Controllable flight ground collision risk discrimination method and system based on Bayesian network model

Abstract

The invention discloses a controllable flight ground collision risk judgment method and system based on a Bayesian network model, and relates to the technical field of controllable flight ground collision risk judgment. And carrying out classification identification on causes of controllable flight ground collision (CFIT) unsafe events. The complex and disordered cause factors are sorted through an interpretation structure model (ISM), and a hierarchical relation schematic diagram capable of clearly showing the mutual causal relation and hierarchy of the cause factors is obtained. A Bayesian network model is established, and key factors in a CFIT event are sorted out from three different angles. According to the analysis result, various external factors and unit capability defects have influences on the CFIT, and the last defensive line for judging key links and preventing and controlling CFIT events can be found. More detailed guidance and training are provided for the key links, and more standardized supervision is carried out.

Description

Technical Field

[0001] This invention relates to the field of controlled flight collision risk assessment technology, and more specifically, to a method and system for controlling flight collision risk assessment based on a Bayesian network model. Background Technology

[0002] Controlled Flight Collide-In-Train (CFIT) is defined as an incident in which an aircraft collides with terrain, water, or an obstacle in flight without any indication of loss of control. Runway safety-related incidents, Loss of Control in Flight (LOC-I), and Controlled Flight Collide-In-Train (CFIT) are three high-risk incident categories.

[0003] Civil aviation accidents are usually not caused by a single factor or by an individual; the factors leading to Controlled Flight Into Land (CFIT) accidents are numerous and complex. Early research on CFIT primarily focused on air-to-ground communication, cockpit workload, and pilot fatigue. However, with further research, while human factors are a major component of CFIT accidents, the underlying regulatory factors and organizational influences cannot be ignored. CFIT is considered a typical accident category with low probability and high consequences. Therefore, it is essential to use scientific and reasonable methods to explore the impact of different factors on CFIT and their interrelationships in order to reduce its occurrence. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for determining the risk of controlled flight impacting the ground based on a Bayesian network model, so as to solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect, embodiments of the present invention provide a method for determining the risk of controlled flight impacting the ground based on a Bayesian network model, including:

[0006] Acquire data on multiple unsafe incidents of controlled flight crashes to the ground;

[0007] The HFACS model was used to identify the causes of unsafe controlled flight collisions to the ground, resulting in multiple causative factors. These causative factors represent the common characteristics of unsafe controlled flight collisions to the ground.

[0008] By interpreting the structural model and performing hierarchical analysis based on multiple causative factors, an interpretive structural model is obtained; the interpretive structural model represents the correlation between different factors.

[0009] The relationships between multiple causal factors in the explanatory structural model are mapped to construct a Bayesian network;

[0010] Using a Bayesian network, risk analysis was performed based on multiple data points of unsafe controlled flight-to-ground collisions and multiple causative factors, resulting in three risk sequences. These risk sequences are used to determine the degree of influence of multiple causative factors on unsafe controlled flight-to-ground collision events.

[0011] Optionally, the explanatory structural model is obtained by performing hierarchical analysis based on multiple causative factors, including:

[0012] An adjacency matrix is ​​established using multiple causative factors as vertices; in the adjacency matrix, if one causative factor has a direct influence on another causative factor, the corresponding edge is 1; if one causative factor does not have a direct influence on another causative factor, the corresponding edge is 0.

[0013] Based on the adjacency matrix, the reachable set, the antecedent set, and the intersection are obtained;

[0014] Based on reachable sets, antecedent sets, and intersections, hierarchical analysis is performed to obtain hierarchical relationships;

[0015] Using the causal factors as vertices, adjacency matrices are used to determine the existence of edges, and hierarchical relationships are used as the directions of edges to construct a directed graph, thus obtaining the explanatory structural model.

[0016] Optionally, the risk analysis is performed using a Bayesian network based on multiple controllable flight-to-ground collision unsafe event data and multiple causative factors to obtain three risk sequences, including:

[0017] The joint probability is obtained based on multiple causative factors;

[0018] The method for calculating the joint probability is as follows:

[0019]

[0020] in, This represents the joint probability of each node. express Parent node, This represents a node in a Bayesian network. express The conditional probability, where n represents the number of nodes in the Bayesian network;

[0021] Multiple controllable flight collision unsafe event data are classified under the states of multiple causative factors to obtain a controllable flight collision unsafe event dataset; the controllable flight collision unsafe event dataset indicates whether there is an unsafe situation corresponding to the causative factors;

[0022] Create a mapping relationship between nodes in the Bayesian network and causative factors in the dataset of unsafe flight crashes to the ground;

[0023] The Bayesian network was trained using an exploratory evolutionary strategy algorithm.

[0024] Using a trained Bayesian network, three methods—diagnostic analysis, sensitivity analysis, and maximum causal path analysis—were employed to detect three risk sequences.

[0025] Optionally, the trained Bayesian network is used to perform detection using three methods: diagnostic analysis, sensitivity analysis, and most likely cause path analysis, resulting in three risk sequences, including:

[0026] The three risk sequences include the first risk sequence, the second risk sequence, and the third risk sequence;

[0027] Using a trained Bayesian network and diagnostic analysis, a first risk sequence is obtained; the first risk sequence represents nodes sorted from largest to smallest posterior probability.

[0028] Using a trained Bayesian network and sensitivity analysis, a second risk sequence is obtained; the second risk sequence represents nodes sorted from largest to smallest posterior probability.

[0029] By using a trained Bayesian network and employing the most causal path analysis, a third risk sequence is obtained; the third risk sequence represents nodes sorted from largest to smallest posterior probability.

[0030] Optionally, the hierarchical analysis based on reachable sets, antecedent sets, and intersections to obtain hierarchical relationships includes:

[0031] Construct a set table based on the reachable set, the preceding set, and the intersection;

[0032] The elements in multiple antecedent sets and multiple sets that are equal in the intersection are taken as the lowest level set;

[0033] Delete elements from the lowest-level set in the collection table;

[0034] The elements in the set that is equal in multiple antecedent sets and multiple intersections are used as the second low-level set;

[0035] This process continues until multiple causal factors are traversed to obtain the hierarchical relationship.

[0036] Optionally, obtaining the reachable set, the preceding set, and the intersection based on the adjacency matrix includes:

[0037] Based on the adjacency matrix, the reachability matrix is ​​obtained; the reachability matrix indicates whether there is a pathway between multiple causative factors.

[0038] The method for obtaining the reachability matrix is ​​as follows:

[0039]

[0040] Where M represents the reachability matrix, D represents the adjacency matrix, I represents the identity matrix, and k represents the number of cycles;

[0041] Mark the columns with a value of 1 in row 1 of the reachability matrix to obtain the reachability set; multiple causative factors correspond to multiple reachability sets.

[0042] The rows with a value of 1 in column 1 of the reachability matrix are marked to obtain the preceding set; multiple causative factors correspond to multiple preceding sets;

[0043] Calculate the intersection of the antecedent set and the reachable set.

[0044] Optionally, the step of identifying the causes of unsafe controlled flight-to-ground collision events using the HFACS model yields multiple causative factors, including:

[0045] The data on unsafe controlled flight crashes to the ground are cleaned to obtain sample data;

[0046] The sample data was analyzed using the HFACS model to obtain multiple causative factors; these causative factors represent the factors that cause insecurity.

[0047] The HFACS model includes HFACS levels, categories, and factors; the HFACS levels include organizational impact, unsafe monitoring, preconditions for unsafe behavior, and unsafe behavior; the categories represent the subdivision of the HFACS levels; and the factors represent data on unsafe events involving controlled flight crashes.

[0048] Optionally, the third risk sequence is obtained by using the trained Bayesian network and the most significant causal path analysis, including:

[0049] Obtain the discrimination node; the discrimination node is the node that determines whether a controllable flight crash into the ground unsafe event will occur;

[0050] By using a trained Bayesian network, the intensity influence value is obtained based on the probability of the parent node to the child node; multiple intensity influence values ​​are obtained for multiple parent nodes of a discriminant node.

[0051] The maximum value among multiple intensity influence values ​​is taken as the maximum intensity influence value;

[0052] The parent node corresponding to the maximum intensity influence value is taken as the third parent node; the third parent node is the parent node of the retained discrimination node;

[0053] By taking the third parent node as a child node and repeating this process, the third risk sequence is obtained.

[0054] Optionally, the first risk sequence is obtained by diagnostic analysis using a trained Bayesian network, including:

[0055] Obtain the discrimination node; the discrimination node is the node that determines whether a controllable flight crash into the ground unsafe event will occur;

[0056] The posterior probability of a node to its parent node is detected and determined using a trained Bayesian network.

[0057] The first risk sequence is obtained by sorting the sequences according to their posterior probabilities from largest to smallest.

[0058] Secondly, embodiments of the present invention provide a controllable flight collision risk assessment system based on a Bayesian network model, comprising:

[0059] The acquisition module is used to acquire data on multiple unsafe events of controlled flight crashes to the ground;

[0060] The causative factor module is used to identify the causes of controlled flight collisions to the ground unsafe events using the HFACS model, and obtain multiple causative factors; the causative factors represent the common characteristics of the occurrence of controlled flight collisions to the ground unsafe events.

[0061] The explanatory structure module is used to perform hierarchical analysis based on multiple causative factors to obtain an explanatory structure model; the explanatory structure model represents the correlation between different factors.

[0062] The Bayesian module is used to map the relationships between multiple causal factors in the explanatory structural model and construct a Bayesian network.

[0063] The risk detection module is used to perform risk analysis based on multiple controllable flight collision unsafe event data and multiple causative factors through a Bayesian network, and obtain three risk sequences. The risk sequences are used to determine the degree of influence of multiple causative factors on controllable flight collision unsafe events.

[0064] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0065] This invention also provides a method and system for determining the risk of Controlled Flight Into-Earth (CFIT) events based on a Bayesian network model. The Bayesian network model is used to quantitatively analyze the risk of CFIT events. Causative factors identified by the HFACS model are used as nodes, and a Bayesian network topology is constructed with reference to the causal relationships in the ISM model. Parameters are applied to the collected and processed sample data for probabilistic inference. Through diagnostic analysis, sensitivity analysis, and maximum causal path analysis, the complex causal chain of CFIT events is further revealed, and targeted improvement measures are proposed. At the organizational level, safety management systems should be strengthened, the level of regulatory training for flight crews should be improved, pilot competency assessment mechanisms should be optimized, crew communication and collaboration capabilities should be enhanced, and control over high-risk environments should be strengthened. At the individual level, it is necessary to enhance the study of manuals and documents, improve risk identification capabilities, strengthen aircraft monitoring and communication capabilities, and improve manual flight skills. Through the probabilistic inference capabilities and clear topology diagram of the Bayesian network, the key causes of CFIT events and their interrelationships can be intuitively displayed, providing theoretical support and practical guidance for the prevention and control of CFIT events, and also providing a scientific basis for aviation safety management. In the future, the range of sample data can be further expanded and the model structure optimized to improve the accuracy and applicability of the model. Attached Figure Description

[0066] Figure 1 This is a flowchart of a controllable flight collision risk assessment method based on a Bayesian network model provided in an embodiment of the present invention.

[0067] Figure 2 This is a schematic diagram of the adjacency matrix in a controllable flight collision risk discrimination method based on a Bayesian network model provided in an embodiment of the present invention.

[0068] Figure 3 This is a schematic diagram of the reachability matrix in a controllable flight collision risk discrimination method based on a Bayesian network model provided in an embodiment of the present invention.

[0069] Figure 4 This is a schematic diagram illustrating the relationship between the reachable set, the preceding set, and the intersection in a controllable flight collision risk discrimination method based on a Bayesian network model provided in an embodiment of the present invention.

[0070] Figure 5 This is a schematic diagram of the hierarchical relationship of causative factors in a controllable flight collision risk discrimination method based on a Bayesian network model provided in an embodiment of the present invention.

[0071] Figure 6 This is a schematic diagram of the explanatory structure model in a controllable flight collision risk discrimination method based on a Bayesian network model provided in an embodiment of the present invention.

[0072] Figure 7 This is a schematic diagram of a dataset of unsafe events related to controlled flight impacts to the ground, provided in an embodiment of the present invention, based on a Bayesian network model-based method for determining the risk of controlled flight impacts to the ground.

[0073] Figure 8 This is a schematic diagram of the top five nodes with the posterior probability in a controllable flight collision risk discrimination method based on a Bayesian network model provided in an embodiment of the present invention.

[0074] Figure 9 This is a schematic diagram of the intensity influence value in a controllable flight impact risk discrimination method based on a Bayesian network model provided in an embodiment of the present invention. Detailed Implementation

[0075] The present invention will now be described in detail with reference to the accompanying drawings.

[0076] Example 1:

[0077] like Figure 1 As shown, this embodiment of the invention provides a method for determining the risk of controlled flight impacting the ground based on a Bayesian network model. The method includes:

[0078] S101: Acquire data on multiple unsafe events involving controlled flight crashes to the ground.

[0079] The Controlled Flight-to-Ground (CFIT) unsafe event data refers to the data on unsafe events resulting from an aircraft that fully meets airworthiness requirements colliding with the ground, mountain, water, or other obstacles in a non-failure and controllable state. The factors are shown in Table 1.

[0080] The Controlled Flight Into-Earth (CFIT) unsafe event data records the general process of the event and the underlying causes.

[0081] In this embodiment, CFIT event data from the Aviation Safety Network from 2000 to 2019 is used as data on controllable flight-to-ground collision unsafe events.

[0082] Using the methods described above, historical data on controllable flight-to-ground collisions provides a detailed and reliable data foundation for the analysis of results of the same type of event at the same event level, helping to carry out safety management work more scientifically and efficiently.

[0083] S102: Using the HFACS model, identify the causes of unsafe controlled flight crashes to the ground and obtain multiple causative factors; the causative factors represent the common characteristics of unsafe controlled flight crashes to the ground.

[0084] The Human Factors Analysis and Classification System (HFACS) model is an effective model for analyzing human factors. The HFACS model includes HFACS levels, categories, and factors. HFACS levels encompass four aspects: organizational influence, unsafe supervision, preconditions for unsafe behavior, and unsafe behavior itself. Categories within each HFACS level represent a more detailed breakdown of the HFACS level. Factors within each HFACS level represent specific behaviors. The HFACS model establishes more detailed causal categories to identify both explicit and implicit causes of accidents.

[0085] The HFACS model is shown in Table 1:

[0086]

[0087] S103: By interpreting the structural model, a hierarchical analysis is performed based on multiple causative factors to obtain the interpreting structural model; the interpreting structural model represents the relationship between different factors.

[0088] In this embodiment, the explanatory structural model is as follows: Figure 6 As shown.

[0089] S104: Map the relationships between multiple causal factors in the structural model to construct a Bayesian network.

[0090] In this embodiment, the Genie software is used to implement Bayesian network modeling of the causative factors of CFIT.

[0091] A Bayesian network is a directed acyclic graph (DAG) model based on a network structure. It uses nodes and arcs to represent random variables and their conditional dependencies to discover potential relationships. In a Bayesian network, nodes represent variables, and arrows represent conditional dependencies between variables. The node pointed to by the arrow is a child node, and the node furthest from the arrow is a parent node.

[0092] In this embodiment, there are 35 causative factors. A Bayesian network is configured with 36 nodes, where 35 nodes correspond to nodes S1…S35 in the Explanatory Structure Model, and S36 indicates whether the CFIT event is safe.

[0093] Using the methods described above, it's unreasonable to directly map the HFACS model, a causal classification system, to a Bayesian network topology because its four categories and subcategories are hierarchical and it doesn't explain the causal relationships between individual causal factors. In the ISM model, arrows represent the interactions between different causal factors, consistent with the meaning of arrows in the Bayesian model. However, the ISM model merely presents the relationships between causal factors in a simple and intuitive way and doesn't play a crucial role in the initial causal collection. Therefore, this paper uses the HFACS model for causal identification and classification, then utilizes the ISM model to determine causal relationships, mapping them to a Bayesian network topology. This significantly improves the reliability and credibility of the analysis.

[0094] S105: Using a Bayesian network, risk analysis is performed based on multiple data points of unsafe controlled flight collisions and multiple causative factors to obtain three risk sequences; these risk sequences are used to determine the degree of influence of multiple causative factors on unsafe controlled flight collision events.

[0095] Using the methods described above, the civil aviation system, as a vast system, is composed of various subsystems such as aircraft, air traffic control, facilities and equipment, and meteorological environment. This system constantly generates CFIT events. Ideally, disassembling the system and patching each vulnerability individually could reduce problems. However, in reality, many factors are unavoidable, and when different systems operate simultaneously, new vulnerabilities often emerge, significantly increasing the risk of some existing ones. Therefore, research on CFIT should not treat its underlying causes as isolated factors for prevention and control, but rather clarify the interrelationships of various factors to understand the CFIT occurrence process holistically. This paper first introduces the Interpretive Structural Model (ISM) and then uses it to establish a hierarchical CFIT causation model based on the 35 identified CFIT causes.

[0096] Optionally, the step of obtaining an explanatory structural model diagram by performing hierarchical analysis based on multiple causative factors through an explanatory structural model includes:

[0097] An adjacency matrix is ​​established using multiple causative factors as vertices; in the adjacency matrix, if one causative factor has a direct influence on another causative factor, the corresponding edge is 1; if one causative factor does not have a direct influence on another causative factor, the corresponding edge is 0.

[0098] In this embodiment, the number of causative factors is 35, and the adjacency matrix is ​​as follows: Figure 2 As shown.

[0099] Wherein, if the causative factor of Si in the adjacency matrix has a direct influence on the causative factor of Sj, the corresponding edge is 1; if the causative factor of Si in the adjacency matrix does not have a direct influence on the causative factor of Sj, the corresponding edge is 0.

[0100] Using the causal factors as vertices, the adjacency matrix is ​​used to determine whether there are edges, and the hierarchical relationship is used as the direction of the edges to construct a directed graph and obtain the explanatory structure model;

[0101] The explanatory structural model is visualized in the form of a directed graph, thus obtaining the explanatory structural model.

[0102] Using the methods described above, an Explanatory Structural Model (ISM) was constructed to conduct a hierarchical analysis of the causes of CFIT (Controlled Flight Into Land) events. The ISM effectively handles multivariate and multi-relationship problems in complex systems, making it particularly suitable for safety issues like CFIT involving multiple subsystems and complex causal relationships. After analyzing the event, the direct relationships between the 35 identified causes were determined in conjunction with expert opinions, and adjacency and reachability matrices were calculated, ultimately forming the ISM for CFIT. During the model analysis, a causal priority model was used to reveal the underlying driving factors in CFIT events. The causal priority model reveals the root causes of CFIT events, primarily focusing on inadequate risk management at the organizational level and the influence of external environmental factors.

[0103] Optionally, the risk analysis is performed using a Bayesian network based on multiple controllable flight-to-ground collision unsafe event data and multiple causative factors to obtain three risk sequences, including:

[0104] The joint probability is obtained based on multiple causative factors.

[0105] The probabilities of multiple causative factors can be obtained by marginalizing them using joint probabilities. The joint probabilities represent the association relationships among multiple causative factors as represented by a Bayesian network.

[0106] The method for calculating the joint probability is as follows:

[0107]

[0108] in, This represents the joint probability of each node. express Parent node, This represents a node in a Bayesian network. express The conditional probability, where n represents the number of nodes in the Bayesian network.

[0109] Among them, if If there is no parent node, then the conditional probability... It degenerates into prior probability. .

[0110] Wherein, the joint probability is the prior probability corresponding to the root node. Conditional probabilities corresponding to other nodes The product is obtained by multiplying.

[0111] Here, the prior probability refers to the probability assigned to a certain event. Conditional probability is interpreted as the likelihood of an event occurring based on the occurrence of a previous event or outcome. Joint probability refers to the probability of two or more events occurring simultaneously.

[0112] The relationship between the conditional probability and the prior probability is as follows:

[0113]

[0114] in, and These refer to the prior probabilities of A and B occurring, respectively. and Both are conditional probabilities. P(A|B) is the probability of A given B, and P(B|A) is the probability of B given A.

[0115] In order to obtain the prior probability of the root node and the conditional probability of the other nodes, in this embodiment, considering that there are many causal factors in this study, it is difficult to achieve through expert judgment, and the expert opinions are highly subjective and will distort the results. Therefore, the corresponding prior probabilities or conditional probabilities are obtained by using the parameter learning function of the Genie software based on historical data of unsafe controlled flight crashes.

[0116] Multiple controllable flight collision unsafe event data are classified under the states of multiple causative factors to obtain a controllable flight collision unsafe event dataset; the controllable flight collision unsafe event dataset indicates whether there will be an unsafe situation corresponding to the causative factors.

[0117] The number of elements in the controlled flight crash unsafe event dataset is equal to the product of the number of controlled flight crash unsafe event data and the number of causal factors.

[0118] The states of the causative factor and the CFIT event include two states: yes and no. "Yes" indicates the existence of an unsafe situation corresponding to the causative factor, while "no" indicates the absence of an unsafe situation corresponding to the causative factor. For example... Figure 7 As shown, in this embodiment, 40 CFIT events are used to construct a dataset of unsafe events that result in controlled flight crashes to the ground.

[0119] Create a mapping relationship between nodes in the Bayesian network and causative factors in the dataset of unsafe flight crashes.

[0120] The task involves determining whether the dataset of unsafe controlled flight crashes matches the dataset in the Bayesian network. If they differ, the difference is highlighted with a yellow background. Manual matching between nodes in the Bayesian network and the dataset of unsafe controlled flight crashes can be performed by dragging and dropping variables and their results.

[0121] The Bayesian network was trained using an exploratory evolutionary strategy algorithm.

[0122] Among them, the Exploratory Evolutionary Strategy (EM) algorithm trains the network based on the initial data.

[0123] Using the method described above, the Exploratory Evolutionary Strategy (EM) algorithm is employed to learn parameters from a dataset containing missing values ​​when learning the probability distribution of a Bayesian network. Parameter initialization is performed first, followed by the use of a uniform distribution. In this case, the algorithm assigns a confidence level of 1, allowing the parameters to choose random values, which introduces some randomness as the algorithm searches for optimal parameter values. Any method of setting initial parameter values ​​simply provides a starting point for the EM algorithm to search for a parameter set that maximizes the data probability of a given model. Allowing random parameter selection is particularly useful when learning parameters with latent variables. In this case, the EM algorithm is more likely to avoid local maxima around a uniform distribution.

[0124] Using a trained Bayesian network, three methods—diagnostic analysis, sensitivity analysis, and maximum causal path analysis—were employed to detect three risk sequences.

[0125] In this embodiment, the occurrence of "lack of crew communication and failure to provide timely and accurate reminders or takeover" in all three reasoning analyses demonstrates its significant impact on CFIT events. Under the combined influence of various external factors and crew capability deficiencies, if the final opportunity for error correction—through crew cooperation and reminders—is missed, it can lead to unsafe behaviors that directly cause CFIT events.

[0126] Optionally, the trained Bayesian network is used to perform detection using three methods: diagnostic analysis, sensitivity analysis, and most likely cause path analysis, resulting in three risk sequences, including:

[0127] The three risk sequences include the first risk sequence, the second risk sequence, and the third risk sequence.

[0128] The first risk sequence is obtained by using a trained Bayesian network and diagnostic analysis.

[0129] Diagnostic analysis refers to analyzing the posterior probability of nodes from effect to cause, and analyzing the important factors leading to CFIT risk. In this embodiment, 36 causative factors are identified. The probability of the CFIT event corresponding to the last causative factor S36 is set to 100% to obtain the posterior probability of each node. ).

[0130] In this embodiment, the top five causative factors in the first risk sequence are selected for focused monitoring to determine whether they pose a risk to CFIT events. For example, when a CFIT event has a 100% occurrence rate, the top five factors with the highest probability of occurrence are: S28. Lack of crew communication, failure to provide timely and accurate alerts or takeover; S20. Aircraft in an unexpected state; S13. Manual flight; S31. Poor PF (Flight Precipitation Officer) correction of various flight parameters; S30. Crew members losing situational awareness or having asynchronous situational awareness. Based on case studies, these factors represent the most common scenarios when CFIT events occur. CFIT events frequently occur during the manual flight phase of the fifth leg. External factors or pilot error cause unexpected deviations in the aircraft. Poor communication between crew members means that even if one crew member detects a deviation, they cannot help the remaining crew members establish situational awareness. This leads to the PF failing to detect and correct the deviation in a timely manner, ultimately resulting in a CFIT event.

[0131] Since the underlying causative factors are mostly external factors causing CFIT, the posterior probabilities of the underlying factors in the first risk sequence are extracted. In this embodiment, the underlying factors and their corresponding posterior probability distributions are S16: 25%, S17: 25%, S14: 22%, S15: 13%, S19: 10%, S21: 10%, S9: 7%, S18: 7%, S3: 5%, and S7: 5%. The top five factors with the highest posterior probabilities among the underlying factors are S14 local training, S15 low visibility, S16 airflow and wind effects, S17 special airport procedures or special terrain, and S19 controller communication and command reasons. The underlying factors are the most fundamental and driving factors in the system. These five factors are also the environmental factors in the preconditions of unsafe behavior in the HFACS model, and are often closely related to other categories of factors. For example, under local training conditions, instructors often pair with trainees to fly under artificial conditions. Trainees are often just undergoing initial conversion training or have never even been in contact with a real aircraft. In such situations, insufficient flight pilot (PF) control skills and poor correction of flight parameters are almost inevitable factors. Trainees and instructors cannot truly check and implement procedures as in a two-person system, leading to a significant increase in the instructor's workload. Therefore, airlines often assign an observer to assist the instructor in monitoring the aircraft. If the observer is unqualified or the instructor is careless, danger can easily arise. Low visibility is a weather condition that is not frequently encountered during operations. Because of this, pilots are less proficient in the division of labor and attention allocation under low visibility conditions. If they focus all their energy on finding visual references during the fifth approach, they can easily neglect monitoring the aircraft's status. Therefore, under low visibility conditions, pilots need to fully utilize their crew resource management capabilities and clearly define their respective responsibilities before flight to successfully complete flight missions under high workload. When airport procedures are special or there is special terrain, pilots' lack of familiarity with or understanding of relevant requirements greatly increases the possibility of errors. This depends on the pilot's pre-flight preparation and whether effective briefings were given to manage relevant threats. Air traffic controller communication and command issues are also common risk factors in aviation accidents. Controllers can make mistakes during flight, and pilots need to enhance their situational awareness to identify abnormal instructions and promptly communicate with controllers to confirm any such instructions. These external factors often increase the pilot's workload, and if the pilot is not adequately prepared, they will struggle to cope with the threats they pose.

[0132] Using a trained Bayesian network and sensitivity analysis, a second risk sequence is obtained; the second risk sequence represents nodes sorted from largest to smallest posterior probability.

[0133] Sensitivity analysis can reflect the quantification of target nodes caused by changes in local parameters of the network model, thereby identifying sensitive factors in the model.

[0134] In this process, node S36 is set as the target node for sensitivity analysis. The first derivative of the posterior probability of the target node with respect to the causative factor is calculated as the sensitivity value. The nodes are then arranged in descending order of sensitivity value to obtain the second risk sequence.

[0135] By using a trained Bayesian network and employing the most causal path analysis, a third risk sequence is obtained; the third risk sequence represents nodes sorted from largest to smallest posterior probability.

[0136] Using the above method, based on the ISM model, Bayesian networks are used for structure learning and parameter learning of event data. The posterior probability, sensitivity, and influence intensity of the nodes corresponding to the causal factors are obtained through inverse reasoning, and the largest causal path is analyzed.

[0137] Optionally, the hierarchical analysis based on reachable sets, antecedent sets, and intersections to obtain hierarchical relationships includes:

[0138] Construct a set table based on the reachable set, the preceding set, and the intersection;

[0139] The elements in multiple antecedent sets and multiple sets that are equal in the intersection are taken as the lowest level set;

[0140] Delete elements from the lowest-level set in the collection table;

[0141] The elements in the set that is equal in multiple antecedent sets and multiple intersections are used as the second low-level set;

[0142] This process continues until multiple causal factors are traversed to obtain the hierarchical relationship.

[0143] The hierarchical relationship is as follows: Figure 5 As shown.

[0144] Optionally, obtaining the reachable set, the preceding set, and the intersection based on the adjacency matrix includes:

[0145] Based on the adjacency matrix, the reachability matrix is ​​obtained; the reachability matrix indicates whether there is a pathway between multiple causative factors.

[0146] The method for obtaining the reachability matrix is ​​as follows:

[0147]

[0148] Where M represents the reachability matrix, D represents the adjacency matrix, I represents the identity matrix, and k represents the number of cycles.

[0149] Where k is a positive integer greater than 0.

[0150] The adjacency matrix is ​​added to the identity matrix, and then exponentially multiplied until the matrix no longer changes, to obtain the reachability matrix M.

[0151] Wherein, the reachability matrix is ​​as follows Figure 3 As shown.

[0152] Mark the columns with a value of 1 in row 1 of the reachability matrix to obtain the reachability set; multiple causative factors correspond to multiple reachability sets.

[0153] The rows with a value of 1 in column 1 of the reachability matrix are marked to obtain the preceding set; multiple causative factors correspond to multiple preceding sets.

[0154] The relationships between the reachable set, the antecedent set, and the intersection are as follows: Figure 4 As shown, the numerical representations of the reachable set, antecedent set, and intersection represent the indices of the causative factors, such as the causative factors themselves. The corresponding reachable set is { , , , , , , , , , , , }

[0155] Calculate the intersection of the antecedent set and the reachable set.

[0156] Using the above methods, the reachability matrix is ​​mainly used to describe whether there are pathways between causative factors, revealing the global influence path between causative factors and reflecting the interaction between causative factors.

[0157] Optionally, the step of identifying the causes of unsafe controlled flight-to-ground collision events using the HFACS model yields multiple causative factors, including:

[0158] The data on unsafe controlled flight crashes to the ground are cleaned to obtain sample data;

[0159] Among them, data on controllable flight-to-ground unsafe events that exclude general events caused by false terrain warnings were selected.

[0160] The above methods ensure the accuracy of the analysis by ensuring the clarity of the sample.

[0161] The sample data was analyzed using the HFACS model to obtain multiple causative factors; these causative factors represent the factors that cause insecurity.

[0162] The HFACS model includes HFACS levels, categories, and factors; the HFACS levels include organizational impact, unsafe monitoring, preconditions for unsafe behavior, and unsafe behavior; the categories represent the subdivision of the HFACS levels; and the factors represent data on unsafe events involving controlled flight crashes.

[0163] Among them, the data on a controlled flight crash incident is as follows: On November 3, 2022, a training flight of a certain airline, an A320, was on the left takeoff and landing route of runway 35 at Yangzhou Airport, at an altitude of 1500 feet. The left seat was occupied by the monitoring pilot (PM) instructor leading the training flight, and the right seat was occupied by a new student from the student team. The fourth turn was delayed, and the aircraft overshot the fifth turn. The instructor was focused on reminding the flight controller (PF) to control the aircraft's status and forgot to set the configuration to full. At the fifth turn, the runway altitude was 260 feet, configuration 3, wheels down. The third seat reminded, "Configuration not set." The monitoring pilot (PM) set configuration to full and gave the command "Go around." At the runway altitude of 216 feet, the flight controller (PF) set the go-around thrust to execute the go-around, triggering the TOO LOW FLAPS warning. After the go-around, the co-pilot took over the takeoff and landing. The aircraft came to a complete stop upon landing, and the training was terminated. "Configuration" usually refers to the aircraft's flight status under a specific configuration, mainly involving the minimum speed requirements for wing anti-icing operations.

[0164] The results of analyzing the aforementioned controllable flight-to-ground unsafe event data using the HFACS model are shown in Table 2:

[0165]

[0166] In this process, multiple factors are numbered as causative factors. For example, slow deviation correction is assigned one number, and deviation from standard operating procedures (SOP) is assigned another number.

[0167] In this embodiment, the CFIT insecurity events from 2019 to 2023 were used, and the method for obtaining the causative factors is shown in Table 3:

[0168] Optionally, the third risk sequence is obtained by using the trained Bayesian network and the most significant causal path analysis, including:

[0169] Obtain a discrimination node; the discrimination node is a node for determining whether a controllable flight crash into the ground unsafe event will occur.

[0170] By using a trained Bayesian network, the intensity influence value is obtained based on the probability of the parent node to the child node; multiple intensity influence values ​​are obtained for multiple parent nodes of a discriminant node.

[0171] Wherein, for the same child node: the sum of the conditional probabilities of multiple parent nodes to the child node multiplied by the marginal probabilities of the parent nodes is the strength influence value. In this embodiment, the strength influence values ​​of some multiple parent nodes to multiple child nodes are as follows: Figure 9 As shown.

[0172] The maximum value among multiple intensity influence values ​​is taken as the maximum intensity influence value;

[0173] The parent node corresponding to the maximum intensity influence value is taken as the third parent node; the third parent node is the parent node of the retained discrimination node.

[0174] By taking the third parent node as a child node and repeating this process, the third risk sequence is obtained.

[0175] The most significant path analysis depends on the influence strength between the connected nodes, representing the distance between various conditional probability distributions on child nodes conditioned on the state of the parent node. Therefore, in this embodiment, the influence strength value is found by weighting the distance according to the marginal probability of the parent node. First, the parent node with the largest influence strength value corresponding to child node S36 is found. Then, this node is treated as a child node, and the parent node with the largest influence strength value is found again. This process is repeated until the root node. Parent nodes with the same influence strength for the same child node are all retained.

[0176] In this embodiment, the three causal paths are S3, S5, S35, S28, S36; S3, S35, S28, S36; and S9, S28, S36. These describe the most likely occurrence of a CFIT event. All three causal paths are third-risk sequences.

[0177] The first causal path is that S3's lack of safety awareness and insufficient understanding of regulations among the flight crew led to inadequate implementation of the S5 manual, causing S35 to deviate from the Standard Operating Procedure (SOP). This resulted in a lack of communication between the S28 crew and the failure to provide timely and accurate alerts or takeover, ultimately triggering the CFIT incident. The second causal path differs in that S3's lack of safety awareness and insufficient understanding of regulations among the flight crew directly led to S35 deviating from the SOP. The third causal path is that S9's insufficient technical assessment of the pilots resulted in a lack of communication between the S28 crew and the failure to provide timely and accurate alerts or takeover, thus triggering the CFIT incident. Therefore, a lack of crew communication and the failure to provide timely and accurate alerts or takeover are the most critical direct factors causing CFIT. This factor mostly occurs during the manual control phase of the fifth-plane flight, when the crew's workload is high. During operation, the following situations often occur: PMs and observers have insufficient monitoring of the aircraft status and cannot detect PF errors; poor communication between crews prevents PFs from understanding the alerts from PMs and observers; PMs and observers, constrained by the captain's authority, discover errors but fail to alert the crew. According to collected case studies, this factor often occurs when pilots (PFs) have weak operational skills, making it difficult for them to detect aircraft deviations and make appropriate corrections. Pilots (PMs), mostly instructors, often fail to conduct effective monitoring and standard callouts, taking over the aircraft too late and leaving insufficient time to correct the aircraft's condition. Insufficient technical assessment of pilots (S9) and deviations from standard operating procedures (SOPs) (S35) are significant contributing factors. Instructors who do not conduct sufficient pre-flight assessments of PFs lose vigilance and are unable to gauge the appropriate level of control. Airlines' SOPs typically include instructions on proper aircraft monitoring, standard callouts for deviations, and requirements for taking over the aircraft. In actual operation, failure to strictly adhere to standard procedures can lead to crew incompatibility, loss of situational awareness, and compromised operational safety due to Bayesian model-based CFIT risk analysis. Insufficient safety awareness within the airline (S3), inadequate understanding of regulations among flight crews, and inadequate implementation of S5 manuals are also important reasons for crew deviations from SOPs. When the safety awareness of the company and the flight crew declines, pilots are prone to complacency and negligence, leading to many deliberate violations. The regulations in the manuals and documents are difficult to implement. In addition, the company's reduced safety awareness may also result in the failure to fully explain and disseminate new management regulations issued by the company for daily safety management to pilots, causing them to deviate from the standard operating procedures (SOPs) due to lack of understanding or insufficient guidance.

[0178] Optionally, the first risk sequence is obtained by diagnostic analysis using a trained Bayesian network, including:

[0179] Obtain a discrimination node; the discrimination node is a node for determining whether a controllable flight crash into the ground unsafe event will occur.

[0180] In this embodiment, the discrimination node is .

[0181] The posterior probability of a node to its parent node is detected and determined using a trained Bayesian network.

[0182] The first risk sequence is obtained by sorting the sequences according to their posterior probabilities from largest to smallest.

[0183] Diagnostic analysis refers to analyzing the posterior probability of nodes from effect to cause, analyzing the important factors leading to CFIT risk. For example, in this embodiment, there are 35 causative factors and 1 node indicating whether a controlled flight impact-to-ground (CFIT) unsafe event will occur. The probability of the corresponding CFIT event occurring is set to 100% to obtain the posterior probability of each node. ).

[0184] Here, the first risk sequence represents the ranking of the posterior probabilities of nodes to identify the key influencing factors leading to CFIT incidents. The results for the top five nodes in terms of posterior probability are as follows: Figure 8 As shown.

[0185] Example 2:

[0186] Alternatively, an outcome-first approach can be used to construct an explanatory structural model. This outcome priority is determined by using R(Si) = A(Si) to determine the highest level L1, then removing the causative factors included in L1 from the table, and again using R(Si) = A(Si) to determine level L2. This process is repeated until the lowest level Ln is obtained. The 35 causative factors of the CFIT event are then hierarchically divided.

[0187] Using the methods described above, the outcome-oriented model was used to illustrate the underlying driving factors and top-level outcome factors in CFIT events. The outcome-oriented model highlights the direct manifestations of CFIT events, such as poor pilot correction of flight parameters and improper handling of go-around procedures. The causal graph generated by the Controlled Flight Into Land (ISM) model also provides a reference for subsequently establishing the structure of a Bayesian network.

[0188] Example 3:

[0189] Based on the above-described controllable flight collision risk assessment method based on a Bayesian network model, this invention also provides a controllable flight collision risk assessment system based on a Bayesian network model, the system comprising:

[0190] The acquisition module is used to acquire data on multiple unsafe events of controlled flight crashes to the ground;

[0191] The causative factor module is used to identify the causes of controlled flight collisions to the ground unsafe events using the HFACS model, and obtain multiple causative factors; the causative factors represent the common characteristics of the occurrence of controlled flight collisions to the ground unsafe events.

[0192] The explanatory structure module is used to perform hierarchical analysis based on multiple causative factors to obtain an explanatory structure model; the explanatory structure model represents the correlation between different factors.

[0193] The Bayesian module is used to map the relationships between multiple causal factors in the explanatory structural model and construct a Bayesian network.

[0194] The risk detection module is used to perform risk analysis based on multiple controllable flight collision unsafe event data and multiple causative factors through a Bayesian network, and obtain three risk sequences. The risk sequences are used to determine the degree of influence of multiple causative factors on controllable flight collision unsafe events.

[0195] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0196] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0197] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. A method for controlled flight into terrain risk discrimination based on a Bayesian network model, characterized in that, The method comprises the following steps: obtaining a plurality of controllable flight crash unsafe event data; identifying the causes of the controllable flight crash unsafe event data through an HFACS model to obtain a plurality of cause factors; the cause factors represent the common characteristics of the controllable flight crash unsafe event; performing hierarchical analysis based on the plurality of cause factors through an interpretative structural model to obtain an interpretative structural model; the interpretative structural model represents the correlation of different factors; mapping the correlation of the plurality of cause factors in the interpretative structural model to construct a Bayesian network; performing risk analysis based on the plurality of controllable flight crash unsafe event data and the plurality of cause factors through the Bayesian network to obtain three risk sequences; the risk sequences are used to determine the sequence of the influence degree of the plurality of cause factors on the controllable flight crash unsafe event.

2. The method according to claim 1, wherein, The hierarchical analysis based on the plurality of cause factors through the interpretative structural model comprises the following steps: establishing an adjacency matrix with the plurality of cause factors as vertices; in the adjacency matrix, if one cause factor has a direct influence on another cause factor, the corresponding edge is 1; if one cause factor has no direct influence on another cause factor, the corresponding edge is 0; obtaining reachable sets, precedence sets and intersections based on the adjacency matrix; performing hierarchical analysis based on the reachable sets, the precedence sets and the intersections to obtain a hierarchical relationship; constructing a directed graph with the cause factors as vertices, using the adjacency matrix to determine whether there is an edge, and using the hierarchical relationship as the direction of the edge to obtain the interpretative structural model.

3. The method of claim 1, wherein the method comprises: The risk analysis based on the plurality of controllable flight crash unsafe event data and the plurality of cause factors through the Bayesian network comprises the following steps: obtaining a joint probability according to the plurality of cause factors; the calculation method of the joint probability is as follows: wherein, denotes a joint probability of each node, denotes a parent node, denotes a node in a Bayesian network, denotes a conditional probability of n, which denotes a number of nodes in a Bayesian network. determining the state of the plurality of controllable flight crash unsafe event data in the plurality of cause factors to obtain a controllable flight crash unsafe event data set; the controllable flight crash unsafe event data set represents whether there is an unsafe situation corresponding to the cause factor; creating a mapping relationship between the nodes in the Bayesian network and the cause factors in the controllable flight crash unsafe event data set; training the Bayesian network by using an exploration evolutionary strategy algorithm; detecting by using three methods of diagnostic analysis, sensitivity analysis and maximum cause path analysis through the trained Bayesian network to obtain three risk sequences respectively.

4. The method of claim 3, wherein the Bayesian network model is a Bayesian network model of controlled flight into terrain risk. The detection by using the three methods of diagnostic analysis, sensitivity analysis and maximum cause path analysis through the trained Bayesian network to obtain three risk sequences respectively comprises the following steps: the three risk sequences comprise a first risk sequence, a second risk sequence and a third risk sequence; the first risk sequence is obtained by using diagnostic analysis through the trained Bayesian network; the first risk sequence represents nodes sorted in descending order of posterior probability; the second risk sequence is obtained by using sensitivity analysis through the trained Bayesian network; the second risk sequence represents nodes sorted in descending order of posterior probability; The third risk sequence is obtained by using maximum cause path analysis based on the trained Bayesian network; the third risk sequence represents nodes sorted in descending order of posterior probability.

5. The method of claim 2, wherein the Bayesian network model is a Bayesian network model of controlled flight into terrain risk. The hierarchical analysis is performed based on the reachable set, the predecessor set and the intersection set to obtain the hierarchical relationship, including: A set table is constructed according to the reachable set, the predecessor set and the intersection set; Elements in the sets equal in the plurality of predecessor sets and the plurality of intersection sets are taken as the lowest level set; Elements in the lowest level set in the set table are deleted; Elements in the sets equal in the plurality of predecessor sets and the plurality of intersection sets are taken as the second lowest level set; The hierarchical relationship is obtained by traversing the plurality of cause factors.

6. The method of claim 1, wherein the method further comprises: The reachable set, the predecessor set and the intersection set are obtained based on the adjacency matrix, including: A reachable matrix is obtained according to the adjacency matrix; the reachable matrix represents whether there is a path between the plurality of cause factors; The reachable matrix is obtained by the following method: Wherein, M represents the reachable matrix, D represents the adjacency matrix, I represents the unit matrix, and k represents the number of cycles; Columns corresponding to the values of 1 in the first row of the reachable matrix are marked to obtain the reachable set; the plurality of cause factors correspond to obtain the plurality of reachable sets; Rows corresponding to the values of 1 in the first column of the reachable matrix are marked to obtain the predecessor set; the plurality of cause factors correspond to obtain the plurality of predecessor sets; The intersection set of the predecessor set and the reachable set is calculated.

7. The method of claim 1, wherein the method further comprises: The plurality of cause factors are obtained by identifying the causes of the controlled flight into terrain unsafe event data through the HFACS model, including: The sample data is obtained by sample cleaning of the controlled flight into terrain unsafe event data; The plurality of cause factors are obtained by analyzing the sample data through the HFACS model; the cause factors represent the factors causing the unsafe event; The HFACS model includes HFACS levels, categories and factors; the HFACS levels include organizational influence, unsafe supervision, unsafe behavior precondition and unsafe behavior; the categories represent the subdivision of the HFACS levels; and the factors represent the controlled flight into terrain unsafe event data.

8. The method of claim 1, wherein the Bayesian network model is a Bayesian network model of controlled flight into terrain risk. The third risk sequence is obtained by using maximum cause path analysis based on the trained Bayesian network, including: A discrimination node is obtained; the discrimination node is a node for discriminating whether a controlled flight into terrain unsafe event will occur; The intensity influence value is obtained according to the probability of the parent node to the child node through the trained Bayesian network; the plurality of parent nodes of the discrimination node correspond to obtain the plurality of intensity influence values; The maximum value in the plurality of intensity influence values is taken as the maximum intensity influence value; The parent node corresponding to the maximum intensity influence value is taken as the third parent node; the third parent node is the parent node of the retained discrimination node; The third parent node is taken as the child node, and the third risk sequence is obtained by sequentially circulating.

9. The method of claim 1, wherein the method comprises: The first risk sequence is obtained by using diagnostic analysis based on the trained Bayesian network, including: A discrimination node is obtained; the discrimination node is a node for discriminating whether a controlled flight into terrain unsafe event will occur; The posterior probability of the discrimination node to the parent node is detected through the trained Bayesian network; According to the posterior probability from large to small, a first risk sequence is obtained.

10. A controlled flight into terrain risk discrimination system based on a Bayesian network model, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire a plurality of controlled flight into terrain unsafe event data. A cause factor module is configured to identify causes of the controlled flight into terrain unsafe event data by using a HFACS model to obtain a plurality of cause factors, wherein the cause factors represent common characteristics of the occurrence of the controlled flight into terrain unsafe event. An interpretive structure module is configured to perform hierarchical analysis based on the plurality of cause factors by using an interpretive structure model to obtain an interpretive structure model, wherein the interpretive structure model represents the correlation between different factors. A Bayesian module is configured to map the correlation between the plurality of cause factors in the interpretive structure model to construct a Bayesian network. A risk detection module is configured to perform risk analysis based on the plurality of controlled flight into terrain unsafe event data and the plurality of cause factors by using the Bayesian network to obtain three risk sequences, wherein the risk sequences are used to determine the sequence of the influence degree of the plurality of cause factors on the controlled flight into terrain unsafe event.

Images