An Analysis Method for the Identification Efficiency of Enemy and Friendly Targets Based on Expert Experience and Bayesian Networks

Abstract

The present invention relates to a method for analyzing the identification effectiveness of enemy and friendly targets based on expert experience and Bayesian networks, belonging to the technical field of target identification. A Bayesian network structure for identifying enemy and friendly targets including 6 parent nodes is established; fuzzy prior knowledge of the strong and weak relationships among the 6 parent nodes in the network is given by experts; the transition matrix MAT is calculated; and the CPT table of the enemy and friendly target identification ability is obtained from the transition matrix. The method of the present invention does not require training data, solves the problem limited by the lack of network data, and improves the accuracy of the analysis of the identification effectiveness of enemy and friendly targets.

Description

Technical Field

[0001] The present invention belongs to the technical field of target recognition, and relates to a method for analyzing the recognition efficiency of enemy and friendly targets based on expert experience and Bayesian network. Background Art

[0002] The recognition of enemy and friendly targets plays an important role in offense and defense, and is an indispensable important link in the offense and defense system. The technologies involved are relatively complex. It is very necessary to analyze its efficiency and find the factors that have the most influence on the recognition ability of enemy and friendly targets. Therefore, it is necessary to analyze the efficiency of enemy and friendly target recognition.

[0003] The most common and simplest method for analyzing the efficiency of key factors in complex systems is the Analytic Hierarchy Process (AHP). AHP decomposes the factors related to decision-making into different levels, and on this basis, qualitative and quantitative analyses are carried out to achieve the purpose of efficiency analysis. However, the Analytic Hierarchy Process only considers the experience of experts and ignores the information contained in the original node data, with relatively high subjectivity and poor objectivity. In order to obtain information from node data in the analysis process and make the results of efficiency analysis more objective, a Bayesian network is generally used to analyze the efficiency of key factors in complex systems.

[0004] A Bayesian network can well express the random uncertainty and correlation existing between variables, and moreover, it can perform relevant reasoning of uncertainty. A Bayesian network can not only achieve forward reasoning, deduce the posterior probability based on the prior probability, that is, reason the result from the cause, but also calculate the prior probability of the node from the posterior probability through the Bayesian formula, that is, reason the cause from the result. This two-way reasoning of causal relationship enables the application of Bayesian network efficiency analysis in many fields, such as target recognition, damage effect assessment, information security assessment, and medical diagnosis. In these fields, the Bayesian network has shown good performance, and the results of efficiency assessment are also relatively ideal.

[0005] When constructing a Bayesian network, the number of states of the Bayesian network nodes is generally set to no more than 3, and the number of parent nodes corresponding to the nodes is generally set to no more than 4. If these values are too large, the number of parameters of the Bayesian network will be extremely large and difficult to train. Taking a Bayesian network with 3 states and 4 parent nodes as an example, the child nodes of this network have 243 parameters. From the cognitive perspective, it is very difficult for us to obtain so many parameters from experts. From the statistical perspective, if we hope to learn parameters from data, we need a very large amount of training data to reliably estimate so many parameters.

[0006] In a complex system, the influence of various factors on the result is very complex, and it is difficult to determine the relationship and the strength of the association between factors through the human brain. For the Bayesian network generally used for the effectiveness analysis of key factors in complex systems, researchers set the possible values of nodes to 2. Taking the weapon attack distance as an example, the general values are {long attack distance, short attack distance}. Using the Bayesian network to conduct effectiveness analysis on nodes is a simple, flexible multi-criteria decision-making method for quantitative analysis of qualitative problems. By setting the parameter values (i.e., conditional probabilities) of each node and then performing effectiveness analysis, the importance of the target node can be quantitatively described.

[0007] In actual use, for some nodes of the network, it is difficult to obtain network training data, or the number of parent nodes and their value ranges are too large, resulting in the difficulty of obtaining the conditional probability table (CPT) by experts, making it very difficult to use the method of effectiveness analysis through the Bayesian network under such conditions. Summary of the Invention

[0008] Technical Problems to be Solved

[0009] The existing methods for evaluating the effectiveness of target recognition can be divided into two types: knowledge-driven and data-driven methods. In the case of no training data, the parameters can only be obtained through expert experience, that is, knowledge-driven. There are many factors involved in the field of target recognition, and the number of parameters required for the constructed Bayesian network is very large. In this case, it is very difficult to construct a probability table through expert experience, and usually only fuzzy prior knowledge of the strength relationship between nodes can be given for such a large number of network parameters. Therefore, the present invention proposes a method for estimating the parameters of a Bayesian network without training data, only by giving the strength relationship of the parent nodes, in the field of enemy and friendly target recognition effectiveness analysis.

[0010] Technical Solution

[0011] An enemy and friendly target recognition effectiveness analysis method based on expert experience and Bayesian network, characterized by the following steps:

[0012] Step 1: Establish a Bayesian network structure, including 6 parent nodes, namely node 1 representing the target attribute, node 2 representing the target state, node 3 representing the maximum number of identifications, node 4 representing the response time of the cluster to target recognition, node 5 representing the recognition distance, and node 6 representing the effective probability of the cluster to target recognition; the child node represents the target enemy and friendly target recognition ability of the effectiveness analysis;

[0013] Step 2: The fuzzy prior knowledge of the strength relationship among the 6 parent nodes in the network is given by experts: the evaluation value of the target attribute is 0.1, the evaluation value of the target state is 0.16, the evaluation value of the maximum number of identifications is 0.14, the evaluation value of the response time of the cluster to target identification is 0.25, the evaluation value of the identification distance is 0.25, and the evaluation value of the effective probability of the cluster to target identification is 0.25, which is formulated as follows:

[0014] π 1 = 0.1, π 2 = 0.16, π 3 = 0.14, π 4 = 0.2, π 5 = 0.25, π 6 = 0.25,

[0015] α = {0.1, 0.16, 0.14, 0.2, 0.25, 0.25}

[0016] Step 3: Calculate the transition matrix MAT

[0017] Give the state matrix, where 1, 2, 3 represent weak, medium, and strong;

[0018]

[0019] where N represents the number of parent nodes;

[0020]

[0021] Step 4: Obtain the CPT from the transition matrix

[0022] par1′ = MAT(:, 1:N)·α

[0023] par3′ = 1 - par1′

[0024] par1 = par1′.*par1′

[0025] par2 = 2×par1′.*par2′

[0026] par3 = par2′.*par2′

[0027] par = [par1, par2, par3]

[0028] where par is the CPT table of the enemy - friendly target recognition ability; where 1, 2, 3 refer to the states of the nodes, weak, medium, and strong; the numbers represent the probabilities of weak enemy - friendly target recognition ability.

[0029] A computer system, characterized in that it includes: one or more processors, a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in claim 1.

[0030] A computer-readable storage medium, characterized in that it stores computer-executable instructions which, when executed, are used to implement the method described in claim 1.

[0031] A computer program, characterized in that it includes computer-executable instructions which, when executed, are used to implement the method described in claim 1.

[0032] Beneficial effects

[0033] An analysis method for the identification efficiency of enemy and friendly targets based on expert experience and Bayesian network proposed by the present invention requires no training data, solves the problem of being limited by the lack of network data, and greatly improves the accuracy of the analysis of the identification efficiency of enemy and friendly targets. It has the following advantages:

[0034] 1. Compared with the analytic hierarchy process, since this method uses the Bayesian network method, after giving the node parameters, the network parameters can be corrected through training data, making the parameters of the Bayesian network more reasonable. Because the nature of the data itself is considered, the analysis results are relatively objective.

[0035] 2. For previous related algorithms, the previous algorithms could only calculate networks with 2 node states, and the applicable range of the related algorithms was too narrow and not universal. The present invention improves the universality and can be applied to Bayesian networks with any number of states.

[0036] 3. Currently existing algorithms generally perform parameter learning through training data or obtain parameters based on expert experience. These two methods can only be used when the node size is small. The present invention has no requirements for the node size and can perform calculations regardless of the number. In this case, if the network parameters are calculated through training data, on the premise that the accuracy is set to 0.1, training data is required. However, the present invention does not require training data to train the parameters and only needs 6 constraints to deduce the network parameters.

[0037] This method enables the parameter learning of the Bayesian network that was originally impossible to be carried out, and expands the applicable range of the Bayesian network.

[0038] 4. If the method of assigning parameters by experts is used, assuming that the average time for an expert to give a parameter is 5 s, in this case, it takes 2187×5 s = 10935 s, that is, 3.0375 hours. For a network with 500 nodes, such efficiency is unacceptable. The present invention only takes 30 seconds.

[0039] Table 10: Comparison of execution time between the present invention and expert experience (6 parent nodes)

[0040]

[0041] When there are only two parent nodes, expert experience requires 27×5 s = 135 s, and the computer of the present invention only needs 0.00001 s plus the time of 10.00001 s for experts to assign parameters, as shown in Table 11.

[0042] Table 11: Comparison of execution time between the present invention and expert experience (2 parent nodes)

[0043] Execution time Expert experience 135s The present invention 0.00001 second + 10 seconds (time given by the expert for the parameter)

[0044] It can be seen that in the conventional case with a small number of nodes, in terms of execution time, the present algorithm still has an advantage compared with the expert experience method.

[0045] In summary, compared with the analytic hierarchy process, the result of the effectiveness analysis of the present invention is more reasonable. Compared with parameter learning through data and the expert experience method, it has superiority in performance, faster calculation speed, smaller required quantity, and universality in the application field. The present invention is applied to the identification of enemy and friendly targets and has a certain advantage compared with the existing algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings are only for the purpose of showing specific embodiments and are not considered as a limitation to the present invention. Throughout the drawings, the same reference signs denote the same components.

[0047] Figure 1 Schematic diagram of Bayesian network (CPT description);

[0048] Figure 2 Schematic diagram of Yang Hui triangle;

[0049] Figure 3 Label description;

[0050] Figure 4 Effectiveness analysis of the enemy and friendly identification ability;

[0051] Figure 5 Schematic diagram of the Bayesian network structure. DETAILED DESCRIPTION OF THE INVENTION

[0052] To make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0053] Generally speaking, a Bayesian network B consists of two parts: a structure G and parameters θ, B = (G, θ). The network structure G is a directed acyclic graph, and each node corresponds to a clear meaning. Let the parent node be π, so the CPT containing each attribute is

[0054]

[0055] The following takes Figure 1 as an example to illustrate the meaning of the CPT.

[0056] Figure 1 As shown, it is a network composed of five nodes. The CPTs of each node are shown in Table 1, Table 2, Table 3, Table 4, and Table 5:

[0057] Table 1: CPT of node Difficulty

[0058] <![CDATA[d 0 > <![CDATA[d 1 > 0.6 0.4

[0059] Table 2: CPT of node Intelligence

[0060] <![CDATA[i 0 > <![CDATA[i 1 > 0.7 0.3

[0061] Table 3: CPT of node Grade

[0062] <![CDATA[g 1 > <![CDATA[g 2 > <![CDATA[g 3 > <![CDATA[i 0 d 0 > 0.3 0.4 0.3 <![CDATA[i 0 d 1 > 0.05 0.25 0.7 <![CDATA[i 1 d 0 > 0.9 0.8 0.02 <![CDATA[i 1 d 1 > 0.5 0.3 0.2

[0063] Table 4: CPT of node SAT

[0064] <![CDATA[s 0 > <![CDATA[s 1 > <![CDATA[i 0 > 0.95 0.05 <![CDATA[i 1 > 0.2 0.8

[0065] Table 5: CPT of node Letter

[0066] <![CDATA[l 0 > <![CDATA[l 1 > <![CDATA[g 1 > 0.1 0.9 <![CDATA[g 2 > 0.4 0.6 <![CDATA[g 3 > 0.99 0.01

[0067] In this network, the sizes of nodes Difficulty, Intelligence, SAT, and Letter are 2, that is, there are two value-taking methods. The size of node Grade is 3, that is, there are three value-taking methods. The CPT table of each node is only related to the current node and its parent node. Let node_size be the size of the current node, {π|π 1 ,π 2,...} is the set of node parent nodes, N is the number of parent nodes, π 1 , π 3 ,... are the sizes of the parent nodes. The size calculation formula of the CPT table is as follows:

[0068]

[0069] Taking Figure 1 the Grade node in the network as an example, its parent nodes are Difficulty and Intelligence, node_size = 3, π Difficulty = π Intelligence = 2

[0070] CPT_size = node_size × π Difficulty × π Intelligence = 3 × 2 × 2 = 12 (0.3)

[0071] The size of CPT_size determines the amount of data for training the node parameters and the difficulty of the probability values given by the expert.

[0072] In practical applications, it often occurs that the number of parent nodes of a certain node exceeds 5. At this time, if the size of the parent node is 3, CPT_size will be very large. When the node size is 3 and the number of parent nodes is 5, CPT_size = 3 × 3 5 = 729. In practical applications, there may even be a situation where there are up to 7 parent nodes. At this time, CPT_size = 3 × 3 7 = 6561. Therefore, it is almost an impossible task to use expert experience. So, a method is needed to quickly give the CPT table on the premise of giving the fuzzy prior knowledge of the strength relationship of the parent nodes in advance. The given CPT table should meet the following conditions:

[0073] 1. Meet the conditions defined by probability theory, that is, the sum of probabilities is 1.

[0074] 2. The given CPT table has no missing items, and each probability value corresponds one by one to all possibilities.

[0075] Figure 2 What is shown is Pascal's triangle, which represents a geometric arrangement of the coefficients of the unknowns a and b in (a + b) n changing with n. Yang Hui, a mathematician in the Southern Song Dynasty of China, discovered it in 1261 and recorded it in "Detailed Explanation of the Nine Chapters of Algorithms". In Europe, Pascal also discovered this rule in 1654.

[0076] The present invention uses the properties of the polynomial expansion formula (explained by the formula below) to complete the estimation of Bayesian parameters:

[0077] a + b = 1

[0078] Then

[0079] (a + b) n = 1

[0080] The values in Pascal's triangle are the coefficients of each term after the above formula is expanded, which are expressed by the formula as follows:

[0081]

[0082] The coefficients are consistent with Pascal's triangle. In this patent, Pascal's triangle can be used to better introduce the coefficients of polynomial expansion.

[0083] Therefore, it is a feasible method to conduct effectiveness analysis on the enemy - friendly target recognition ability through the Bayesian method, and it has the possibility of implementation, with simple operation and easy to use.

[0084] The present invention proposes a method for estimating Bayesian network parameters based on the strength relationship of the influence of parent nodes on target nodes. It aims to solve the problem that it is too difficult to give CPT parameters when the number of parent nodes is more than 5, so as to use expert experience and Bayesian network for effectiveness analysis of key factors in complex systems.

[0085] Suppose there is a node named node, node_size = n, and there are N parent nodes. The set of its parent nodes is {π|π i , i = 1 ~ N}, and the values of each node are 1 ~ n. The influence (affect) of the parent nodes on the node node given by expert experience is

[0086] Suppose the node is node, node_size = ns.

[0087] Traverse all possibilities of the node:

[0088] π 1 1 π 2 1 ; π 3 1 ...π N-1 1 π N 1 node 1

[0089] π 1 1 π 2 1 π 3 1 ...π N-1 1 π N1 node 2 ...

[0090] π 1 1 π 2 1 π 3 1 ...π N-1 1 π N 1 node n

[0091] π 1 1 π 2 1 π 3 1 ...π N-1 1 π N 2 node 1

[0092] π 1 1 π 2 1 π 3 1 ...π N-1 1 π N 2 node 2 ...

[0093] The symbolic meanings of the above expressions are as Figure 3 shown

[0094] The above expressions are compiled into Table 6:

[0095] Table 6: Brief Explanation of the Value of Each Node

[0096]

[0097]

[0098] Thus, the value of the CPT table can be represented by p(111...111) = P 111...111 where P 111...111 represents the probability value when the values of each node are 111...111.

[0099] The left side of Table 6 can be represented by an n N+1 matrix:

[0100] INDEX = {index ij | i = 1 to n N , J = 1 to n} (0.4)

[0101] Obviously, the above formula can be understood as the index value of the CPT table.

[0102] Next, introduce a transition matrix MAT for the following logical operations:

[0103] MAT = {mat ij | i = 1 to n N , j = 1 to n} (0.5)

[0104]

[0105] Then:

[0106] par1′ = {par1′ i | i = 1 to n N} (0.7)

[0107] par1′ = MAT(:, 1:N) · α (0.8)

[0108] MAT(:, 1:N) represents the matrix formed by the 1st to Nth columns of matrix MAT.

[0109] parn′ = 1 - par1′

[0110] Get

[0111]

[0112]

[0113] ...

[0114]

[0115] par1, par2,..., parn are the obtained CPT tables, where par1 represents the conditional probability of node = 1, par2 represents the conditional probability of node = 2, and parn represents the conditional probability of node = n.

[0116] Define par as the generated parameter, then

[0117] par = [par1, par2... parn] (0.13)

[0118] In the formula, par1, par2... parn are column vectors of size n N and par is n NAn n×n matrix, and par is the CPT table to be obtained.

[0119] par = {par ij | i = 1 to n N , j = 1 to n} (0.14)

[0120] The corresponding relationship with INDEX is shown in Table 7.

[0121] Table 7: The corresponding relationship between the generated parameter par and INDEX

[0122]

[0123] Use the above Bayesian network to conduct effectiveness analysis on the enemy - friendly target recognition ability. As Figure 4 shown, the evaluation target of this effectiveness evaluation system is the enemy - friendly recognition ability, which has 6 related factors, namely target attribute, target status, maximum recognition number, cluster's response time to target recognition, recognition distance, and cluster's effective probability of target recognition. The value of each node can be set to 3 types, with the values being {strong, medium, weak}. For example, a strong maximum recognition number means a relatively large maximum recognition number, medium means a moderate maximum recognition number, and weak means a relatively small maximum recognition number. The physical meanings of other nodes are similar.

[0124] Explanation of relevant information of the Bayesian network:

[0125] 1. The structure of the Bayesian network is as Figure 5 shown. Node 1 represents the target attribute, node 2 represents the target status, node 3 represents the maximum recognition number, node 4 represents the cluster's response time to target recognition, node 5 represents the recognition distance, node 6 represents the cluster's effective probability of target recognition, and node node represents the target enemy - friendly target recognition ability of the effectiveness analysis.

[0126] 2. Assume that all nodes have three states. Taking the maximum recognition number as an example, its three states are few, medium, and many. We set the probabilities of these three states to 1 / 3, and the remaining nodes are processed similarly. That is, the CPT of each node is {1 / 3, 1 / 3, 1 / 3}.

[0127] Calculation objective: Through the fuzzy prior knowledge of the strength relationship between the parent nodes given by experts, calculate the CPT table of node node, obtain the complete Bayesian network parameters, and then conduct effectiveness analysis through the Bayesian network to be consistent with the strength relationship between the parent nodes given by experts.

[0128] Step 1: Experts give the fuzzy prior knowledge of the strength relationship between 6 parent nodes in the network

[0129] Experts use multi-factor decision-making methods such as AHP to give the strength relationship of the parent nodes and perform summation normalization. The results show that the evaluation value of the target attribute is 0.1, the evaluation value of the target state is 0.16, the evaluation value of the maximum number of identifications is 0.14, the evaluation value of the cluster's response time to target identification is 0.25, the evaluation value of the identification distance is 0.25, and the evaluation value of the effective probability of the cluster's target identification is 0.25, which is formulated as follows.

[0130] π 1 = 0.1, π 2 = 0.16, π 3 = 0.14, π 4 = 0.2, π 5 = 0.25, π 6 = 0.25,

[0131] α = {0.1, 0.16, 0.14, 0.2, 0.25, 0.25}

[0132] Step 2: Calculate the transition matrix MAT

[0133] First, we calculate the size of the conditional probability table for the target recognition capabilities of our own and the enemy:

[0134] CPT_size = 3 7 = 2187

[0135] Next, we give the state matrix, using 1, 2, and 3 to represent weak, medium, and strong.

[0136]

[0137] Step 3: Obtain the CPT from the transition matrix

[0138] par1′ = MAT(:, 1:N)·α

[0139] par3′ = 1 - par1′

[0140] par1 = par1′.*par1′

[0141] par2 = 2×par1′.*par2′

[0142] par3 = par2′.*par2′

[0143] par = [par1, par2, par3]

[0144] The par is the CPT table for the IFF (Identification Friend or Foe) ability (see Appendix). Among them, 1, 2, and 3 refer to the states of the nodes, weak, medium, and strong. Taking the first data as an example, it means that when the target attribute is weak, the target state is weak, the maximum number of identifications is small, the reaction time of the cluster to target identification is long, the identification distance is short, and the effective probability of the cluster to target identification is low, the probability of weak IFF ability is 0.667.

[0145] Appendix: (Partial display of the results of the IFF ability effectiveness analysis case)

[0146] Note: π represents the parent node, node represents the child node, and the Arabic numerals represent the node values. The original table has 729 rows of parameters, and only 36 rows of parameters are shown here.

[0147]

[0148] After obtaining the complete parameters of the Bayesian network, then using the Bayesian network for effectiveness analysis can obtain the following results.

[0149]

[0150] It can be found that the obtained results are consistent with those given by the experts.

[0151] In this case, CPT_size = 3 7 = 2187. It is relatively difficult to use the method of assigning parameters by experts. Using the present invention, it only takes less than 1 s to complete the assignment. In this way, the expert experience can be used for the effectiveness analysis of IFF, and the results of the effectiveness analysis can be obtained more conveniently.

[0152] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. An analysis method for the recognition efficiency of enemy and friendly targets based on expert experience and Bayesian network, characterized in that the steps are as follows: Step 1: Establish a Bayesian network structure, including 6 parent nodes, namely Node 1 representing target attributes, Node 2 representing target status, Node 3 representing the maximum number of recognitions, Node 4 representing the response time of the cluster to target recognition, Node 5 representing the recognition distance, and Node 6 representing the effective probability of the cluster to target recognition; the child node represents the target enemy and friendly target recognition ability of the efficiency analysis; Step 2: The expert gives the fuzzy prior knowledge of the strength relationship between the 6 parent nodes in the network: the evaluation value of target attributes is 0.1, the evaluation value of target status is 0.16, the evaluation value of the maximum number of recognitions is 0.14, the evaluation value of the response time of the cluster to target recognition is 0.25, the evaluation value of the recognition distance is 0.25, and the evaluation value of the effective probability of the cluster to target recognition is 0.

25. It is formulated as follows: π 1 ＝0.1,π 2 ＝0.16,π 3 ＝0.14,π 4 ＝0.2,π 5 ＝0.25,π 6 ＝0.25, α＝{0.1,0.16,0.14,0.2,0.25,0.25} Step 3: Calculate the transition matrix MAT Give the state matrix, using 1, 2, 3 to represent weak, medium, and strong; where, N represents the number of parent nodes; Step 4: Obtain the conditional probability table CPT from the transition matrix par1' = MAT(:, 1:N)·α par3' = 1 - par1' par1 = par1'.*par1' par2 = 2×par1'.*par2' par3 = par2'.*par2' par = [par1, par2, par3] where, par is the CPT table of the enemy and friendly target recognition ability; where 1, 2, 3 refer to the states of the nodes, weak, medium, and strong; the numbers represent the probabilities of weak enemy and friendly target recognition ability.

2. A computer system, characterized in that it includes: One or more processors, a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in claim 1.

3. A computer-readable storage medium, characterized in that it stores computer-executable instructions, and the instructions are used to implement the method described in claim 1 when executed.

4. A computer program, characterized in that it includes computer-executable instructions, and the instructions are used to implement the method described in claim 1 when executed.

Images