Water traffic accident cause chain analysis method and device based on HFACS and Bayesian network

Abstract

The invention discloses an HFACS and Bayesian network-based water traffic accident cause chain analysis method and device, and the method combines the theoretical framework of HFACS and Bayesian network, builds a Bayesian network model for water traffic accident cause analysis, carries out the chi-square test, sensitivity analysis and other quantitative analysis, carries out the multi-dimensional analysis of accident causes, and improves the analysis precision. The cause chains of different types of accidents are deeply analyzed, and the root cause of the accident can be deeply revealed, so that the accident rate is effectively reduced, and the personnel safety, property and water area environment are protected.

Description

Technical Field

[0001] The present invention relates to the technical field of accident risk analysis, and in particular to an analysis method and device for a water traffic accident causal chain based on HFACS and Bayesian network. Background Art

[0002] As an important mode of transportation, water transportation plays an irreplaceable role. Water transportation has the advantages of large carrying capacity, low cost, and high transportation efficiency, and occupies an important position in international trade and personnel transportation. However, it is accompanied by frequent water transportation accidents, which poses a huge threat to people's life, property safety and the environment. The occurrence of water transportation accidents is not only caused by a single factor, but the result of the interaction of multiple factors. Therefore, in-depth analysis of the cause chain of water transportation accidents is of great significance for preventing accidents. Summary of the invention

[0003] Therefore, based on the above background, the present invention combines the HFACS model with the Bayesian network to develop an analysis method and device for the causal chain of water traffic accidents based on HFACS and Bayesian network, so as to provide a basis for improving the safe operation level of water transportation.

[0004] The technical solution of the present invention is as follows:

[0005] The analysis method of the cause chain of water traffic accidents based on HFACS and Bayesian network includes the following steps:

[0006] S1: Obtain water traffic accident data;

[0007] S2: According to the characteristics of water traffic accidents, the HFACS model for water traffic accident risk analysis is established. The HFACS model divides the causal factors of water traffic accidents into five levels, namely organizational influence, poor supervision, preconditions for unsafe behavior, unsafe behavior, and accident level;

[0008] S3: Extract the causal factors of water traffic accidents related to the five levels of the HFACS model based on traffic accident data;

[0009] S4: After determining the topological structure of the Bayesian network model using the structural learning method, the maximum likelihood estimation method is used to solve the network parameters. According to the solution results, the conditional probability table of each node in the Bayesian network model is obtained, that is, the probability distribution of each node given the state of its parent node;

[0010] S5: According to the conditional probability table obtained in step S4, the factors with the top three occurrence probabilities in each level are taken as the key factors among the causative factors of the water traffic accident;

[0011] S6: Analyze the sensitivity of the causal factors of water traffic accidents extracted in step S3 by using the sensitivity analysis of the Bayesian network, obtain sensitivity parameters, and obtain sensitivity factors;

[0012] S7: constructing a causal path of water traffic accidents according to the key factors and sensitivity factors obtained in step S5 and step S6 respectively;

[0013] S8: Use the causal reasoning method of Bayesian network to obtain the global causal chain of the causal factors of different types of water traffic accidents.

[0014] Furthermore, the water traffic accident data of step S1 is investigation report information of the water traffic accident, and the traffic accident investigation report information includes the accident type and the cause of the accident.

[0015] Furthermore, the sample size of the water traffic accident data satisfies the requirements for conducting a Pearson chi-square test.

[0016] Furthermore, in step S4, the chi-square test is used as a tool of the structural learning method, specifically the chi-square test is used to analyze the statistical correlation in the water traffic accident data, identify the potential connections between different causal factors, and construct the topological structure of the Bayesian network model.

[0017] Furthermore, the causal path is constructed in step S7: based on the sensitivity analysis results of step S6, the causal factors with insignificant effects are deleted, and the remaining influencing factors are constructed into a causal path according to the Bayesian network nodes constructed in step S4.

[0018] The insignificantly affecting causal factors are the node sensitivity parameters less than 0.01; or the causal factors outside the three nodes with relatively high sensitivity in each level.

[0019] Furthermore, in step S8, by setting the target node as a known condition, the posterior probability of each evidence node is calculated according to the following formula (5) to infer the possible cause of the accident:

[0020] P(A|B)=P(B|A)*P(A) / P(B) (5).

[0021] Further, the operation of the maximum likelihood estimation method in step S4 is as follows:

[0022] Suppose the node variable X in the Bayesian network i There are k values, corresponding to the parent node Pa(X i ) has j combinations, then the parameters of the network can be expressed as

[0023] θ ijk =(X i = k|Pa(X i )=j) (6)

[0024] The above formula (6) refers to the i-th node variable X i The parent node Pa(X i ) takes the value j, X i The conditional probability of taking value k is parameter θ ijk The value of .

[0025] Define sample H l The characteristic function of satisfy

[0026]

[0027] Sample H l Satisfy X i = k, Pa (X i )=j, the number of samples is denoted as μ ijk ;

[0028] Let parameter θ ijk The maximum likelihood estimate of

[0029]

[0030] In the above formula, m is the node variable X in the Bayesian network. i There are m possible values.

[0031] Based on the same inventive concept, the present invention also provides an electronic device, comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, and the program instructions executing the above method when read and executed by the one or more processors.

[0032] Based on the same inventive concept, the present invention also provides a computer-readable medium on which a computer program is stored, wherein the program implements the above method when executed by a processor.

[0033] The effective effects achieved by adopting the present invention are:

[0034] The present invention combines the theoretical framework of HFACS with Bayesian network, creatively divides the causal factors of HFACS into five levels, especially the accident level, and on this basis, constructs a Bayesian network model for causal analysis of water traffic accidents, and carries out quantitative analysis such as chi-square test and sensitivity analysis, conducts multi-dimensional analysis of the causes of accidents, deeply analyzes the causal chains of different types of accidents, and can deeply reveal the root causes of accidents, thereby effectively reducing the accident rate and protecting personnel safety, property and water environment.

[0035] Instruction Manual

[0036] Attached Figure 1 It is the hierarchical relationship of the HFACS model system;

[0037] Attached Figure 2 This is an accident type distribution diagram of an embodiment of the present invention.

[0038] Attached Figure 3 This is a diagram showing the C2*D1 chi-square test result of an embodiment of the present invention.

[0039] Attached Figure 4 This is a diagram of the Bayesian network calculation results of the causes of water traffic accidents in the GENIE software of an embodiment of the present invention.

[0040] Attached Figure 5 These are the top three occurrence probabilities of the causal factors at each level in the embodiment of the present invention.

[0041] Attached Figure 6 This is a network diagram for sensitivity analysis of wind disaster accidents in the GeNIe software according to an embodiment of the present invention.

[0042] Attached Figure 7 These are the top three sensitivity factors of different types of accidents in the embodiments of the present invention. DETAILED DESCRIPTION

[0043] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with its embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] The present invention combines the theoretical framework of HFACS and Bayesian network to analyze the cause chain of water traffic accidents. Next, HFACS and Bayesian network are introduced:

[0045] 1) HFACS model

[0046] HFACS (Human Factors Analysis and Classification System) is an analysis framework widely used in aviation, healthcare and safety management, which aims to deeply analyze the key role of human factors in accidents. Through its multi-level and systematic structure, HFACS systematically classifies and analyzes multiple dimensions in accidents to accurately identify the root causes of accidents.

[0047] The core of the HFACS model consists of four levels: Figure 1As shown in the figure: unsafe behavior, preconditions of unsafe behavior, poor supervision, and organizational influence. First, the unsafe behavior level focuses on the errors and violations that occur when individuals perform tasks due to insufficient skills, wrong decisions, or neglect of standard procedures; second, the precondition level of unsafe behavior analyzes the abnormal psychological state of individuals and unfavorable working environment conditions that provide the basis for wrong behaviors; third, the poor supervision level focuses on negligence and impropriety at the supervisory and management levels, such as poor monitoring, wrong decisions, and insufficient resource support; finally, the organizational influence level covers broader and systematic factors within the organization, such as policy defects, limited resources, organizational cultural bias, and imbalance between production efficiency and safety. The HFACS model shows significant advantages in the analysis of the cause chain of water traffic accidents. Its systematic framework allows researchers to comprehensively and deeply explore the multiple causes of accidents and reveal the internal logic between the causes through a hierarchical structure. This analysis method not only improves the pertinence and effectiveness of accident prevention, but also provides a scientific basis for proposing effective prevention countermeasures, thereby promoting the continuous improvement of the level of water traffic safety.

[0048] 2) Bayesian Network

[0049] Bayesian network is an extension of Bayesian method, also known as causal network and belief network. It is one of the most effective theoretical models for expressing and analyzing uncertain events and performing probabilistic reasoning. In terms of accident cause analysis, it can intuitively express the relationship between causal factors and infer the potential laws that lead to accidents.

[0050] The Bayesian network is a probabilistic graphical model used to represent dependencies between variables. It is presented in the form of a directed acyclic graph (DAG), which can intuitively understand and analyze the causal relationship and conditional dependency between variables. In this model, nodes represent random variables, directed edges represent the relationships between variables, and these relationships are expressed through conditional probabilities.

[0051] Bayesian network is a probabilistic graph model, and its theoretical basis is mainly probability theory and graph theory. Probability theory describes the probabilistic relationship between different variables and is the basis of Bayesian network.

[0052] In order to better explain and apply Bayesian networks, several related basic concepts and formulas are introduced below.

[0053] (1) Prior Probability

[0054] Prior probability is an initial assessment of the probability that a random event or variable will take a particular value before any relevant data or evidence is observed. In Bayesian statistics and Bayesian networks, prior probability reflects an initial belief about the state of an event or variable based on historical data, expert knowledge, or background information.

[0055] (2) Posterior Probability

[0056] The posterior probability is a reassessment of the probability that a random event or variable will take on a particular value after new data or evidence is observed. The calculation of the posterior probability is based on Bayes' theorem, which combines the prior probability (i.e., the probability before the new data is observed) and the likelihood function (i.e., the probability of the observed data given the event). The posterior probability is the updated belief about the state of the event or variable given the new evidence or observation.

[0057] (3) Conditional probability

[0058] The conditional probability formula is a basic concept in probability theory, which describes the probability of another event occurring when a certain event has already occurred. The conditional probability formula is defined as follows: Let A and B be two events, and P(B)>0 (that is, the probability of event B occurring is not zero), then the probability of event A under the condition that event B has already occurred is denoted as P(A|B).

[0059] (4) Joint Probability

[0060] The joint probability formula describes the probability of two or more events occurring at the same time. For two events A and B, the joint probability formula is:

[0061] P(A∩B)=P(A,B) (1)

[0062] However, if events A and B are independent, their joint probability can be found by multiplying their individual probabilities:

[0063] P(A∩B)=P(A)·P(B) (2)

[0064] In Bayesian networks, joint probabilities are usually not given directly, but are calculated indirectly through conditional probability tables, which describe the probability of a child node taking each possible value given the value of the parent node.

[0065] (5) Total probability formula

[0066] It is an important theorem in probability theory, which describes that the probability of a complex event can be calculated by the probabilities and conditional probabilities of a series of simple events.

[0067] Assume the entire sample space is S, if events E1, E2, E3, ...E n is a complete set of events (i.e., they are mutually exclusive, and E1∪E2∪E3…∪E n =S), and P(E i )>0(i=1,2,3…,n), then for any event A:

[0068]

[0069] (6) Bayesian formula

[0070] The formula of Bayes' Theorem is used to describe the relationship between two conditional probabilities.

[0071] Assume the entire sample space is S, A is the event in it, if the events E1, E2, E3, ...E n is a complete set of events (i.e., they are mutually exclusive, and E1∪E2∪E3…∪E n =S), and P(A)>0, P(E i )>0(i=1,2,3…,n), then:

[0072]

[0073] Embodiment 1: The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network of the present invention comprises the following steps:

[0074] S1: Obtain water traffic accident data;

[0075] The water traffic accident data is the investigation report information of the water traffic accident, and the traffic accident investigation report information includes the accident type and the cause of the accident.

[0076] The investigation report information of the water traffic accident may come from an authoritative agency, such as the China Maritime Safety Administration and its directly affiliated agencies.

[0077] The sample size of the water traffic accident data is sufficient for the Pearson chi-square test. The minimum sample size for the Pearson chi-square test is greater than or equal to 40.

[0078] S2: According to the characteristics of water traffic accidents, the HFACS model for water traffic accident risk analysis is established. The HFACS model divides the causal factors of water traffic accidents into five levels, namely organizational influence, poor supervision, preconditions for unsafe behavior, unsafe behavior, and accident level;

[0079] In the HFACS model of this embodiment, the accident layer can be understood as the classification of the accidents that have occurred, and these accident types themselves reflect the results caused by the combined action of factors at all levels. Therefore, the accident type can be regarded as a summary result, showing which factors interacted to cause the accident under specific backgrounds and conditions, and the relationship between the accident layer and the causal factors can be regarded as a causal chain.

[0080] S3: Extract the causal factors of water traffic accidents related to the five levels of the HFACS model based on traffic accident data;

[0081] In this step, the chi-square test is used as a tool for the structural learning method. Specifically, the chi-square test is used to analyze the statistical correlation in the water traffic accident data, identify the potential connections between different causal factors, and construct the topological structure of the Bayesian network model.

[0082] S4: After using the structural learning method to determine the topological structure of the Bayesian network model, the maximum likelihood estimation method is used to solve the network parameters.

[0083] According to the solution results, the conditional probability table of each node in the Bayesian network model is obtained, that is, the probability distribution of each node given the state of its parent node;

[0084] The operation of the maximum likelihood estimation method in this step is as follows:

[0085] Suppose the node variable X in the Bayesian network i There are k values, and the corresponding parent node Pa(Xi) has j combinations, so the parameters of the network can be expressed as

[0086] θ ijk =(X i = k|Pa(X i )=j) (6)

[0087] The above formula (6) refers to the i-th node variable X i The parent node Pa(X i ) takes the value j, X i The conditional probability of taking value k is parameter θ ijk The value of .

[0088] Define sample H l The characteristic function of satisfy

[0089]

[0090] Sample H l Satisfy X i = k, Pa (X i )=j, the number of samples is denoted as μ ijk ;

[0091] Let parameter θ ijk The maximum likelihood estimate of

[0092]

[0093] In the above formula, m is the node variable X in the Bayesian network. iThere are m possible values.

[0094] This step can be calculated using genie software.

[0095] S5: According to the conditional probability table obtained in step S4, the factors with the top three occurrence probabilities in each level are taken as the key factors among the causative factors of the water traffic accident;

[0096] In this step, the top three nodes with the highest probability of occurrence in each level are sorted into a key factor table (such as Figure 5 As shown, Figure 5 The main factors affecting water traffic accidents can be clearly shown, which is convenient for subsequent analysis and discussion. The key factors extracted in this step will become the basis for analysis and modeling in the subsequent steps. In subsequent work, it is necessary to further explore the interrelationships between these identified key factors and their comprehensive impact on accident risks.

[0097] S6: Analyze the sensitivity of the causal factors of water traffic accidents extracted in step S3 by using the sensitivity analysis of the Bayesian network, obtain sensitivity parameters, and obtain sensitivity factors;

[0098] The core formula of sensitivity analysis

[0099]

[0100] S θ is the posterior probability of the target node P(X i |Pa(X i ))Sensitivity value of network parameter θ;

[0101] θ is a parameter of the Bayesian network, usually a parameter in the conditional probability table (CPT).

[0102] This step can be calculated using genie software.

[0103] S7: constructing a causal path of water traffic accidents according to the key factors and sensitivity factors obtained in step S5 and step S6 respectively;

[0104] This step is based on the sensitivity analysis results of step S6. The causal factors with insignificant effects are deleted, and the remaining influencing factors are constructed into causal paths according to the Bayesian network nodes constructed in step S4.

[0105] The above-mentioned insignificant causal factors are those with node sensitivity parameters less than 0.01; or the causal factors are outside the three nodes with relatively high sensitivity in each level.

[0106] S8: Use the causal reasoning method of Bayesian network to obtain the global causal chain of the causal factors of different types of water traffic accidents.

[0107] In this step, the target node is set as a known condition, and then the posterior probability of each evidence node is calculated according to the following formula (5) to infer the possible cause of the accident:

[0108] P(A|B)=P(B|A)*P(A) / P(B) (5).

[0109] Example 2: The present invention uses the investigation reports on water traffic accidents in my country obtained from the official website of the China Maritime Safety Administration and its directly affiliated institutions from the beginning of the record to June 2023 as water traffic accident data to analyze the cause chain of water traffic accidents. The water traffic accident investigation reports obtained above are screened, and duplicate data and reports with incomplete information are excluded to ensure the credibility and representativeness of the data. The water traffic accident investigation reports obtained after the above screening include the type of accident and the cause of the accident.

[0110] The above is mainly based on the water traffic accident reports published on the official website of the Maritime Safety Administration of the People's Republic of China, which records in detail various types of accidents including collision, stranding, reefing, contact, fire / explosion, wind disaster, and self-sinking. However, when deeply analyzing the cause chain of various types of accidents, it is found that the sample data of reefing and stranding accidents are relatively small. In view of the impact of the amount of sample data on accuracy, this embodiment excludes the accident types of reefing and contact.

[0111] This embodiment focuses on collision, contact, fire / explosion, wind disaster, self-sinking and other types of accidents. These types of accidents have relatively sufficient sample data in public reports.

[0112] 1) Statistics

[0113] After systematic data statistics and screening, there are 756 investigation reports on water traffic accidents along the coast of China, which are used as valid data for this embodiment. For example, for collision accidents, each ship is used as an independent research sample. After strict data statistics, this study included a total of 964 ship samples to ensure the accuracy and reliability of the research results. Several common accidents are classified below, and their respective occurrence frequencies are recorded, as shown in Table 1:

[0114] Table 1: Accident type description and frequency table

[0115]

[0116]

[0117] Figure 2The statistics show that there are significant differences in the proportion of different types of accidents in the total. Collision accidents, other accidents and self-sinking accidents account for a higher proportion, indicating that these accidents are more likely to occur in reality. Relatively speaking, contact accidents, fire and explosion accidents and wind disaster accidents account for a lower proportion, which means that these accidents are less likely to occur in statistics.

[0118] 2) Establishment of HFACS

[0119] In the in-depth analysis of water traffic accidents, although the HFACS model provides a systematic causal logic framework, it still shows some limitations when directly applied to the classification of accident factors in water traffic. First, the HFACS model mainly relies on the research results in the aviation field in setting the factor level, which may ignore the uniqueness of the cause structure of water traffic accidents. There are significant differences between water traffic and air traffic in terms of environment, operation, equipment, etc., which may lead to differences in the causes of accidents between the two. Secondly, the HFACS model lacks clarity in accident classification, which is particularly important for the analysis of water traffic accidents.

[0120] There are many types of water traffic accidents, such as collisions, groundings, and wind disasters, and each type of accident has its own unique causal mechanism. Therefore, in the analysis of water traffic accidents, constructing a special "accident layer" to clearly classify various types of accidents is an indispensable part of the analysis framework. Furthermore, at the level of "unsafe behavior premise", in addition to internal factors, the influence of the external environment should also be noted. This influence does not directly constitute a causal relationship with the upper-level factors, but as an independent, external force, it has a significant impact on the lower-level factors (i.e., the specific manifestations of unsafe behavior). Finally, when describing the relationship between factors, the HFACS model assumes a linear causal relationship between adjacent levels, that is, the upper-level factors directly and singly trigger the lower-level factors, and finally connect them into an accident chain. However, in the actual situation of water traffic accidents, the relationship between the various factors may be more complex and diverse. Nonlinear causal relationships are more common in water traffic accidents, that is, one factor may affect multiple results at the same time, or multiple factors may act on the same result together. The existence of this complexity requires us to examine the interactions and correlations between the various factors in more detail when analyzing water traffic accidents.

[0121] In order to more effectively apply the HFACS model to the analysis of marine traffic accidents, the present invention proposes a new marine traffic accident analysis and classification system (MTAACS). The system has made several key improvements (adding accident layers and defining linear relationships of external environments) on the basis of the HFACS model to fully consider the characteristics and complexity of marine traffic accidents.

[0122] The system divides the causal factors of water traffic accidents into the following five levels: organizational influence, poor supervision, preconditions for unsafe behavior, unsafe behavior, and accident level. Factors at the same level are set as parallel relationships, that is, there is no direct cause-and-effect relationship. This setting is intended to avoid oversimplifying the complexity of water traffic accidents and allow analysts to more comprehensively consider the interaction of various factors at the same level. At the same time, the relationship between factors at adjacent levels is described as nonlinear. This means that one factor may affect multiple lower-level factors at the same time, or multiple factors may act on the same lower-level factor together. This nonlinear relationship more accurately reflects the complexity and diversity between various factors in water traffic accidents, and provides a more in-depth and comprehensive analytical perspective for accident prevention.

[0123] Compared with the traditional HFACS model, this embodiment creatively adds an accident layer to the traditional four-level causal factors. The accident layer can be understood as the classification of the accidents that have occurred, and these accident types themselves reflect the results caused by the combined effects of factors at each level. Therefore, the accident type can be regarded as a summary result, showing which factors interact to cause the accident under specific backgrounds and conditions, and the relationship between the accident layer and the causal factors can be regarded as a causal chain.

[0124] According to the investigation report of water traffic accidents, relevant causal factors at various levels are extracted. These factors may involve organizational influence, poor supervision, preconditions for unsafe behavior, unsafe behavior, and accident layer. The accident risk causal factors extracted in this study are all derived from the description in the accident investigation report and can be divided into 38 accident causal factors, as shown in Table 2.

[0125] Table 2: Integrated factors of water traffic accidents

[0126]

[0127] 3) Construction of Bayesian network for causes of water traffic accidents

[0128] When conducting a chain analysis of the causes of water traffic accidents, structural learning becomes a crucial preliminary stage, and its core purpose is to clarify the topological structure of the Bayesian network. As the basis for building a Bayesian network model, structural learning determines how the nodes in the model are connected to each other and the dependencies between them.

[0129] In this process, the constrained search method plays a decisive role. This method uses a variety of techniques in statistics and information theory, such as chi-square test, mutual information and G2 test. In this study, the chi-square test is selected as the core analysis tool for structural learning. The chi-square test is used to analyze the statistical correlation in the accident data and identify the potential connections between different causal factors, thus providing a basis for constructing the topological structure of the Bayesian network model.

[0130] For example, taking C2 (lack of safety awareness) and D1 (improper lookout) as examples, the chi-square test is used to test whether there is a significant statistical association between these two factors. The chi-square test is used to analyze the potential connection between C2 (lack of safety awareness) and D1 (improper lookout), and the results are as follows: Figure 3 As shown, Figure 3 The accident layer is not drawn in the figure. The figure shows that the expected frequency (Expected Count) is significantly greater than 5 (specifically 116.19), and the total sample analysis volume reaches 964, which meets the conditions for Pearson's Chi-Squared Test. The test results show that the P value is less than 0.05. According to the conventional statistical criteria, this result rejects the null hypothesis (that C2 and D1 are independent of each other) and accepts the alternative hypothesis that there is a statistically significant relationship between C2 (lack of safety awareness) and D1 (improper lookout).

[0131] 3) Analysis of the cause chain of water traffic accidents

[0132] After the topological structure of the Bayesian network is determined through the structural learning method, the next key step is to solve the network parameters.

[0133] In this process, Maximum Likelihood Estimation (MLE), as a commonly used statistical method, is widely used in the parameter learning of Bayesian networks. This method is based on existing accident data and estimates the conditional probability table (CPT) of each node by calculating the likelihood function of the data and finding the parameters that maximize the value of the function. The result of the parameter solution will give the conditional probability table of each node in the Bayesian network model, that is, the probability distribution of each node given the state of its parent node.

[0134] Next, based on the conditional probability table obtained above, we further analyze the dependencies between different factors and their influence.

[0135] ① Identification of key factors in water traffic accidents

[0136] The Bayesian network constructed for the cause analysis of water traffic accidents plays a core role in the identification of key factors. Identifying and analyzing these key factors is essential for understanding the causes of water traffic accidents, predicting accident risks, and formulating effective preventive measures and emergency response strategies. Figure 4 The node probability of each node in the , extract the factors of the nodes with the top three occurrence probabilities in each level, and organize them into a key factor table, such as Figure 5 shown.

[0137] ②Sensitivity analysis of causes of water traffic accidents

[0138] Sensitive factors can be identified through sensitivity analysis of Bayesian networks. The analysis aims to determine the degree of influence of network parameters on the posterior probability of specific target nodes, thereby identifying the key factors that lead to various types of accidents.

[0139] The core algorithm of Bayesian network sensitivity analysis (global sensitivity analysis algorithm) is to quantify the influence of parameter changes on the posterior probability by calculating the differential of the posterior probability of the target node with respect to the network parameters.

[0140]

[0141] S θ is the posterior probability of the target node P(X i |Pa(X i ))Sensitivity value of the network parameter θ.

[0142] θ is a parameter of the Bayesian network, usually a parameter in the conditional probability table (CPT).

[0143] In specific applications, in the GeNle software, by setting the state of the target node, sensitivity analysis can be performed, which can intuitively show the importance of the upper-level factors to the lower-level factors. Combined with the calculation results of the sensitivity parameters, further quantitative analysis can be made.

[0144] Taking the wind disaster accident as an example, E4 in the accident layer is set as the target node and sensitivity analysis is performed. The color of the Bayesian network node changes to intuitively display the location of sensitive factors. The red node is the node where the relevant parameter is located. The darker the color, the more sensitive the relevant parameter of the node. Figure 6In the figure, we can intuitively see that theoretical knowledge, work experience, skill level, other lack of ability (C1), strong wind, waves, and current (C7), and improper typhoon prevention measures (D12) are the most relevant factors for wind disasters. The intensity of each factor needs to be combined with the sensitivity parameter for quantitative evaluation. The maximum, minimum, and average values ​​of the sensitivity of the relevant parameters of each node can be obtained by calculation using the GeNle software. Among them, the maximum value is consistent with the color depth of the associated node, that is, the maximum value of the sensitivity analysis is regarded as the main criterion for evaluating the sensitivity of the associated node.

[0145] Through the sensitivity analysis of the causal factors of various types of water traffic accidents, the sensitivity parameter data in Table 3 were obtained. These parameters reveal the relative importance of the causal factors in different accident types.

[0146] Table 3: Results of sensitivity analysis of collision accidents

[0147]

[0148] Through Table 3 and Figure 3 ,For collision accidents, the analysis results show that insufficient safety awareness (C2) has the largest sensitivity parameter absolute value, which indicates that in collision accidents, the crew’s safety awareness level has the most significant impact on the occurrence of the accident. Therefore, improving the crew’s safety awareness is of great significance to reducing collision accidents.

[0149] Table 4: Results of sensitivity analysis of touch accidents

[0150]

[0151] Through Table 4 and Figure 7 For collision accidents, improper collision avoidance behavior (D4) becomes the factor with the largest absolute value of sensitivity parameter among the causal factors. This reflects that during the navigation process of ships, correct collision avoidance behavior is crucial to preventing collision accidents.

[0152] Table 5: Results of sensitivity analysis of fire / explosion accidents

[0153]

[0154] Through Table 5 and Figure 7 ,In the fire / explosion accidents, the absolute value of the sensitivity parameter of improper use of equipment (D7) is the largest, indicating that improper use of equipment is one of the main causes of fire / explosion accidents. Therefore, strengthening the standardization and safety of equipment use is the key to reducing fire / explosion accidents.

[0155] Table 6: Results of sensitivity analysis of wind disaster accidents

[0156]

[0157] Through Table 6 and Figure 7 For wind disaster accidents, the absolute value of the sensitivity parameter of improper typhoon prevention measures (D12) is the largest, indicating that when facing severe weather such as typhoons, whether the typhoon prevention measures taken by ships are appropriate is directly related to the incidence of wind disaster accidents. Therefore, strengthening the formulation and implementation of ship typhoon prevention measures is crucial to reducing the risk of wind disaster accidents.

[0158] Table 7: Sensitivity analysis results of self-sinking accidents

[0159]

[0160] Through Table 7 and Figure 7 In the case of self-sinking accidents, the absolute value of the sensitivity parameter of improper emergency measures (D11) is the largest, indicating that when a ship encounters an emergency, whether the emergency measures taken are appropriate plays a decisive role in preventing self-sinking accidents. Therefore, improving the emergency handling capabilities of crew members and formulating effective emergency measures are important ways to reduce self-sinking accidents.

[0161] Table 8: Results of sensitivity analysis of other accidents

[0162]

[0163] Through Table 8 and Figure 7 ,For other accident types, insufficient safety awareness (C2) also has the largest absolute value of the ,sensitivity parameter, which once again emphasizes the general importance of improving crew safety ,awareness in reducing water traffic accidents.

[0164] ③Cause path of water traffic accidents

[0165] Based on the key factors and sensitivity factors of water traffic accidents obtained above, the causal path of water traffic accidents is constructed next. Based on the results of sensitivity analysis, the causal factors in the causal chain analysis model of water traffic accidents are strictly screened. Specifically, those causal factors that have no significant impact on various types of accidents (i.e., node sensitivity parameters are less than 0.01) are eliminated.

[0166] However, in the analysis of the causal factors of the collision accidents, a special situation was found: the sensitivity of most of the causal factors was less than 0.01. In order to comprehensively and deeply analyze the causal chain of the collision accidents, a special strategy was adopted: when constructing the causal path, three nodes with relatively high sensitivity in each level were selected. That is, the factors with the top three sensitivity parameters in different types of water accidents were selected as sensitivity factors.

[0167] Table 9: Cause chain of different types of accidents

[0168]

[0169] As can be seen from Table 9, the number of causal chains for various types of water traffic accidents varies: there are 12 collision accidents, 3 contact accidents, 6 fire and explosion accidents, 2 wind disaster accidents, 4 self-sinking accidents, and 7 other accidents. The causal chain of collision accidents shows complexity and diversity, covering the intertwined influence of human errors (such as improper lookout and improper operation) and environmental factors (such as insufficient visibility and complex water flow conditions). In contrast, the causal chain of contact accidents is relatively simple, with improper collision avoidance behavior as the core, and is also affected by environmental factors to a certain extent. The causal chain of fire and explosion accidents reveals the potential risks of multiple links from equipment failure to improper human operation, showing its unique complexity. The causal chain of wind disaster accidents is mainly closely related to extreme weather conditions, especially the insufficient wind resistance of ships in bad weather. The causal chain of self-sinking accidents involves improper cargo stowage and ship operation performance issues. The causal chains of other accident categories are more extensive and complex, reflecting the diversity and complexity of the causes of water traffic accidents.

[0170] ④ Global causal chain analysis of water traffic accidents

[0171] The causal reasoning method of Bayesian network is used to analyze the global causal chain of the causal factors of water traffic accidents. Global causal chain analysis is a causal reasoning method based on Bayesian network. This method sets the target node (i.e. the occurrence status of different accident types) as a known condition, and then calculates the posterior probability of each evidence node to infer the possible cause of the accident.

[0172] Taking the wind disaster accident as an example, the occurrence state of the target node is set to 1, and then the posterior probabilities of its four parent nodes - failure to use safe speed, improper lookout, improper collision avoidance behavior, and improper typhoon prevention measures are derived. The results show that the probability of improper typhoon prevention measures is the highest, reaching 69%. Further analysis of the three parent nodes of improper typhoon prevention measures - lack of theoretical knowledge / work experience / skill level / other abilities, lack of safety awareness, and excessive wind / waves / currents based on the calculated posterior probability shows that the posterior probabilities of excessive wind, waves, and currents are the highest. Using the same method, a global causal chain analysis is performed on other accident types, and finally the global causal chain of each accident type is obtained, as shown in Table 10.

[0173] Table 10: Global causal chains of different types of accidents

[0174] Type of accident Global Causal Chain Collision A1-B4-C1-D1-E1 Contact accident A5-B2-C2-D1-E2 Fire and explosion accidents A5-B1-C2-D7-E3 Typhoon accident C7-D12-E4 Sinking accident A1-B4-C1-D3-E5 Other accidents A1(A5)-B4(B2)-C1(C2)-D3-E5

[0175] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. The analysis method of the cause chain of water traffic accidents based on HFACS and Bayesian network is characterized by: The steps include: S1: Obtain water traffic accident data; S2: According to the characteristics of water traffic accidents, the HFACS model for water traffic accident risk analysis is established. The HFACS model divides the causal factors of water traffic accidents into five levels, namely organizational influence, poor supervision, preconditions for unsafe behavior, unsafe behavior, and accident level; S3: Extract the causal factors of water traffic accidents related to the five levels of the HFACS model based on traffic accident data; S4: After determining the topological structure of the Bayesian network model using the structural learning method, the maximum likelihood estimation method is used to solve the network parameters. According to the solution results, the conditional probability table of each node in the Bayesian network model is obtained, that is, the probability distribution of each node given the state of its parent node; S5: According to the conditional probability table obtained in step S4, the factors with the top three occurrence probabilities in each level are taken as the key factors among the causative factors of the water traffic accident; S6: Analyze the sensitivity of the causal factors of water traffic accidents extracted in step S3 using a global sensitivity analysis algorithm of a Bayesian network to obtain sensitivity factors; S7: constructing a causal path of water traffic accidents according to the key factors and sensitivity factors obtained in step S5 and step S6 respectively; S8: Use the causal reasoning method of Bayesian network to obtain the global causal chain of the causal factors of different types of water traffic accidents.

2. The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network according to claim 1 is characterized in that: The water traffic accident data of step S1 is derived from the investigation report information of the water traffic accident, and the traffic accident investigation report information includes the accident type and the cause of the accident.

3. The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network according to claim 2 is characterized in that: The sample size of the water traffic accident data is sufficient for conducting a Pearson chi-square test.

4. The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network according to claim 1 is characterized in that: In step S4, the chi-square test is used as a tool of the structural learning method. Specifically, the chi-square test is used to analyze the statistical correlation in the water traffic accident data, identify the potential connections between different causal factors, and construct the topological structure of the Bayesian network model.

5. The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network according to claim 1 is characterized in that: Construction of the causal path in step S7: Based on the sensitivity analysis results of step S6, the causal factors with insignificant effects are deleted and the remaining influencing factors are constructed into a causal path according to the Bayesian network nodes constructed in step S4.

6. The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network according to claim 5 is characterized in that: The insignificantly affecting causal factors are the node sensitivity parameters less than 0.01; or the causal factors outside the three nodes with relatively high sensitivity in each level.

7. The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network according to claim 1 is characterized in that: In step S8, the target node is set as a known condition, and then the posterior probability of each evidence node is calculated according to the following formula (5) to infer the possible cause of the accident: P(A|B)=P(B|A)*P(A) / P(B) (5).

8. The method for analyzing the cause chain of water traffic accidents based on HFACS and Bayesian network according to claim 1 is characterized in that: The operation of the maximum likelihood estimation method in step S4 is as follows: Suppose the node variable X in the Bayesian network i There are k values, corresponding to the parent node Pa(X i ) has j combinations, then the parameters of the network can be expressed as θ ijk =(X i =k|Pa(X i )=j) (6) The above formula (6) refers to the i-th node variable X i The parent node Pa(X i ) takes the value j, X i The conditional probability of taking value k is parameter θ ijk The value of Define sample H l The characteristic function of satisfy Sample H l Satisfy X i = k, Pa (X i )=j, the number of samples is denoted as μ ijk ; Let parameter θ ijk The maximum likelihood estimate of In the above formula, m is the node variable X in the Bayesian network. i There are m possible values.

9. An electronic device, It is characterized in that include: one or more processors; And a memory associated with the one or more processors, the memory is used to store program instructions, and when the program instructions are read and executed by the one or more processors, the method according to any one of claims 1-8 is executed.

10. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images