A Threat Assessment Method for Aerial Targets Based on Variable Weighted Cloud Bayesian Networks

Abstract

This invention provides an aerial target threat assessment method based on a variable-weighted cloud Bayesian network. This method addresses the issue that threat assessments often rely on quantitative analysis while neglecting qualitative aspects such as target threat levels. It categorizes target threats as high, medium, or low. The method includes the following steps: First, a threat assessment model of aerial target attributes is constructed based on a Bayesian network. Second, a cloud model is introduced into the Bayesian network to represent the correlation probabilities between network nodes. Third, battlefield situational information is combined with a variable-weighted method using an improved Gaussian representation to determine the weights of target attributes for threat assessment. Finally, based on the correlation probabilities and target attribute weights, the target threat level is determined using cloud model algorithms. This invention addresses the qualitative problem of target threat levels, effectively completing target threat assessment, obtaining target threat level decisions, and further improving the utilization rate of battlefield information.

Description

Technical Field

[0001] This invention relates to the field of high-level information fusion technology, and in particular to a method for assessing aerial target threats based on variable weighted cloud Bayesian networks. Background Technology

[0002] Target threat assessment is a crucial aspect of intelligence analysis, aiming to evaluate the degree of threat posed by an adversary target to our operational activities. In target threat assessment, battlefield information often possesses ambiguity and randomness. For example, intelligence information such as the location, quantity, and weaponry of adversary targets may exhibit a degree of uncertainty. To more accurately assess target threats, this uncertainty needs to be transformed into a relationship between qualitative concepts and quantitative values. Specifically, methods such as fuzzy mathematics and stochastic processes can be employed to convert uncertain battlefield intelligence into numerical values ​​such as probabilities and membership degrees, enabling quantitative analysis and decision-making.

[0003] Research and applications of aerial target threat assessment in complex scenarios have attracted widespread attention from experts and scholars both domestically and internationally, resulting in extensive exploratory research and the proposal of numerous target threat assessment methods. Current aerial target threat assessment methods are generally divided into two categories: dynamic threat assessment methods and static threat assessment methods. Regarding dynamic threat assessment methods, Xu et al. utilized an ELM neural network to assess aerial target threats. By training the neural network with a large amount of air combat data, using target attribute information as the input layer and target threat as the output layer, they achieved highly accurate assessment results. However, this method requires training the neural network with a large amount of data to ensure its prediction accuracy. In real battlefield situations, the availability of enemy aerial target data is often very limited, significantly restricting the practical application of this method. Li Huiqiang et al. proposed an aerial target threat assessment method based on dynamic Bayesian networks. By establishing a dynamic Bayesian network model of aerial target threats and using a state transition probability matrix to correlate the target threat at the previous moment with the target threat at the current moment, they achieved dynamic assessment of target threats over a time series. However, this method assigns equal weights to each target attribute, failing to fully consider the actual situational context. Furthermore, the state transition is significantly affected by the posterior probability of the target threat at the previous moment, leading to decreased assessment accuracy. Regarding static threat assessment methods, Zhang et al. used the Analytic Hierarchy Process (AHP) to calculate the weights of each target attribute. This method treats the complex aerial target threat assessment problem as a system, divides target-related attributes into interconnected levels, and then quantitatively represents the importance of target attributes, determining their relative weights. However, battlefield information is fuzzy, and traditional AHP cannot represent this fuzzy information. Subsequent research has increasingly focused on handling fuzzy information. Zhang Yinyan used cloud model theory to assess aerial target threats, representing aerial target information using a cloud model. This combined the fuzziness and randomness of battlefield information, achieving an uncertain transformation between qualitative concepts and quantitative values. However, when using cloud models to represent target attribute information, this method relies heavily on expert experience and does not take into account the correlation between target attributes, thus ignoring objective information about the target and failing to effectively utilize battlefield information, leading to misjudgments by decision-makers.

[0004] While most of the aforementioned threat assessment methods can obtain a specific threat value for a target, they are essentially quantitative methods of target threat assessment and cannot determine the target threat level. This makes it difficult for our decision-makers to rank the threats of multiple targets with similar threat values, thus affecting battlefield decision-making. Therefore, it is necessary for us to study this qualitative method of target threat assessment. Summary of the Invention

[0005] The purpose of this invention is to provide an aerial target threat assessment method based on variable weighted cloud Bayesian networks, which can effectively obtain qualitative information on target threat levels and effectively improve the utilization rate of battlefield information.

[0006] The technical solution adopted in this invention is as follows:

[0007] A method for assessing aerial target threats based on variable weighted cloud Bayesian networks includes the following specific steps:

[0008] Step 1: Analyze the qualitative information of different target threats to obtain target threat attributes, and construct an aerial target threat assessment model based on this. Specifically:

[0009] Step 1.1: Describe the relationships between nodes using a Bayesian network structure G. Specifically:

[0010] G = (V, E)

[0011] Where V = {V1, V2, ..., V} n} is the set of nodes in a directed acyclic graph, which is an abstract description of the object it represents. E is the set of directed edges, which represent the relationships between the nodes.

[0012] Step 1.2: Based on the node set V of the directed acyclic graph obtained in Step 1.1, use the conditional probability P to reflect the local probability distribution set of the relationships between nodes. Specifically:

[0013]

[0014] Among them, each node V i With V i The parent node given is not V i Any subset of nodes formed by descendant nodes is conditionally independent, Pa(V) i ) represents V i The direct parent node.

[0015] Step 1.3: Analyze the qualitative information of different target threats to obtain target threat attributes, and construct an aerial target threat assessment model based on this. Specifically:

[0016] Based on steps 1.1 and 1.2, the topology G of the Bayesian network can be obtained. Therefore, in aerial target threat assessment, all incoming aerial targets that intend to attack our sensor sites and may pose a threat or even cause large-scale damage require timely countermeasures from our decision-makers. Currently, aerial target attributes are mainly divided into two aspects: benefit-oriented attributes and cost-oriented attributes. Benefit-oriented attributes include target type T, speed S, jamming capability J, and maneuverability M, etc. These attributes are directly proportional to the target threat TL; that is, the stronger the attack of the target type and the faster its movement speed, the greater the threat to our sites. Cost-oriented attributes include target altitude H, distance D, and flight path shortcut R, etc. These attributes are inversely proportional to the target threat TL; that is, the higher the target altitude H and the farther the distance D, the smaller the threat to our sites. An aerial target threat assessment model is constructed based on the benefit-oriented and cost-oriented attributes of aerial target threats.

[0017] Step 2: Introduce the cloud model into the Bayesian network. Use the cloud model to represent the probability of association between nodes in the Bayesian network, thereby expressing the degree of association between nodes. Specifically:

[0018] Step 2.1: Linguistic variables are used to handle fuzzy information about target attributes that cannot be quantitatively represented by Bayesian networks. Specifically:

[0019] Let S = {s} i Let |i=1,2,…,2τ+1,τ∈N} be a set of linguistic terms, where s i Let τ represent a time period, and N represent the set of positive integers. The set of linguistic terms must satisfy the following conditions:

[0020] 1) Orderliness: When i > k, s i >s k ;

[0021] 2) Negation operator: s i =neg(s k ), where i+k=2τ+1.

[0022] Step 2.2: Represent the language values ​​using a cloud model, thereby completing the cloud model transformation of the target attribute threat level. Specifically:

[0023] Based on battlefield environment and expert experience, this invention classifies target threat attribute levels into five levels: very high, high, medium, low, and very low. Let U = [X min ,X max Let S be the universe of discourse for the cloud model, where S = {s1, s2, ..., s}. 2τ+1 Given a set of linguistic terms, 2τ+1 normal cloud models can be obtained using the golden section method. For example, when τ=2, a set of linguistic terms with 5 labels can be obtained:

[0024] S = {s1 = Very low (VL), s2 = Low (L), s3 = Medium (M), s4 = High (H), s5 = Very high (VH)}

[0025] Step 2.3: Network node correlation probability P(TL) j The calculation of ) is as follows:

[0026] The Bayesian network representing the cloud model can be obtained based on steps 2.1 and 2.2. The correlation probability of nodes in a Bayesian network is also called the prior probability. This invention utilizes expert rating to obtain the prior probability of a target threat. Assuming a target's threat level is divided into j levels, and M decision-makers rate the target's threat level, then the prior probability P(TL) of each level of the target is... j )for:

[0027]

[0028] in, X j This represents the number of decision-makers who consider the target threat to be at level j.

[0029] Step 3: Combining battlefield situation information, determine the target threat attribute weights using the improved Gaussian representation of variable weights. Specifically:

[0030] Traditional Gaussian representations, when used for weight calculations, often result in small differences between the weights of different attributes. However, in real battlefield environments, the importance of target attributes often varies significantly. Small differences in attribute weights lead to decreased discriminative power between targets, making it impossible to effectively differentiate the threat levels of each target. This invention improves the exponential part of the Gaussian representation, resulting in more discriminative attribute weights. The specific calculation formula is as follows:

[0031]

[0032]

[0033]

[0034] Where, μ n σ is the mean. n Let ω be the standard deviation, i = 1, 2, ..., n, and m = n / 2 be the number of target attributes. i =[ω1,ω2,…,ω m Normalization will yield the ascending weight θ of the target attribute. i :

[0035]

[0036] Where i = 1, 2, ..., m. At different times, the decision-maker ranks the importance of the target threat attributes, thus obtaining the target threat attribute weight w at each time point. i .

[0037] Step 4: Initialize parameter settings using the correlation probability between Bayesian network nodes and the target threat attribute weights, given the maximum number of rounds T for the algorithm to run. max The cloud model algorithm is used to determine the target threat level. Specifically:

[0038] Steps 2 and 3 yield the node association probability P(TL). j and the target threat attribute weights w at each time point i Combining the two yields the posterior probability P(TL) of the target threat level. j |v1,v2,…v n ):

[0039]

[0040] Among them, P(TL) j Let P(v) be the prior probability of the target threat. i |TL j ) represents the conditional probability of each node, w i The weights for each attribute.

[0041] Finally, the algorithm runs T max In each round, the highest posterior probability of the target threat level is selected from the calculation results of the cloud model operation rules. Then, the decision-maker judges the threat level of the target based on the magnitude of the target threat posterior probability.

[0042] The beneficial effects of this invention are:

[0043] By employing the aforementioned technical solutions, this invention addresses the difficulties in ranking and low accuracy of target threat assessment by proposing an aerial target threat assessment method based on a variable-weighted cloud Bayesian network. The proposed method constructs a threat assessment model of aerial target attributes using a Bayesian network. Then, a cloud model is introduced into the Bayesian network to represent the correlation probabilities between Bayesian network nodes. Simultaneously, battlefield situational information is combined, and a variable-weighting method with an improved Gaussian expression is used to determine the weights of target attributes for threat assessment. Finally, based on the correlation probabilities and target attribute weights, the cloud model's operational rules are used to obtain the target threat level decision. This proposed method effectively obtains qualitative information on target threat levels, significantly improves the utilization rate of battlefield information, effectively processes fuzzy information in the battlefield environment, and provides decision-makers with more reliable target information. Attached Figure Description

[0044] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a flowchart of the present invention; Detailed Implementation

[0046] like Figure 1 As shown, the present invention includes the following steps:

[0047] Step 1: Analyze the qualitative information of different target threats to obtain target threat attributes, and construct an aerial target threat assessment model based on this. Specifically:

[0048] Step 1.1: Describe the relationships between nodes using a Bayesian network structure G. Specifically:

[0049] G = (V, E)

[0050] Where V = {V1, V2, ..., V} n} is the set of nodes in a directed acyclic graph, which is an abstract description of the object it represents. E is the set of directed edges, which represent the relationships between the nodes.

[0051] Step 1.2: Based on the node set V of the directed acyclic graph obtained in Step 1.1, use the conditional probability P to reflect the local probability distribution set of the relationships between nodes. Specifically:

[0052]

[0053] Among them, each node V i With V i The parent node given is not V i Any subset of nodes formed by descendant nodes is conditionally independent, Pa(V) i ) represents V i The direct parent node.

[0054] Step 1.3: Analyze the qualitative information of different target threats to obtain target threat attributes, and construct an aerial target threat assessment model based on this. Specifically:

[0055] Based on steps 1.1 and 1.2, the topology G of the Bayesian network can be obtained. Therefore, in aerial target threat assessment, all incoming aerial targets that intend to attack our sensor sites and may pose a threat or even cause large-scale damage require timely countermeasures from our decision-makers. Currently, aerial target attributes are mainly divided into two aspects: benefit-oriented attributes and cost-oriented attributes. Benefit-oriented attributes include target type T, speed S, jamming capability J, and maneuverability M, etc. These attributes are directly proportional to the target threat TL; that is, the stronger the attack of the target type and the faster its movement speed, the greater the threat to our sites. Cost-oriented attributes include target altitude H, distance D, and flight path shortcut R, etc. These attributes are inversely proportional to the target threat TL; that is, the higher the target altitude H and the farther the distance D, the smaller the threat to our sites. An aerial target threat assessment model is constructed based on the benefit-oriented and cost-oriented attributes of aerial target threats.

[0056] Step 2: Introduce the cloud model into the Bayesian network. Use the cloud model to represent the probability of association between nodes in the Bayesian network, thereby expressing the degree of association between nodes. Specifically:

[0057] Step 2.1: Linguistic variables are used to handle fuzzy information about target attributes that cannot be quantitatively represented by Bayesian networks. Specifically:

[0058] Let S = {s} i Let |i=1,2,…,2τ+1,τ∈N} be a set of linguistic terms, where s i Let τ represent a time period, and N represent the set of positive integers. The set of linguistic terms must satisfy the following conditions:

[0059] 1) Orderliness: When i > k, s i >s k ;

[0060] 2) Negation operator: s i =neg(s k ), where i+k=2τ+1.

[0061] Step 2.2: Represent the language values ​​using a cloud model, thereby completing the cloud model transformation of the target attribute threat level. Specifically:

[0062] Based on battlefield environment and expert experience, this invention classifies target threat attribute levels into five levels: very high, high, medium, low, and very low. Let U = [X min ,X max Let S be the universe of discourse for the cloud model, where S = {s1, s2, ..., s}. 2τ+1 Given a set of linguistic terms, 2τ+1 normal cloud models can be obtained using the golden section method. For example, when τ=2, a set of linguistic terms with 5 labels can be obtained:

[0063] S = {s1 = Very low (VL), s2 = Low (L), s3 = Medium (M), s4 = High (H), s5 = Very high (VH)}

[0064] Step 2.3: Network node association probability P(TL) j The calculation of ) is as follows:

[0065] The Bayesian network representing the cloud model can be obtained based on steps 2.1 and 2.2. The correlation probability of nodes in a Bayesian network is also called the prior probability. This invention utilizes expert rating to obtain the prior probability of a target threat. Assuming a target's threat level is divided into j levels, and M decision-makers rate the target's threat level, then the prior probability P(TL) of each level of the target is... j )for:

[0066]

[0067] in, X j The number of decision-makers who consider the target threat to be at level j.

[0068] Step 3: Combining battlefield situation information, determine the target threat attribute weights using the improved Gaussian representation of variable weights. Specifically:

[0069] Traditional Gaussian representations, when used for weight calculations, often result in small differences between the weights of different attributes. However, in real battlefield environments, the importance of target attributes often varies significantly. Small differences in attribute weights lead to decreased discriminative power between targets, making it impossible to effectively differentiate the threat levels of each target. This invention improves the exponential part of the Gaussian representation, resulting in more discriminative attribute weights. The specific calculation formula is as follows:

[0070]

[0071]

[0072]

[0073] Where, μ n σ is the mean. n Let ω be the standard deviation, i = 1, 2, ..., n, and m = n / 2 be the number of target attributes. i =[ω1,ω2,…,ω m Normalization will yield the ascending weight θ of the target attribute. i :

[0074]

[0075] Where i = 1, 2, ..., m. At different times, the decision-maker ranks the importance of the target threat attributes, thus obtaining the target threat attribute weight w at each time point. i .

[0076] Step 4: Initialize parameter settings using the correlation probability between Bayesian network nodes and the target threat attribute weights, given the maximum number of rounds T for the algorithm to run. max The cloud model algorithm is used to determine the target threat level. Specifically:

[0077] Steps 2 and 3 yield the prior probability P(TL) of the node. j and the target threat attribute weights w at each time point i Combining the two yields the posterior probability P(TL) of the target threat level. j |v1,v2,…v n ):

[0078]

[0079] Among them, P(TL) j Let P(v) be the prior probability of the target threat. i |TL j ) represents the conditional probability of each node, w i The weights for each attribute.

[0080] Finally, the algorithm runs T max In each round, the highest posterior probability of the target threat level is selected from the calculation results of the cloud model operation rules. Then, the decision-maker judges the threat level of the target based on the magnitude of the target threat posterior probability.

[0081] By employing the above technical solutions, this invention addresses the problems of difficulty in threat assessment ranking and low threat assessment accuracy by proposing an aerial target threat assessment method based on a variable-weighted cloud Bayesian network. First, a threat assessment model of aerial target attributes is constructed based on a Bayesian network. Second, a cloud model is introduced into the Bayesian network to represent the correlation probability between Bayesian network nodes. Third, combining battlefield situational information, a variable-weighting method with an improved Gaussian expression is used to determine the weights of target attributes for threat assessment. Finally, based on the correlation probability and target attribute weights, the cloud model's operational rules are used to obtain the target threat level decision.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing aerial target threat based on variable weighted cloud Bayesian networks, characterized in that: It includes the following steps: Step 1: Analyze the qualitative information of different target threats to obtain target threat attributes, and construct an aerial target threat assessment model based on this. The construction of the aerial target threat assessment model based on Bayesian networks in Step 1 specifically includes the following steps: Step 1.1: Describe the relationships between nodes using a Bayesian network structure G. Specifically: G = (V, E) Where V = {V1, V2, ..., V} n } is the set of nodes in a directed acyclic graph, which is an abstract description of the object it represents. E is the set of directed edges, which represent the relationships between the nodes. Step 1.2: Based on the node set V of the directed acyclic graph obtained in Step 1.1, use the conditional probability P to reflect the local probability distribution set of the relationships between nodes. Specifically: Among them, each node V i With V i The parent node given is not V i Any subset of nodes formed by descendant nodes is conditionally independent, Pa(V) i ) represents V i The direct parent node; Step 1.3: Analyze the qualitative information of different target threats to obtain target threat attributes, and construct an aerial target threat assessment model based on this. Specifically: Based on steps 1.1 and 1.2, the topology G of the Bayesian network can be obtained, and an air target threat assessment model can be constructed based on the benefit and cost attributes of air target threats. Among them, the benefit-type attributes include target type T, speed S, jamming capability J, and maneuverability M. These attributes are directly proportional to the target threat TL, that is, the stronger the offensiveness of the target type and the faster the movement speed, the greater the threat of the target to our position. Cost-related attributes include target altitude H, distance D, and route shortcut R. These attributes are inversely proportional to target threat TL, meaning that the higher the target altitude H and the farther the distance D, the smaller the threat to us. Step 2: Introduce the cloud model into the Bayesian network, and use the cloud model to represent the probability of association between nodes in the Bayesian network, thereby expressing the degree of association between nodes; Step 3: Combining battlefield situation information, determine the target threat attribute weights using an improved Gaussian representation with variable weighting; the specific calculation formula for the improved Gaussian representation with variable weighting in Step 3 is as follows: Where, μ n σ is the mean. n Let ω be the standard deviation, i = 1, 2, ..., n, and m = n / 2 be the number of target attributes; i =[ω1,ω2,…,ω m Normalization will yield the ascending weight θ of the target attribute. i : Where i = 1, 2, ..., m; at different times, the decision-maker ranks the importance of the target threat attributes to obtain the target threat attribute weight w at each time. i ; Step 4: Utilize the correlation probability between Bayesian network nodes and the target threat attribute weights, and use cloud model algorithms to determine the target threat level.

2. The aerial target threat assessment method based on variable weighted cloud Bayesian network according to claim 1, characterized in that: Step 2 involves introducing the cloud model into the Bayesian network, specifically including the following steps: Step 2.1: Process fuzzy information about target attributes that cannot be quantitatively represented by Bayesian networks using linguistic variables. Specifically: Let S = {s} i Let |i=1,2,…,2τ+1,τ∈N} be a set of linguistic terms, where s i Let τ represent the possible values ​​of the linguistic variable, τ represent a time period, and N represent the set of positive integers; the linguistic term set needs to satisfy the following conditions: 1) Orderliness: When node i > k, s i >s k ; 2) Negation operator: s i =neg(s k ), where i+k=2τ+1; Step 2.2: Represent the language values ​​using a cloud model, thereby completing the cloud model transformation of the target attribute threat level. Specifically: Based on battlefield environment and expert experience, the target threat level is divided into five levels: very high, high, medium, low, and very low; let U = [X min ,X max Let S be the universe of discourse for the cloud model, where S = {s1, s2, ..., s}. 2τ+1 } is a set of linguistic terms, and 2τ+1 normal cloud models can be obtained using the golden section method.

3. The aerial target threat assessment method based on variable weighted cloud Bayesian network according to claim 1, characterized in that: Step 2 uses a cloud model to represent the probability of association between nodes in the Bayesian network, thereby expressing the degree of association between nodes. This includes the following steps: Based on steps 2.1 and 2.2, the Bayesian network representing the cloud model can be obtained. The correlation probability of nodes in the Bayesian network is also called the prior probability. The prior probability of the target threat TL is obtained by using expert rating: Assuming that the threat level of a target is divided into j levels, and M decision-makers rate the threat level of the target, then the prior probability P(TL) of each level of the target threat is... j )for: in, X j This represents the number of decision-makers who consider the target threat to be at level j.

4. The aerial target threat assessment method based on variable weighted cloud Bayesian network according to claim 1, characterized in that: Step 4 specifically includes the following steps: The prior probabilities P(TL) of each target threat level are obtained through steps 2 and 3. j and the target threat attribute weights w at each time point i By combining attribute weights with prior probabilities, the posterior probability P(TL) of the target threat level is obtained. j |v1,v2,…v n Specifically: Among them, P(TL) j Let P(v) be the conditional probability of each level of target threat. i |TL j ) represents the conditional probability of each node, w i Weights for each attribute; Finally, the algorithm runs T max In each round, the highest posterior probability of the target threat level is selected from the calculation results of the cloud model operation rules. Then, the decision-maker judges the threat level of the target based on the magnitude of the target threat posterior probability.

Images