Method and device for constructing locomotive steward accident multi-factor cause model

Abstract

The invention provides a method and a device for constructing a locomotive steward accident multi-factor cause model. The method comprises the following steps of: constructing an influence factor system of safe driving of a locomotive steward; based on the direct influence relation matrix of the influence factor system, calculating to obtain a comprehensive influence matrix of the influence factor system; generating an adversarial hierarchical topological graph based on the comprehensive influence matrix; and taking the antagonistic hierarchical topological graph as an initial graph structure of a Bayesian network, and constructing an accident multi-factor cause model. According to the method provided by the invention, the accident multi-factor cause model is finally constructed by constructing the multi-dimensional influence factor system, quantitatively calculating the comprehensive influence relationship, generating the antagonistic hierarchical topological graph and scientifically, deeply, comprehensively and accurately revealing the complex cause behind the accident of the locomotive steward; the limitation that a traditional analysis method focuses on a single factor or a linear relation is overcome, then accurate decision support is provided for railway safety management, and the effectiveness of railway accident prevention is further improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of railway transportation safety, and in particular to a method and device for constructing a locomotive attendant accident multifactor causation model. BACKGROUND

[0002] Accidents caused directly or indirectly by locomotive attendant factors still occur from time to time, which constitutes an important risk source of railway operation safety. In-depth analysis of the complex causation mechanism behind the locomotive attendant accident has extremely important practical significance for realizing accurate accident prevention, improving safety management efficiency, and ensuring long-term stability of railway transportation. Traditional accident analysis often focuses on isolated investigation of a single factor or simple linear attribution.

[0003] However, the railway transportation system is a highly complex human-machine-environment-management dynamic coupling system, and the operation behavior of the locomotive attendant and its safety is influenced by multiple levels and multiple dimensions of factors. Therefore, merely focusing on the individual level of the attendant's mistake or a single isolated technical failure cannot comprehensively and deeply reveal the systemic root of the accident, nor can it provide sufficient theoretical support for targeted prevention measures. SUMMARY

[0004] The present application provides a method and device for constructing a locomotive attendant accident multifactor causation model to solve the defect that the prior art only focuses on the individual level of mistake and cannot comprehensively and deeply reveal the systemic root of the accident.

[0005] The present application provides a method for constructing a locomotive attendant accident multifactor causation model, comprising: constructing an influence factor system of safe driving of a locomotive attendant; calculating a comprehensive influence matrix of the influence factor system based on a direct influence relationship matrix of the influence factor system; generating an adversarial hierarchical topology graph based on the comprehensive influence matrix; constructing an accident multifactor causation model by taking the adversarial hierarchical topology graph as an initial graph structure of a Bayesian network.

[0006] According to the method for constructing a locomotive attendant accident multifactor causation model provided by the present application, the generation of an adversarial hierarchical topology graph based on the comprehensive influence matrix comprises: calculating an accessible matrix based on the comprehensive influence matrix; performing hierarchical and regional division on the accessible matrix according to extraction rules to calculate a skeleton matrix; generating the adversarial hierarchical topology graph based on the skeleton matrix.

[0007] According to the locomotive attendant accident multi-factor causation model construction method provided by the application, the reachable matrix is calculated based on the comprehensive influence matrix, and the reachable matrix comprises: Based on the comprehensive influence matrix, the centrality and reason degree of each influence factor are calculated. Based on the screening threshold and the centrality and reason degree of each influence factor, the initial relationship matrix of the influence factor system is constructed. Based on the initial relationship matrix and the unit matrix, the overall relationship matrix is constructed, and the reachable matrix is calculated by multiplying the overall relationship matrix.

[0008] According to the locomotive attendant accident multi-factor causation model construction method provided by the application, the reachable matrix is calculated based on the comprehensive influence matrix, and the reachable matrix comprises: The influence factors in the adversarial hierarchical topology graph are mapped as nodes, and the influence relationship is mapped as a directed arc, so as to construct the initial graph structure. The state of each node in the initial graph structure is discretized to determine the variable value space of each node. Based on the locomotive attendant accident background data and the variable value space of each node, the conditional probability table of each node is learned and generated to construct the accident multi-factor causation model.

[0009] According to the locomotive attendant accident multi-factor causation model construction method provided by the application, the reachable matrix is calculated based on the comprehensive influence matrix, and the reachable matrix comprises: Based on the influence degree between each influence factor in the influence factor system and the remaining influence factors, the direct influence relationship matrix is constructed. The direct influence relationship matrix is normalized to calculate a normalized influence matrix. The normalized influence matrix is subjected to matrix operation to obtain the comprehensive influence matrix.

[0010] The application also provides a locomotive attendant accident analysis method, comprising: Obtaining locomotive attendant accident data to be analyzed; The locomotive attendant accident data to be analyzed is input into the accident multi-factor causation model to obtain the accident factor analysis result output by the accident multi-factor causation model. The accident multi-factor causation model is constructed based on any one of the above locomotive attendant accident multi-factor causation model construction methods.

[0011] The application further provides a locomotive attendant accident multi-factor causation model construction device, comprising: An influencing factor system construction unit constructs an influencing factor system of safe driving of the locomotive attendant; A matrix calculation unit calculates a comprehensive influence matrix of the influencing factor system based on a direct influence relationship matrix of the influencing factor system; A topological graph generation unit generates an adversarial hierarchical topological graph based on the comprehensive influence matrix; A model construction unit constructs an accident multi-factor causation model by taking the adversarial hierarchical topological graph as an initial graph structure of a Bayesian network.

[0012] The application further provides a locomotive attendant accident analysis device, comprising: An acquisition unit acquires locomotive attendant accident data to be analyzed; An analysis unit inputs the locomotive attendant accident data to be analyzed into an accident multi-factor causation model to obtain an accident factor analysis result output by the accident multi-factor causation model; The accident multi-factor causation model is constructed based on any one of the locomotive attendant accident multi-factor causation model construction methods.

[0013] The application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the locomotive attendant accident multi-factor causation model construction method or the locomotive attendant accident analysis method according to any one of the above.

[0014] The application further provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable on a processor to implement the locomotive attendant accident multi-factor causation model construction method or the locomotive attendant accident analysis method according to any one of the above.

[0015] The application further provides a computer program product comprising a computer program, wherein the computer program is executable on a processor to implement the locomotive attendant accident multi-factor causation model construction method or the locomotive attendant accident analysis method according to any one of the above.

[0016] The method and apparatus for constructing a multi-factor causation model for locomotive crew accidents provided by this invention, by constructing a multi-dimensional system of influencing factors, quantifying and calculating the comprehensive influence relationship, generating an adversarial hierarchical topology diagram, and finally constructing a multi-factor causation model for accidents, can scientifically, deeply, comprehensively and accurately reveal the complex causes behind locomotive crew accidents. It overcomes the limitations of traditional analysis methods that focus on single factors or linear relationships, and thus provides a precise decision support tool for railway safety management, further improving the effectiveness of railway accident prevention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the method for constructing a multi-factor causative model of locomotive crew accidents provided by the present invention; Figure 2 A schematic diagram of the influencing factor system provided for this invention; Figure 3 This is the second flowchart illustrating the method for constructing a multi-factor causative model of locomotive crew accidents provided by the present invention. Figure 4 This is a flowchart illustrating the locomotive crew accident analysis method provided by the present invention; Figure 5 This is a schematic diagram of the structure of the device for constructing a multi-factor causation model of locomotive crew accidents provided by the present invention; Figure 6 This is a schematic diagram of the locomotive crew accident analysis device provided by the present invention; Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] It should be noted that railway transportation undertakes the important tasks of passenger and freight transportation, and its operation safety is directly related to the safety of life and property and the social and economic development. As the core operator of train operation and the direct executor of the last line of defense for safety, the operation behavior and safety status of locomotive crew members have a decisive impact on preventing train accidents. However, despite the continuous improvement of railway technical equipment level and the increasing perfection of safety management system, accidents directly or indirectly caused by locomotive crew members still occur from time to time, such as overrunning signals, speeding, misrouting, incomplete lookout, etc., which constitute important risk sources for railway operation safety.

[0021] Traditional accident analysis usually focuses on single factors. However, the causes of railway safety accidents are usually multiple. Especially in specific typical scenarios, such as operation in bad weather, complex line sections, night driving, slow construction sections, intensive operation in stations, high-density train operation, emergency response to emergencies, etc., various internal and external risk factors are more likely to be activated, superimposed and even produce non-linear amplification effects, which may ultimately lead to accidents. These scenarios often have characteristics such as high task complexity, large information load, tight time pressure, many environmental disturbances, and low decision-making error tolerance, posing severe challenges to the physiological and psychological states, cognitive judgment abilities, operation proficiency and emergency response levels of crew members. Merely focusing on the mistakes at the individual level of crew members or a certain isolated technical failure is difficult to comprehensively and profoundly reveal the systematic root causes of accidents, and it is also impossible to provide sufficient theoretical support for targeted preventive measures.

[0022] Currently, although a large number of studies have been carried out in the fields of railway safety, human factors engineering, accident causation theory, etc., the research on the multi-factor interaction mechanism and the construction of systematic causation models for locomotive crew accidents in typical scenarios is still insufficient. Existing studies either focus on general human error analysis, lacking in-depth exploration of the particularity of specific scenarios; or focus on objective factors such as equipment and environment, and the description of the cognitive behavior patterns of crew members in specific situations and their interaction with other elements of the system is not fine enough; or the proposed models are too macroscopic to guide the refined risk prevention and control practice in specific scenarios. Therefore, there is an urgent need to construct a comprehensive causation model that can integrate the characteristics of typical scenarios, multi-dimensional causation factors and their interaction relationships.

[0023] To address the above problems, the present invention provides a method for constructing a multi-factor causation model for locomotive crew accidents to identify key factors and their interrelationships, construct a model that can comprehensively and profoundly analyze locomotive accident factors, and thus provide scientific decision-making support for railway safe driving. Figure 1 is one of the flow schematic diagrams of the method for constructing a multi-factor causation model for locomotive crew accidents provided by the present invention. As Figure 1 shown, the method includes: Step 110: Construct a system of factors influencing safe driving by locomotive crew members.

[0024] Here, the influencing factor system refers to a comprehensive and multi-dimensional collection of all relevant factors that may affect the safe driving behavior and state of locomotive crew members. This system serves as the foundational framework for subsequent causal analysis and model building, ensuring the comprehensiveness and systematic nature of the analysis and avoiding the omission of key risk sources.

[0025] Specifically, this influencing factor system can be constructed through extensive literature analysis, interviews with senior experts in the field of railway safety, and in-depth review of historical accident cases. These methods can initially identify numerous potential factors related to locomotive crew driving safety. It should be noted that these factors can be summarized and classified from multiple dimensions, such as human factors, vehicle factors, environmental factors, and management factors. To ensure the scientific rigor and practicality of the system, the initially identified factors can be screened and optimized, for example, by eliminating factors with weak correlations or vague descriptions, and merging factors with overlapping connotations, ultimately forming a clearly structured and hierarchical influencing factor system.

[0026] In one embodiment, Figure 2 A schematic diagram of the influencing factor system provided by the present invention, as shown below. Figure 2 As shown, the influencing factors include four primary indicators: human factors, vehicle factors, social factors, and external factors, as well as 23 secondary indicators, including: fatigue level, driver's emotional state and mental health, driver's decision-making and cognitive abilities, driver's attitude and sense of responsibility, poor driving behavior, physiological state, safety awareness, information and communication, train equipment status, train design and manufacturing, safety technology and innovation, intelligent driving assistance systems, maintenance and upkeep, social background, economic status and work environment, education level and cultural background, social relationships and interpersonal communication, management systems and norms, laws and regulations, weather conditions, road conditions, animals and pedestrians, and other vehicles.

[0027] The following table lists the influencing factors and their descriptions provided by this invention: It can be seen that the influencing factors include four primary indicators: human factors, vehicle factors, social factors, and external factors, as well as 23 secondary indicators. Each influencing factor is explained to achieve comprehensive coverage of the influencing factors.

[0028] It should be noted that by constructing a system of influencing factors, a complete and reliable input is provided for subsequent quantitative analysis, ensuring that the model can cover the main dimensions of accident occurrence from the root, and laying the foundation for improving the accuracy of accident prevention.

[0029] Step 120: Based on the direct influence relationship matrix between any influencing factor and the other influencing factors in the influencing factor system, calculate the comprehensive influence matrix of the influencing factor system.

[0030] Here, the direct influence matrix is ​​a mathematical matrix used to quantitatively describe the direct and explicit influence relationship between any two factors in a system of influencing factors. It should be noted that the element values ​​in the matrix represent the strength of the direct effect of one factor on another. This matrix can quantify the initial causal relationship between factors and serves as the basis for calculating the overall impact.

[0031] Furthermore, the comprehensive influence matrix here is calculated using a specific algorithm based on the direct influence matrix. It reflects the overall influence relationship between any two factors in the system, including all direct and indirect transmission paths. In essence, the comprehensive influence matrix can be used to reveal deep, implicit connections between factors and comprehensively measure the overall influence of each factor in the entire system.

[0032] Specifically, a direct influence matrix can be established based on expert evaluation. For example, a questionnaire can be designed to invite multiple experts in the field to rate the degree of direct influence between each pair of factors in the influencing factor system. The rating can use a multi-level scaling method, such as the 0-4 scale, where 0 represents no influence and 4 represents a strong influence. By summarizing and averaging the ratings from all experts, the direct influence matrix can be constructed.

[0033] Subsequently, the direct impact matrix can be normalized to eliminate the influence of dimensions, and mathematical methods such as the decision laboratory method can be used for calculation to finally obtain the comprehensive impact matrix. This calculation process takes into account the transmission and cumulative effects of the impact in the system network.

[0034] It should be noted that transforming expert experience and knowledge into quantitative mathematical models not only identifies the direct effects between factors but also delves deeper into the indirect effects. This enables the scientific and objective measurement of the true influence of each factor in a complex system, providing data support for identifying key factors.

[0035] Step 130: Based on the comprehensive influence matrix, generate an adversarial hierarchical topology graph.

[0036] Here, the adversarial hierarchical topology diagram is a directed, acyclic hierarchical structure diagram that visually displays the dependencies between influencing factors, the transmission paths of influence, and the hierarchy to which the factors belong. The adversarial aspect lies in its construction process's ability to stratify factors from different perspectives, such as prioritizing causes or effects, thereby revealing the structural characteristics of the system more comprehensively. Understandably, the adversarial hierarchical topology diagram can be used to transform complex numerical matrix relationships into an intuitive graphical model, clearly revealing which are the surface-level direct causes and which are the deep-seated root causes.

[0037] Specifically, based on the comprehensive influence matrix obtained in the previous step, an influence threshold can be set first to filter out the more significant influence relationships in the system, thereby simplifying the network structure and forming a reachability matrix. Then, methods such as the adversarial interpretive structural model are applied to hierarchically divide the reachability matrix. By iteratively extracting top-level factors that are not affected by other factors or have minimal influence, and bottom-level factors that do not affect other factors or have minimal influence, all factors are gradually assigned to different levels, and the factor nodes in each level are connected according to the influence relationships, finally drawing an adversarial hierarchical topology diagram.

[0038] It should be noted that by generating an adversarial hierarchical topology graph, the complex interaction relationships between factors are clearly presented, revealing the path and level of risk transmission from root factors to direct factors. This allows decision-makers to intuitively understand the mechanism of the accident and provides a reasonable initial structure for the subsequent construction of Bayesian networks.

[0039] Step 140: Using the adversarial hierarchical topology graph as the initial graph structure of the Bayesian network, a multi-factor causation model of the accident is constructed.

[0040] Here, the initial graph structure refers to the preliminary network topology used when constructing the Bayesian network model, which defines the initial dependencies between nodes in the network. Furthermore, the multi-factor causal model of the accident described here is the final, completed probabilistic graphical model based on a Bayesian network. This model not only includes the causal structure between factors but also quantifies the strength of these causal relationships through conditional probability tables, enabling probabilistic inference, risk prediction, and sensitivity analysis.

[0041] Specifically, the adversarial hierarchical topology graph generated in the previous step can be directly converted into the initial graph structure of a Bayesian network. Subsequently, parameter learning is performed on the Bayesian network, which involves determining the conditional probability table for each node in the initial graph structure. This conditional probability table can be determined by analyzing historical accident data, conducting simulation experiments, or utilizing expert knowledge again, and is used to quantify the probability that child nodes will take different states when parent nodes are in different states. After completing parameter learning, a complete and usable multi-factor causal model of an accident is obtained.

[0042] Thus, the qualitative, structured topology map is upgraded into a dynamic, probabilistic, quantitative model. The multi-factor causation model of accidents can integrate multi-source information to simulate the impact of changes in different factors on the probability of the final accident, thereby enabling dynamic extrapolation of accident risks and simulation evaluation of the effectiveness of intervention measures.

[0043] The method provided in this invention, by constructing a multi-dimensional system of influencing factors, quantifying and calculating the comprehensive influence relationship, generating an adversarial hierarchical topology diagram, and finally constructing a multi-factor causal model of accidents, can scientifically, deeply, comprehensively and accurately reveal the complex causes behind locomotive crew accidents. It overcomes the limitations of traditional analysis methods that focus on a single factor or linear relationship, and thus provides a precise decision support tool for railway safety management, further improving the effectiveness of railway accident prevention.

[0044] Based on any of the above embodiments, step 130 includes: Based on the comprehensive influence matrix, the reachability matrix is ​​calculated; According to the extraction rules, the reachability matrix is ​​divided into levels and regions, and the skeleton matrix is ​​calculated. Based on the skeleton matrix, the adversarial hierarchical topology graph is generated.

[0045] Here, the reachability matrix refers to a Boolean matrix, whose elements consist of 0s and 1s, used to describe whether an influence path exists between any two factors in the influencing factor system. If the element in the i-th row and j-th column of the matrix is ​​1, it means that factor i can influence factor j through one or more paths; if it is 0, it means that no influence path exists. It should be noted that the purpose of the reachability matrix is ​​to transform the quantified comprehensive influence matrix with different intensities of information into a structural relationship matrix that only focuses on whether influence relationships exist, laying the foundation for subsequent system structure analysis and hierarchical division.

[0046] Specifically, the first step is to set an influence threshold. This threshold can be determined based on expert experience or through statistical analysis of the values ​​in the comprehensive influence matrix, such as by taking the mean or the mean plus one standard deviation. The purpose is to filter out weaker influence relationships in the system and highlight the main contradictions. Elements in the comprehensive influence matrix greater than or equal to this threshold are set to 1, and elements less than the threshold are set to 0, resulting in an initial Boolean matrix. Subsequently, this initial Boolean matrix is ​​added to an identity matrix, and Boolean operations are used to perform continuous exponentiation on the resulting matrix until the matrix no longer changes, i.e., the matrix converges. This finally converged matrix is ​​the reachability matrix.

[0047] Next, the classic extraction rules from the interpretative structural model can be used. For any factor i in the reachability matrix, find its reachability set R(i) and its antecedent set A(i). If the intersection of R(i) and A(i) equals its reachability set R(i), i.e., R(i) ∩ A(i) = R(i), then factor i is determined to be the highest-level factor in the system. All extracted highest-level factors are removed from the matrix, and the above extraction process is repeated in the remaining submatrices to determine the second level, third level, etc., until all factors are assigned to their corresponding levels.

[0048] Understandably, after completing the hierarchical division, the skeleton matrix is ​​constructed. This can be done by checking the reachability matrix to eliminate all connections arising from transitivity. For example, if factor A influences factor B, factor B influences factor C, and factor A directly influences factor C, then the connection from A to C is considered redundant and eliminated. After this elimination, the skeleton matrix is ​​obtained.

[0049] Furthermore, the skeleton matrix obtained in the previous step can be used as a direct basis for drawing. For example, all influencing factors can be treated as nodes and arranged according to the previously defined hierarchy. For instance, the bottom-level factors can be placed at the bottom of the graph, the top-level factors at the top, and other level factors in the middle. Then, directed edges between nodes are drawn based on the element values ​​in the skeleton matrix. If the element in the i-th row and j-th column of the skeleton matrix is ​​1, then a directed arrow is drawn from the node corresponding to factor i to the node corresponding to factor j. Thus, a clear, hierarchical, and intuitively reflective adversarial hierarchical topology diagram can be obtained.

[0050] It should be noted that by transforming abstract matrix data into an intuitive adversarial hierarchical topology, researchers can quickly grasp the core logical relationships between influencing factors and the main paths of accident evolution without needing to delve into complex mathematical calculations, greatly enhancing the readability and practicality of the analysis results.

[0051] The method provided in this invention, which generates a topology graph by calculating the reachability matrix, applying extraction rules to divide the hierarchy, and calculating the skeleton matrix, effectively avoids the subjectivity and arbitrariness of manually drawing topology graphs. It ensures the scientific nature, objectivity, and simplicity of the transformation process from comprehensive influence data to the final graphical model, thus providing a strong guarantee for constructing a more accurate and reliable initial structure of Bayesian networks.

[0052] Based on any of the above embodiments, and based on the comprehensive influence matrix, the reachability matrix is ​​calculated, including: Based on the comprehensive influence matrix, the centrality and causality of each influencing factor are calculated. Based on the screening threshold and the centrality and causality of each influencing factor, an initial relationship matrix of the influencing factor system is constructed. Based on the initial relation matrix and the identity matrix, a global relation matrix is ​​constructed, and the reachability matrix is ​​calculated by multiplying the global relation matrix together.

[0053] Here, centrality is an indicator used to measure the overall importance of a single influencing factor within the entire system of influencing factors. It can be obtained by adding the sum of the influence that factor exerts on all other influencing factors to the sum of the influences that factor receives from all other influencing factors. Understandably, an influencing factor with higher centrality means that it interacts more frequently and is more closely related to other factors in the system, making it a key node in the system.

[0054] The causality degree here is an indicator used to determine whether a single influencing factor leans towards being a cause or an effect in the system. It can be obtained by calculating the difference between the total influence that the influencing factor exerts on other influencing factors and the total influence it receives from other influencing factors. Understandably, a positive causality degree indicates that the influencing factor's external influence is greater than its internal influence, tending to be the root cause of other problems in the system, thus belonging to a causal factor; conversely, a negative value indicates that it is more of a result of being influenced by other factors, belonging to an effect factor.

[0055] Specifically, firstly, based on the calculated centrality of each influencing factor, the summation of each row in the comprehensive influence matrix yields the total influence exerted by each factor, denoted as the influence degree of the corresponding influencing factor in that row. Then, the summation of each column in the comprehensive influence matrix yields the total impact of each influencing factor, denoted as the degree of influence of the corresponding influencing factor in that column. For any influencing factor, its centrality is the sum of its influence degree and its degree of influence, and its causality is the difference between its influence degree and its degree of influence.

[0056] In one embodiment, centrality and causality can be calculated using the following formulas, as shown below: In the formula, Indicates the degree of influence of factor i; This indicates the total number of influencing factors. This indicates the degree of influence between influencing factor i and influencing factor j; This indicates the degree to which influencing factor i is affected; Indicates the centrality of influencing factor i; This indicates the degree of causality of factor i.

[0057] Next, based on an assessment of the overall system complexity, or according to analytical experience, combined with the centrality and causality distributions of each factor calculated in the previous step, and the numerical characteristics of the comprehensive influence matrix itself, a global screening threshold can be determined. Subsequently, each element in the comprehensive influence matrix is ​​traversed, and its value is compared with the screening threshold. If the element value is greater than or equal to the screening threshold, it is recorded as 1 in the corresponding position of the initial relationship matrix, indicating that the influence relationship is significant; if it is less than the screening threshold, it is recorded as 0, indicating that the influence relationship can be ignored.

[0058] For example, a decision matrix can be constructed using centrality and causality. By applying a screening threshold, each element in the decision matrix can be filtered to obtain an initial relation matrix. This initial relation matrix can be obtained using the following formula, as shown below: ; In the formula, This represents the element in the i-th row and j-th column of the initial relation matrix; Indicates the decision threshold; This represents the element in the i-th row and j-th column of the decision matrix.

[0059] To further consider the influence of the influencing factors themselves, a global relation matrix can be constructed based on the initial relation matrix and the identity matrix. For example, the initial relation matrix can be added to an identity matrix of the same order using Boolean addition: 0+0=0, 0+1=1, 1+0=1, 1+1=1, thus constructing the global relation matrix. This global relation matrix can be calculated using the following formula, as shown below: ; In the formula, Represents the overall relational matrix; Represents the initial relation matrix; Represents the identity matrix.

[0060] Finally, the overall relation matrix is ​​multiplied consecutively to obtain the reachability matrix. For example, Boolean multiplication rules can be used to perform continuous self-multiplication operations on the overall relation matrix, i.e., second power, third power, etc., until the result of a certain power operation is exactly the same as the result of the previous power operation, i.e., the matrix reaches a stable state. This finally stable matrix is ​​the desired reachability matrix. The reachability matrix can be calculated using the following formula, as shown below: In the formula, Represents the reachability matrix; represents the overall relation matrix; k represents the power of the exponent.

[0061] The method provided in this invention simplifies the influence network by introducing quantitative indicators such as centrality, causality, and screening threshold, and uses rigorous Boolean matrix operations to derive reachability relationships, significantly improving the objectivity and accuracy of the reachability matrix construction process, and laying a reliable foundation for generating adversarial hierarchical topology graphs with higher reliability.

[0062] Based on any of the above embodiments, step 140 includes: The initial graph structure is constructed by mapping the influencing factors in the adversarial hierarchical topology graph to nodes and the influencing relationships to directed arcs. The state of each node in the initial graph structure is discretized to determine the value space of the variables of each node; Based on the background data of locomotive crew accidents and the variable value space of each node, a conditional probability table is learned and generated for each node, and the multi-factor causal model of the accident is constructed.

[0063] Here, the initial graph structure refers to the topological skeleton of the Bayesian network. It graphically defines the conditional dependencies between various influencing factors in the multi-factor causation model of an accident, and serves as the foundation for subsequent probabilistic parameter learning. In essence, the initial graph structure directly determines the causal logic that the model can express.

[0064] Specifically, each influencing factor in the generated adversarial hierarchical topology graph can be converted into a random variable node in a Bayesian network. Simultaneously, each directed arrow representing an influence relationship in the topology graph can be converted into a directed arc connecting the corresponding node in the Bayesian network. The direction of these directed arcs is consistent with the direction of the arrows in the topology graph, pointing from the causal factor to the result factor. Thus, through this mapping, an initial Bayesian network graph structure with the same topological structure as the adversarial hierarchical topology graph can be obtained.

[0065] It should be noted that using a hierarchical topology graph to provide a highly reasonable and causal logical initial structure for the Bayesian network effectively avoids the huge computational overhead and potential structural errors caused by blindly learning the network structure from data, ensuring that the theoretical basis of the multi-factor causation model of the accident is consistent with the actual mechanism.

[0066] Next, the nodes in the initial graph structure are discretized to determine the variable value space for each node. Specifically, this can be determined based on the nature of the influencing factors. For discrete categorical variables, such as weather conditions, the value space can be directly defined as {sunny, rainy, windy}, etc. For continuous numerical variables, such as continuous driving time, equal-width binning or equal-frequency binning can be used to divide them into several intervals, such as {less than 4 hours, 4-6 hours, more than 6 hours}. For abstract factors that are difficult to quantify directly, such as psychological stress, they can be divided into several states such as {low stress, moderate stress, high stress} based on industry standards, expert questionnaire ratings, or relevant psychological scales. After this processing is completed for all nodes, each variable has a clear and finite value space.

[0067] Finally, based on the background data of locomotive crew accidents, conditional probability tables are learned and generated for each node, thus constructing a multi-factor causation model for the accidents. The background data of locomotive crew accidents here can include historical data and expert experience.

[0068] Specifically, firstly, a large amount of historical accident cases involving locomotive crew members, safety hazard reports, and daily operation records are collected and organized as background data for locomotive crew member accidents. Then, the nodes in the background data for locomotive crew member accidents are represented by the variable value space of each defined node.

[0069] Then, for each node in the network, its conditional probability table is computed using this database and statistical learning algorithms, such as maximum likelihood estimation or Bayesian estimation. In a Bayesian network, each node has a conditional probability distribution given its parent nodes. Here, Pa(X) is the set of parent nodes of node X. Let the set of variables... , For variables The set of parent nodes, then The conditional probability is: ; In the formula, each node is represented by 23 variables, with the variable category representing the node state, and the relationship between different node states is expressed using conditional probability. Indicates; n represents the total number of nodes; Represents a node The set of direct parent nodes; This indicates that when the parent node takes a specific state, The conditional probability distribution.

[0070] Furthermore, the state information U of the corresponding node can be updated by adjusting the probability of a certain category of the variable, and the probability changes of other nodes can be inferred using Bayesian theory. ; In the formula, This represents the posterior probability distribution of the variable set V after U is observed; Let U represent the prior probability distribution, and let V represent the initial belief about V before observing U. Let U represent the marginal probability (normalization constant) of evidence U, ensuring that the sum of the posterior probabilities is 1; This represents the set of observed evidence variables used to update the probabilities of other nodes in the network.

[0071] It should be noted that once all nodes in the network have generated their corresponding conditional probability tables, a complete and usable multi-factor causation model for an accident is constructed.

[0072] Based on any of the above embodiments, step 120 includes: Based on the degree of influence between each influencing factor and the other influencing factors in the influencing factor system, the direct influence relationship matrix is ​​constructed. The direct influence matrix is ​​normalized to obtain the normalized influence matrix; Perform matrix operations on the normalized influence matrix to obtain the comprehensive influence matrix.

[0073] Here, the direct influence matrix is ​​a numerical matrix used to initially quantify the strength of the direct interaction between any two factors in a system of influencing factors. Each element in the matrix represents the magnitude of the direct, immediate, and uninterrupted influence of one factor on another.

[0074] Specifically, an expert survey method can be used. For example, a questionnaire can be designed and distributed to multiple experts in fields such as railway safety, locomotive driving, and human factors engineering. The questionnaire requires experts to rate the degree of direct influence of each pair of factors in the influencing factor system based on their professional knowledge and practical experience. Here, a clear scale can be used for the rating, such as a 5-point scale from 0 to 4, where 0 represents no direct influence at all, 1 represents a very weak influence, 2 represents a moderate influence, 3 represents a relatively strong influence, and 4 represents a very strong influence. After collecting the rating questionnaires from all experts, the ratings of each expert on the same influence relationship can be summarized, for example, by calculating their arithmetic mean, thus obtaining a direct influence relationship matrix that integrates the opinions of multiple experts.

[0075] Next, the direct influence matrix is ​​normalized to obtain the normalized influence matrix. Specifically, the sum of elements in each row of the direct influence matrix is ​​calculated, resulting in a list containing the total direct influence of each factor. Then, the largest row sum is found from this list. Finally, each element in the direct influence matrix is ​​divided by this found largest row sum. This row sum normalization method yields the normalized influence matrix. All elements in the matrix are scaled to between 0 and 1.

[0076] Here, the normalized influence matrix can be calculated using the following formula, as shown in the following equation: In the formula, Represents the normalized influence matrix; This indicates the direct impact relation matrix; This indicates a direct influence on any element in the relationship matrix, specifically representing the influencing factors. Influencing factors The degree of influence is defined as follows: when i = j, (in ); n represents the total number of influencing factors.

[0077] It should be noted that standardized mathematical processing ensures the consistency and comparability of the impact metrics, laying the foundation for the stable convergence and correct calculation of the subsequent comprehensive impact matrix, and avoiding calculation errors that may be caused by different scales of the original data.

[0078] Furthermore, matrix operations are performed on the normalized influence matrix to obtain the comprehensive influence matrix. Here, the comprehensive influence matrix can be calculated using the following formula, as shown below: In the formula, Represents the comprehensive impact matrix; Represents the identity matrix.

[0079] Based on any of the above embodiments Figure 3 This is the second flowchart illustrating the method for constructing a multi-factor causative model of locomotive crew accidents provided by this invention. Figure 3 As shown, the method includes: First, in the DEMATEL phase, the core objective is to quantify the comprehensive influence relationships among various influencing factors. In this phase, a direct influence matrix is ​​first constructed using expert scoring to preliminarily describe the strength of the direct effects between factors. To eliminate dimensions and standardize the data, this method supports various normalization processes, such as row sum and maximum value methods. After obtaining the normalized matrix, matrix operations are used to calculate a comprehensive influence matrix that reflects all direct and indirect influences. Based on this comprehensive influence matrix, four key indicators for each factor are further calculated: influence degree, degree of influence, centrality, and causation degree. A centrality-causation degree scatter plot can be generated to visually classify all factors into roles, such as causal factors and outcome factors.

[0080] Next, in the AISM phase, the core objective is to construct a clear hierarchical model based on the quantification results from DEMATEL. In this phase, firstly, the reachability matrix is ​​calculated using the comprehensive influence matrix from the previous phase, thereby identifying the transmission paths between all factors in the system. Subsequently, a skeleton matrix is ​​constructed to eliminate redundant influence relationships, simplifying the complex system structure. During the hierarchical division process, this method supports UP (outcome-first) and DOWN (cause-first) extraction rules, ultimately generating an adversarial hierarchical topology diagram that intuitively displays the risk transmission paths and factor hierarchical dependencies.

[0081] Finally, in the Bayesian network stage, the core objective is to transform the aforementioned deterministic causal structure into a probabilistic model capable of dynamic analysis. In this stage, firstly, the hierarchical topology graph generated by AISM is used as the initial graph structure of the Bayesian network, i.e., a directed acyclic graph. Then, by combining historical accident data, simulation data, or expert knowledge for parameter learning, a conditional probability table is generated for each node (influencing factor) in the network, thereby accurately quantifying the strength of dependencies between factors. The completed Bayesian network model can support dynamic probabilistic inference, sensitivity analysis of key factors, risk assessment in specific scenarios, and simulation analysis of the effects of different intervention measures, ultimately providing data-driven decision support for safety management.

[0082] The method provided in this invention integrates three analysis methods: DEMATEL, AISM, and Bayesian networks. It systematically identifies and quantifies the interactions among 23 human factors affecting the safe driving of locomotive crew members, constructing a dynamically interpretable accident causation model. By calculating the causal degree, centrality, and conditional probability of each factor, it reveals the risk transmission path and key intervention nodes, enabling dynamic extrapolation of risk probabilities and optimization of prevention and control strategies in multi-factor coupled scenarios. This invention can provide data-driven decision support for railway safety management. Railway track maintenance safety is a crucial link in ensuring railway traffic safety. By analyzing the causes of locomotive crew member accidents in typical scenarios and proposing corresponding prevention and control measures, the accuracy and systematic nature of accident prevention can be effectively improved, which is of great significance for reducing safety hazards and preventing safety accidents.

[0083] Based on any of the above embodiments, the present invention also provides a locomotive crew accident analysis method to achieve accurate and comprehensive accident cause analysis. Figure 4 This is a flowchart illustrating the locomotive crew accident analysis method provided by the present invention, as shown below. Figure 4 As shown, the method includes: Step 410: Obtain the locomotive crew accident data to be analyzed.

[0084] Here, the locomotive crew accident data to be analyzed refers to a set of objective information collected through investigation, evidence gathering, and recording methods for a specific accident, near miss, or safety hazard incident, reflecting the state of various relevant influencing factors at the time of the incident. It is understood that the locomotive crew accident data to be analyzed can serve as evidence for accident diagnosis and reasoning.

[0085] Specifically, for an accident case to be analyzed, accident investigators can collect information from multiple sources. For example, they can review the train's safety recorder data at the time of the accident to determine information such as train speed, signal status, and braking operations; retrieve the crew's shift and rest records to assess their fatigue level; review the vehicle's maintenance history to understand whether there are potential equipment malfunctions; and obtain weather reports from the accident site to clarify the environmental conditions at the time. This collected raw information is then transformed according to the defined value space for each node variable to form a set of definite state observation values, thus obtaining the locomotive crew accident data to be analyzed. For example, if the investigation finds that the crew member worked continuously for 7 hours before the accident, the state of the "continuous driving time" node can be set to "more than 6 hours".

[0086] Step 420: Input the locomotive crew accident data to be analyzed into the multi-factor causation model of the accident to obtain the accident factor analysis results output by the multi-factor causation model of the accident. The multi-factor causation model of the accident is constructed based on any of the above-mentioned methods for constructing a multi-factor causation model of locomotive crew accidents.

[0087] Specifically, the acquired accident data of the locomotive crew members to be analyzed can be input into the multi-factor causation model of the accident. The multi-factor causation model of the accident uses the accident data of the locomotive crew members to be analyzed as observation evidence. Based on the known evidence, probability propagation and updates are carried out in the entire network to calculate the posterior probability distribution of all unobserved nodes, thus obtaining the accident factor analysis results.

[0088] It should be noted that by acquiring objective data from specific accidents and inputting it into a pre-built multi-factor causation model for probabilistic reasoning, more multi-dimensional and scientific accident factor analysis results are generated. This enables a more scientific and in-depth revelation of the key causal chains and root causes of accidents, thus providing a strong basis for decision-making for in-depth accident investigations and the precise formulation of future preventive measures.

[0089] The method provided in this invention, by constructing a multi-dimensional system of influencing factors, quantifying and calculating the comprehensive influence relationship, generating an adversarial hierarchical topology diagram, and finally constructing a multi-factor causal model of accidents, can scientifically, deeply, comprehensively and accurately reveal the complex causes behind locomotive crew accidents. It overcomes the limitations of traditional analysis methods that focus on a single factor or linear relationship, and thus provides a precise decision support tool for railway safety management, further improving the effectiveness of railway accident prevention.

[0090] Based on any of the above embodiments Figure 5 This is a schematic diagram of the structure of the device for constructing a multi-factor causative model of locomotive crew accidents provided by the present invention, as shown in the figure. Figure 5 As shown, the device includes: Unit 510, which constructs a system of influencing factors for safe driving by locomotive crew members; The matrix calculation unit 520 calculates the comprehensive influence matrix of the influencing factor system based on the direct influence relationship matrix of the influencing factor system. The topology graph generation unit 530 generates an adversarial hierarchical topology graph based on the comprehensive influence matrix. The model building unit 540 uses the adversarial hierarchical topology graph as the initial graph structure of the Bayesian network to build a multi-factor causal model of the accident.

[0091] The device provided in this invention, by constructing a multi-dimensional system of influencing factors, quantifying and calculating the comprehensive influence relationship, generating an adversarial hierarchical topology diagram, and finally constructing a multi-factor causal model of accidents, can scientifically, deeply, comprehensively and accurately reveal the complex causes behind locomotive crew accidents. It overcomes the limitations of traditional analysis methods that focus on a single factor or linear relationship, and thus provides a precise decision support tool for railway safety management, further improving the effectiveness of railway accident prevention.

[0092] Based on any of the above embodiments, the topology graph generation unit is specifically used for: Based on the comprehensive influence matrix, the reachability matrix is ​​calculated; According to the extraction rules, the reachability matrix is ​​divided into levels and regions, and the skeleton matrix is ​​calculated. Based on the skeleton matrix, the adversarial hierarchical topology graph is generated.

[0093] Based on any of the above embodiments, the topology graph generation unit is further specifically used for: Based on the comprehensive influence matrix, the centrality and causality of each influencing factor are calculated. Based on the screening threshold and the centrality and causality of each influencing factor, an initial relationship matrix of the influencing factor system is constructed. Based on the initial relation matrix and the identity matrix, a global relation matrix is ​​constructed, and the reachability matrix is ​​calculated by multiplying the global relation matrix together.

[0094] Based on any of the above embodiments, the model building unit is specifically used for: The initial graph structure is constructed by mapping the influencing factors in the adversarial hierarchical topology graph to nodes and the influencing relationships to directed arcs. The state of each node in the initial graph structure is discretized to determine the variable value space of each node; Based on the background data of locomotive crew accidents and the variable value space of each node, a conditional probability table is learned and generated for each node, and the multi-factor causal model of the accident is constructed.

[0095] Based on any of the above embodiments, the matrix calculation unit is specifically used for: Based on the degree of influence between each influencing factor and the other influencing factors in the influencing factor system, the direct influence relationship matrix is ​​constructed. The direct influence matrix is ​​normalized to obtain the normalized influence matrix; Perform matrix operations on the normalized influence matrix to obtain the comprehensive influence matrix.

[0096] Based on any of the above embodimentsFigure 6 This is a schematic diagram of the locomotive crew accident analysis device provided by the present invention, as shown below. Figure 6 As shown, the device includes: Acquisition unit 610 acquires the locomotive crew accident data to be analyzed; The analysis unit 620 inputs the locomotive crew accident data to be analyzed into the multi-factor causation model of the accident, and obtains the accident factor analysis results output by the multi-factor causation model of the accident; The multi-factor causation model of the accident is constructed based on any of the above-mentioned methods for constructing a multi-factor causation model of locomotive crew accidents.

[0097] The device provided in this invention, by constructing a multi-dimensional system of influencing factors, quantifying and calculating the comprehensive influence relationship, generating an adversarial hierarchical topology diagram, and finally constructing a multi-factor causal model of accidents, can scientifically, deeply, comprehensively and accurately reveal the complex causes behind locomotive crew accidents. It overcomes the limitations of traditional analysis methods that focus on a single factor or linear relationship, and thus provides a precise decision support tool for railway safety management, further improving the effectiveness of railway accident prevention.

[0098] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a method for constructing a multi-factor causal model of locomotive crew accidents. This method includes: constructing a system of influencing factors for safe driving by locomotive crew; calculating a comprehensive influence matrix of the influencing factor system based on the direct influence relationship matrix of the influencing factor system; generating an adversarial hierarchical topology graph based on the comprehensive influence matrix; and using the adversarial hierarchical topology graph as the initial graph structure of a Bayesian network to construct a multi-factor causal model of the accident.

[0099] A locomotive crew accident analysis method can also be implemented, the method including: acquiring locomotive crew accident data to be analyzed; inputting the locomotive crew accident data to be analyzed into a multi-factor accident causation model, and obtaining the accident factor analysis results output by the multi-factor accident causation model; the multi-factor accident causation model is constructed based on any of the above-mentioned locomotive crew accident multi-factor causation model construction methods.

[0100] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the construction method of the multi-factor causal model of locomotive crew accidents provided by the above methods. The method includes: constructing a system of influencing factors for safe driving of locomotive crew; calculating a comprehensive influence matrix of the influencing factor system based on the direct influence relationship matrix of the influencing factor system; generating an adversarial hierarchical topology graph based on the comprehensive influence matrix; and constructing a multi-factor causal model of the accident by using the adversarial hierarchical topology graph as the initial graph structure of a Bayesian network.

[0102] A locomotive crew accident analysis method can also be implemented, the method including: acquiring locomotive crew accident data to be analyzed; inputting the locomotive crew accident data to be analyzed into a multi-factor accident causation model, and obtaining the accident factor analysis results output by the multi-factor accident causation model; the multi-factor accident causation model is constructed based on any of the above-mentioned locomotive crew accident multi-factor causation model construction methods.

[0103] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for constructing a multi-factor causal model of locomotive crew accidents provided by the methods described above. The method includes: constructing a system of influencing factors for safe driving by locomotive crew; calculating a comprehensive influence matrix of the influencing factor system based on the direct influence relationship matrix of the influencing factor system; generating an adversarial hierarchical topology graph based on the comprehensive influence matrix; and constructing a multi-factor causal model of the accident by using the adversarial hierarchical topology graph as the initial graph structure of a Bayesian network.

[0104] A locomotive crew accident analysis method can also be implemented, the method including: acquiring locomotive crew accident data to be analyzed; inputting the locomotive crew accident data to be analyzed into a multi-factor accident causation model, and obtaining the accident factor analysis results output by the multi-factor accident causation model; the multi-factor accident causation model is constructed based on any of the above-mentioned locomotive crew accident multi-factor causation model construction methods.

[0105] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0106] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of constructing a locomotive crew accident multifactor causation model, comprising: The method comprises the following steps: constructing an influence factor system of locomotive attendant safe driving; calculating a comprehensive influence matrix of the influence factor system based on a direct influence relationship matrix of the influence factor system; generating an adversarial hierarchical topology graph based on the comprehensive influence matrix; constructing an accident multifactor causation model based on the adversarial hierarchical topology graph as an initial graph structure of a Bayesian network.

2. The method of claim 1, wherein: The method of generating an adversarial hierarchical topology graph based on the comprehensive influence matrix comprises the following steps: calculating a reachable matrix based on the comprehensive influence matrix; dividing the reachable matrix into levels and areas according to extraction rules to calculate a skeleton matrix; generating the adversarial hierarchical topology graph based on the skeleton matrix.

3. The method of claim 2, wherein: The method of calculating a reachable matrix based on the comprehensive influence matrix comprises the following steps: calculating the centrality and reason degree of each influence factor based on the comprehensive influence matrix; constructing an initial relationship matrix of the influence factor system based on a screening threshold and the centrality and reason degree of each influence factor; constructing an overall relationship matrix based on the initial relationship matrix and a unit matrix, and calculating the reachable matrix by multiplying the overall relationship matrix.

4. The method of constructing a locomotive crew accident multifactor causation model according to any one of claims 1 to 3, wherein, The method of constructing an accident multifactor causation model based on the adversarial hierarchical topology graph as an initial graph structure of a Bayesian network comprises the following steps: mapping the influence factors in the adversarial hierarchical topology graph as nodes and the influence relationships as directed arcs to construct the initial graph structure; performing state discretization processing on each node in the initial graph structure to determine the variable value space of each node; learning and generating a conditional probability table for each node based on locomotive attendant accident background data and the variable value space of each node to construct the accident multifactor causation model.

5. The method of constructing a locomotive crew accident multifactor causation model according to any one of claims 1 to 3, wherein, The method of calculating a comprehensive influence matrix of the influence factor system based on a direct influence relationship matrix of the influence factor system comprises the following steps: constructing the direct influence relationship matrix based on the influence degree between each influence factor in the influence factor system and the remaining influence factors; performing normalization processing on the direct influence relationship matrix to calculate a normalized influence matrix; performing matrix operation on the normalized influence matrix to obtain the comprehensive influence matrix.

6. A locomotive crew incident analysis method characterized by, The method comprises the following steps: obtaining locomotive attendant accident data to be analyzed; inputting the locomotive attendant accident data to be analyzed into an accident multifactor causation model to obtain an accident factor analysis result output by the accident multifactor causation model; The accident multifactor causation model is constructed based on the construction method of the locomotive attendant accident multifactor causation model in any one of the above claims 1 to 5.

7. A device for constructing a locomotive crew accident multifactor causation model, characterized by, The method comprises the following steps: an influence factor system construction unit for constructing an influence factor system of locomotive attendant safe driving; a matrix calculation unit for calculating a comprehensive influence matrix of the influence factor system based on a direct influence relationship matrix of the influence factor system; a topology graph generation unit for generating an adversarial hierarchical topology graph based on the comprehensive influence matrix; a model construction unit for constructing an accident multifactor causation model based on the adversarial hierarchical topology graph as an initial graph structure of a Bayesian network.

8. A locomotive crew incident analysis device characterized by, The method comprises the following steps: An acquisition unit acquires locomotive crew accident data to be analyzed; An analysis unit inputs the locomotive crew accident data to be analyzed into an accident multifactor causation model to obtain an accident factor analysis result output by the accident multifactor causation model; The accident multifactor causation model is constructed based on the construction method of the locomotive crew accident multifactor causation model in any one of claims 1 to 5.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the construction method of the locomotive crew accident multifactor causation model in any one of claims 1 to 5 or the locomotive crew accident analysis method in claim 6 when executing the computer program.

10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the construction method of the locomotive crew accident multifactor causation model in any one of claims 1 to 5 or the locomotive crew accident analysis method in claim 6 when executed by the processor.

Images