration in the field of radiation source target threat level assessment. At present, there are two main types of algorithms in this field:

[0003] (1) Evaluation method based on mathematical analytical model

[0004] The core of the evaluation method based on mathematical analytical model is to establish a corresponding mathematical analytical model to simulate the situation, then construct the relevant factor matrix, and compare the matrix operation results to obtain the threat level of the radiation source target. Typical algorithms include multi-attribute decision theory, hierarchical analysis method, etc. These methods are fast and have simple model structures, but when faced with complex electromagnetic environments, the accuracy of their mathematical analytical models will be difficult to guarantee.

[0005] (2) Evaluation method based on artificial intelligence theory

[0006] The evaluation method based on artificial intelligence theory uses sample data and combines expert knowledge to establish a network model for evaluating the threat level of radiation source targets, and then infers the threat level of each radiation source target based on the network model. Typical algorithms include Bayesian network methods and neural network algorithms. Such algorithms can be used to process nonlinear models and have good modeling accuracy when the sample size is sufficient. However, this makes them have high requirements for sample data, which is often difficult to meet in complex and changeable electromagnetic environments.

[0007] Among many methods, the Bayesian network radiation source target threat level assessment method stands out in the field of radiation source target threat level assessment due to its powerful uncertainty modeling and reasoning capabilities, modeling advantages under small sample data, and full use of prior knowledge. Many experts and scholars have achieved good results in the research of this algorithm. The learning accuracy of the Bayesian network structure largely determines the accuracy of the threat level assessment model, which is crucial to the radiation source target threat level assessment results. However, the traditional Bayesian network structure learning method often cannot guarantee sufficient learning accuracy when dealing with the radiation source target threat level assessment needs in complex and changeable electromagnetic environments. In some application scenarios, the structure of the Bayesian network is directly designed by experts manually. This process is not only time-consuming and labor-intensive, but also easily affected by subjective bias, resulting in limited generalization ability and accuracy of the model. Summary of the invention

[0008] In view of this, the embodiments of the present application propose a radiation source target threat level assessment method, device, electronic device and storage medium, aiming to improve the accuracy of establishing a radiation source target threat level assessment model by improving the learning accuracy of the Bayesian network structure, thereby completing the radiation source target threat level assessment task more efficiently and accurately.

[0009] To achieve the above-mentioned purpose, an embodiment of the present application provides a method for assessing the threat level of a radiation source target, comprising: obtaining the threat factors of the radiation source, and randomly constructing multiple node sequences based on the threat factor construction data of the radiation source, and constructing an integrated group based on the multiple node sequences; searching the neighborhood of the initial node sequence in the integrated group based on a pre-constructed node sequence optimization algorithm to obtain a local optimal integrated group; transforming each of the node sequences in the local optimal integrated group based on a pre-constructed K2 algorithm to obtain multiple local optimal network structures; obtaining each structure matrix according to each local optimal network structure; processing each of the structure matrices based on a pre-constructed model averaging algorithm to obtain an integrated matrix, processing the integrated matrix based on a pre-constructed integrated restart strategy to obtain an optimized node sequence, restarting the search using the optimized node sequence, terminating the search after satisfying the iteration stop condition, obtaining the optimal node sequence, and converting the optimal node sequence into a Bayesian network structure; processing the sample data and the Bayesian network structure based on the maximum likelihood estimation method to obtain the network parameters of the Bayesian network structure; processing the Bayesian network structure and the network parameters based on the clique tree propagation algorithm to obtain a threat level assessment result.

[0010] Optionally, the threat factors include target type, target intention, target mission capability, target speed, target heading angle, target distance, target altitude, target operating mode and threat level index.

[0011] Optionally, the pre-constructed node sequence optimization algorithm searches the neighborhood of the initial node sequence in the integrated population to obtain a local optimal integrated population, including: updating the node sequence in the integrated population based on a preset first formula; starting the search with a random node sequence based on a neighborhood search algorithm, and updating each node sequence in the integrated population during the search process to obtain the current optimal integrated population; the preset first formula is:

[0012]

[0013] Among them, T represents the node order space, D represents the sample data, Yes The optimized node order, is the node order obtained by the λth search when optimizing the i-th node order, represents the scoring function, Pa represents the parent node, X k represents a node, and n represents the maximum node sequence.

[0014] Optionally, the obtaining of each structural matrix according to each local optimal network structure includes: calculating the parent node set of each node based on a preset second formula, and determining all subgraphs of each node based on each node and the corresponding parent node set; determining the local optimal structure based on all subgraphs of each node; converting the local optimal structure into a matrix form using a preset rule to obtain a structural matrix corresponding to each local optimal structure; the preset second formula is:

[0015]

[0016] The expression of the preset rule is:

[0017]

[0018] Among them, M(p,q)=1 means that in the directed acyclic graph, q To X p The directed edge of X q Yes X p The parent node of X; M(p,q) = 0 means that in the directed acyclic graph, q To X p There is no edge in the direction of X q Not X p The parent node of .

[0019] Optionally, the processing of each of the structure matrices based on the pre-built model averaging algorithm to obtain an integrated matrix includes: processing all of the local optimal structures using a preset aggregate function to obtain an integrated matrix;

[0020] The expression of the aggregate function is:

[0021]

[0022] in, represents the weight of the local optimal structure in the ensemble, W arc Represents the local optimal structure G of each edge of the directed acyclic graph i The weight in ⊙ is the Hadamard product, M i Represents the structure matrix of node i.

[0023] Optionally, the method restarts the search using the optimized node sequence, ends the search after satisfying the iteration stop condition, obtains the optimal node sequence, and converts the optimal node sequence into a Bayesian network structure, including: starting to execute the ENINOBS algorithm based on a preset initial condition, determining a set of disturbance variables based on a disturbance factor, wherein the preset initial condition includes a maximum number of restarts, an integrated disturbance factor, an influence factor, and an integrated population size; determining a parent node set of the disturbance variable set in the integrated matrix based on each of the disturbance variables and the influence factor; randomly extracting an initial node sequence and searching the neighborhood of the initial node sequence based on the INOBS algorithm; after a round of search, comparing each disturbance variable in the disturbance variable set with all the disturbance variables ranked above it in the current node sequence. Any node after the parent node is swapped, and the search is restarted after a new node sequence is obtained; during the search process, if the node sequence score obtained by a search is less than the score of the current node sequence, the current node sequence is kept unchanged; if the node sequence score obtained by a search is greater than or equal to the score of the current node sequence, the current node sequence is replaced; if the node sequence score obtained during the search is greater than the score of the lowest-scoring node sequence in the integrated group, it is replaced; if the size of the integrated group is smaller than the preset integrated group size, the node sequence is added to the integrated group; execute the above steps until the number of restarts is greater than the maximum number of restarts, and convert the optimal node sequence finally obtained into a network structure output based on the K2 algorithm, otherwise continue to execute the restart algorithm until the number of restarts is greater than the maximum number of restarts.

[0024] To achieve the above-mentioned purpose, an embodiment of the present application further provides a device for assessing the threat level of a radiation source target, comprising: an acquisition module, used to acquire the threat factors of the radiation source, and based on the threat factor construction data of the radiation source, randomly construct a plurality of node sequences, and construct an integrated group based on the plurality of the node sequences; a group construction module, used to search the neighborhood of the initial node sequence in the integrated group based on a pre-constructed node sequence optimization algorithm to obtain a local optimal integrated group; a structure construction module, used to transform each of the node sequences in the local optimal integrated group based on a pre-constructed K2 algorithm to obtain a plurality of local optimal network structures; a matrix construction module, used to obtain the local optimal network structures corresponding to each of the local optimal network structures; to each structure matrix; a network structure generation module, used to process each of the structure matrices based on a pre-built model averaging algorithm to obtain an integrated matrix, process the integrated matrix based on a pre-built integrated restart strategy to obtain an optimized node sequence, restart the search using the optimized node sequence, end the search after the iteration stop condition is met, obtain the optimal node sequence, and convert the optimal node sequence into a Bayesian network structure; a parameter estimation module, used to process sample data and the Bayesian network structure based on a maximum likelihood estimation method to obtain network parameters of the Bayesian network structure; a threat inference module, used to obtain a threat level assessment result by inferring the Bayesian network structure and the network parameters based on a clique tree propagation algorithm.

[0025] To achieve the above-mentioned purpose, an embodiment of the present application also provides a server, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned radiation source target threat level assessment method.

[0026] To achieve the above objectives, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the above-mentioned radiation source target threat level assessment method.

[0027] The embodiments of the present application propose a method, device, electronic device and storage medium for assessing the threat level of a radiation source target. The method obtains the threat factors of the radiation source, and constructs data based on the threat factors of the radiation source to randomly construct multiple node sequences, and constructs an integrated group based on the multiple node sequences; searches the neighborhood of the initial node sequence in the integrated group based on a pre-constructed node sequence optimization algorithm to obtain a local optimal integrated group; transforms each node sequence in the local optimal integrated group based on a pre-constructed K2 algorithm to obtain multiple local optimal network structures; obtains each structure matrix according to each local optimal network structure; processes each structure matrix based on a pre-constructed model averaging algorithm to obtain an integrated matrix, processes the integrated matrix based on a pre-constructed integrated restart strategy to obtain an optimized node sequence, restarts the search using the optimized node sequence, iterates, ends the search after the iteration stop condition is met, obtains the optimal node sequence, and converts the optimal node sequence into a Bayesian network structure; processes sample data and the Bayesian network structure based on the maximum likelihood estimation method to obtain network parameters of the Bayesian network structure, and obtains the threat level assessment result through the Bayesian network structure and network parameter reasoning based on the clique tree propagation algorithm. The invention uses a Stacking-based restart strategy to make constrained adjustments to the local optimal sequences in the integrated matrix obtained by multiple searches, so that the new node sequence can retain its advantageous characteristics while jumping out of the local optimum, and more effectively achieve the purpose of restarting, thereby improving the accuracy of the Bayesian network threat level assessment model. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is the process of the radiation source target threat level assessment method provided in one embodiment of the present application. Figure 1 ;

[0029] Figure 2 This is the process of the radiation source target threat level assessment method provided in one embodiment of the present application. Figure 2 ;

[0030] Figure 3 It is an integrated learning structure of a radiation source target threat level assessment method provided in one embodiment of the present application;

[0031] Figure 4 It is a Bayesian network structure diagram learned based on the ENINOBS method of a radiation source target threat level assessment method provided by an embodiment of the present application;

[0032] Figure 5 A threat level assessment flow chart of a radiation source target threat level assessment method provided in one embodiment of the present application;

[0033] Figure 6 It is a structural block diagram of a radiation source target threat level assessment device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0034] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not be construed as any limitation, and the various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0035] In a complex electromagnetic environment, when constructing a radiation source target threat level assessment model through a Bayesian network, it is often difficult to ensure the accuracy and timeliness of the radiation source target threat level assessment results due to insufficient learning accuracy of the network structure. To address this problem, the present invention provides a method for establishing a radiation source target threat level assessment model based on a Bayesian network based on ensemble learning, aiming to improve the learning accuracy of the Bayesian network structure through ensemble learning technology, thereby improving the application of the Bayesian network in the radiation source target threat level assessment and increasing the accuracy and reliability of the assessment.

[0036] At present, local search is used as a common search mode due to its wide application. At the same time, the method of searching for the optimal Bayesian network structure in a node order space with lower complexity also shows its advantages. In the present invention, the optimal Bayesian network structure is continuously approached by local search in the node order space, thereby establishing a core foundation for building an accurate and reliable threat level assessment model.

[0037] When performing a local search for a Bayesian network structure in the node sequence space, the search is first carried out with an initial sequence, and the neighborhood of the initial sequence is explored through the optimization algorithm. Finally, the optimal node sequence obtained by the search is converted into a Bayesian network structure through the K2 algorithm. It can be seen that in the process of local search, the solution obtained by the initial search is often not the global optimal solution, but a local optimal solution limited by the ability of the search algorithm. In order to get closer to the global optimal solution, it is necessary to generate a new node sequence for multiple searches, and gradually approach the global optimal solution by changing the search direction until the global optimal solution is found. When restarting the search, choosing a suitable node sequence is crucial to the search direction, and existing restart strategies mostly use random methods, which may lead to invalid restarts and poor learning effects. Reference Figure 3As a classic strategy in ensemble learning, Stacking uses the outputs of multiple learners as new features to train the upper learner, thus realizing the hierarchical superposition of learners. That is, the primary learner is first trained with the initial sample data set, and then the secondary learner is trained with the output of the primary learner. The present invention proposes a restart strategy based on Stacking, which uses ensemble learning theory to integrate multiple local search results, and makes constrained adjustments to the current local optimal sequence, so that the new node sequence retains its advantageous features while jumping out of the local optimum, and more effectively achieves the purpose of restarting, thereby improving the accuracy of the Bayesian network threat level assessment model.

[0038] An embodiment of the present application proposes a radiation source target threat level assessment method, which is applied to an electronic device, wherein the electronic device can be a terminal or a server. This embodiment and the following embodiments are all described by taking the server as an example. The implementation details of the radiation source target threat level assessment method proposed in this embodiment are described in detail below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution.

[0039] The specific process of the radiation source target threat level assessment method proposed in this embodiment can be as follows: Figure 1 and Figure 2 As shown, the radiation source target threat level assessment method may include:

[0040] S10, obtaining threat factors of the radiation source, and constructing data based on the threat factors of the radiation source to randomly construct multiple node sequences, and constructing an integrated group based on the multiple node sequences;

[0041] Among them, the node sequence refers to the sequence composed of various threat elements, namely nodes, which has strong restrictions on the creation of the Bayesian network structure: the threat elements include at least one of the target type, target intention, target mission capability, target speed, target heading angle, target distance, target altitude, target working mode and threat level index.

[0042] It should be noted that Bayesian networks usually deal with discrete variables, which requires discretization of data. For qualitative indicators, discretization can be performed directly based on experience; for quantitative indicators, it is crucial to choose an appropriate discretization method, otherwise it may lead to erroneous reasoning. In the present invention, an arithmetic discretization method is used to discretize the data. This is a simple and direct discretization algorithm. Its basic principle is to divide continuous data values ​​into multiple intervals according to certain arithmetic intervals, and then map these intervals to different discrete values. The specific steps are as follows:

[0043] (1) Determine the discretization goal: clarify the purpose of discretization, such as to simplify the model, improve computational efficiency, or to deal with noise and outliers in the data;

[0044] (2) Select an appropriate arithmetic interval: Select an appropriate arithmetic interval based on the characteristics of the data and the discretization goal. The size of the arithmetic interval will directly affect the discretization results and the accuracy of subsequent analysis.

[0045] (3) Divide data intervals: Divide continuous data values ​​into multiple intervals according to the selected arithmetic intervals. The values ​​in each interval are considered similar and mapped to the same discrete value.

[0046] (4) Mapping discrete values: Assign a discrete value to each interval, usually an integer or a letter label. For example, the smallest interval can be mapped to 1, the second smallest to 2, and so on.

[0047] Among them, threat factors can be used as nodes. After numbering each node, a node sequence is obtained. The set formed by these node sequences is called an integrated group.

[0048] The parent of a node always comes from the variable that comes before it in the node order. For a given node order >, the set of possible parents of a node can be defined as:

[0049] U i,＜ ={U:U<X i ,|U|≤k}

[0050] Among them, U<X i Represents X i The set of all possible parents, which contains only those in the order X i The previous node, k is the in-degree. The best parent set of each node is:

[0051]

[0052] Once a node order is determined, the optimal network structure corresponding to the node order can be quickly obtained by the above formula;

[0053] S20, searching the neighborhood of the initial node sequence in the integrated group based on a pre-built node sequence optimization algorithm to obtain a local optimal integrated group;

[0054] In an embodiment of the present application, step S20 may include the following execution process:

[0055] Update the node order in the integrated group based on a preset first formula;

[0056] Based on the neighborhood search algorithm, the search starts with a random node sequence, and the node sequence in the integrated group is updated during the search process to obtain the current optimal integrated group;

[0057] The first preset formula is:

[0058]

[0059] Among them, T represents the node order space, D represents the sample data, Yes The optimized node order, is the node order obtained by the λth search when optimizing the i-th node order, score(·) represents the scoring function, Pa represents the parent node, and X k represents a node, and n represents the maximum node sequence.

[0060] Exemplarily, some node orders are randomly selected to form an ensemble population E:

[0061]

[0062] In the formula, s i represents the initial state of the i-th node sequence in the integrated population (i.e. itself), and m is the size of the integrated population.

[0063] Subsequently, an optimization algorithm is used, that is, a neighborhood search algorithm is used to update the node order in the ensemble:

[0064] The first preset formula is:

[0065]

[0066] Among them, T represents the node order space, D represents the sample data, It is right in E The optimized node order, is the node order obtained by the λth search when optimizing the order of the i-th node in E, that is, the search node order, represents a scoring function, which is used to measure the matching degree between a certain network structure and data. In this invention, BIC scoring is used, Pa represents the parent node, X k represents a node, and n represents the maximum node sequence.

[0067] explore Commonly used neighborhood search algorithms include OBS method, INOBS method, etc.:

[0068]

[0069] In the formula, α is the position parameter of the node to be inserted (INOBS method will be introduced later). For simplicity, f is selected INOBS (·) Perform subsequent derivation to obtain the updated integrated population E:

[0070]

[0071] S30, based on the pre-built K2 algorithm, transform the node order in the local optimal integrated group to obtain multiple local optimal network structures

[0072] Specifically, in this step, the node order is converted to a DAG form, i.e., a directed acyclic graph form, for more efficient model averaging. A common conversion method is the K2 algorithm, which starts with an undirected graph containing all nodes but no edges, then checks the variables one by one according to the node order, selects the best parent set for each variable, and then adds the corresponding directed edges to the undirected graph. Specifically, for a given node order <, the algorithm determines the parent set for each variable in the sequence. For example, for variable X i , it will be located in X i Filter out X from the previous variables i The set of parents that maximizes the score is then added to the structure graph as directed edges.

[0073] For example, after the conversion is completed, the integrated population E will be updated to:

[0074]

[0075] Where G i represent The local optimal structure of Represents the node Xk and its parent node set The sub-graph The preset second formula may be:

[0076]

[0077] S40, obtaining each structural matrix according to each of the local optimal network structures;

[0078] In order to evaluate the credibility of the parent-child relationship between nodes, the structural matrix M is introduced i ∈{0,1} n×n As the structure diagram G i The following rules need to be followed when converting DAG into a matrix. The preset rule expression is:

[0079]

[0080] Among them, M(p,q)=1 means that in the directed acyclic graph, q To X p The directed edge of X q Yes X p The parent node of X; M(p,q) = 0 means that in the directed acyclic graph, q To X p There is no edge in the direction of X q Not Xp The parent node of .

[0081] S50, processing each structural matrix based on a pre-constructed model average algorithm to obtain an integrated matrix, processing the integrated matrix based on a pre-constructed integrated restart strategy to obtain an optimized node sequence, restarting the search using the optimized node sequence, iterating in this way, ending the search after satisfying the iteration stop condition, obtaining an optimal node sequence, and converting the optimal node sequence into a Bayesian network structure;

[0082] In an embodiment of the present application, step S30 may include the following execution steps:

[0083] Starting to execute the ENINOBS algorithm based on the preset initial conditions, and determining the set of disturbance variables based on the disturbance factor, wherein the preset initial conditions include the maximum number of restarts, the integrated disturbance factor, the influence factor, and the integrated population size;

[0084] Determine the parent node set of the disturbance variable set in the integrated matrix based on each disturbance variable and the influencing factor;

[0085] In the current node sequence, each disturbance variable in the disturbance variable set is exchanged with any node that is arranged after all its parent nodes;

[0086] An initial node sequence is randomly selected and the neighborhood of the initial node sequence is searched based on the INOBS algorithm;

[0087] After a round of search, swap the position of each perturbation variable in the perturbation variable set with any node after all its parent nodes in the current node order, and restart the search after obtaining a new node order;

[0088] During the search process, if the node order score obtained by a search is less than the score of the current node order, the current node order remains unchanged;

[0089] If the score of the node sequence obtained by a search is greater than or equal to the score of the current node sequence, the current node sequence is replaced;

[0090] If the score of the node sequence obtained during the search process is greater than the score of the lowest-scoring node sequence in the integrated group, it will be replaced; if the size of the integrated group is smaller than the preset integrated group size, the node sequence will be added to the integrated group;

[0091] Execute the above steps until the number of restarts is greater than the maximum number of restarts, and convert the final optimal node sequence into a network structure output based on the K2 algorithm, otherwise continue to execute the restart algorithm until the number of restarts is greater than the maximum number of restarts.

[0092] Exemplarily, all local optimal structures are processed using a preset aggregate function to obtain an integrated matrix;

[0093] The expression of the aggregate function is:

[0094]

[0095] Here, Representative structure diagram G i The weight in the ensemble, W arc is the weight of each edge in all structures, and ⊙ is the Hadamard product. The definition of a single structure G i The weight of will increase as its score increases ( Corresponding to the optimal structure, Corresponding to the worst structure). i Represents the structure matrix of node i. arc The definition of indicates that if some edges appear in most structured graphs, their weights will be higher.

[0096] The following is a detailed description of the restart strategy:

[0097] By applying the integration strategy to generate a matrix W(W∈R n×n ), which measures the impact of the parent-child relationship between different nodes on the score. The larger the value of W(i,j), the more likely it is that the node X j Set as Node X i The parent nodes of X are more likely to generate high-scoring structures, and W is referred to as the integration matrix here. In addition, the present invention also considers introducing an impact factor β, which represents the minimum standard for a pair of node relationships to have a positive impact on the score of the entire Bayesian network structure, that is, when W(i,j)>β, it can be considered that node X j As Node X i The parent node of is conducive to obtaining a higher-scoring Bayesian network structure. Next, the integration matrix W and the accuracy threshold β are used to sort the local optimal sequence. c Perform constrained perturbation operations to generate new node sequences. The specific perturbation steps include:

[0098] 1) Randomly select n·p e variables As the disturbance variable, p e It is called the integrated disturbance factor, and then find the rows corresponding to these disturbance variables in the integrated matrix, and select the columns with values ​​greater than β. The index values ​​of these selected columns represent the numbers of certain nodes, which can have a positive impact on the score as the parent nodes of the disturbance variables, so that the parent node set π of the disturbance variables that are helpful to the score can be determined * , that is, the third parent node set:

[0099]

[0100] 2) For each disturbance variable Determine π * In the current node sequence < c The variable X at the last position in the k max , and then the disturbance variable X l and c Any row in X k max The subsequent nodes (i.e., node 1) are swapped to obtain the fourth parent node.

[0101] The restart strategy proposed in this invention is called EN strategy, and EN strategy is combined with the currently popular optimization algorithm INOBS method (the formed algorithm is called ENINOBS method) to learn the Bayesian network structure. INOBS method searches in the node order space by inserting the operator, that is, inserting the i-th node into the j-th position in the node order:

[0102]

[0103] Here i can be less than j. For a given initial sequence, using the insertion operator can produce (n-1) 2 The basic steps of the ENINOBS method are as follows:

[0104] Initialization: Set the maximum number of restarts max , integrated disturbance factor p e , influence factor β, integrated population size γ; randomly extract the initial sequence <0 and set it as the current optimal sequence, that is, use the initial node sequence to replace node 2 in the fourth parent node, that is, < * =<0;

[0105] Search: Start the search using the INOBS method and set the search results to < b and * In comparison, if score(< b )>score(< * ), then update the optimal node sequence < * = b In addition, if there is no individual in the integrated population or the number of individuals is less than γ, then < b Join the integrated group, that is, E=E∪< b ; Otherwise, < b The worst node order in E l In comparison, if score(< b )>score(< l ), then use < b Replacement l location;

[0106] Restart: If the current restart count re>remax , end the search and go to step (4), otherwise, use the EN restart strategy to search for the local optimal sequence < b To disturb. Generate restart sequence< re , with< re Enter step (2) as the initial sequence;

[0107] Conversion: Use K2 algorithm to convert the optimal node sequence < * Convert to network structure and output, the output network structure reference Figure 4 shown.

[0108] S60, processing the sample data and the Bayesian network structure based on the maximum likelihood estimation method to obtain network parameters of the Bayesian network structure;

[0109] S70, processing the Bayesian network structure and the network parameter inference based on the clique tree propagation algorithm to obtain a threat level assessment result.

[0110] In summary, the present invention proposes to apply the ensemble learning theory to the construction of the Bayesian network radiation source target threat level assessment model, abandon the traditional initial node order random selection mode in the Bayesian network structure learning under the node order space, and creatively combine the Stacking idea in the ensemble learning theory to model average the excellent node orders obtained by multiple searches, and constrain the local optimal node order through the result of the model average to obtain a higher quality node order for restarting the search. The newly generated node order not only breaks away from the local optimum, but also retains the favorable characteristics of multiple high-quality node orders, and has more search potential, thereby improving the learning accuracy of the Bayesian network structure to a certain extent, and then making the constructed threat level assessment model more accurate and the assessment result more convincing. In short, the present invention proposes a Bayesian network radiation source target threat level assessment model construction method based on ensemble learning theory, which can accurately establish a target threat level assessment model and establish a core foundation for completing the radiation source target threat level assessment.

[0111] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this application.

[0112] Next, the application of the present invention is explained in more depth through application examples in electronic countermeasure systems, especially the threat level assessment process. In this example, the electronic countermeasure system includes a command and control system and three electronic countermeasure payloads. The command and control system is the center of the entire electronic countermeasure system, responsible for coordinating the three payloads to perform reconnaissance and interference operations. The system workflow is as follows: the payload first receives a reconnaissance command and performs a reconnaissance mission on a designated area. Once the electromagnetic radiation emitted by the target is captured by the payload, the collected radiation source information is then transmitted to the command and control system, which performs in-depth processing of the reconnaissance information, including information fusion, target identification and other tasks. Based on the in-depth analysis of the reconnaissance data, the command and control system autonomously generates corresponding interference strategies, and issues instructions through the commander, so that the payload can carry out precise interference.

[0113] Threat level assessment plays a key role in the information fusion process of the command and control system, and deeply affects the generation of countermeasure strategies. After successfully capturing the electromagnetic radiation of the target, the electronic countermeasure payload first performs preliminary processing to obtain information such as signal frequency, pulse repetition period, bandwidth, amplitude, etc. At the same time, more signal features are extracted through further sorting and identification. In addition, the system will accurately measure the speed, direction and distance of the target. All this information is summarized and transmitted back to the command and control system for more in-depth analysis and processing to explore important factors related to the target threat level assessment:

[0114] Target type: The radiation released by different types of radiation source targets differs in frequency, pulse repetition period, pulse width, modulation type, etc. The target type can be matched and identified in combination with the radiation source target threat library. In this example, the radiation source targets targeted by the electronic countermeasure system are digital and analog radio stations, UAV measurement and control, reconnaissance radars, and airborne / unmanned SAR.

[0115] Target working mode: The signal waveforms of different radiation source targets in different working modes have different characteristics. Further analysis of these characteristics can help to preliminarily determine the working mode of the target;

[0116] Target intention: The target intention can be preliminarily inferred through reconnaissance information (target type, signal information, altitude, heading angle and speed). In this example, the target intention is closely related to the target type.

[0117] Target mission capability: After determining the target type (based on the target radiation information, i.e. frequency, pulse repetition period, etc. combined with the target threat library), different target types have different performance, functions and quantities, which affects their mission capability.

[0118] In this example, there are 8 threat factors of the target, namely target type, target intention, target mission capability, target speed, target heading angle, target distance, target altitude and target working mode, which are represented by x1, x2, ..., x8. There is also a threat level indicator x9, which makes the data dimension reach 9 dimensions. The number of samples is set to 2000, and the specific threat level indicator is determined by investigating the military threat level assessment criteria and expert experience. Based on these data, the method of the present invention is used to evaluate the target threat level. The detailed process is shown in Figure 5 .

[0119] Step 1: Data Preprocessing

[0120] Qualitative indicators such as target type and target intention are directly discretized, and quantitative targets such as target speed and target height are discretized by arithmetic discretization method. The discretization results are as follows:

[0121] Target Type

[0122] x1={0,1,2,3}={digital and analog radio, UAV measurement and control, reconnaissance radar, airborne / unmanned SAR}

[0123] Target intention x2 = {0, 1, 2} = {communication, imaging, surveillance, fire control guidance};

[0124] Target task capability x3 = {0, 1, 2} = {weak, medium, strong};

[0125] Target speed x4 = {0, 1, 2} = {slow, medium, fast};

[0126] Target heading angle x5 = {0,1} = {small, large};

[0127] Target distance x6 = {0, 1, 2} = {near, medium, far};

[0128] Target height x7 = {0, 1, 2} = {low, medium, high};

[0129] Target working mode x8 = {0, 1, 2, 3} = {communication, search, tracking, locking};

[0130] Target threat level x9 = {0, 1, 2} = {low, medium, high}.

[0131] Step 2: Build a threat level assessment model

[0132] The aforementioned ENINOBS method is used to learn the Bayesian network structure from the sample data. The parameter settings of the ENINOBS method are shown in Table 1:

[0133] Table 1 ENINOBS method parameter settings

[0134]

[0135]

[0136] Learning outcomes such as Figure 4 Then, the MLE algorithm is used to learn the network parameters based on the determined network structure and sample data, thus completing the construction of the entire Bayesian network. This network is the threat level assessment model.

[0137] Step 3: Reasoning about the threat level assessment results

[0138] According to the threat level assessment model established in step 2, the clique tree propagation algorithm is used to calculate the threat value of the target. Now, three different sets of target situation information are selected and input into the network to infer the corresponding threat level. The three sets of situation information are:

[0139] Target 1: x1=0, x2=0, x3=0, x4=1, x5=0, x6=2, x7=0, x8=0;

[0140] Target 2: x1=1, x2=2, x3=1, x4=2, x5=0, x6=1, x7=1, x8=1;

[0141] Inference is performed on the following posterior probabilities:

[0142] P(x 10 =i|x1=0,x2=0,x3=0,x4=1,x5=0,x6=2,x7=0,x8=0), i=0,1,2;

[0143] P(x 10 =i|x1=1,x2=2,x3=1,x4=2,x5=0,x6=1,x7=1,x8=1),i=0,1,2

[0144] The inference values ​​of the above posterior probabilities are shown in Table 2:

[0145] Table 2 Reasoning results

[0146]

[0147] By comparison, the target threat levels are ranked as follows:

[0148] Goal 2 > Goal 1.

[0149] Based on the above embodiments, the present application also provides a radiation source target threat level assessment device, referring to Figure 6, the device 100 may include an acquisition module 1001, a group construction module 1002, a structure construction module 1003, a matrix construction module 1004, a network structure generation module 1005, a parameter estimation module 1006 and a threat reasoning module 1007, wherein the acquisition module 1001 is used to obtain the threat factors of the radiation source, and randomly construct a plurality of node sequences based on the threat factor construction data of the radiation source, and construct an integrated group based on the plurality of node sequences; the group construction module 1002 is used to search the neighborhood of the initial node sequence in the integrated group based on a pre-constructed node sequence optimization algorithm to obtain a local optimal integrated group; the structure construction module 1003 is used to transform each of the node sequences in the local optimal integrated group based on a pre-constructed K2 algorithm to obtain a plurality of local optimal network structures; The matrix construction module 1004 is used to obtain each structure matrix according to each local optimal network structure; the network structure generation module 1005 is used to process each structure matrix based on the pre-constructed model average algorithm to obtain an integrated matrix, process the integrated matrix based on the pre-constructed integrated restart strategy to obtain an optimized node sequence, restart the search using the optimized node sequence, iterate in this way, end the search after satisfying the iteration stop condition, obtain the optimal node sequence, and convert the optimal node sequence into a Bayesian network structure; the parameter estimation module 1006 is used to process the sample data and the Bayesian network structure based on the maximum likelihood estimation method to obtain the network parameters of the Bayesian network structure; the threat reasoning module 1007 is used to obtain the threat level assessment result through the Bayesian network structure and the network parameters based on the clique tree propagation algorithm. .

[0150] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiments.

[0151] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0152] Another embodiment of the present application proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the radiation source target threat level assessment method in the above-mentioned method embodiments.

[0153] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators and power management circuits together, which are all well known in the art, so this article will not describe them further. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna, and further, the antenna also receives the data and transmits the data to the processor.

[0154] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0155] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program, which implements the above method embodiment when executed by a processor.

[0156] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory), RAM (Random Access Memory), disk or optical disk and other media that can store program codes.

[0157] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A method for assessing the threat level of a radiation source target, characterized in that: include: Acquire threat factors of the radiation source, and randomly construct multiple node sequences based on the threat factor construction data of the radiation source, and construct an integrated group based on the multiple node sequences; Searching the neighborhood of the initial node sequence in the integrated population based on a pre-built node sequence optimization algorithm to obtain a local optimal integrated population; Based on the pre-built K2 algorithm, the node sequences in the local optimal integrated population are transformed to obtain multiple local optimal network structures; Obtaining each structural matrix according to each of the local optimal network structures; Processing each of the structural matrices based on a pre-constructed model averaging algorithm to obtain an integrated matrix, processing the integrated matrix based on a pre-constructed integrated restart strategy to obtain an optimized node sequence, restarting the search using the optimized node sequence, terminating the search after satisfying an iteration stop condition, obtaining an optimal node sequence, and converting the optimal node sequence into a Bayesian network structure; Processing the sample data and the Bayesian network structure based on the maximum likelihood estimation method to obtain network parameters of the Bayesian network structure; The Bayesian network structure and the network parameter reasoning are processed based on the clique tree propagation algorithm to obtain the threat level assessment result.

2. The method for assessing the threat level of a radiation source target according to claim 1, characterized in that: The threat factors include target type, target intention, target mission capability, target speed, target heading angle, target distance, target altitude, target working mode and threat level index.

3. The method for assessing the threat level of a radiation source target according to claim 2, characterized in that: The pre-built node order optimization algorithm searches the neighborhood of the initial node order in the integrated population to obtain a local optimal integrated population, including: Update the node sequence in the integrated population based on a preset first formula; Based on the neighborhood search algorithm, a search is started with a random node sequence, and the sequence of each node in the integrated group is updated during the search process to obtain the current optimal integrated group; The preset first formula is: Among them, T represents the node order space, D represents the sample data, Yes The optimized node order, is the node order obtained by the λth search when optimizing the i-th node order, represents the scoring function, Pa represents the parent node, X k represents a node, and n represents the maximum node order.

4. The method for assessing the threat level of a radiation source target according to claim 3, characterized in that: The structure matrices are obtained according to the local optimal network structures, including: Calculate the parent node set of each node based on a preset second formula, and determine all subgraphs of each node based on each node and the corresponding parent node set; Determine a local optimal structure based on all subgraphs of each of the nodes; The local optimal structure is converted into a matrix form using a preset rule to obtain a structure matrix corresponding to each local optimal structure; The preset second formula is: The expression of the preset rule is: Among them, M(p,q)=1 means that in the directed acyclic graph, q To X p The directed edge of X q Yes X p The parent node of X; M(p,q) = 0 means that in the directed acyclic graph, q To X p There is no edge in the direction of X q Not X p The parent node of .

5. The method for assessing the threat level of a radiation source target according to claim 4, characterized in that: The pre-built model averaging algorithm is used to process each of the structure matrices to obtain an integrated matrix, including: Processing all the local optimal structures using a preset aggregate function to obtain an integrated matrix; The expression of the aggregate function is: in, represents the weight of the local optimal structure in the ensemble, W arc Represents the local optimal structure G of each edge of the directed acyclic graph i The weights in ⊙ are the Hadamard products, M i Represents the structure matrix of node i.

6. The method for assessing the threat level of a radiation source target according to claim 1, characterized in that: The method of restarting the search using the optimized node sequence, ending the search after satisfying the iteration stop condition, obtaining the optimal node sequence, and converting the optimal node sequence into a Bayesian network structure includes: Starting to execute the ENINOBS algorithm based on preset initial conditions, and determining a set of disturbance variables based on a disturbance factor, wherein the preset initial conditions include a maximum number of restarts, an integrated disturbance factor, an impact factor, and an integrated population size; Determining a parent node set of the disturbance variable set in the integrated matrix based on each of the disturbance variables and the influencing factors; An initial node sequence is randomly selected and the neighborhood of the initial node sequence is searched based on the INOBS algorithm; After a round of search, each disturbance variable in the disturbance variable set is swapped with any node that is arranged after all its parent nodes in the current node order, and the search is restarted after a new node order is obtained; During the search process, if the node order score obtained by a search is less than the score of the current node order, the current node order remains unchanged; If the score of the node sequence obtained by a search is greater than or equal to the score of the current node sequence, the current node sequence is replaced; If the score of the node sequence obtained during the search process is greater than the score of the lowest-scoring node sequence in the integrated group, it will be replaced; if the size of the integrated group is smaller than the preset integrated group size, the node sequence will be added to the integrated group; Execute the above steps until the number of restarts is greater than the maximum number of restarts, and convert the optimal node sequence finally obtained into a network structure output based on the K2 algorithm, otherwise continue to execute the restart algorithm until the number of restarts is greater than the maximum number of restarts.

7. A radiation source target threat level assessment device, characterized in that: include: An acquisition module, used for acquiring threat factors of a radiation source, and randomly constructing a plurality of node sequences based on the threat factor construction data of the radiation source, and constructing an integrated group based on the plurality of node sequences; A group building module is used to search the neighborhood of the initial node order in the integrated group based on a pre-built node order optimization algorithm to obtain a local optimal integrated group; A structure construction module, used to transform each of the node sequences in the local optimal integrated population based on a pre-constructed K2 algorithm to obtain multiple local optimal network structures; A matrix construction module, used to obtain each structural matrix according to each of the local optimal network structures; A network structure generation module, used for processing each of the structure matrices based on a pre-built model averaging algorithm to obtain an integrated matrix, processing the integrated matrix based on a pre-built integrated restart strategy to obtain an optimized node sequence, restarting the search using the optimized node sequence, terminating the search after satisfying an iteration stop condition, obtaining an optimal node sequence, and converting the optimal node sequence into a Bayesian network structure; A parameter estimation module, used for processing sample data and the Bayesian network structure based on the maximum likelihood estimation method to obtain network parameters of the Bayesian network structure; The threat reasoning module is used to obtain a threat level assessment result by reasoning through the Bayesian network structure and the network parameters based on the clique tree propagation algorithm.

8. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the radiation source target threat level assessment method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for assessing the threat level of a radiation source target according to any one of claims 1 to 6 can be implemented.

Images