Dynamic portrait construction and intelligent management method and system for railway crew members

Abstract

The invention provides a dynamic portrait construction and intelligent management method and system for railway crew members, and relates to the technical field of computers, and the method comprises the steps: obtaining the multi-source original data of the railway crew members; performing data fusion according to the multi-source original data, and constructing an initial crew knowledge graph; performing feature aggregation according to the initial crew member knowledge graph to obtain a standard crew member knowledge graph; performing time sequence feature extraction according to the standard crew knowledge graph to obtain dynamic behavior features; carrying out matching and correlation analysis on the dynamic behavior characteristics and loads and response conditions in different working scenes to obtain a portrait label set; and making a decision according to the portrait label set, comparing the real-time emergency disposal data with a preset threshold value and associating a knowledge graph to position a training missing root, and obtaining an optimization management suggestion. According to the invention, dynamic and accurate portrait construction and intelligent management decision making for crew members are realized.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method and system for constructing and intelligently managing dynamic profiles of railway attendants. Background Technology

[0002] In the field of intelligent railway transportation management, accurate characterization and evaluation of the work status and professional capabilities of train attendants are crucial for improving passenger service quality and operational safety. Traditional personnel profiling technology provides a basic framework for personnel evaluation by integrating multi-dimensional data to generate labeled feature models. In railway attendant management practice, existing technical solutions mainly collect structured scheduling, attendance, and basic assessment data by connecting to a single business system, and rely on preset static rules to perform statistical analysis and simple classification of the data, thereby generating limited static evaluation labels such as "work efficiency" and "attendance compliance" in tabular form to assist management decisions. However, this method is limited by the single and unstructured nature of the data source. The lack of information and cross-system business connections makes it difficult to achieve in-depth perception of complex and dynamic operational scenarios. Its label generation mechanism, based on fixed rules and simple statistics, cannot capture the evolution of crew behavior over time and its complex relationship with specific passenger events and workloads, resulting in static, one-sided, and lagging profiles. At the application level, existing technologies can only provide statistical presentations of historical data, lacking the ability to respond to real-time anomalies and provide dynamic early warnings of potential risks. Furthermore, they cannot perform cross-dimensional tracing and accurate root cause location when problems occur, making it difficult to provide real-time, accurate, and explainable intelligent support for refined management decisions such as crew scheduling, specialized training, and process optimization.

[0003] Based on the shortcomings of the existing technologies, there is an urgent need for a dynamic profile construction and intelligent management method and system for railway crew members. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for constructing dynamic profiles and intelligent management of railway crew members, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides a method for constructing dynamic profiles and intelligent management of railway crew members, including: Acquire multi-source raw data of railway crew members, including service process data, passenger feedback data, and personal qualification data. Data fusion is performed based on the multi-source raw data. By extracting four types of entities—crew members, duty tasks, security management events, and assessment items—and defining cross-system semantic relationships, an initial crew knowledge graph is constructed. Based on the initial cabin crew knowledge graph, feature aggregation is performed. By calculating the attention weight between the duty shift and security management events and weighting and aggregating the neighbor features, a standard cabin crew knowledge graph is obtained. Based on the standard cabin crew knowledge graph, temporal features are extracted. By analyzing the dynamic evolution of service behavior data in the time dimension, the temporal dependency between service response and passenger events is captured, and dynamic behavior features are obtained. The dynamic behavioral characteristics are matched and correlated with the load and response conditions under different working scenarios to obtain a set of profile tags; Decisions are made based on the set of profile tags. By comparing real-time emergency response data with preset thresholds and linking them with a knowledge graph, the root causes of training deficiencies are identified, and optimization management suggestions are obtained.

[0005] Secondly, this application also provides a dynamic profile construction and intelligent management system for railway crew members, including: The acquisition module is used to acquire multi-source raw data of railway attendants, including service process data, passenger feedback data, and personal qualification data. The fusion module is used to perform data fusion based on the multi-source raw data. By extracting four types of entities—crew members, duty tasks, safety management events, and assessment items—and defining cross-system semantic relationships, an initial crew knowledge graph is constructed. The aggregation module is used to perform feature aggregation based on the initial cabin crew knowledge graph. By calculating the attention weight between the duty shift and security management events and weighting and aggregating neighbor features, a standard cabin crew knowledge graph is obtained. The extraction module is used to extract time-series features based on the standard cabin crew knowledge graph. By analyzing the dynamic evolution of service behavior data in the time dimension, it captures the time-series dependency between service response and passenger events to obtain dynamic behavior features. The analysis module is used to match and correlate the dynamic behavioral characteristics with the load and response conditions under different working scenarios to obtain a set of profile tags; The decision-making module is used to make decisions based on the set of profile tags. By comparing real-time emergency response data with preset thresholds and associating them with a knowledge graph, it can locate the root cause of training deficiencies and obtain optimization management suggestions.

[0006] The beneficial effects of this invention are as follows: This invention constructs a knowledge graph by integrating multi-source data from railway crew, and captures dynamic behavioral patterns by aggregating features and temporal analysis based on attention mechanisms, thereby generating scenario-based tags. Ultimately, it achieves dynamic and accurate profiling of crew members and intelligent management decisions. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating a method for constructing and intelligently managing dynamic profiles of railway attendants, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a dynamic profile construction and intelligent management system for railway attendants as described in an embodiment of the present invention.

[0009] The diagram is labeled as follows: 901, Acquisition Module; 902, Fusion Module; 903, Aggregation Module; 904, Extraction Module; 905, Analysis Module; 906, Decision Module. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present 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 the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0011] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0012] Example 1: This embodiment provides a method for constructing dynamic profiles and intelligent management of railway crew members.

[0013] See Figure 1 The figure shows that the method includes steps S100 to S600.

[0014] Step S100: Obtain multi-source raw data of railway attendants, including service process data, passenger feedback data and personal qualification data; Understandably, this step involves acquiring the basic data needed to construct a crew profile through system integration. Service process data originates from the passenger service and production control platform, obtaining structured data such as train schedules, service response records, and real-time passenger flow statistics through its open data interfaces. Passenger feedback data is generated by integrating text records from the passenger survey system and complaint platform, forming unstructured data containing service evaluations. Personal qualification data is extracted from the crew business interaction system, covering basic personnel information such as crew members' shift arrangements, attendance status, and performance evaluation scores. Simultaneously, multimodal data from across business areas, including emergency response records and safety incident reports, is accessed from the comprehensive safety management information system. This systematic collection of multi-source, heterogeneous data lays the data foundation for the subsequent construction of a comprehensive and multi-dimensional crew profile.

[0015] Step S200: Based on the multi-source raw data, perform data fusion, extract four types of entities: crew members, duty tasks, security management events, and assessment items, and define cross-system semantic relationships to construct an initial crew knowledge graph; It should be noted that step S200 addresses the heterogeneity and dispersion of multi-source data by identifying four core entities: flight attendants, duty tasks, security management events, and assessment items. Based on business logic, it defines the semantic relationships between these entities, integrating data fragments that were originally scattered across different business systems into an initial knowledge graph with a unified semantic framework. This initially breaks down information silos at the data structure level.

[0016] Step S300: Perform feature aggregation based on the initial cabin crew knowledge graph. Calculate the attention weights between duty tasks and security management events and aggregate neighbor features in a weighted manner to obtain a standard cabin crew knowledge graph. Understandably, step S300 uses an attention mechanism to quantify and weight the entity relationships in the initial knowledge graph, paying particular attention to the correlation strength between key business nodes such as duty tasks and safety management events. By strengthening the influence of important connections through feature aggregation, the graph can more accurately reflect the actual interaction relationships of various elements in the duty operation, thereby improving the accuracy and usability of knowledge representation.

[0017] Step S400: Extract time-series features based on the standard cabin crew knowledge graph. By analyzing the dynamic evolution of service behavior data in the time dimension, capture the time-series dependency between service response and passenger events to obtain dynamic behavior features. It should be noted that step S400 targets the temporal characteristics of cabin crew operations. By analyzing the changing patterns of service behavior data over time, it captures the response patterns and behavioral inertia of cabin crew members when facing different types of passenger events, thereby transforming static business records into sequence information that can reflect the dynamic characteristics of individual work patterns.

[0018] Step S500: Match and correlate the dynamic behavioral characteristics with the load and response conditions under different work scenarios to obtain a set of profile tags; Understandably, step S500 interprets the extracted dynamic behavioral features within specific work scenarios. By matching and correlating behavioral patterns with different load conditions and response standards, the abstract temporal features are transformed into a quantifiable and interpretable labeling system, enabling the crew's profile to truly reflect their actual performance under different work pressures.

[0019] Step S600: Make decisions based on the profile tag set, compare real-time emergency response data with preset thresholds and associate with knowledge graphs to locate the root cause of training deficiencies, and obtain optimization management suggestions.

[0020] Finally, step S600 connects the profiling results with actual management decisions. By comparing real-time monitoring data with historical information in the knowledge graph, it can not only promptly detect service anomalies, but also trace the business root causes of problems, providing managers with clear and targeted improvement suggestions, thereby achieving a closed loop from data perception to management execution.

[0021] Further, step S200 includes steps S210 to S230.

[0022] Step S210: Perform entity recognition processing based on multi-source raw data. By parsing the duty route, emergency response records and assessment item scoring data, identify and extract four types of entities and their attributes: crew members, duty tasks, safety management events, and assessment items, to obtain an entity set. Step S220: Perform relation definition processing based on the entity set, define cross-system semantic relationships between entities based on preset service management business rules, and obtain the initial relation network; Step S230: Perform graph construction processing based on the initial relationship network. By merging the entity set with the initial relationship network and completing and resolving conflicts of entity attributes according to business rules, an initial cabin crew knowledge graph is constructed.

[0023] Specifically, step S210 first performs entity recognition processing on the raw data from different business systems. By analyzing specific business data such as crew members' shift information, emergency response records in safety management, and performance evaluation item scores, four core business entities—crew members, shift tasks, safety management events, and evaluation items—and their key attributes are identified and extracted, thus forming a structured entity set. Preferably, this entity recognition processing is implemented using a sequence labeling model (BERT-BiLSTM-CRF model) that combines a pre-trained language model, a bidirectional long short-term memory network, and a conditional random field layer. The hidden layer dimension of the bidirectional long short-term memory network is set to 256, and the conditional random field layer is trained through 100 iterations to optimize label transfer constraints. The pre-trained language model first performs deep semantic encoding on the text data to obtain rich contextual representations; the bidirectional long short-term memory network further captures long-distance bidirectional dependency features in the text; finally, the conditional random field layer learns the transfer constraints between labels to ensure the global optimality of the output entity label sequence. By setting specific network hidden layer dimensions and conducting sufficient iterative training, this model can accurately identify and extract four types of entities—"crew members," "safety management events," "attendance events," and "performance evaluation events"—from unstructured or semi-structured text such as crew scheduling logs, event reports, and performance evaluation comments. For each identified entity, the model simultaneously extracts its key business attributes. For example, for the "crew member" entity, it extracts their employee ID, work group, and training history; for the "safety management event" entity, it extracts the event number, event nature, and handling duration. This process effectively achieves a structured transformation from raw, heterogeneous business text to structured entity knowledge.

[0024] The entity extraction formula is: ; In the formula, The core entity set extracted; This is the entity extraction function; This is a multi-source dataset that integrates data from three major systems. These are the model parameters.

[0025] The property initialization formula is: ; In the formula, Representing entities The initial set of attributes; This is a property mapping function; Represents an entity.

[0026] Based on this, step S220 performs relationship definition processing according to the entity set. Its core lies in formally defining cross-system semantic relationships between entities based on business rules in the cabin crew management field, such as "cabin crew members perform duty tasks," "duty tasks are associated with security events," and "assessment items evaluate cabin crew performance," connecting previously isolated data points into an initial relationship network that reflects business logic. Finally, step S230 performs graph construction processing within this relationship network framework. By merging the entity set with the relationship network and supplementing entity attributes according to business rules (e.g., supplementing missing skill levels based on advanced training completion status) and resolving conflicts (arbitrating based on the authority of the data source when different systems record the same event inconsistently, such as "training system data weight 0.7 > cabin crew system 0.3"), a semantically consistent initial cabin crew knowledge graph with clear business relationships is ultimately constructed, providing a unified and high-quality data foundation for subsequent in-depth analysis and feature aggregation.

[0027] Further, step S300 includes steps S310 to S330.

[0028] Step S310: Perform feature projection processing based on the initial cabin crew knowledge graph. By performing a linear transformation on the initial feature vectors of entities in the graph, the entity features of different dimensions are mapped to a unified feature space to obtain the projected entity feature representation. Step S320: Perform attention calculation processing based on the projected entity feature representation. By analyzing the business correlation strength between the value multiplication task and the emergency response event, calculate the attention weight coefficient between adjacent entities to obtain the correlation strength distribution between entities. Step S330: Perform feature weighted aggregation processing based on the association strength distribution. Based on the preset attention weight coefficient, perform weighted summation on the projection features of neighboring entities to generate an enhanced feature representation for each entity and update the graph node features to obtain the standard cabin crew knowledge graph.

[0029] Specifically, step S310 first performs feature projection processing. Its core is to use a linear transformation matrix to map the initial feature vectors of various entities (such as flight attendants and duty tasks) in the graph, which originally had different dimensions and scales, to a unified and comparable low-dimensional feature space. This process solves the feature heterogeneity problem caused by different data sources and lays the foundation for subsequent deep correlation analysis.

[0030] The linear transformation formula is expressed as: ; In the formula, Represents a node Features after linear transformation; Represents a shared parameter matrix. 10 is the initial dimension, and 20 is the output dimension; node The initial eigenvectors.

[0031] Based on this, step S320 performs attention calculation processing. This processing does not treat all entity connections equally, but specifically targets entity pairs with strong business logic relationships, such as multiplication tasks and emergency response events. It calculates the association strength coefficient between them through a learnable attention mechanism and performs normalization aggregation. This coefficient quantifies the importance of a neighboring entity to the feature representation of the target entity in a specific business context (such as during an emergency response process), thereby obtaining a fine-grained association strength distribution map.

[0032] The formula for calculating attention is: ; In the formula, Represents a node and neighboring nodes Attention relevance score; This is an activation function used to solve the gradient vanishing problem; For attention parameters, The attention parameter has a dimension of 40; Represents a node and nodes The transformed feature vectors are concatenated; This is the transpose of the matrix.

[0033] The normalized aggregation formula is: ; ; This represents the normalized attention weights, reflecting the node's... For nodes The importance of; Represents a node The set of neighboring nodes; Represents a node The final aggregation characteristics; It is a non-linear activation function.

[0034] In the feature weighted aggregation process of step S330, a three-head attention mechanism is adopted for optimization to further improve the robustness of the model and the richness of feature representation. This mechanism deploys three independent attention computing units in parallel, enabling each unit to learn different association patterns within the same entity neighborhood. In a preferred embodiment, the first attention head focuses on analyzing the spatiotemporal association features between duty tasks and security events, the second head focuses on capturing the performance association patterns between crew members and assessment items, and the third head focuses on cross-shift operation pattern associations. Each head performs feature weighted aggregation based on its learned attention weights to generate corresponding subspace feature representations. Finally, the feature representations calculated by the three heads are concatenated and fused to form a comprehensive enhanced feature representation. This three-head parallel computing structure enables the model to collaboratively learn neighborhood features from multiple business dimensions such as spatiotemporal distribution, performance association, and operation patterns, effectively enhancing the comprehensiveness and discriminativeness of feature representations and providing a richer feature foundation for subsequent analysis.

[0035] Further, step S400 includes steps S410 to S430.

[0036] Step S410: Based on the standard crew knowledge graph, perform time series data construction and processing. By extracting service response records and passenger event sequences associated with crew members, and aligning them according to the time window of the shift, structured multidimensional time series data is obtained. Step S420: Perform dynamic pattern analysis based on multidimensional time series data. By analyzing the delay and continuous change patterns of service response indicators relative to event trigger points under different passenger transport scenarios, time series feature patterns are obtained. The time series feature patterns include short-term fluctuations and long-term trend characteristics of service behavior. Step S430: Perform dependency modeling based on temporal feature patterns. By calculating the correlation between service response features and passenger event features at different time lags, capture the temporal dependency between the two and obtain dynamic behavior features.

[0037] Specifically, step S410 first involves constructing and preprocessing time-series data. The core of this process is extracting time-series data associated with specific crew members from the data map, such as key indicators like service frequency per hour and average response time. Subsequently, this raw data is standardized using the Min-Max normalization formula to eliminate the influence of unit dimensions. Finally, the processed data is organized into 24-hour time windows to construct a structured three-dimensional input sequence. The dimensions correspond to the number of samples, time step, and feature dimension, respectively, which prepares the data for subsequent sequence modeling.

[0038] Building upon this, step S420 performs dynamic pattern analysis, implemented by a recurrent neural network model containing an LSTM hidden layer and a fully connected output layer. The core of the model is a gated computation unit containing a forget gate, an input gate, and an output gate. At each time step, the model determines the information to retain from the previous cell state based on the forget gate calculation formula; it determines how much of the current input information is used to update the state based on the input gate and candidate state calculation formulas; then, it calculates the new cell state using the cell state update formula; finally, it obtains the hidden state output for the current time step based on the output gate calculation formula and the hidden state update formula. This series of calculations enables the model to adaptively learn and memorize long-term dependencies in the service behavior sequence, thereby automatically extracting temporal feature patterns that characterize short-term fluctuations and long-term trends in behavior.

[0039] The formula for calculating the forget gate is: ; The formula for calculating the input gate is: ; The formula for calculating candidate states is: ; The cell state update formula is: Output gate: Hidden state .

[0040] In the formula, This represents the output of the forget gate, ranging from 0 to 1. The closer the value is to 1, the more historical information is retained. The sigmoid activation function maps the output to [0,1]. This is the forget gate weight matrix; Forget gate bias term; express The state of the hidden layer is always hidden; express Input data continuously; (This represents the concatenation of the hidden layer state and the input data). This represents the output of the forget gate, ranging from 0 to 1. The closer the value is to 1, the more historical information is retained. This represents the concatenation of the hidden layer state and the input data; This indicates the input gate output, ranging from 0 to 1. The closer the value is to 1, the more current information is retained. Indicates the state of candidate cells. The activation function outputs to [-1, 1]. , The input gate and candidate state weight matrix are used. , For the corresponding bias term; express Cellular state at any given moment (long-term memory); This indicates element-wise multiplication; express Cell state at any given moment; This indicates the output gate output, ranging from 0 to 1, which controls the proportion of cell state output. express The state of the hidden layer is always hidden; This represents the output gate weight matrix; Indicates the output gate bias term; Finally, step S430 performs dependency modeling, which quantifies the dynamic correlation strength between service behavior and external events by calculating the correlation between service response features and passenger event features at different time lags. For example, it analyzes the changing patterns of service response speed within a specific time window after a peak passenger flow event, thereby accurately capturing the complex temporal dependency between the two and ultimately outputting a feature representation that comprehensively reflects the dynamic characteristics of crew behavior. This step trains the model by minimizing the mean squared error loss between the predicted output and the true value; the loss function formula is used to measure the accuracy of the model's predictions. The model parameters are updated using the Adam optimizer, with its learning rate and number of iterations set to a predetermined value (preferably: learning rate 0.001, 100 iterations). Through this optimization process, the model can ultimately accurately capture the complex temporal dependency between service response features and passenger event features, thereby outputting a feature representation that comprehensively reflects the dynamic characteristics of crew behavior.

[0041] The formula for calculating the loss function is: ; In the formula, This represents the mean squared error loss value; Indicates the number of samples; Indicates the first Predicted values ​​for individual samples (e.g., peak service load); Indicates the first The true value of each sample.

[0042] Further, step S500 includes steps S510 to S530.

[0043] Step S510: Perform scenario condition matching processing based on dynamic behavior characteristics. By matching service response time and event handling frequency characteristics with preset passenger flow load thresholds and emergency handling standards, preliminary scenario matching results are obtained. Step S520: Perform multi-dimensional correlation analysis based on the preliminary scenario matching results. By analyzing the correlation between the value multiplication task type and the assessment item score, identify the degree of matching between service performance and business requirements, and obtain the correlation analysis results. Step S530: Based on the association analysis results, perform tag generation processing. By combining the scene matching results and the association analysis results, generate dynamic tags that represent the service capabilities of flight attendants according to preset tag rules, and obtain a profile tag set.

[0044] Specifically, step S510 first performs scenario condition matching processing. The core of this process is to compare and match quantitative behavioral characteristics obtained from time-series analysis, such as service response time and event handling frequency, with pre-set standard thresholds for different scenarios based on business experience (such as peak-hour passenger load thresholds and standardized emergency response durations). For example, the average response time of a train attendant during a sudden surge in passenger flow is compared with the "large passenger flow emergency response standard" to determine whether their performance in a specific scenario meets the standard, obtaining a preliminary matching result based on a single indicator and the scenario. Building on this, step S520 performs multi-dimensional correlation analysis processing. This processing aims to go beyond the judgment of a single indicator by deeply analyzing the statistical correlation and implicit patterns between the train attendant's duty type (such as whether they serve on key trains or inter-regional trains) and their scores on various assessment items (such as service etiquette and operational standards) to comprehensively evaluate the degree of matching between their service performance and complex, multi-dimensional business requirements. For example, analyzing the scores of flight attendants performing high-requirement duties on the "emergency response" item reveals whether they are generally correlated, thereby identifying the structural relationship between business requirements and actual capabilities, and obtaining more in-depth correlation analysis results. Finally, step S530 performs tag generation processing, which merges the matching results of step S510 based on specific scenarios with the correlation analysis results of step S520 reflecting the overall business matching degree, and automatically synthesizes a series of semantic tags that can dynamically and comprehensively characterize the service capabilities and characteristics of flight attendants, such as "peak load ≥ 0.8 and response time ≤ 60 seconds → strong peak response capability" and "peak load ≥ 0.8 and response time > 120 seconds → weak peak response capability", ultimately forming a structured profile tag set.

[0045] Further, step S600 includes steps S610 to S630.

[0046] Step S610: Perform anomaly detection processing based on the profile tag set and real-time collected service data. By comparing the real-time response time, complaint frequency and preset thresholds based on historical data statistics, identify service anomaly events and obtain anomaly detection results. Step S620: Based on the anomaly detection results, perform root cause analysis and processing. By querying the related knowledge graph, identify the training records, duty records, and similar event handling procedures of the crew members associated with the anomaly event, locate the root cause of missing training content or deviation in process execution, and obtain the analysis results. Step S630: Based on the analysis results, generate decision suggestions by matching training resources and scheduling adjustment strategies in the business rule base to generate targeted special training plans or process optimization suggestions, thus obtaining optimized management suggestions.

[0047] Specifically, step S610 first performs anomaly detection processing, the core of which is to continuously monitor the real-time collected cabin crew business data stream, such as the current service response time, complaint frequency, and cabin crew employee number, and compare these real-time indicators with preset dynamic thresholds derived from historical data statistical analysis in real time. It should be noted that the system's preset warning thresholds (such as emergency response time greater than 180 seconds, or complaint frequency reaching 3 times per hour) are set based on the statistical quantiles of historical data. When the real-time data meets any threshold condition, an warning is triggered, thereby completing the transformation from massive data to key abnormal signals and obtaining detection results that identify abnormal events and their associated employee numbers. Based on this, step S620 performs root cause analysis. This process automatically queries detailed information related to abnormal events by associating with the constructed standard crew knowledge graph. For example, it queries the crew member's complete training records and duty history based on the abnormal employee number, and calls the graph interface to obtain standardized handling procedures for similar security incidents. Through comparative analysis, the system can automatically locate the root cause. For example, if it is found that the relevant training records of the personnel are missing, the cause is located as "missing training content"; if the training records are complete but the execution is found to be non-standard when compared with the standard handling procedure, the cause is located as "deviation in process execution", thus obtaining analysis results with clear attribution. Finally, step S630 generates decision recommendations by intelligently matching the root cause analysis results with preset resources and strategies in the business rule base. For example, for the root cause of "lack of training", it automatically matches the "special training plan" template and resources; for "process execution deviation", it matches the "process reinforcement guidance" strategy, thereby generating targeted management outputs that include specific employee numbers, anomaly details, root cause location and clear recommendations. For example, "Employee number 2023001 emergency response time 210 seconds, root cause is lack of training, it is recommended to start special emergency response training", ultimately forming optimized management recommendations that can guide actual actions.

[0048] Example 2: like Figure 2 As shown, this embodiment provides a dynamic profile construction and intelligent management system for railway crew members. The system includes: The acquisition module 901 is used to acquire multi-source raw data of railway crew members, including service process data, passenger feedback data and personal qualification data. The fusion module 902 is used to perform data fusion based on multi-source raw data. By extracting four types of entities—crew members, duty tasks, safety management events, and assessment items—and defining cross-system semantic relationships, an initial crew knowledge graph is constructed. The aggregation module 903 is used to perform feature aggregation based on the initial cabin crew knowledge graph. It calculates the attention weight between the duty and security events and aggregates the neighbor features in a weighted manner to obtain the standard cabin crew knowledge graph. The extraction module 904 is used to extract time-series features based on the standard cabin crew knowledge graph. By analyzing the dynamic evolution of service behavior data in the time dimension, it captures the time-series dependency between service response and passenger events to obtain dynamic behavior features. Analysis module 905 is used to match and correlate dynamic behavioral characteristics with load and response conditions under different work scenarios to obtain a set of profile tags; The decision module 906 is used to make decisions based on the profile tag set. By comparing real-time emergency response data with preset thresholds and linking them with a knowledge graph, it can locate the root cause of training deficiencies and obtain optimization management suggestions.

[0049] In one specific embodiment of this application, the fusion module 902 includes: The first fusion unit is used to perform entity recognition processing based on multi-source raw data. By parsing the duty route, emergency response records and assessment item scoring data, it identifies and extracts four types of entities and their attributes: crew members, duty tasks, safety management events, and assessment items, and obtains an entity set. The second fusion unit is used to perform relation definition processing based on the entity set, and to define cross-system semantic relationships between entities based on preset service management business rules to obtain an initial relation network; The third fusion unit is used to perform graph construction processing based on the initial relationship network. By fusion of the entity set and the initial relationship network, and by completing and resolving conflicts of entity attributes according to business rules, an initial cabin crew knowledge graph is constructed.

[0050] In one specific embodiment of this application, the aggregation module 903 includes: The first aggregation unit is used to perform feature projection processing based on the initial passenger service knowledge graph. By performing a linear transformation on the initial feature vectors of entities in the graph, entity features of different dimensions are mapped to a unified feature space to obtain the projected entity feature representation. The second aggregation unit is used to perform attention calculation processing based on the projected entity feature representation. By analyzing the business correlation strength between the value multiplication task and the emergency response event, it calculates the attention weight coefficient between adjacent entities and obtains the correlation strength distribution between entities. The third aggregation unit is used to perform feature weighted aggregation processing based on the association strength distribution. It performs weighted summation of the projected features of neighboring entities based on preset attention weight coefficients, generates an enhanced feature representation of each entity, and updates the graph node features to obtain a standard passenger service knowledge graph.

[0051] In one specific embodiment of this application, the extraction module 904 includes: The first extraction unit is used to construct and process time-series data based on the standard crew knowledge graph. It extracts service response records and passenger event sequences associated with crew members and aligns them according to the time window of the shift to obtain structured multidimensional time-series data. The second extraction unit is used to perform dynamic pattern analysis processing based on multidimensional time series data. By analyzing the delay and continuous change patterns of service response indicators relative to event trigger points under different passenger transport scenarios, time series feature patterns are obtained. The time series feature patterns include short-term fluctuations and long-term trend characteristics of service behavior. The third extraction unit is used to perform dependency modeling based on temporal feature patterns. By calculating the correlation between service response features and passenger event features at different time lags, the temporal dependency between the two is captured, and dynamic behavioral features are obtained.

[0052] In one specific embodiment of this application, the analysis module 905 includes: The first analysis unit is used to perform scenario condition matching processing based on dynamic behavioral characteristics. By matching service response time and event handling frequency characteristics with preset passenger flow load thresholds and emergency handling standards, preliminary scenario matching results are obtained. The second analysis unit is used to perform multi-dimensional correlation analysis based on the preliminary scenario matching results. By analyzing the correlation between the value multiplication task type and the assessment item score, it identifies the degree of matching between service performance and business requirements and obtains the correlation analysis results. The third analysis unit is used to generate tags based on the association analysis results. By combining the scene matching results and the association analysis results, dynamic tags representing the service capabilities of flight attendants are generated according to preset tag rules, resulting in a profile tag set.

[0053] In one specific embodiment of this application, the decision module 906 includes: The first decision unit is used to perform anomaly detection processing based on the profile tag set and real-time collected service data. By comparing the real-time response time, complaint frequency and preset thresholds based on historical data statistics, it identifies service anomaly events and obtains anomaly detection results. The second decision-making unit is used to perform root cause analysis based on the anomaly detection results. By querying the related knowledge graph, it can locate the root cause of missing training content or deviation in process execution by identifying the trainees' training records, duty resumes and similar incident handling procedures associated with the anomaly. The third decision-making unit is used to generate decision recommendations based on the analysis results. By matching training resources and scheduling adjustment strategies in the business rule base, it generates targeted special training plans or process optimization suggestions, thus obtaining optimized management suggestions.

[0054] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing dynamic profiles and intelligent management of railway attendants, characterized in that, include: Obtain multi-source raw data of railway attendants, including service process data, passenger feedback data, and personal qualification data; Data fusion is performed based on the multi-source raw data. By extracting four types of entities—crew members, duty tasks, security management events, and assessment items—and defining cross-system semantic relationships, an initial crew knowledge graph is constructed. Based on the initial cabin crew knowledge graph, feature aggregation is performed. By calculating the attention weight between the duty shift and security management events and weighting and aggregating the neighbor features, a standard cabin crew knowledge graph is obtained. Based on the standard cabin crew knowledge graph, temporal features are extracted. By analyzing the dynamic evolution of service behavior data in the time dimension, the temporal dependency between service response and passenger events is captured, and dynamic behavior features are obtained. The dynamic behavioral characteristics are matched and correlated with the load and response conditions under different working scenarios to obtain a set of profile tags; Decisions are made based on the set of profile tags. By comparing real-time emergency response data with preset thresholds and linking them with a knowledge graph, the root causes of training deficiencies are identified, and optimization management suggestions are obtained.

2. The method for constructing and intelligently managing dynamic profiles of railway crew members according to claim 1, characterized in that, Based on the aforementioned multi-source raw data, data fusion is performed. By extracting four types of entities—crew members, duty tasks, security incidents, and assessment items—and defining cross-system semantic relationships, an initial crew knowledge graph is constructed, including: Based on the multi-source raw data, entity recognition processing is performed. By parsing the duty routes, emergency response records, and assessment item scoring data, four types of entities and their attributes, namely, crew members, duty tasks, safety management events, and assessment items, are identified and extracted to obtain an entity set. The entity set is processed to define relationships, and cross-system semantic relationships between entities are defined based on preset service management business rules to obtain an initial relationship network. The initial relationship network is used for graph construction. By merging the entity set with the initial relationship network and completing and resolving conflicts of entity attributes according to business rules, an initial cabin crew knowledge graph is constructed.

3. The method for constructing and intelligently managing dynamic profiles of railway crew members according to claim 1, characterized in that, Based on the initial cabin crew knowledge graph, feature aggregation is performed. By calculating the attention weights between duty tasks and security management events and weighted aggregating neighbor features, a standard cabin crew knowledge graph is obtained, including: Based on the initial passenger service knowledge graph, feature projection processing is performed. By performing a linear transformation on the initial feature vectors of entities in the graph, entity features of different dimensions are mapped to a unified feature space to obtain the projected entity feature representation. Attention calculation is performed based on the projected entity feature representation. By analyzing the business association strength between the multiplication task and the emergency response event, the attention weight coefficient between adjacent entities is calculated to obtain the association strength distribution between entities. Based on the association strength distribution, feature weighting and aggregation processing is performed. The projected features of neighboring entities are weighted and summed based on preset attention weight coefficients to generate an enhanced feature representation for each entity. The graph node features are then updated to obtain a standard passenger service knowledge graph.

4. The method for constructing and intelligently managing dynamic profiles of railway crew members according to claim 1, characterized in that, Based on the aforementioned standard cabin crew knowledge graph, temporal features are extracted. By analyzing the dynamic evolution of service behavior data over time, the temporal dependency between service response and passenger events is captured, resulting in dynamic behavioral features, including: Based on the standard crew knowledge graph, time-series data is constructed and processed. By extracting service response records and passenger event sequences associated with crew members, and aligning them according to the time window of the shift, structured multidimensional time-series data is obtained. Dynamic pattern analysis is performed on the multidimensional time series data. By analyzing the delay and continuous change patterns of service response indicators relative to event trigger points under different passenger transport scenarios, time series feature patterns are obtained. The time series feature patterns include short-term fluctuations and long-term trend characteristics of service behavior. Dependency modeling is performed based on the temporal feature pattern. By calculating the correlation between service response features and passenger event features at different time lags, the temporal dependency between the two is captured, and dynamic behavioral features are obtained.

5. The method for constructing and intelligently managing dynamic profiles of railway crew members according to claim 1, characterized in that, The dynamic behavioral characteristics are matched and correlated with the load and response conditions under different working scenarios to obtain a set of profile tags, including: Based on the dynamic behavioral characteristics, scenario condition matching is performed. By matching service response time and event handling frequency characteristics with preset passenger flow load thresholds and emergency handling standards, preliminary scenario matching results are obtained. Based on the preliminary scenario matching results, multi-dimensional correlation analysis is performed. By analyzing the correlation between the task type and the score of the assessment item, the degree of matching between service performance and business requirements is identified, and the correlation analysis results are obtained. Based on the association analysis results, tag generation is performed. By combining the scene matching results and the association analysis results, dynamic tags representing the service capabilities of flight attendants are generated according to preset tag rules, resulting in a profile tag set.

6. A dynamic profile construction and intelligent management system for railway crew members, characterized in that, include: The acquisition module is used to acquire multi-source raw data of railway attendants, including service process data, passenger feedback data, and personal qualification data. The fusion module is used to perform data fusion based on the multi-source raw data. By extracting four types of entities—crew members, duty tasks, safety management events, and assessment items—and defining cross-system semantic relationships, an initial crew knowledge graph is constructed. The aggregation module is used to perform feature aggregation based on the initial cabin crew knowledge graph. By calculating the attention weight between the duty shift and security management events and weighting and aggregating neighbor features, a standard cabin crew knowledge graph is obtained. The extraction module is used to extract time-series features based on the standard cabin crew knowledge graph. By analyzing the dynamic evolution of service behavior data in the time dimension, it captures the time-series dependency between service response and passenger events to obtain dynamic behavior features. The analysis module is used to match and correlate the dynamic behavioral characteristics with the load and response conditions under different working scenarios to obtain a set of profile tags; The decision-making module is used to make decisions based on the set of profile tags. By comparing real-time emergency response data with preset thresholds and associating them with a knowledge graph, it can locate the root cause of training deficiencies and obtain optimization management suggestions.

7. The dynamic profile construction and intelligent management system for railway crew members according to claim 6, characterized in that, The fusion module includes: The first fusion unit is used to perform entity recognition processing based on the multi-source raw data. By parsing the duty route, emergency response records and assessment item scoring data, it identifies and extracts four types of entities and their attributes: crew members, duty tasks, safety management events, and assessment items, to obtain an entity set. The second fusion unit is used to perform relationship definition processing based on the entity set, define cross-system semantic relationships between entities based on preset service management business rules, and obtain an initial relationship network; The third fusion unit is used to perform graph construction processing based on the initial relationship network. By fusing the entity set with the initial relationship network and completing and resolving conflicts of entity attributes according to business rules, an initial crew knowledge graph is constructed.

8. The dynamic profile construction and intelligent management system for railway crew members according to claim 6, characterized in that, The aggregation module includes: The first aggregation unit is used to perform feature projection processing based on the initial passenger service knowledge graph. By performing a linear transformation on the initial feature vectors of entities in the graph, entity features of different dimensions are mapped to a unified feature space to obtain the projected entity feature representation. The second aggregation unit is used to perform attention calculation processing based on the projected entity feature representation. By analyzing the business association strength between the value multiplication task and the emergency response event, it calculates the attention weight coefficient between adjacent entities and obtains the association strength distribution between entities. The third aggregation unit is used to perform feature weighted aggregation processing based on the association strength distribution, and to perform weighted summation of the projection features of neighboring entities based on preset attention weight coefficients to generate an enhanced feature representation of each entity, and update the graph node features to obtain a standard passenger service knowledge graph.

9. The dynamic profile construction and intelligent management system for railway crew members according to claim 6, characterized in that, The extraction module includes: The first extraction unit is used to construct and process time-series data based on the standard crew knowledge graph. By extracting service response records and passenger event sequences associated with crew members, and aligning them according to the time window of the shift, structured multidimensional time-series data is obtained. The second extraction unit is used to perform dynamic pattern analysis processing based on the multidimensional time series data. By analyzing the delay and continuous change patterns of service response indicators relative to event trigger points under different passenger transport scenarios, a time series feature pattern is obtained. The time series feature pattern includes short-term fluctuations and long-term trend characteristics of service behavior. The third extraction unit is used to perform dependency modeling based on the time-series feature pattern. By calculating the correlation between service response features and passenger event features at different time lags, the time-series dependency between the two is captured to obtain dynamic behavior features.

10. The dynamic profile construction and intelligent management system for railway crew members according to claim 6, characterized in that, The analysis module includes: The first analysis unit is used to perform scenario condition matching processing based on the dynamic behavior characteristics. By matching the service response time and event handling frequency characteristics with preset passenger flow load thresholds and emergency handling standards, a preliminary scenario matching result is obtained. The second analysis unit is used to perform multi-dimensional correlation analysis based on the preliminary scenario matching results. By analyzing the correlation between the value multiplication task type and the assessment item score, it identifies the degree of matching between service performance and business requirements and obtains the correlation analysis results. The third analysis unit is used to generate tags based on the association analysis results. By combining the scene matching results and the association analysis results, dynamic tags representing the service capabilities of flight attendants are generated according to preset tag rules to obtain a profile tag set.

Images