Method for predicting accident rate of unmanned aerial vehicle system considering human error correlation and closed-loop feedback

Abstract

The present application provides a kind of unmanned aerial vehicle system accident rate prediction method considering human error correlation and closed loop feedback, it is related to unmanned aerial vehicle system safety assessment technical field, comprising the following steps: S1: determine analysis range;S2: the influencing factor of unmanned aerial vehicle system is classified;S3: the hierarchical relationship of influencing factor is structured;S4: obtain the feedback loop existing in unmanned aerial vehicle system;S5: obtain the causal relationship graph between each influencing factor of unmanned aerial vehicle system;S6: determine the quantitative relationship between variables when not considering feedback;S7: determine the quantitative relationship of node with dynamic feedback characteristics and continuous and discrete hybrid characteristics;S8: transform stock flow diagram;S9: drive model simulation, obtain the prediction value of unmanned aerial vehicle system accident probability.The present application avoids the disadvantages of insufficient accuracy caused by relying on single modeling in traditional accident prediction method, builds a complete theoretical prediction system, avoids artificial judgment intervention, and the model prediction accuracy is higher.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) system safety assessment technology, specifically to a method for predicting the accident rate of UAV systems that considers human error correlation and closed-loop feedback. Background Technology

[0002] Unmanned aerial vehicle (UAV) systems are integrated systems composed of aircraft, control stations, communication links, and mission payloads, and are widely used in various fields such as military, firefighting, agriculture, and logistics. A complete UAV system mainly includes four core components: The system comprises an aircraft subsystem, a control station subsystem, a communication link system, and a mission payload system. Compared to other systems, unmanned aerial vehicle (UAV) systems have two significant characteristics: First, accidents involving unmanned aerial vehicles (UAVs) are highly correlated with human factors, with approximately 75%-80% of UAV crashes related to human error, including insufficient skills, poor teamwork, and lack of situational awareness. This is drastically different from the role of "humans" in traditional equipment, where they typically play a peripheral role in operation or maintenance. In UAV systems, personnel are the core subsystem; errors such as misjudging altitude or inadequate maintenance can directly lead to accidents through complex interaction chains.

[0003] Secondly, the core functions of the UAV system, such as scheduling, mission planning, and status monitoring, are all executed in a coordinated manner by the Flight Management System (FCMS), through which operators and maintenance personnel interact with the UAV. For example, when the UAV exhibits abnormal attitude, the FCMS dynamically issues an alarm to the operator, who then adjusts the control commands based on the alarm information. These commands are processed by the FCMS to drive the UAV to correct its attitude. For maintenance personnel, the FCMS will indicate any abnormal UAV status, allowing them to perform repairs according to the prompts, thus completing the maintenance of the UAV system.

[0004] Considering the above two characteristics, when predicting the accident rate of unmanned aerial vehicle (UAV) systems, two new features need to be considered: human error correlation and dynamic feedback.

[0005] Human error correlation can be categorized into direct and indirect correlations. Indirect human error correlation primarily stems from the shared impact of the FCMS (Functional Control System) on different personnel. The FCMS provides operational and maintenance prompts, which influence the experience and safety awareness of different personnel, thus affecting their individual probability of human error. Direct human error correlation is mainly mediated by the equipment's technical condition. The equipment's technical condition is primarily affected by maintenance personnel, whose maintenance skills determine the machine failure rate. If maintenance personnel make mistakes leading to inadequate maintenance, the probability of machine failure will increase. The equipment's technical condition also affects operator error; for example, when equipment malfunctions and becomes unavailable during a mission, operators may need to implement alternative solutions, significantly increasing the mission difficulty and thus increasing the probability of operator error. Therefore, in long-cycle missions, a complex human error correlation network is formed among maintenance personnel, operators, drones, and the FCMS.

[0006] The dynamic feedback capability stems from the influence of the FCMS (Front-Load Management System). The FCMS collects system operational data, personnel operational behavior, and drone operational data to monitor the system's status. Based on this, feedback to relevant personnel, reminding them to conduct training and establish relevant regulations, can enhance personnel's safety awareness, thereby indirectly improving the overall safety of the drone system. Because drone accident rates are dynamically changing, the feedback from the FCMS also changes dynamically. Dynamic feedback is a typical characteristic of drone systems.

[0007] How to establish an effective constraint mechanism that takes into account the correlation between human error and dynamic feedback has become a research hotspot in the prediction of unmanned aerial vehicle (UAV) system accident rates. Currently, UAV system safety assessment methods mainly include event logic modeling, systematic analysis modeling, multi-agent modeling, and system dynamics simulation modeling.

[0008] Considering the human error correlation and closed-loop feedback characteristics of long-cycle tasks, the above methods each have their drawbacks: event logic modeling methods cannot describe dynamic feedback characteristics; systematic analysis modeling methods are mainly qualitative or semi-quantitative analysis and cannot output the changes of safety indicators over time; although multi-agent simulation and system dynamics modeling can model the dynamic feedback characteristics of the system, multi-agent simulation adopts a bottom-up modeling approach and lacks a macroscopic modeling perspective of the system; the causal and quantitative relationships of influencing factors in system dynamics methods rely on expert judgment and lack theoretical guidance. Summary of the Invention

[0009] To address the shortcomings of the existing technology, the present invention aims to provide a method for predicting the accident rate of unmanned aerial vehicle (UAV) systems that considers human error correlation and closed-loop feedback. This method combines event logic modeling and system dynamics methods to model the human error correlation and closed-loop feedback characteristics in long-cycle tasks, thereby determining the UAV accident evolution path and its probability of occurrence, and supporting safety assessments during the design phase.

[0010] Specifically, on the one hand, the present invention provides a method for predicting the accident rate of an unmanned aerial vehicle (UAV) system that considers human error correlation and closed-loop feedback, characterized by comprising the following steps: S1: Define the scope of analysis, clarify the research objectives and boundaries; S2: Classify the influencing factors of unmanned aerial vehicle (UAV) systems using human factors analysis and classification methods; S3: Introduce an explanatory structural model to structure the hierarchical relationships of influencing factors and determine the correlations between various influencing factors, including the following sub-steps: S31: Construct the adjacency matrix ; S32: Calculate the reachability matrix The calculation formula is as follows: ; In the formula, I represents the identity matrix, whose order is equal to that of the adjacency matrix. If the order is the same, then the adjacency matrix will be... Add the matrix to the identity matrix I and then perform Boolean algebra operations on the matrix itself until the result remains unchanged. At this point, the calculated result is the reachability matrix. n represents the number of operations until the result no longer changes; S33: Divide the hierarchical structure: based on the reachability matrix The reachable set is obtained by partitioning. , the previous set and common factors set If the reachability set of influencing factors is the same as the set of common factors, then they belong to the same level in the reachability matrix. Cross out the rows and columns corresponding to the same level of influencing factors, and repeat this process to obtain the next level of influencing factors until all influencing factors are crossed out, and draw a hierarchical structure diagram of the influencing factors of UAV system safety. S4: Determine the feedback path of the UAV system, and combine the correlation between the various influencing factors in S3 to obtain the feedback loops existing in the UAV system; S5: Based on the feedback loops existing in the UAV system in S4, obtain the causal relationship diagram between the various influencing factors of the UAV system; S6: Based on the causal relationship diagram in S5, determine the quantitative relationship between variables without considering feedback by using a hybrid causal logic method; S7: Based on the causal relationship graph in S5, determine the quantitative relationship between nodes with dynamic feedback characteristics and a mixture of continuous and discrete characteristics; S8: Combining the quantitative relationships determined in S6 and S7, the causal relationship diagram in S5 that qualitatively expresses the correlation between variables is transformed into a stock flow diagram calculated quantitatively. S9: Input the expression of the quantitative relationship determined by S6 and S7 into the variables corresponding to the stock flow diagram to drive the model simulation and obtain the predicted value of the accident probability of the UAV system.

[0011] Furthermore, the factors affecting unmanned aerial vehicle systems in S2 include latent effects, unsafe surveillance, warning signs of unsafe behavior, and unsafe behavior itself.

[0012] Furthermore, the feedback paths of the UAV system in S4 include: feedback from the UAV accident rate to the flight management system, feedback from the UAV accident rate to the safety awareness of maintenance personnel, and feedback from the UAV accident rate to the safety awareness of operators.

[0013] Furthermore, in S6, the quantitative relationships between variables include: causal logic modeling of human error and its correlation, causal logic modeling of machine failure, and accident evolution logic modeling.

[0014] Furthermore, methods for modeling human error and its correlation with causal logic include: Calculate the final human error probability: ; In the formula, and These represent the final human error probability and the base human error probability, respectively. This represents the maximum influence value of the i-th experience performance formation factor. This represents the proportion of the impact assessment of the i-th experience performance formation factor; Performance factors influencing the probability of human error include safety awareness and experience, which are calculated as follows: Calculate the factors that influence changes in safety awareness: ; In the formula, SA represents personnel safety awareness, OCTS represents the degree of importance the flight management system attaches to safety, ROCtS represents the reference value of the degree of importance the flight management system attaches to safety, OtoP represents the impact index of the flight management system on personnel safety, and T1 represents the time required for personnel safety awareness to change. Calculate the empirical variation factor: ; In the formula, Ex represents the experience level of the personnel, and Tr represents the level of technical training provided by the flight management system.

[0015] Furthermore, methods for causal logic modeling of machine failures include: Calculate the probability of system failure when the failure logic is AND gate or OR gate; Calculate the actual mean time between failures (MTBF) of a machine. : ; In the formula, The mean time between failures (MTBF) represents the machine's design time. These are variables affected by the probability of human error. Calculate the failure probability F(x) of each subsystem: F(x) = 1 - exp(-t / MTBF); In the formula, t is the subsystem running time, and MTBF represents the mean time between failures (MTBF), which is obtained by statistical methods.

[0016] Furthermore, methods for modeling the logic of accident evolution include accident occurrence probability. for: ; In the formula, This represents the probability of failure of the UAV mission payload system; The probability of human error where the operator is unaware of the risk; The probability of human error representing incorrect operation by the operator; This represents the probability of failure in the drone control system.

[0017] Furthermore, methods for determining the quantitative relationship between nodes include: dividing the personnel safety awareness performance shaping factor into three levels: low, medium, and high, then the proportion of personnel safety awareness in the assessment... for: ; In the formula, , and The impact ratio of the representative personnel's safety awareness performance shaping factor at different levels is determined.

[0018] Furthermore, the methods for driving model simulation in S9 include: determining the initial values ​​of level variables, simulation time, and simulation step size based on the available data range, and predicting the accident probability value of the UAV system.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention addresses the human error correlation and dynamic feedback in long-cycle missions by proposing a system dynamics-based method for evaluating the safety of unmanned aerial vehicle (UAV) systems. It incorporates human factor analysis and classification methods to identify key factors affecting UAV system safety at the system level. An interpretive structural model (ESM) is introduced to structure the hierarchical relationships of these factors, and qualitative models are built to identify the correlations between them. Based on this qualitative modeling, a hybrid causal logic method is used to model the human error correlation and causal logic relationships of influencing factors in long-cycle missions to determine the quantitative relationships between variables. For the closed-loop feedback characteristics present in long-cycle UAV systems, a stock flow diagram considering feedback from the flight management system is established using system dynamics. Quantitative relationships are input into the variables corresponding to this stock flow diagram to obtain the predicted accident probability of the UAV system. This method establishes a UAV accident rate prediction model that can predict UAV accident rates, avoiding the shortcomings of traditional accident prediction methods that rely on single quantitative or qualitative models, which result in insufficient accuracy. It also overcomes the drawback of system dynamics methods where the causal and quantitative relationships of influencing factors depend on expert judgment. A complete theoretical prediction system is constructed, and the model's prediction accuracy is high compared to actual UAV data. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the human error correlation model for long-cycle tasks that considers closed-loop feedback, as disclosed in this invention. Figure 2 This is a causal logic modeling diagram of human error correlation and influencing factors in long-cycle missions without considering feedback from the flight management system, as disclosed in this invention. Figure 3 This is a causal relationship diagram between various influencing factors of the unmanned aerial vehicle system in S5 of the present invention; Figure 4 This is the stock flow diagram determined in S8 of the present invention; Figure 5 This is a flowchart of the human-machine system safety modeling method based on system dynamics in this invention; Figure 6 This is a hierarchical structure diagram of the factors influencing drone accidents in this invention; Figure 7 This is a causal relationship diagram of the factors affecting drone accidents in this invention; Figure 8 This invention uses a hybrid causal logic method to determine the relationship graph between variables when feedback relationships are not included. Figure 9 This is a stock flow diagram of the UAV accident rate prediction model in this invention; Figure 10 This is a comparison chart of predicted data and actual data in this invention.

[0021] The attached image uses English abbreviations, the meanings of which are as follows: PSF stands for Performance Shaping Factor. ESD stands for Event Sequence Diagram, a time series diagram. FT stands for Fault Tree. BN stands for Bayesian Network. SD stands for System Dynamics. HFACS stands for Human Factor Analysis and Classification System. ISM stands for Interpretive Structural Modeling. HCL stands for Hybrid Causal Logic. MTBF stands for Mean Time Before Failure. APOA stands for Assessed proportion of affect. Detailed Implementation

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0023] First, the modeling principle of the unmanned aerial vehicle (UAV) system accident probability prediction method of this invention is introduced: 1. Overview of the human error correlation model for long-cycle tasks considering closed-loop feedback; Human error correlation in long-cycle missions primarily stems from the common-factor influence of the flight management system (FMS) on different personnel. The FMS provides operational and maintenance prompts; if these prompts are frequent, the experience-based performance formation factors and safety awareness performance formation factors of different personnel will change accordingly, thus affecting their individual human error probabilities. Furthermore, for operators, an additional equipment technical condition performance formation factor must be considered. Equipment technical condition also affects operator human error; for example, when equipment malfunctions and becomes unavailable during the mission, operators may need to implement alternative solutions, significantly increasing mission difficulty and thus increasing the probability of operator human error. Equipment technical condition is mainly affected by maintenance personnel; the maintenance skills of maintenance personnel determine the machine failure rate. If maintenance personnel make mistakes leading to inadequate maintenance, the probability of machine failure will increase. Therefore, in long-cycle missions, a complex human error correlation network is formed among maintenance personnel, operators, machines, and the flight management system. Figure 1 In the diagram, the black lines represent the correlation between human error and the causal relationships between human, machine, environment, and flight management system factors in long-term missions.

[0024] Analysis of the correlation between human error and system error reveals that the flight management system (FMS) has the core influence. The FMS collects data on system operation, personnel behavior, and UAV operation to monitor the system's status. Based on this, by providing feedback to relevant personnel and reminding them to conduct training and establish relevant regulations, it can enhance personnel's safety awareness, thereby indirectly improving the overall safety of the UAV system. At this point, the system simultaneously contains continuous variables such as changes in failure rate and discrete variables such as performance formation factor levels, making it a hybrid system. Figure 1 In the diagram, the red line represents the feedback of the system's safety status to the flight management system, and the double-ring nodes represent discrete nodes, mainly some performance-forming factor nodes that affect human error.

[0025] 2. Modeling the correlation between human error and long-cycle tasks; Among the influencing factors analyzed in this invention, the causal relationships between machine factors typically follow deterministic logic, meaning their interactions and influences are predictable. However, when human factors are involved, the causal relationships exhibit uncertainty; that is, the occurrence of some factors does not necessarily lead to the occurrence of others. In the hybrid causal logic method, Bayesian networks are typically used to model the causal logic of human error and its correlation, while fault trees are used to model the causal logic of machine failure. Furthermore, the occurrence of both human error and machine failure affects the accident evolution process; in the hybrid causal logic method, time series graphs are typically used to model the accident evolution logic. The machine failure probability calculated by fault trees and the human error probability calculated by Bayesian networks can support the calculation of relevant nodes in the time series graph. In summary, a causal logic model of human error correlation and influencing factors for long-cycle tasks based on the hybrid causal logic method can be obtained, such as… Figure 2 As shown.

[0026] 3. Modeling the closed-loop feedback characteristics in long-cycle tasks; For closed-loop feedback characteristics, modeling in system dynamics is relatively simple, which is a line connecting the node "accident rate" to the node "the degree of safety importance attached to the flight management system". Currently, the feedback process of the accident rate to the flight management system is generally considered to follow the following process: (1) The flight management system has an accident rate reference value. If the actual accident rate exceeds the reference value, it will have a significant impact on the degree of safety importance attached to the flight management system. Therefore, the accident rate participates in the feedback process in a relative manner. Its formula is shown in Equation 1 in Table 1. (2) The flight management system has different levels of learning from accidents. Therefore, there is an accident learning index of the flight management system to describe the impact of accidents on the flight management system. According to existing research, the impact of the accident rate on the flight management system is an exponential relationship. Its formula is shown in Serial No. 2 in Table 1. (3) After learning from accidents, the flight management system will adjust the target value of the degree of safety importance attached to the flight management system, but the target value will not exceed the maximum value of the degree of safety importance attached to the flight management system. Its formula is shown in Serial No. 3 and Serial No. 4 in Table 1. (4) The flight management system integrates the target value of the flight management system's emphasis on safety with the current emphasis on safety, and takes into account the adjustment time of the flight management system's emphasis on safety, to determine the change in the flight management system's emphasis on safety at each simulation step. This adjustment time depends on the specific circumstances of the flight management system and varies from several days to several months. The formula is shown in Serial No. 5 in Table 1. (5) Finally, using the INTEG function, the changes in the flight management system's emphasis on safety at each time slice are accumulated to the initial value to obtain the current emphasis on safety of the flight management system. The formula is shown in Serial No. 6 in Table 1. Combining the above six formulas, the calculation formula for the impact of the accident rate on the flight management system feedback is obtained.

[0027] Table 1. Function Expressions in the FCMS Security Priority Module

[0028] The following are the specific steps of the UAV accident rate prediction method considering human error correlation and closed-loop feedback proposed in this invention: S1: Define the scope of analysis; Defining the scope of analysis requires clearly defining the research objectives and boundaries. For unmanned aerial vehicle (UAV) systems, the research objective can generally be set as the operational status or safety level of the UAV system. The research boundaries can be defined in conjunction with the research objectives. For example, for UAV systems, the research scope can be defined as the impact of operation, maintenance, and management processes on the safety of the UAV system.

[0029] S2: Identify factors affecting the safety of human-machine systems based on human factors analysis and classification methods; After defining the scope of analysis, it is necessary to identify the factors affecting the UAV. Human-machine systems often involve numerous factors, thus requiring a classification system to support the identification of these influencing factors. Here, we introduce the Human Factors Analysis and Classification System (HFAS) classification method. The HFAS classification method is divided into four levels: latent effects, unsafe monitoring, warning signs of unsafe behavior, and unsafe behavior. Latent effects focus on describing the impact of the flight management system's status and behavior on system safety; these are the hidden causal factors affecting system safety. Unsafe monitoring refers to the flight management system's failure to effectively monitor and promptly correct errors during the actual process, including inadequate supervision and insufficient risk control and correction. Warning signs of unsafe behavior refer to factors that induce unsafe behavior by personnel, including personnel and equipment factors. Unsafe behavior refers to potential unsafe behaviors by personnel, including operational errors and violations.

[0030] S3: Determine the correlation and structure of influencing factors based on the interpretive structural model method; Currently, after identifying the relevant influencing factors of unmanned aerial vehicle (UAV) systems, researchers mainly rely on subjective judgment to determine the correlations between these factors. However, such subjective judgments inevitably suffer from inaccuracies or omissions. To address this issue, this embodiment introduces an interpretive structural model (ESM). The ESM method is a structured approach for analyzing hierarchical relationships among elements of complex systems. Its core is to transform the direct and indirect relationships between elements into a multi-level topological hierarchy graph through matrix operations, such as adjacency matrices and reachability matrices. ESMs have been widely applied in social sciences, management, engineering, and other fields to help researchers and decision-makers identify and analyze the influence and interactions of various factors within a system.

[0031] S4: Determine the system feedback loop based on feedback analysis; The main feedback paths in unmanned aerial vehicle (UAV) systems can be divided into two types: learning and adaptation feedback and management feedback. Learning and adaptation feedback primarily targets individuals, who gradually accumulate experience during operation. Simultaneously, unsafe incidents require individuals to examine their own actions, thereby improving their operational behavior and preventing similar incidents from recurring. This feedback mechanism typically involves individual reflection and adjustment of operational behavior, with effects reflected in improved individual operational skills and enhanced safety awareness. Management feedback, on the other hand, focuses more on adjustments and interventions at the flight management system level. When the flight management system identifies safety issues or accident risks, it will prompt the development of regulations and the implementation of safety training to provide feedback and guidance to individuals or teams, correcting unsafe behaviors and improving overall safety levels. This feedback mechanism involves the flight management system's ability to identify, analyze, and respond to safety issues, as well as the enforcement and supervision of regulations. In system dynamics, feedback paths are generally represented by red lines, i.e., causal bonds in system dynamics. For feedback paths like flight management system feedback, which have a longer time frame and a certain delay effect, the causal bond also needs to be marked with a "Delay" symbol represented by two vertical lines.

[0032] After identifying the feedback path, the feedback loops present in the system can be determined by combining the hierarchical causal structure in S3. In nature, feedback loops in a system are divided into two categories: positive feedback loops and negative feedback loops. Positive feedback loops reinforce system behavior, while negative feedback loops regulate system behavior, acting as stabilizers. Positive and negative feedback loops are typically determined by the number of causal bonds. For example, if variables A and B are positively correlated, a "+" sign is used to indicate the end of the causal bond when they are connected by a causal bond. Conversely, a "-" sign is used. Generally, a feedback loop with an even number of negative causal bonds is considered a positive feedback loop, and a feedback loop with an odd number of negative causal bonds is considered a negative feedback loop.

[0033] S5: Determine the cause-and-effect diagram; A causal relationship diagram can be obtained through the analysis of S3 and S4, such as... Figure 3 As shown.

[0034] S6: Determine the quantitative relationship between variables without considering feedback based on the hybrid causal logic method; There are three types of quantitative relationships: causal logic modeling of human error and its correlation, causal logic modeling of machine failure, and accident evolution logic modeling. For causal logic modeling of human error and its correlation, human error assessment and reduction techniques are typically used for calculation. For machine failure logic, different machine failures are generally used as top nodes. The system is decomposed into subsystems based on functional decomposition, and different subsystem failures are connected using AND or OR gates to describe the system failure modes. For the accident evolution process, it generally starts with an initial event, followed by several key event branches. The occurrence of some events is conditional on the occurrence of others. These events include human error events and machine failure events, and their probabilities can be calculated using the next layer of fault trees and Bayesian networks.

[0035] S7: Determine the quantitative relationship between nodes with dynamic feedback characteristics and a mixture of continuous and discrete characteristics; After considering closed-loop feedback characteristics, the variables are expanded in the time domain. At this point, most variables are converted to continuous variables. However, performance shaping factors, because they use rating scales, must be described using discrete variables. In system dynamics, the IF-THEN-ELSE function can be used to write different values ​​of the discrete variables into the variables.

[0036] S8: Determine the stock flow diagram; Based on the quantitative descriptions in S6 and S7, the causal relationship diagram can be converted into a stock-flow diagram. For example, Figure 3 The corresponding stock flow diagram can be established as follows: Figure 4 As shown.

[0037] S9: Safety evaluation of human-machine systems; Based on the stock flow diagram, a human-machine system safety assessment can be conducted. Variables that still need to be determined in the simulation include: the initial values ​​of the level variables; otherwise, the system dynamics simulation cannot begin recursion; and the simulation step size. The value of is required; otherwise, system dynamics simulation cannot be calculated, and the simulation step size... The value of will affect the accuracy of the calculation; variables must be calculated in order, that is, the level variables must be calculated first before the rate variables can be calculated.

[0038] The above steps allow for a safety assessment of the unmanned aerial vehicle (UAV) system. The results of this assessment can provide decision support for the management team. By analyzing the assessment results, the management team can better identify investment priorities, resource allocation, and risk management strategies, thereby improving the scientific rigor and effectiveness of decision-making. The modeling process of the human-machine system safety modeling method based on system dynamics proposed in this invention is shown below. Figure 5 System dynamics modeling is performed using the modeling tool Vensim.

[0039] The following preferred embodiment illustrates the specific steps of the UAV accident rate prediction method that considers human error correlation and closed-loop feedback disclosed in this invention: Example 1 S1: Define the scope of analysis; Defining the scope of analysis requires clarifying the research objectives and boundaries. For human-machine system safety analysis, the research objective can generally be set as the operational state or safety level of the human-machine system. The research boundaries can be defined in conjunction with the research objectives.

[0040] Specifically regarding unmanned aerial vehicles (UAVs), an UAV system is a complex human-machine system influenced by multiple factors. "Humans" includes operators and maintenance personnel, "machines" primarily refers to the UAV and its flight management system, and "environment" refers to the physical environment affecting UAV flight and personnel maintenance and operation. The direct causes of UAV accidents are usually unsafe acts by personnel and unsafe conditions of the machine. The environment and flight management system typically do not directly affect safety events but indirectly influence system safety by affecting safe personnel behavior and safe machine conditions.

[0041] S2: Identify factors affecting the safety of human-machine systems based on human factors analysis and classification methods; After defining the scope of the analysis, it is necessary to identify the factors affecting the human-machine system. Human-machine systems often involve numerous factors, thus requiring a classification system to support the identification of these influencing factors. Here, we introduce human factors analysis and classification methods. These methods are divided into four levels: latent influences, unsafe monitoring, warning signs of unsafe behavior, and unsafe behavior. Flight management system influences focus on describing the impact of the flight management system's status and behavior on system safety; these are latent causal factors affecting system safety. Unsafe monitoring refers to the flight management system's failure to effectively monitor and promptly correct errors during the actual process, including inadequate monitoring and insufficient risk control and correction. Warning signs of unsafe behavior refer to factors that induce unsafe behavior by personnel, including personnel and equipment factors. Unsafe behavior refers to potential unsafe behaviors by personnel, including operational errors and violations.

[0042] For unmanned aerial vehicle (UAV) systems, the influencing factors identified using human factors analysis and classification methods include the following categories: The primary factor influencing the latent impact level is the degree of importance the flight management system (FMS) places on safety. This emphasis reflects the shared values, beliefs, and behavioral norms within the FMS. A safety-first approach effectively promotes safe behaviors; conversely, neglecting safety increases the likelihood of accidents.

[0043] The primary factor influencing the level of unsafe supervision is technical training. The purpose of technical training is to improve the operational skills of operators and the maintenance capabilities of maintenance personnel.

[0044] Factors influencing early warning signs of unsafe behavior include the experience level of operators and maintenance personnel, operator safety awareness, and the technical condition of the equipment. Personnel experience accumulates with tasks and training. Technical training increases operator safety awareness. Maintenance capabilities determine the failure rate of the UAV system. As for equipment factors, the UAV system can be decomposed according to its functions. The UAV system mainly includes the mission payload system and the flight control system, each of which consists of many subsystems.

[0045] Unsafe acts. Unsafe acts here mainly refer to human error by operators and human error by maintenance personnel. Human error by operators refers to situations where pilots make mistakes due to insufficient skill or machine malfunction. Human error by maintenance personnel refers to situations where maintenance personnel fail to perform their duties properly due to lack of experience and safety awareness.

[0046] The influencing factors identified based on human factors analysis and classification methods are shown in Table 2.

[0047] Table 2. Influencing factors identified using the HFACS method

[0048] S3: Determine the correlation and structure of influencing factors based on the interpretive structural model method; Currently, after identifying the relevant influencing factors of unmanned aerial vehicle (UAV) systems, researchers mainly rely on subjective judgment to determine the correlations between these factors. However, such subjective judgments inevitably suffer from inaccuracies or omissions. To address this issue, the Interpretive Structural Model (ISM) method is introduced. ISM is a structured approach for analyzing hierarchical relationships among elements of complex systems. Its core is the use of matrix operations, such as adjacency matrices and reachability matrices, to transform the direct and indirect relationships between elements into a multi-level topological hierarchy. ISM has been widely applied in social sciences, management, engineering, and other fields to help researchers and decision-makers identify and analyze the influence and interactions of various factors within a system. If the complete set of influencing factors identified in S2 is... Therefore, the correlation and structure of influencing factors can be determined according to the following steps: S31: Construct the adjacency matrix If influencing factors right If it has an impact, then it is in the adjacency matrix. middle, If there is no effect, then Adjacency matrix It consists of only 0 and 1.

[0049] S32: Calculate the reachability matrix The reachability matrix represents whether a path exists between two influencing factors. The reachability matrix can be calculated using the chain multiplication method, which involves adding the adjacency matrix to the identity matrix and then repeatedly performing Boolean algebra operations with itself until the result no longer changes. The formula for calculating the reachability matrix is ​​shown in equation (1).

[0050] (1) S33: Divide the hierarchical structure. Based on the reachability matrix. The reachable set can be obtained by partitioning. , the previous set and common factors set Among them, the reachable set Represents the reachability matrix Influencing factors The set of influencing factors corresponding to the value 1 in the current row; the antecedent set. Represents the reachability matrix Influencing factors The set of influencing factors corresponding to the value 1 in the column; common factor set. This indicates the influence factors. The reachable set is the intersection of the reachable set and the previous set. If the reachable set of an influencing factor is the same as the common factor set, then these influencing factors belong to the same level. The rows and columns corresponding to these influencing factors are removed from the reachable matrix, and this process is repeated to obtain the next level's set of influencing factors until all influencing factors have been removed.

[0051] The drone safety model focuses on the drone accident rate as its output. Therefore, in addition to the factors affecting drone safety identified in Table 2, the drone accident rate needs to be added as a factor, denoted as A. Thus, the complete set of influencing factors is... Applying the explanatory structural model method: S31: Construct the adjacency matrix. Determine the universal set. The correlation between the influencing factors can be used to obtain the adjacency matrix. : ; S32: Calculate the reachability matrix For the adjacency matrix Perform a Boolean product operation to obtain the reachability matrix. : ; S33: Divide the hierarchical structure. Based on the reachability matrix. The results of the first hierarchical processing are shown in Table 3. It can be seen that only the reachability set of influencing factor A is equal to the common factor set; therefore, influencing factor A is in the first level. (In the reachability matrix...) Cross out the row and column containing influencing factor A, and perform a second hierarchical processing. The results are shown in Table 4. It can be seen that the reachability set of E1 is equal to the common factor set, therefore influencing factor E1 belongs to the second level. Following this pattern, the hierarchical division is shown in Table 5.

[0052] Table 3 Results of the first hierarchical processing

[0053] Table 4 Results of the second hierarchical processing

[0054] Table 5. Hierarchical Division

[0055] Based on the hierarchical divisions in S33 and combined with the reachability matrix, a hierarchical structure diagram of the factors influencing UAV safety can be drawn, such as... Figure 6 As shown.

[0056] S4: Determine the system feedback loop based on feedback analysis; The main feedback paths in the system can be divided into two types: learning and adaptation feedback and management feedback. Learning and adaptation feedback primarily targets individuals, who gradually accumulate experience during operation. Simultaneously, unsafe events require individuals to examine their own actions, thereby improving their operational behavior and preventing similar incidents from recurring. This feedback mechanism typically involves individual reflection and adjustment of operational behavior, with its effect manifested in improved individual operational skills and enhanced safety awareness. Management feedback, on the other hand, focuses more on adjustments and interventions at the flight management system level. When the flight management system identifies safety issues or accident risks, it provides feedback and guidance to individuals or teams through regulations, safety training, and other means to correct unsafe behaviors and improve overall safety levels. This feedback mechanism involves the flight management system's ability to identify, analyze, and respond to safety issues, as well as the enforcement and supervision of regulations. In system dynamics, feedback paths are generally represented by red lines, known as causal bonds. For feedback paths like flight management system feedback, which have a longer time frame and a certain delay effect, the causal bond also needs to be marked with a "Delay" symbol represented by two vertical lines.

[0057] After identifying the feedback path, the feedback loops present in the system can be determined by combining the hierarchical causal structure in S3. In nature, feedback loops in a system are divided into two categories: positive feedback loops and negative feedback loops. Positive feedback loops reinforce system behavior, while negative feedback loops regulate system behavior, acting as stabilizers. Positive and negative feedback loops are typically determined by the number of causal bonds. If variables A and B are positively correlated, a "+" sign is used to indicate the end of the causal bond when they are connected by a causal bond. Conversely, a "-" sign is used. Generally, a feedback loop with an even number of negative causal bonds is considered a positive feedback loop, and a feedback loop with an odd number of negative causal bonds is considered a negative feedback loop.

[0058] For drones, there are three feedback paths within the drone system: feedback from the drone accident rate to the flight management system, feedback from the drone accident rate to maintenance personnel's safety awareness, and feedback from the drone accident rate to operators' safety awareness. The feedback from the drone accident rate to the flight management system's level of safety awareness constitutes the flight management system-level feedback path. When the drone accident rate increases, the flight management system recognizes the seriousness of the safety issue, thereby increasing its focus on safety. This may trigger a series of safety improvement measures, such as strengthening technical training and improving regulations, ultimately reducing the drone accident rate. This feedback path reflects the flight management system's ability to learn from accidents and make adaptive changes. The feedback from the drone accident rate to maintenance personnel's safety awareness and to operators' safety awareness constitute the individual-level feedback paths. An increase in the accident rate prompts maintenance and operators to enhance their safety awareness, perform tasks more cautiously, and reduce human error, thus lowering the accident rate. This feedback path reflects the individual's ability to learn from experience and adjust their behavior.

[0059] Based on the above three feedback paths, and combined with Figure 6 The hierarchical structure in the data allows for the analysis of 15 feedback loops in the human-machine system, which are: Negative feedback loop 1, A: UAV system accident ↑ → P3: Operator safety awareness ↑ → E1: Operator error ↓ → A: UAV system accident ↓; Negative feedback loop 2, A: UAV system accident ↑ → O1: FCMS's level of safety emphasis ↑ → P3: Operator's safety awareness ↑ → E1: Operator error ↓ → A: UAV system accident ↓; Negative feedback loop 3, A: UAV system accident ↑ → P3: Operator safety awareness ↑ → P1: Operator experience ↑ → E1: Operator error ↓ → A: UAV system accident ↓; Negative feedback loop 4, A: UAV system accident ↑ → P4: Maintenance personnel safety awareness ↑ → E2: Maintenance personnel error ↓ → P5: UAV malfunction ↓ → A: UAV system accident ↓; Negative feedback loop 5, A: UAV system accident ↑ → P4: Maintenance personnel safety awareness ↑ → P2: Maintenance personnel experience ↑ → E2: Maintenance personnel error ↓ → P5: UAV malfunction ↓ → A: UAV system accident ↓; Negative feedback loop 6, A: UAV system accident ↑ → P4: Maintenance personnel safety awareness ↑ → E2: Maintenance personnel error ↓ → P5: UAV malfunction ↓ → E1: Operator error ↓ → A: UAV system accident ↓; Negative feedback loop 7, A: UAV system accident ↑ → O1: FCMS's level of safety emphasis ↑ → P3: Operator's safety awareness ↑ → P1: Operator's experience ↑ → E1: Operator error ↓ → A: UAV system accident ↓; Negative feedback loop 8, A: UAV system accident ↑ → O1: FCMS's level of safety emphasis ↑ → R1: Technical training ↑ → P1: Operator experience ↑ → E1: Operator error ↓ → A: UAV system accident ↓; Negative feedback loop 9, A: UAV system accident ↑ → O1: FCMS's level of safety emphasis ↑ → P4: Maintenance personnel's safety awareness ↑ → E2: Maintenance personnel's error ↓ → P5: UAV malfunction ↓ → A: UAV system accident ↓; Negative feedback loop 10, A: UAV system accident ↑ → P4: Maintenance personnel safety awareness ↑ → P2: Maintenance personnel experience ↑ → E2: Maintenance personnel error ↓ → P5: UAV malfunction ↓ → E1: Operator error ↓ → A: UAV system accident ↓; Negative feedback loop 11, A: UAV system accident ↑ → O1: FCMS's level of safety emphasis ↑ → P4: Maintenance personnel's safety awareness ↑ → P2: Maintenance personnel's experience ↑ → E2: Maintenance personnel's error ↓ → P5: UAV malfunction ↓ → A: UAV system accident ↓; Negative feedback loop 12, A: UAV system accident ↑ → O1: FCMS's level of safety awareness ↑ → P4: Maintenance personnel's safety awareness ↑ → E2: Maintenance personnel's error ↓ → P5: UAV malfunction ↓ → E1: Operator's error ↓ → A: UAV system accident ↓; Negative feedback loop 13, A: UAV system accident ↑ → O1: FCMS's level of safety emphasis ↑ → R1: Technical training ↑ → P2: Maintenance personnel experience ↑ → E2: Maintenance personnel error ↓ → P5: UAV malfunction ↓ → A: UAV system accident ↓; Negative feedback loop 14, A: UAV system accident ↑ → O1: FCMS's level of safety emphasis ↑ → P4: Maintenance personnel's safety awareness ↑ → P2: Maintenance personnel's experience ↑ → E2: Maintenance personnel's error ↓ → P5: UAV malfunction ↓ → E1: Operator's error ↓ → A: UAV system accident ↓; Negative feedback loop 15, A: UAV system accident ↑ → O1: FCMS safety emphasis ↑ → R1: Technical training ↑ → P2: Maintenance personnel experience ↑ → E2: Maintenance personnel error ↓ → P5: UAV malfunction ↓ → E1: Operator error ↓ → A: UAV system accident ↓; In this context, ↑ represents an increase, ↓ represents a decrease, and → represents being affected by this variable.

[0060] S5: Determine the cause-and-effect diagram; Based on the analysis results of comprehensive qualitative modeling, a causal relationship diagram among the influencing factors of unmanned aerial vehicle (UAV) system accidents can be obtained, such as... Figure 7 As shown, the qualitative modeling is now complete.

[0061] S6: Determine the quantitative relationship between variables without considering feedback based on the hybrid causal logic method; The correlations of influencing factors determined by the mixed causal logic approach are as follows: Figure 8 As shown, the introduction is as follows: (1) Modeling the causal logic of human error and its correlation; First, all human error quantification methods are based on human error assessment and reduction methods. According to the human error assessment and reduction methods, the calculation formula for human error is shown in equation (2): (2) In the formula, and Represents the final human error probability and the base human error probability. This represents the maximum impact value of the i-th performance-forming factor. This represents the impact assessment ratio of the i-th performance formation factor. For maintenance personnel, experience and safety awareness are the two main performance formation factors, while for operators, the equipment technical condition performance formation factor also needs to be considered.

[0062] Regarding the correlation between human error and safety, it is mainly achieved by flight management system factors influencing personnel experience and safety awareness. In this study, the degree of safety emphasis placed on the flight management system is used as a variable to characterize safety variables at the flight management system level. Regarding the relationship between the degree of safety emphasis placed on the flight management system and personnel safety awareness, if we use OCtS to represent the degree of safety emphasis placed on the flight management system, ROCtS to represent the reference value of the degree of safety emphasis placed on the flight management system, OtoP to represent the index of the flight management system's impact on personnel safety, and T1 to represent the time required for a change in personnel safety awareness, then the change in personnel safety awareness... It can be expressed as in equation (3): (3) As shown in equation (3), the flight management system has a reference value for the level of safety importance, and the flight management system will adjust its impact on personnel safety awareness based on this reference value. Meanwhile, the change in personnel safety awareness is a gradual process, and this change will consume time T1. Regarding the relationship between the flight management system's level of safety importance and personnel experience, if Tr represents the level of technical training provided by the flight management system, then the change in personnel experience... It can be expressed as in equation (4): (4) As can be seen from equation (4), if the flight management system provides sufficient reminders and the relevant technical training level is high, personnel can become more familiar with the equipment operation procedures, master more troubleshooting and handling techniques, and thus improve their experience level. At the same time, the enhancement of individual safety awareness can prompt personnel to pay more attention to the accumulation of daily experience in their daily work, so as to reduce erroneous operations.

[0063] For unmanned aerial vehicle (UAV) systems, the parameters that still need to be determined in the above formula include the nominal human error probability. The maximum impact value of performance formation factors And the proportion of impact assessment of performance formation factors. : Nominal human error probability This can be derived from Table 6. For human error, specifically the failure of operators to perceive risk, the corresponding description in Table 6 is "C1: Simple response to alarms in clearly indicated situations, requiring simple diagnosis; the response may involve directly performing simple operations or initiating other operations that require separate evaluation." Therefore, its... The value is 0.0007. For maintenance errors by maintenance personnel, the corresponding description in Table 6 is "B1: Routine inspection of equipment status," therefore its... The value is 0.03. For operator error, the corresponding description in Table 6 is "A3: Start or reconfigure the system according to procedure with feedback from the local control panel," therefore its... The value is 0.003.

[0064] Table 6. P(NHE) for different tasks

[0065] Maximum impact value This can be obtained using the NARA method. According to the NARA method, the performance formation factors for personnel safety awareness... The value is 2, representing the performance formation factor based on experience. The performance formation factor for equipment technical status is 8. The value is 4.

[0066] Impact assessment ratio There is currently no unified reference for the value of APOA. Here, based on existing research and experience evaluation, we provide the APOA values ​​for some performance formation factors, as shown in Table 7.

[0067] Table 7. APOA values ​​for different performance formation factors

[0068] For the correlation between human error and safety, equations (3) and (4) are mainly used for calculation. For equation (3), the parameters that need to be determined are the reference value of the flight management system's emphasis on safety, the coefficient of the flight management system's influence on personnel safety, and the time required for personnel safety awareness to change. According to existing research, the reference value of the flight management system's emphasis on safety is generally assumed to be 0.8, while the coefficient of the flight management system's influence on personnel safety is generally between 0.3 and 0.5, which is assumed to be 0.4 here. As for the time required for personnel safety awareness to change, there is no direct reference, so publicly available data from the internet is used here. According to publicly available data from the internet, the rectification time after an accident is about 2 to 3 months, so the time required for personnel safety awareness to change is uniformly set at 3 months.

[0069] (2) Causal logic modeling of machine failures; For machine fault logic, different machine faults are generally used as top nodes. Based on functional decomposition, the system is decomposed into subsystems, and different subsystem faults are connected using AND or OR gates to describe the system fault modes. Let... Let the failure probability of subsystem 11 be denoted as . Let represent the failure probability of subsystem 12. Then, when the failure logic is an AND gate, what is the probability of system 1 failing? It can be expressed as equation (5): (5) And set Let the failure probability of subsystem 21 be denoted as . Let represent the failure probability of subsystem 22. Then, when the failure logic is an "OR" gate, the probability of system 2 failing can be expressed as equation (6): (6) Furthermore, the subsystem failure probability can be calculated from the Mean Time Between Failures (MTBF). In actual use, machines typically cannot achieve the designed MTBF and are affected by the performance of maintenance personnel. If using... The machine design MTBF represents the actual machine performance. It can be expressed as equation (7): (7) In the formula, the influence coefficient of maintenance level on MTBF is... This is a variable influenced by the maintenance skill level of the maintenance personnel. In this study, the probability of human error by maintenance personnel is used to represent their maintenance skill level. When the probability of human error by maintenance personnel is in different intervals... The values ​​are also different.

[0070] Specifically regarding unmanned aerial vehicle (UAV) systems, according to publicly available literature, UAVs generally consist of a payload system and a flight control system. A failure in the payload system will prevent it from detecting risks. Structurally, the payload system is mainly divided into two subsystems: the payload subsystem and the information transmission subsystem. The payload subsystem plays a crucial role in information acquisition and processing, and can be further subdivided into information acquisition equipment and information countermeasure equipment. The information acquisition equipment is responsible for collecting data from the external environment, while the information countermeasure equipment focuses on processing and countering potential interference or threats. The information transmission subsystem ensures the effective transmission and reception of information. It consists of several key components, including an antenna servo system, antenna transceivers, and data processing equipment. The antenna servo system is responsible for the precise pointing and stable tracking of the antenna, the antenna transceivers handle signal transmission and reception, and the data processing equipment analyzes and processes the collected data to ensure the accuracy and reliability of the information. On the other hand, the flight control system also plays a vital role in the operation of the UAV, primarily responsible for precise flight control. The flight control system can be further subdivided into two main subsystems: the control subsystem and the navigation subsystem. The control subsystem is primarily responsible for the UAV's flight operations and path planning. It includes a flight planning system and a power plant system. The flight planning system is responsible for developing the UAV's flight path and mission plan, while the power plant system provides the necessary power support to ensure the UAV flies along the predetermined route and speed. The navigation subsystem focuses on providing the UAV with accurate positioning and navigation information, including an inertial navigation system, a satellite navigation system, and an electro-optical navigation system. The inertial navigation system uses sensors such as accelerometers and gyroscopes to determine the UAV's position and attitude, the satellite navigation system achieves precise positioning by receiving GPS or other satellite navigation signals, and the electro-optical navigation system uses optical sensors and image processing technology to provide navigation information. These subsystems work together to ensure the UAV executes its mission according to the predetermined route and plan, while adapting to various complex flight environments. (This section is visible...) Figure 8 Fault logic of the mission payload system and fault logic of the flight control system.

[0071] To calculate the failure probability of each subsystem, the design MTBF and average sortie time of the UAV need to be determined. For MTBF, referring to publicly available statistical data, the MTBF of other subsystems without data support was estimated, resulting in the design MTBF of each system at the start of the simulation, as shown in Table 8. The MTBF also changes during the simulation period. Although the overall MTBF data for the UAV system is missing, based on existing data analysis, the UAV system's MTBF increased from 58 to 74 over five years, an average annual increase of 5.5%. Here, it is assumed that the growth trend of the subsystem MTBF is consistent with that of the UAV system, also maintaining an annual growth of 5.5%. As for the average sortie time, combining publicly available sortie data for a certain UAV, the total sortie time and number of sorties for some fiscal years are shown in Table 9. For missing fiscal year data, the average sortie time of the preceding and following fiscal years was used for calculation. Finally, the failure probability of each subsystem follows an exponential distribution (…). Based on the above assumptions, the failure probability of each subsystem can be calculated.

[0072] Table 8. MTBF values ​​for each subsystem of the UAV

[0073] Table 9. Deployment data of a certain UAV

[0074] Furthermore, according to equation (7), the maintenance skill level of maintenance personnel also affects MTBF. As for the influence coefficient of maintenance skill level on MTBF... Currently, statistical support is lacking, and data from other sources must be relied upon. According to existing research, the level of maintenance in the aerospace and defense sectors has an impact of approximately 20-30% on MTBF. Since the simulation began after the formal establishment of the UAV unit, by which time the UAVs had been in use for a considerable period and a relatively systematic maintenance plan had been developed, we assume here that the level of maintenance personnel has a 20% impact on MTBF. Specifically, when the probability of human error by maintenance personnel is between 0 and 0.001, The value is 0.95; when the probability of human error by maintenance personnel is between 0.001 and 0.01, The value is 0.90; when the probability of human error by maintenance personnel is between 0.01 and 0.015, The value is 0.85; when it is greater than 0.015, The value is 0.8.

[0075] Regarding the logical structure between machines, except for the inertial navigation system, satellite navigation system, and electro-optical navigation system which use an "OR" structure, the logic between the remaining subsystems uses an "AND" structure. Combining these structures, the failure probabilities of the flight control system and mission payload system can be calculated.

[0076] (3) Accident evolution logic modeling; For unmanned aerial vehicle (UAV) systems, the accident evolution follows a logic of UAV detection – operator perception – operator decision-making – UAV control. First, the UAV relies on its mission payload system to detect internal and external risks. Then, the operator receives information from the machine, performs perception and analysis to determine the nature and severity of the accident. Next, the operator makes appropriate decisions based on their experience and judgment, deciding how to respond to the accident. Finally, these decisions are rapidly implemented automatically by the UAV's actuators, effectively controlling and mitigating the risk. An error in any of these stages can lead to an accident. This section is as follows... Figure 8 The accident evolution section is shown.

[0077] Regarding the quantitative calculation formula for the accident evolution process, if the failure probability of the UAV mission payload system is... The probability of human error where the operator is unaware of the risk is: The probability of human error due to incorrect operation by the operator is: Probability of drone control system failure The probability of the accident occurring is: (8) S7: Determine the quantitative relationship between nodes with dynamic feedback characteristics and a mixture of continuous and discrete characteristics; After considering the closed-loop feedback characteristics, the variables will be expanded in the time domain. At this time, most variables will be converted into continuous variables. However, since the performance formation factor uses a rating scale, it must be described by discrete variables. In equation (2), the impact assessment ratio varies depending on the level of the PSF. In system dynamics, the IF-THEN-ELSE function can be used to write different values ​​of the impact assessment ratio into the variables. For example, the personnel safety awareness performance formation factor can be divided into three levels: low, medium, and high. If SA is used to represent personnel safety awareness, the functional expression of the impact assessment ratio of personnel safety awareness can be expressed as equation (9): (9) In the formula, , and The evaluation proportions of the impact of the safety awareness performance factor at different levels are determined. A two-level IF-THEN-ELSE language is used here. In the first level IF-THEN-ELSE, if the safety awareness PSF... SC The initial level is low, and then SA_APOA takes the value APOA. SA1 If security awareness PSF SCFor medium to high security levels, the second-level IF-THEN-ELSE is activated. In the second-level IF-THEN-ELSE, if security awareness PSF... SC When the level is medium, then SA_APOA takes the value APOA. SA2 If security awareness PSF SC When the level is high, then SA_APOA takes the value APOA. SA3 .

[0078] The quantitative modeling is now complete.

[0079] S8: Determine the stock flow diagram; By combining the quantitative relationships determined in S6 and S7, the causal relationship diagram that qualitatively expresses the correlation between variables can be transformed into a stock-flow diagram that can be quantitatively calculated, such as... Figure 9 As shown, the variables of this model are shown in Table 10.

[0080] Table 10. Key Variables in the Drone Accident Rate Prediction Model

[0081] S9: Safety evaluation of unmanned aerial vehicle systems; Inputting the variable expressions from quantitative modeling into the corresponding variables in the stock flow diagram can drive the model simulation. Finally, the simulation timeframe still needs to be determined. Here, the simulation timeframe is chosen as fiscal year 2003 to fiscal year 2010, a total of 84 months, with a simulation step size of one month. It is worth noting that, according to publicly available data, a certain UAV officially entered service in 1995, but was in the testing phase from 1995 to 2000, with the total number of units consistently remaining in single digits, such as 6 units in fiscal year 1996 and 7 units in fiscal year 1999. By the end of fiscal year 2002, the number of UAVs had reached double digits, reaching 21 units, and thereafter entered service at an average rate of 20 units per fiscal year. It can be said that after fiscal year 2003, the UAV system began to enter a phase of large-scale deployment and use. Therefore, choosing fiscal year 2003 to fiscal year 2010 as the simulation timeframe can effectively reflect the changes in the accident rate of the UAV during this period and its influencing factors.

[0082] According to publicly available reports, accident rate data for a certain UAV between fiscal years 2003 and 2010 can be obtained. Using this as a benchmark, the simulation results of the model can be verified. Figure 10 The table shows a comparison between predicted and actual data. Table 11 shows the error between the actual accident rate and the model's predicted accident rate. Compared with the actual data, the root mean square error (RMSE) in Table 11 is 2.74. It can be seen that the model data and simulation data maintain a high degree of consistency, which verifies the rationality of the established model.

[0083] Table 11 Errors between Predicted and Actual Data

[0084] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback, characterized in that, It includes the following steps: S1: Define the scope of analysis, clarify the research objectives and boundaries; S2: Classify the influencing factors of unmanned aerial vehicle (UAV) systems using human factors analysis and classification methods; S3: Introduce an explanatory structural model to structure the hierarchical relationships of influencing factors and determine the correlations between various influencing factors, including the following sub-steps: S31: Construct the adjacency matrix ; S32: Calculate the reachability matrix The calculation formula is as follows: ； In the formula, I represents the identity matrix, whose order is equal to that of the adjacency matrix. If the order is the same, then the adjacency matrix will be... Add the matrix to the identity matrix I and then perform Boolean algebra operations on the matrix itself until the result remains unchanged. At this point, the calculated result is the reachability matrix. n represents the number of operations until the result no longer changes; S33: Divide the hierarchical structure: based on the reachability matrix The reachable set is obtained by partitioning. , the previous set and common factors set If the reachability set of influencing factors is the same as the set of common factors, then they belong to the same level in the reachability matrix. Cross out the rows and columns corresponding to the same level of influencing factors, and repeat this process to obtain the next level of influencing factors until all influencing factors are crossed out, and draw a hierarchical structure diagram of the influencing factors of UAV system safety. S4: Determine the feedback path of the UAV system, and combine the correlation between various influencing factors to obtain the feedback loops existing in the UAV system; S5: Based on the feedback loops existing in the UAV system, obtain the causal relationship diagram between the various influencing factors of the UAV system; S6: Based on the causal relationship diagram, determine the quantitative relationship between variables without considering feedback by using a hybrid causal logic method; S7: Based on the causal relationship graph, determine the quantitative relationship between nodes with dynamic feedback characteristics and a mixture of continuous and discrete characteristics; S8: Combining the quantitative relationships determined in S6 and S7, the causal relationship diagram in S5 that qualitatively expresses the correlation between variables is transformed into a stock flow diagram calculated quantitatively. S9: Input the expression of the quantitative relationship determined by S6 and S7 into the variables corresponding to the stock flow diagram to drive the model simulation and obtain the predicted value of the accident probability of the UAV system.

2. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in claim 1, characterized in that, Factors affecting unmanned aerial vehicle (UAV) systems in S2 include latent effects, unsafe surveillance, warning signs of unsafe behavior, and unsafe behavior itself.

3. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in claim 1, characterized in that, The feedback paths of the UAV system in S4 include: feedback from the UAV accident rate to the flight management system, feedback from the UAV accident rate to the safety awareness of maintenance personnel, and feedback from the UAV accident rate to the safety awareness of operators.

4. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in claim 1, characterized in that, In S6, the quantitative relationships between variables include: causal logic modeling of human error and its correlation, causal logic modeling of machine failure, and accident evolution logic modeling.

5. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in claim 4, characterized in that, Methods for modeling human error and its correlation with causal logic include: Calculate the final human error probability: ; In the formula, and These represent the final human error probability and the base human error probability, respectively. This represents the maximum influence value of the i-th experience performance formation factor. This represents the proportion of the impact assessment of the i-th experience performance formation factor; Performance factors influencing the probability of human error include safety awareness and experience, which are calculated as follows: Calculate the factors that influence changes in safety awareness: ； In the formula, SA represents personnel safety awareness, OCTS represents the degree of importance the flight management system attaches to safety, ROCtS represents the reference value of the degree of importance the flight management system attaches to safety, OtoP represents the impact index of the flight management system on personnel safety, and T1 represents the time required for personnel safety awareness to change. Calculate the empirical variation factor: ； In the formula, Ex represents the experience level of the personnel, and Tr represents the level of technical training provided by the flight management system.

6. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in claim 4, characterized in that, Methods for causal logic modeling of machine failures include: Calculate the probability of system failure when the failure logic is AND gate or OR gate; Calculate the actual mean time between failures (MTBF) of a machine. : ； In the formula, The mean time between failures (MTBF) represents the machine's design time. These are variables affected by the probability of human error. Calculate the failure probability F(x) of each subsystem: F(x) = 1 - exp(-t / MTBF); In the formula, t is the subsystem running time, and MTBF represents the mean time between failures (MTBF), which is obtained by statistical methods.

7. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in claim 4, characterized in that, Methods for modeling the logic of accident evolution include accident occurrence probability. for: ； In the formula, This represents the probability of failure of the UAV mission payload system; The probability of human error where the operator is unaware of the risk; The probability of human error representing incorrect operation by the operator; This represents the probability of failure in the drone control system.

8. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in claim 1, characterized in that, Methods for determining the quantitative relationship between nodes include: dividing the personnel safety awareness performance shaping factor into three levels: low, medium, and high, then the proportion of personnel safety awareness in the assessment. for: ； In the formula, , and The impact ratio of the representative personnel's safety awareness performance shaping factor at different levels is determined.

9. The method for predicting the accident rate of an unmanned aerial vehicle (UAV) system considering human error correlation and closed-loop feedback as described in any one of claims 1-8, characterized in that, The methods for driving model simulation in S9 include: determining the initial values ​​of level variables, simulation time, and simulation step size based on the available data range, and predicting the accident probability value of the UAV system.

Images