An improved bayesian network-based safety evaluation method for reconnaissance-strike integrated unmanned aerial vehicle

Abstract

In view of the problems of complex and diverse combat tasks of reconnaissance and attack integrated unmanned aerial vehicle, difficult to obtain test data, strong subjectivity of safety evaluation, etc., a safety evaluation method of reconnaissance and attack integrated unmanned aerial vehicle based on improved Bayesian network is proposed, which aims to evaluate the safety of reconnaissance and attack integrated unmanned aerial vehicle through the model. The method mainly includes: step 1. Constructing the safety evaluation index system of reconnaissance and attack integrated unmanned aerial vehicle; step 2. Improving the Bayesian network evaluation model. The present application firstly proposes to use the entropy method to improve the G1 method to obtain the prior probability of the root node, and uses the EM algorithm to obtain the conditional probability of the child node. The improved Bayesian network has the unique advantages of processing complexity and multivariate conditional probability distribution, which significantly enhances the adaptability and accuracy of the model in the environment of incomplete data, and can be applied to more extensive application scenarios.

Description

TECHNICAL FIELD

[0001] The application relates to the field of safety evaluation of unmanned aerial vehicles, and in particular to application of an improved Bayesian network evaluation method in safety evaluation of a reconnaissance and attack integrated unmanned aerial vehicle. BACKGROUND

[0002] The reconnaissance and attack integrated unmanned aerial vehicle integrates functions such as battlefield reconnaissance, remote attack, intelligence collection and electronic countermeasure, can accurately attack small and key targets, can provide ground target positioning and indication, can support air-ground joint operation and electronic countermeasure, and plays an important role in modern military operations. Since the reconnaissance and attack integrated unmanned aerial vehicle integrates multiple task functions and has high system complexity, it needs to coordinate multiple subsystems when performing tasks, which increases the operation difficulty and potential risks. The complexity of the system causes the reconnaissance and attack integrated unmanned aerial vehicle to be easily suppressed and deceived by the enemy in a strong countermeasure battlefield environment, so that the unmanned aerial vehicle faces safety threats such as loss of control, crash and forced landing in the use process. In the face of potential safety threats, effective evaluation of the use safety of the reconnaissance and attack integrated unmanned aerial vehicle and guidance of combat decision and safe use according to the evaluation results are the key to ensure the effective play of the combat effectiveness of the reconnaissance and attack integrated unmanned aerial vehicle.

[0003] On the other hand, the reconnaissance and attack integrated unmanned aerial vehicle undertakes diversified combat tasks such as reconnaissance and surveillance and attack on the ground, which usually need to be performed in complex and changeable battlefield environments, and the real battlefield environment is often difficult to be completely tested or simulated through a test platform, which leads to certain difficulties in obtaining test data of the reconnaissance and attack integrated unmanned aerial vehicle, so that the use safety evaluation process thereof highly depends on the experience and subjective judgment of experts, thereby affecting the fairness and objectivity of the evaluation. Therefore, it is crucial to select a use safety evaluation method that is in line with the characteristics of the combat tasks of the reconnaissance and attack integrated unmanned aerial vehicle. SUMMARY

[0004] In view of the current situation that the combat tasks of the reconnaissance and attack integrated unmanned aerial vehicle are complex and diversified, test data are difficult to obtain, and safety evaluation is highly subjective, the application provides a use safety evaluation method of the reconnaissance and attack integrated unmanned aerial vehicle based on an improved Bayesian network, which specifically comprises the following steps:

[0005] Step 1. Constructing a use safety evaluation index system of the reconnaissance and attack integrated unmanned aerial vehicle;

[0006] The combat phase of the reconnaissance and attack integrated unmanned aerial vehicle is specifically divided into five phases: 1. airport preparation; 2. take-off and climb; 3. entering the battlefield; 4. performing tasks; and 5. returning to the airport;

[0007] Based on the basic principles and methods of index system construction, combined with the opinions of experts in related fields, the index system is designed as a four-layer structure; The following key indicators are selected: command and control ability, emergency handling ability and health fatigue degree of ground station operators; Safety of anti-interference devices, reconnaissance payloads and weapon payloads; Safety of power systems, sensing systems and flight platforms and their own hardware and software failures; Endurance time, cruising speed and endurance height of UAVs; Adaptability of UAVs to harsh terrain, weather and electromagnetic environment; Anti-data link deception, anti-electromagnetic interference, anti-information attack, anti-satellite navigation deception and anti-sound wave interference capabilities of UAVs; Data encryption, information transmission and radar detection safety capabilities of UAVs; Network communication, track tracking and cooperative attack safety between multiple UAVs; Target positioning, attack and anti-interception capabilities of UAVs; Fault detection and self-repairing capabilities of UAVs;

[0008] Step 2. Improve the Bayesian network evaluation model;

[0009] (1) Bayesian network;

[0010] If the observed data D is known, let be the parameter to be estimated, then:

[0011] (1)

[0012] In the formula, is the posterior probability, is the prior probability, is the likelihood function, is the total probability of observation data D under all possible parameters, represents the observation data, represents the context content;

[0013] Formula (1) is transformed as:

[0014] (2)

[0015] In the formula, if the context content is ignored , the parameter to be estimated is denoted as A, and the observation data D is denoted as B, which represents the conditional probability of event B occurring under the condition that event A has occurred;

[0016] The Bayesian network is represented as:

[0017] (3)

[0018] In the formula, G represents the Bayesian network, and <> represents the set of Bayesian networks, Let be a set of nodes, where n is the number of nodes; g represents the dependencies between nodes and is an n×n matrix; p is a conditional probability table, representing the probability of each node x. i The conditional probability, where i is an integer ranging from 1 to n;

[0019] The joint probability is known from the chain method as follows:

[0020] (4)

[0021] In the formula, Denotes the joint probability. express The probability of an event occurring Indicates in Under the conditions of the event The probability of occurrence;

[0022] Bayesian networks consist of a set of variables The structure consists of each node corresponding to one variable in a set of variables; the joint probability is written as the product of local probabilities:

[0023] (5)

[0024] In the formula, Denotes the joint probability. parent node Next child node The conditional probability, It is a node The parent node;

[0025] (2) Determine prior probabilities using the G1 method based on entropy-enhanced order relation analysis;

[0026] Specifically as follows:

[0027] Step 1: Let the evaluation object be A, and the corresponding n evaluation indicators. There are m experts Participate in the evaluation, among which Let m represent the number of experts, and M represent the set of experts; if the evaluation index... The importance of the evaluation is greater than Then it is recorded as , This is a standard symbol, meaning that its importance is greater than;

[0028] Step 2: Experts select the most important indicators from the set of indicators, denoted as... ;

[0029] Step 3: Experts then select the most important indicator from the remaining n-1 indicators, denoted as... ;

[0030] Step4: Repeat the previous step until the last index; get the sequence relationship of all indexes after reordering:

[0031] (6)

[0032] Step5: Calculate the proportion of the jth index under the ith expert:

[0033] (7)

[0034] In the formula:

[0035] is the proportion of the jth index for the ith expert, the larger the value, the more important the index, and the greater the importance of evaluation, i=1,2,...,m, j=1,2,...,n;

[0036] is the evaluation of the jth index by the ith expert;

[0037] is the sum of the evaluation of the jth index by all experts;

[0038] Step6: Calculate the entropy value of the jth index

[0039] (8)

[0040] Step7: Determine the ratio of the importance of adjacent indexes:

[0041] (9)

[0042] In the formula, is the ratio of the entropy values of adjacent indexes, k=n,n-1,...,2;

[0043] Step8: Calculate the index weight, the weight of the last index is:

[0044] (10)

[0045] According to:

[0046] (11)

[0047] Backward calculation to get the weight of all indexes;

[0048] Step9: Set all levels of indexes as s, , ​respectively, L represents a set of grades of the index, and the membership of each index with respect to the grade is determined:

[0049] (12)

[0050] wherein, is the membership of the jthindex to the grade i, N is the total number of evaluators, and k is the number of evaluators whose scoring result of the jthindex belongs to the grade i;

[0051] Step 10: Calculate the prior probability of the root node belonging to each grade:

[0052] (13)

[0053] wherein, P represents the prior probability of the root node, is the weight of the jthindex in the n indexes, is the membership of the jthindex to the grade i;

[0054] (3) Determine the conditional probability based on the EM algorithm;

[0055] E step-expected step: under the current Bayesian model parameter estimate, the posterior probability of the hidden variable is calculated, that is, the expectation of the hidden variable, as the estimated value of the hidden variable now:

[0056] (14)

[0057] wherein, represents the observed variable in the ithsample, which corresponds to the observed node in the Bayesian network; represents the hidden variable in the ithsample, which corresponds to the unobserved node in the Bayesian network; represents the posterior distribution of the hidden variable under the known observation data and the current parameter;

[0058] M step-maximization step: according to the expected value calculated in the E step, the parameter is updated to maximize the log-likelihood function:

[0059] (15)

[0060] wherein, represents the maximum likelihood estimation function under the current parameter, i is each sample; in the M step, the posterior probability calculated in the E step is used to re-estimate the model parameter so as to maximize the maximum likelihood estimation function;

[0061] In the M step, the hidden variables need to be considered all possible values, and the expected value is calculated by weighting according to the posterior probability of them under the current parameters ; this means that, for each sample i, not only the actually observed data but also the influence of all possible values of the hidden variables is taken into account;

[0062] The M step decomposes the objective function into the sum of the logarithms of the conditional probabilities of the nodes, and the parameters of each node are updated independently. The E step and the M step are iterated until the parameters converge. Through the correspondence of formula (15), the EM algorithm can transform the uncertainty of the hidden variables into the basis for updating the conditional probability parameters, gradually optimize the Bayesian network model, and finally approximate the true distribution of the data.

[0063] The advantages of the present application are as follows:

[0064] (1) According to the combat characteristics of the reconnaissance and attack integrated unmanned aerial vehicle, it is divided into five combat stages, the safety risk factors in the use process of different stages are identified, and the reconnaissance and attack integrated unmanned aerial vehicle use safety evaluation index system in strong confrontation environment is constructed;

[0065] (2) Based on the index system, the topological structure of the Bayesian network is determined, the entropy method is used to improve the G1 method to obtain the prior probability of the root node for the first time, and the EM algorithm is used to obtain the conditional probability of the child node for the first time; only the two probabilities are calculated, the use safety of the unmanned aerial vehicle can be evaluated by using the Bayesian network;

[0066] (3) Simulation verification shows that the improved Bayesian network significantly enhances the adaptability and accuracy of the model in the environment of incomplete data, and can be applied to more extensive application scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 is the combat stage of the reconnaissance and attack integrated unmanned aerial vehicle;

[0068] Figure 2 is the combat execution process of the reconnaissance and attack integrated unmanned aerial vehicle;

[0069] Figure 3 is the use safety evaluation index system of the reconnaissance and attack integrated unmanned aerial vehicle;

[0070] Figure 4 is the use safety evaluation flowchart;

[0071] Figure 5 is the local Bayesian network of the C3 node;

[0072] Figure 6 is the simulation result graph, wherein Figure 6 (a) shows the use safety evaluation result graph, Figure 6(b) show the reverse reasoning result, Figure 6 (a) show the impact strength analysis result;

[0073] Figure 7 To evaluate the results of contrast. DETAILED DESCRIPTION

[0074] The present application will be further described in conjunction with examples and drawings.

[0075] The present application is based on improving the Bayesian network to evaluate the use safety of the reconnaissance-strike integrated unmanned aerial vehicle. Specifically, based on the combat characteristics of the reconnaissance-strike integrated unmanned aerial vehicle, it is divided into 5 combat stages, and the index system of the use safety evaluation of the reconnaissance-strike integrated unmanned aerial vehicle in the strong confrontation environment is constructed; based on the index system, the topological structure of the Bayesian network is determined, and the prior probability of the root node is obtained by using the entropy method to improve the G1 method, and the conditional probability of the child node is obtained by using the EM algorithm. Specifically as follows.

[0076] Step 1. Construct the index system of the use safety evaluation of the reconnaissance-strike integrated unmanned aerial vehicle;

[0077] A scientific and systematic index system is the basis of the use safety evaluation of the reconnaissance-strike integrated unmanned aerial vehicle, and plays a decisive role in the whole evaluation process. The present application focuses on the reconnaissance-strike integrated unmanned aerial vehicle, integrates its combat mission characteristics and the structural characteristics of large and medium-sized unmanned aerial vehicles, and according to the division of combat stages, the index system of the use safety evaluation of the reconnaissance-strike integrated unmanned aerial vehicle in the strong confrontation environment is constructed.

[0078] Combat stage refers to the division of the entire combat process into a series of relatively independent but interrelated stages according to different requirements and targets of the task in military operations. These stages help to systematize the management and execution of combat missions, and improve the combat efficiency and success rate. The combat stages of the reconnaissance-strike integrated unmanned aerial vehicle can be further divided into the following 5 stages, as shown in Figure 1 .

[0079] 1. Airport preparation.

[0080] The reconnaissance-strike integrated unmanned aerial vehicle mainly involves a comprehensive inspection of the unmanned aerial vehicle in this stage to ensure that it is in good condition, and according to the task requirements, loads corresponding weapons and sensors (such as high-definition photoelectric / infrared cameras, synthetic aperture radar, air-to-ground missiles, guided bombs, etc.), sets the flight route and combat plan.

[0081] 2. Take-off and climb

[0082] The take-off climbing stage is the first step for the reconnaissance and attack integrated UAV to enter the battlefield, which needs to consider multiple aspects such as endurance time, speed, height, and environmental factors. By reasonably planning the flight route and endurance time, reasonably adjusting the flight speed and height, and comprehensively evaluating the environmental factors, the reconnaissance and attack integrated UAV can ensure the best performance and safety in the take-off climbing stage.

[0083] 3. Entering the battlefield

[0084] The battlefield environment adaptation stage is the process of dynamically perceiving and adapting to the battlefield environment during the execution of the task by the reconnaissance and attack integrated UAV. The reconnaissance and attack integrated UAV usually integrates advanced reconnaissance, attack, and defense systems, and needs to perform tasks in a more complex combat environment. In order to ensure the safe execution of the reconnaissance and attack task, a series of advanced technical means and measures need to be taken to improve its anti-interference, anti-deception, and anti-attack capabilities.

[0085] 4. Executing the task

[0086] The task command and control stage refers to the real-time command and control link during the execution of the task by the reconnaissance and attack integrated UAV, to ensure the smooth progress of the task. The reconnaissance and attack integrated UAV needs to obtain real-time battlefield information, including the position and dynamics of enemy targets, adjust the attack strategy according to the battlefield information, ensure accurate attack, and transmit reconnaissance information to the ground command center in real time to support decision-making.

[0087] 5. Returning to the airport

[0088] The return stage is the process of the UAV safely returning to the base after completing the task. Since the reconnaissance and attack integrated UAV faces a more complex battlefield environment during the execution of the task, its return process needs to pay more attention to safety. The task flow corresponding to each stage is shown in Figure 2 .

[0089] In the reconnaissance and attack integrated UAV combat execution flow shown in Figure 2 , since the UAV enters the battlefield and executes the task at the same time, and the anti-countermeasure capability of the UAV in the strong confrontation environment is difficult to reflect in the execution flowchart, the battlefield environment adaptation stage and the task execution stage are combined into one stage. However, in the use of the safety evaluation index system, these are two stages that need to be evaluated separately, as they evaluate the performance and safety of the UAV from different angles, so they are divided into two stages as independent evaluation indexes.

[0090] Based on the analysis of the reconnaissance and attack integrated UAV combat use process, the risk factors in each combat stage are identified, and based on the basic principles and methods of index system construction, combined with the insights of experts in related fields, the following key indexes are finally selected to evaluate the safety of the reconnaissance and attack integrated UAV. For example,Figure 3 As shown in the figure, the index system is designed as a four-layer structure, and each index is numbered according to the top-down and left-to-right order principle. These indexes are the command and control ability, emergency handling ability and health fatigue level of the ground station operator; the safety of the anti-interference device, reconnaissance load and weapon load; the safety of the power system, sensing system and flight platform and its own software and hardware failure; the endurance time, cruising speed and endurance height of the UAV; the adverse terrain, weather and electromagnetic environment adaptability of the UAV; the anti-data link deception, anti-electromagnetic interference, anti-information attack, anti-satellite navigation deception and anti-sound wave interference ability of the UAV; the data encryption, information transmission and radar detection safety ability of the UAV; the networking communication, track tracking and cooperative attack safety between multiple UAVs; the target positioning, attack and anti-interception ability of the UAV; and the fault detection and self-repairing ability of the UAV. The use safety of the reconnaissance and attack integrated UAV is evaluated according to these indexes. The subsequent Bayesian network is also constructed according to these indexes.

[0091] The preparation stage covers three major elements. First, the professional ability and health status of the ground control personnel; second, the safety of equipment configuration, which specifically involves the reconnaissance ability, weapon load configuration and anti-interference device effectiveness; and finally, the ability to ensure the safety of the basic flight, which includes the stability of the power system, the accuracy of the sensing system, the reliability of the flight platform and the safety performance of the software and hardware system.

[0092] The take-off stage includes two dimensions. One is the flight performance of the reconnaissance and attack integrated UAV, covering its endurance time, flight speed and height; the other is the environmental adaptability of the UAV, specifically in adverse weather, complex terrain and electromagnetic environment.

[0093] The battlefield environment adaptation stage includes three aspects. First, the communication safety, that is, the ability of the reconnaissance and attack integrated UAV to resist electromagnetic interference, deceptive information and information attack; second, the navigation safety, including the ability to resist navigation deception and sound wave interference; and finally, the information safety, covering the data encryption technology, the robustness of signal transmission and the radar detection effectiveness.

[0094] Regarding the execution task stage, it mainly involves two major fields of the reconnaissance and attack integrated UAV, namely, multi-machine cooperative combat and command and control. Multi-machine cooperation needs to consider the reliability of networking communication, the accuracy of track tracking and the effectiveness of cooperative attack; command and control needs to focus on the accuracy of target positioning, attack effectiveness and anti-interception ability.

[0095] Finally, the return stage, which needs to monitor the possible faults of the reconnaissance and attack integrated UAV in flight in real time, and has certain self-repairing ability to ensure safe return.

[0096] Step 2. Improve the Bayesian network evaluation model;

[0097] (1) Bayesian Network. Bayesian Network (BN) is a statistical inference tool based on probabilistic graphical model, which is used to represent the conditional dependence between variables. It describes the causal relationship between variables in the form of Directed Acyclic Graphs (DAG), where nodes represent random variables and edges represent the dependence between variables.

[0098] Bayes' theorem is a basic theorem in probability theory, which describes how to update the probability of an event after observing new evidence. Given the observed data D, let be the parameter to be estimated, then

[0099] (1)

[0100] where is the posterior probability, is the prior probability, is the likelihood function, represents the observed data, represents the context content. Bayes' formula shows that when one conditional probability is difficult to obtain, and another conditional probability is relatively easy to obtain, the inverse conditional probability can be calculated from one conditional probability.

[0101] Equation (1) can be transformed as:

[0102] (2)

[0103] represents the conditional probability of event B occurring given that event A has occurred. It can be inferred that the probability of event A occurring given that event B has occurred.

[0104] Bayesian network can be represented as:

[0105] (3)

[0106] where G represents the Bayesian network, <> represents the set of Bayesian networks, is all the nodes; g represents the dependence between nodes, which is an n x n matrix; p is the conditional probability table, which represents the conditional probability of each node x i , i is an integer between 1 and n (n is the number of nodes).

[0107] By the chain rule, the joint probability is:

[0108] (4)

[0109] where represents the joint probability, denotes the probability of an event occurring, denotes the probability of an event occurring in a given condition.

[0110] A Bayesian network is composed of a set of variables , each node corresponds to a variable in the set of variables. Because each node is independent of the nodes other than its parent nodes, the use of conditional independence can greatly reduce the amount of calculation, that is, the joint probability can be written as the product of local probability:

[0111] (5)

[0112] where is the parent node of node .

[0113] (2) The prior probability is determined by the G1 method improved based on entropy value. The order relation analysis method, also known as the G1 method, is a kind of subjective weighting method. It is an order evaluation method, which determines the importance degree between two indexes through experts, and determines the weight of the index. The order relation is easy to understand, can be applied to quantitative or qualitative data of various types, and can clearly show the relative order between objects, which is convenient for further analysis. However, the G1 method is often affected by the subjective bias of the evaluator, and the entropy method is based on the objective calculation of the data itself. The combination of the entropy method and the order relation analysis method can reduce the influence of human factors on the analysis results and improve the objectivity of the results. The main operation steps are as follows:

[0114] The following 10 steps are the steps of the improved G1 method used in the Bayesian network. The present application first proposes to improve the G method by using the entropy method, and then obtains the prior probability of the root node of the Bayesian network by using the membership weighted method.

[0115] Step 1: Let the evaluation object be A, corresponding to n evaluation indexes , m experts participate in the evaluation, where respectively represent the m experts, and M represents the set of experts. If the evaluation importance degree of the evaluation index is greater than , it is recorded as , is a standard symbol, which means that the importance degree is greater than.

[0116] Step 2: The expert selects the most important index from the index set, denoted as .

[0117] Step 3: The expert selects the most important index from the remaining n-1 indexes, denoted as​ .

[0118] Step4: Repeat the last step (i.e. Step3) until the last index. In this way, we get the sequence relationship of all indexes after reordering:

[0119] (6)

[0120] Step5: Calculate the proportion of the jth index under the ith expert (assuming there are 5 experts and 5 indexes. The jth index under the ith expert means: the proportion of the 3rd index evaluated by the 2nd expert. At this time, i=2, j=3):

[0121] (7)

[0122] where:

[0123] is the proportion of the jth index under the ith expert (the proportion indicates the importance of a certain index, the greater the value, the more important the index, and the greater the importance of the evaluation).(i=1,2,...,m, j=1,2,...,n)

[0124] is the evaluation of the jth index by the ith expert.(i=1,2,...,m, j=1,2,...,n)

[0125] is the sum of the evaluations of the jth index by all experts.

[0126] Step6: Calculate the entropy value of the jth index :

[0127] (8)

[0128] Step7: Determine the ratio of the importance of adjacent indexes:

[0129] (9)

[0130] where, is the ratio of the entropy values of adjacent indexes, also known as the modality operator.(k=n, n-1,...,2)

[0131] In the sequence relationship method, the modality operator is an important concept, which is mainly used to express the degree of affirmation in language. Currently, the 9-level operator method is commonly used in China, which divides the objects of evaluation or decision into 9 levels, each level corresponds to an operator or fuzzy set.

[0132] Traditional sequence relationship method often relies on subjective judgment in determining mood operators, leading to subjectivity and inconsistency of evaluation results. The introduction of entropy method determines the weight through the information of the data itself, avoiding the drawbacks of subjective weighting.

[0133] Step8: Calculate the weight of the last index

[0134] (10)

[0135] According to:

[0136] (11)

[0137] The weight of all indicators can be obtained by back calculation.

[0138] Step9: Set the total number of index grades to s, , respectively represent the grades of the index, L represents the set of index grades, and the membership degree of each index relative to the grade is determined:

[0139] (12)

[0140] where, is the membership degree of the jth index to the ith grade, N is the total number of evaluators, and k is the number of evaluators whose score of the jth index belongs to the ith grade.

[0141] Step10: Calculate the prior probability of the root node belonging to each grade:

[0142] (13)

[0143] where P represents the prior probability of the root node, is the weight of the jth index in the n indicators, is the membership degree of the jth index to the ith grade.

[0144] (3) EM algorithm is a method for parameter estimation of models containing latent variables. This algorithm is mainly used in the case of missing data, and through iterative optimization of model parameters. EM algorithm provides a powerful tool for handling missing data and latent variables, especially in Bayesian networks, by effectively estimating conditional probabilities, improving the predictive ability and application effect of the model. EM algorithm includes two main steps:

[0145] Step1: Expectation step (E step): In Bayesian networks, the conditional probability of a child node is represented by the parameters in its conditional probability table (CPT). Specifically, these parameters are part of the model parameter set, usually denoted by the symbol​ denotes the posterior probability of the hidden variable estimation, the posterior probability of the hidden variable , i.e. the expectation of the hidden variable, is calculated as the current estimate of the hidden variable:

[0146] (14)

[0147] where, denotes the hidden variable, denotes the observed variable, is the model parameter, denotes the expectation of the hidden variable , denotes the posterior distribution of the hidden variable given the observed data and the current parameter.

[0148] Step 2: Maximization step (M-step): update the parameter according to the expectation calculated in the E-step to maximize the log-likelihood function.

[0149] (15)

[0150] The E-step mainly focuses on calculating the expectation of the hidden variable , while the M-step focuses on re-estimating the parameters of the model according to these expectations. These two steps are alternated until the algorithm converges to a local optimal solution .

[0151] 3. Example analysis

[0152] The application selects a reconnaissance and attack integrated unmanned aerial vehicle. Before performing a task, the overall use safety level of the unmanned aerial vehicle is evaluated to infer the possibility of successful task performance. In addition, there are many uncertain factors in the process of performing the task, and the strategy needs to be adjusted at this time. According to GJB900A “General Requirements for Equipment Safety”, the risk severity level is divided into four levels, namely catastrophic, serious, mild and slight. Therefore, the use safety of the reconnaissance and attack integrated unmanned aerial vehicle is divided into four levels. The use safety level A indicates that the safety is very high, and only a slight risk exists. The use safety level B indicates that the safety is general, but a small accident may occur. The use safety level C indicates that the safety level is low, and a more serious accident may occur. The use safety level D indicates that the safety is extremely low, and once a problem occurs, it will cause serious loss and irreversible harm.

[0153] Take the basic support factor C3 of the index system of the unmanned combat aircraft as an example, which is a root node of the Bayesian network. At this time, C3 has four parent nodes, which are the safety of the power system D7, the safety of the sensor system D8, the safety of the flight platform D9 and the fault of the software and hardware D10. Five experts in the field of unmanned aircraft safety were invited to evaluate this time, and the scoring interval of each index was [0, 100]. After discussion, it was stipulated that the scoring interval [90, 100] was good, recorded as Good; the scoring interval [80, 90) was general, recorded as Medium; and the score below 80 was poor, recorded as Bad. The expert scoring results are shown in Table 1.

[0154]

[0155] The calculation results are shown in Figure 5 The probabilities of C3 index belonging to Good, Medium and Bad are 28.6%, 51.4% and 20% respectively. Using the BNT toolbox in MATLAB, the parameters of the Bayesian network can be obtained using the EM algorithm. Table 2 is the conditional probability table of B1 in the preparation stage, which has three parent nodes, namely the safety of human factors C1, the safety of equipment deployment C2 and the safety of basic support C3. Taking this as an example, the conditional probability distribution of all child nodes in the Bayesian network can be obtained.

[0156]

[0157] (1) Simulation results.

[0158] The commonly used Bayesian network simulation software includes GeNIe, Netica, Hugin, BayesiaLab, etc. Compared with other software, the advantages of GeNIe are its ease of use, powerful function, flexibility, open source and cross-platform support, so GeNIe is chosen for simulation calculation. The obtained parameters are input into the Bayesian network, and the simulation results are as follows Figure 6(a) shown. From the figure, it can be seen that the probabilities of the UAV using safety levels A, B, C and D are 35%, 43%, 13% and 9% respectively (the actual results are 35.1%, 43.3%, 13.1% and 8.5%, accurate to the individual digit of the percentage in the figure), and the overall use safety level is B, i.e. the use safety is general. The probabilities of the pre-preparation stage (B1), the take-off stage (B2), the battlefield environment adaptation stage (B3), the task command control stage (B4), the return stage (B5) being excellent are 51%, 44%, 45%, 50% and 52% respectively, and the probabilities of being general are 33%, 33%, 35%, 31% and 36% respectively. The values of Good and Medium in B1, B2, B3, B4 and B5 are subtracted respectively and represented by a, b, c, d and e, a = 18%, b = 11%, c = 10%, d = 19% and e = 16%, and the greater the difference, the better the performance. The values of b and c are relatively small, indicating that the safety of the take-off stage and the battlefield environment adaptation stage of the UAV needs to be improved.

[0159] (2) Simulation analysis.

[0160] By setting the LevelA probability of A to 100%, the model can be updated, and the results are shown in Figure 6 (b). From the figure, it can be seen that the values of Good of all nodes have been improved. For several nodes of A, the value of B1 for Good has been increased from the original 51% to 56%, an increase of 5 percentage points; the value of B2 for Good has been increased from the original 44% to 47%, an increase of 3 percentage points; the value of B3 for Good has been increased from the original 45% to 52%, an increase of 7 percentage points. The value of B4 for Good has been increased from the original 50% to 54%, an increase of 4 percentage points. This phenomenon further illustrates that the battlefield environment adaptation stage has a greater impact on the use safety of the reconnaissance and attack integrated UAV. The safety capability of the battlefield environment adaptation stage is mainly affected by C6 (communication security), C7 (navigation security) and C8 (information security), indicating that in order to improve the safety capability of the UAV, the security capability of the communication link of the UAV needs to be ensured.

[0161] The influence strength represents the degree of dependence between nodes, and the thicker the connection line between two nodes, the greater the influence strength. The analysis results of the influence strength are shown in Figure 6(c) shown in the figure. As can be seen from the figure, the node paths with the greatest influence intensity in the above risk network are: ① C3→B1; ② C5→B2; ③ C10→B4; ④ C11→B5. In the preparation stage, the basic support function of the reconnaissance-strike integrated UAV has a greater impact on it; environmental factors account for a larger proportion in the take-off stage; when the UAV is executing a task, if the target positioning failure rate is relatively high, it will lead to an increase in the use time and in turn increase the safety risk probability. Finally, in the return stage, the reconnaissance-strike integrated UAV needs to monitor possible faults in real time and have certain repair capabilities to ensure safe return.

[0162] In the preparation stage, the basic support function of the reconnaissance-strike integrated UAV needs to be checked, and corresponding weapons and sensors need to be loaded. As can be seen from the simulation results, the basic support function has relatively low safety, and all key devices and systems need to be thoroughly checked before executing a task; in the take-off stage, it can be seen that the flight performance of the reconnaissance-strike integrated UAV is relatively poor, but the environmental adaptability is good. This requires more detailed task planning to ensure that the reconnaissance-strike integrated UAV completes the task efficiently and safely; after the reconnaissance-strike integrated UAV enters the battlefield environment, the suppression, deception and other interference of the enemy will pose a threat to its use safety, and this stage has a greater impact on the use safety of the reconnaissance-strike integrated UAV. As can be seen from the simulation results, the communication safety, navigation safety and information safety of the UAV have similar degrees of influence, but the safety of the navigation system is relatively low, and more advanced navigation technology and equipment need to be used to improve the positioning accuracy and anti-interference ability of the UAV; the command and control stage and the return stage of the reconnaissance-strike integrated UAV have relatively small impact on the overall use safety of the reconnaissance-strike integrated UAV, and the use safety in the two stages is relatively high. Therefore, in the command and control stage and the return stage, the UAV is relatively stable and safe.

[0163] In summary, the basic support function of the reconnaissance-strike integrated UAV has relatively low safety, the flight performance is limited, and the anti-interference ability of the navigation system is weak. All key devices need to be thoroughly checked before executing a task, more advanced navigation technology and equipment need to be used, and more detailed task planning is needed during task execution to ensure the use safety of the reconnaissance-strike integrated UAV.

[0164] (3) Comparison of evaluation results.

[0165] The traditional Bayesian network directly determines the network parameter value through expert scoring and statistical methods, and the improved Bayesian network calculates the parameters through the entropy method to optimize the G1 method and the EM algorithm. The comparison of the two is shown in Figure 7 . As can be seen, although the evaluation results obtained by the two methods are consistent, the improved Bayesian network improves the efficiency and flexibility through automatic calculation and optimization algorithm, making the grade difference more obvious, and can be applied to reconnaissance-strike integrated UAVs with similar use safety, reducing the fuzziness of the evaluation results and being suitable for more extensive application scenarios.

[0166] According to the combat characteristics of reconnaissance-strike integrated UAV, it is divided into five combat stages, and the safety risk factors in different stages are identified. The safety evaluation index system of reconnaissance-strike integrated UAV is constructed in strong confrontation environment. Based on the index system, the topology structure of Bayesian network is determined. The prior probability of root node is obtained by using entropy method to improve G1 method. The conditional probability of sub-node is obtained by using EM algorithm. Finally, the data is imported into GeNie software for simulation. The probability distribution of different safety levels of reconnaissance-strike integrated UAV is obtained. The reverse reasoning and influence intensity analysis of the model are carried out to determine the key factors leading to safety accidents of reconnaissance-strike integrated UAV. The simulation results show that the improved Bayesian network has unique advantages in dealing with complexity and multi-variable conditional probability distribution, which significantly enhances the adaptability and accuracy of the model in the environment of incomplete data, and can be applied to more extensive application scenarios.

Claims

1. An improved method for evaluating the safety of a detect-and-attack integrated unmanned aerial vehicle using a Bayesian network, characterized in that, Specifically includes the following steps: Step 1. Constructing the safety evaluation index system of reconnaissance-strike integrated UAVs; The combat phase of reconnaissance-strike integrated UAVs is divided into five stages:

1. Airport preparation; 2. Take-off and climb; 3. Enter the battlefield; 4. Execute the task; 5. Return to the airport; Based on the basic principles and methods of index system construction, combined with the insights of experts in related fields, the index system is designed as a four-layer structure; The following key indicators are selected: command and control ability, emergency handling ability, and health and fatigue level of ground station operators; Safety of anti-interference devices, reconnaissance payloads, and weapon payloads; Safety of power systems, sensing systems, and flight platforms, as well as their own software and hardware failures; Endurance time, cruising speed, and endurance height of UAVs; Adaptability of UAVs to harsh terrain, weather, and electromagnetic environment; Anti-data link deception, anti-electromagnetic interference, anti-information attack, anti-satellite navigation deception, and anti-sound wave interference capabilities of UAVs; Data encryption, information transmission, and radar detection safety capabilities of UAVs; Network communication, track tracking, and cooperative attack safety between multiple UAVs; Target positioning, attack, and anti-interception capabilities of UAVs; Fault detection and self-repair capabilities of UAVs; Step 2. Improve the Bayesian network evaluation model; (1) Bayesian network; If the observation data D is known, let If the observation data D is known, let (1) wherein, is the posterior probability, is the prior probability, is the likelihood function, is the total probability of the observation data D under all possible parameters, denotes the observation data, denotes the context content; Equation (1) is transformed as: (2) In the formula, if the context is ignored The parameters to be estimated Let A be the observed data D and B be the observed data D. Let B represent the conditional probability of event B occurring given that event A has occurred. The Bayesian network is represented as: (3) In the formula, G represents a Bayesian network, and <> represents a Bayesian network set, is a node set, n is the number of nodes, g represents the dependency relationship between nodes, is an n*n matrix, p is a conditional probability table, represents the conditional probability of each node x i , and i is an integer with a value of 1~n. Through the chain method, the joint probability is: (4) wherein denotes the joint probability, denotes the probability of the event, denotes the probability of the event given the condition that the event occurs; A Bayesian network is composed of a set of variables Each node corresponds to one of the variables in the set; the joint probability is written as a product of local probabilities: (5) wherein denotes the joint probability, is the parent node the lower child node the conditional probability, is the parent node of the node . (2) Determine the prior probability based on the improved order relation analysis G1 method based on entropy value; Specifically as follows: Step 1: Set the evaluation object as A, corresponding to n evaluation indexes , m experts participate in the evaluation , where respectively represents m experts, M represents the set of experts; if the evaluation importance of the evaluation index is greater than , it is recorded as , is a standard symbol, which means that the importance is greater than Step 2: The expert selects the most important indicators from the indicator set, denoted as ; Step 3: The expert selects the most important indicator from the remaining n-1 indicators, denoted as ; Step4: Repeat the previous step until the last index; Get the reordering sequence of all indexes: (6) Step5: Calculate the proportion of the jth index under the ith expert: (7) In the formula: is the proportion of the ith expert to the jth index, the greater the value, the more important the index, and the greater the importance of the evaluation, i=1, 2,..., m, j=1, 2,..., n; Eijis the evaluation of the ith expert for the jth indicator; Sj is the sum of all experts' evaluations for the jth indicator; Step 6: Calculate the entropy value of the jth index : (8) Step7: Determine the ratio of the importance of adjacent indicators: (9) wherein is the ratio of the entropy values of the adjacent indicators, k = n, n - 1,..., 2; Step 8: Calculate the index weight, the weight of the last index is: (10) According to: (11) Inverse the weight of all indexes; Step 9: Let all the grades of the index be divided into s, , respectively, L represents the set of index grades, and the membership of each index with respect to the grade is determined: (12) wherein is the membership of the jth index belonging to the grade i, N is the total number of evaluators, and k is the number of evaluators whose scoring result of the jth index belongs to the grade i. Step10: Calculate the prior probability of the root node belonging to each level: (13) In the formula, P represents the prior probability of the root node, is the weight of the jth index in the n indexes, is the membership degree of the jth index belonging to the i th level. (3) Determine the conditional probability based on the EM algorithm; E-step - Expectation step: Estimate the posterior probability of the hidden variables given the current Bayesian model parameters and compute the hidden variables as their current estimates as the expectation of the hidden variables given the current model parameters. (14) wherein, represents the observed variables in the ith sample, corresponding to the observed nodes in the Bayesian network; represents the latent variables in the ith sample, corresponding to the unobserved nodes in the Bayesian network; represents the posterior distribution of the latent variables given the observed data and the current parameters; M-step - maximization step: update parameters to maximize the log-likelihood function: based on the expected values computed in the E-step ​ (15) wherein represents the maximum likelihood estimation function under the current parameters, i is each sample; in the M step, the model parameters are re-estimated by using the posterior probability calculated in the E step so as to maximize the maximum likelihood estimation function; In the M-step, the hidden variables need to be considered all possible values, and weighted by their posterior probabilities under the current parameters This means that for each sample i, not only the actually observed data but also the influence of all possible values of the hidden variables is taken into account; M-step decomposes the objective function into the sum of the logarithms of each node's conditional probability, and updates each node's parameters independently. E-step and M-step processes are repeated iteratively until the parameters converge; Through the corresponding relationship of equation (15), the EM algorithm can convert the uncertainty of hidden variables into the basis for updating conditional probability parameters, gradually optimize the Bayesian network model, and finally approximate the true distribution of data.

Images