Intelligent service performance management integrated platform

Abstract

The present application relates to traffic operation management and computer data processing technical field, disclose a kind of wisdom attendant performance management integrated platform, comprising: interface control module, for task queue asynchronous distribution and background thread pool multithreading parallel scheduling;Data cleaning module, for original table cell filling, name conversion and missing rate filtering;Data fusion module extracts deduction value, realizes seamless fusion with structured driver node;Equal assessment module, for standardization mapping and quota division;Adaptive reconstruction module, for equal boundary score conflict secondary objective index is introduced to disturb rearrangement;Streaming protection and log module, to abnormal analysis with regular missing label;State repair module, based on label execution log reverse tracing, hot update secondary extraction and target vehicle set local cascade recalculation refresh.The present application realizes the efficient, high fault tolerance, absolute fair accounting of large capacity multi-source heterogeneous attendant data set.

Description

Technical Field

[0001] This invention relates to the fields of transportation operation management and computer data processing technology, specifically to an integrated platform for intelligent passenger service performance management. Background Technology

[0002] Train attendant performance refers to the comprehensive work performance and output of train attendants within rail transit, public transportation, and other operating companies, encompassing aspects such as daily driving, safe operation, attendance, and compliance with rules and regulations. Scientific, dynamic, and systematic performance management of train attendants is not only the fundamental institutional foundation for ensuring transportation operation safety and improving team service quality, but also a core management tool for companies to optimize human resource allocation and rationally implement salary incentive mechanisms.

[0003] Traditional performance evaluation of flight attendants is completed by manually collecting, summarizing, and checking paper records or discrete electronic spreadsheets from each team. However, when faced with large-scale flight attendant datasets from multiple teams, managers need to open different attendance files and team forms one by one, manually move, compare, and perform algebraic addition and subtraction calculations, making the entire accounting process extremely cumbersome and inefficient. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an integrated intelligent passenger service performance management platform, which solves the problem of extremely cumbersome and inefficient traditional technical calculation processes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an integrated intelligent passenger service performance management platform, comprising: The interface control module is used to capture performance calculation trigger instructions through event listeners and submit them to the task queue, so that the background thread pool can distribute them to the background worker threads for asynchronous calculation. The data cleaning module is used to convert the original table file into a two-dimensional data matrix, perform cell splitting and filling and regular expression matching of fleet names, and remove invalid rows through compliance scanning; The data fusion module is used to cascade structured data with employee ID as the primary key, dynamically build an adaptive sliding window, and call compound regular expressions to extract deduction values ​​to update the original performance score. The grade assessment module is used to extract the extreme values ​​of the team scores and apply the min-max normalization method to calculate the standardized scores, and then perform grade division and verify the grade boundary scores; The adaptive reconstruction module is used to send a secondary reconstruction callback instruction when the verification fails, extract multi-level objective technical indicators and apply a multi-level decay weight factor reconstruction model to calculate the fine-tuning score for local rearrangement. The streaming protection and logging module is used to perform event-driven line-by-line reading and memory control of data, synchronously generate traceability log files, and add regular expression missing tags when extraction fails. The state repair module is used to hot update the rules in the dynamic configuration dictionary, and to perform targeted source retrieval based on the missing regular expression tags to perform local state overlay, and to use the cascade control unit to schedule the target fleet set for cascade refresh.

[0006] Preferably, the interface control module includes: The foreground user interface thread unit is used to render the graphical user interface and submit the captured performance calculation trigger instructions to the task queue. The background thread pool scheduling unit is used to dynamically activate a corresponding number of concurrent worker threads based on the number of tasks in the queue and the number of processor cores. The color encoding processing unit is used to receive node status signals sent by the background worker thread in real time, and according to the preset mapping relationship between the status flag and the color matrix, call the color encoding processor to render the running status data into the corresponding preset color.

[0007] Preferably, the data cleaning module includes: The matrix filling unit is used to convert the original table file into a two-dimensional data matrix and write the stored value of the valid cell into each corresponding independent cell after splitting and merging the area; The name conversion unit is used to scan the text columns in the two-dimensional data matrix and use a regular expression matching engine to uniformly replace non-standard fleet text with standard fleet identifiers. The missing value filtering unit is used to calculate the missing value rate of a single row, and to mark the row data as removed when the missing value rate is greater than or equal to the judgment threshold, and when the content of a single row matches a preset blacklist.

[0008] Preferably, the data fusion module includes: The structured cascading unit is used to retrieve and cascade the regular assessment scores, safe driving mileage without failure, and clock-in timestamps of the crew drivers using the employee number as the primary key, and assigns the regular assessment scores to the original performance score field. An adaptive window unit is used to determine the start and end boundary coordinates of an adaptive sliding window centered on the target work number position by bidirectional scanning of regular expressions and iteration of boundary symbols. The unstructured extraction unit is used to extract the deduction scores within the adaptive sliding window covering semantic block using compound regular expressions, and deduct them from the original performance score of the corresponding driver node.

[0009] Preferably, the grade assessment module includes: The extreme value extraction and boundary protection unit is used to scan the final raw performance scores of all drivers in a specific fleet to extract the maximum and minimum values, and when the two are equal, it forces the standardized score of all drivers in the fleet to be the preset maximum score and stops the subsequent calculation. The linear normalization unit is used to apply min-max normalization logic to uniformly map the original performance scores of the flight attendants to a preset standard range when the maximum value is greater than the minimum value, so as to obtain a standardized score with a preset number of decimal places. The quota allocation and verification unit is used to sort the standardized score sets of all drivers in descending order, divide the performance grades based on the preset quota ratio, and use the rule constraint to make conditional judgments on the grade boundary scores.

[0010] Preferably, the adaptive reconstruction module includes: The linkage response unit is used to receive the boundary score absolute equality conflict signal and send back the secondary reconstruction callback instruction when the boundary score does not meet the constraint. The feature multi-level retrieval unit is used to respond to the secondary reconstruction callback instruction, and retrieve the corresponding driver node based on the employee number of the passenger driver in the boundary parallel state, and extract the safe driving fault-free mileage as the first-level indicator and the check-in timestamp millisecond value as the second-level indicator across the table. The perturbation fine-tuning unit is used to perform algebraic reconstruction on the extracted safe driving fault-free mileage and check-in timestamps by combining the first-level and second-level perturbation weight factors, and obtain the fine-tuning score for local descending reordering of drivers with boundary parallel conflicts.

[0011] Preferably, the streaming protection and logging module includes: The event flow control unit is used to read the crew dataset line by line in an event stream, and suspend the reading and call the garbage collection operator when the dynamic memory safety factor calculated based on the memory monitor is less than or equal to the critical trigger threshold. The full-stack traceability log unit is used to run synchronously on the core computing nodes to capture the input parameters, intermediate computing status, key operator outputs and worker ID primary key mapping relationships of each node in real time, and generate structured traceability log files. The abnormal feature labeling unit is used to write the employee number and the original text fragment as abnormal metadata into the traceability log file and add a regular expression missing label when the responsible employee number is matched and the deduction value cannot be extracted within the adaptive sliding window.

[0012] Preferably, the state repair module includes: The hot update configuration unit is used to decouple the regular expression rules used by the core parsing from the underlying logic execution engine as an independent dynamic configuration dictionary, and to overwrite the configuration dictionary in real time based on the modification events sent by the user interface. The targeted tracing and retrieval unit is used to perform pattern matching in the traceability log file with the missing regular expression tag as the target after the dynamic configuration dictionary is hot-updated and an incremental repair instruction is received, and to extract the associated employee number and the original text fragment that failed to be parsed by reverse tracing. The local cascading refresh unit is used to re-input the original text fragment into the updated rule dictionary for secondary extraction and to overwrite the original performance score of the corresponding node. It also uses the cascading control unit to schedule the target fleet to perform local recalculation state activation.

[0013] Preferably, the linear normalization unit includes: Using the min-max normalization method, the minimum original performance score of all crew drivers in the fleet is subtracted from the original performance score of the current fleet driver, divided by the difference between the maximum and minimum scores, and then multiplied by the upper limit of the preset standard interval to obtain the standardized score.

[0014] Preferably, the local cascading refresh unit includes: The set of affected passenger drivers with missing regular expression labels is compared with the total set of passenger drivers in each fleet of the entire network. If a passenger driver in a specific fleet is included in the set of affected passenger drivers, then that specific fleet is assigned to the target fleet set.

[0015] This invention provides an integrated intelligent passenger service performance management platform. It has the following beneficial effects: 1. This invention revolutionizes the inefficient traditional manual table-by-table verification model by using the asynchronous task queue distribution mechanism of the interface control module and the multi-threaded parallel scheduling of the background thread pool. The system can automatically schedule multi-core computing power across modules to achieve streaming reading and second-level asynchronous parallel computing of large-scale multi-team crew datasets, compressing the tedious manual verification process that originally took several hours into a very short time to complete the merged processing, thus freeing up the labor of management personnel.

[0016] 2. Based on matrix filling units and adaptive sliding window positioning mechanism, this invention can automatically parse and convert non-standard multi-source heterogeneous tables and unstructured text notification documents. The system accurately anchors and extracts deduction scores through compound regular expressions. The fully automated data processing eliminates various occasional human errors caused by manual data handling, entry, and verification, resulting in a qualitative leap in the accuracy of performance accounting and fundamentally avoiding salary disputes caused by calculation errors.

[0017] 3. This invention unifies the assessment standards and linear standardized mapping logic of the entire network, eliminating the interference of inconsistent assessment scales between different fleets. When faced with conflicts of absolutely equal scores at the grade boundary, it can automatically introduce multi-level objective secondary technical indicators such as safe driving fault-free mileage and millisecond-level check-in timestamps for algebraic reconstruction and fine-tuning. This allows for precise breaking of parallel deadlocks without disrupting the main scoring structure, eliminating human intervention and boundary ambiguity, and significantly improving the credibility and absolute fairness of the assessment.

[0018] 4. This invention establishes a robust system self-healing and traceability mechanism. The full-stack traceability log unit synchronously captures intermediate states on the computing nodes and automatically adds missing regular expression tags for evidence when encountering abnormal text during parsing. When the rule dictionary is hot-updated, the system can use this tag as the target to trace back and extract the original text. It then uses the cascaded control unit to schedule a specific target fleet set to perform localized precise recalculation, eliminating the need for full-network recalculation or manual intervention. This provides technical support for full-process transparency and traceability while ensuring high system fault tolerance, maintenance-free operation, and no risk of crashes. Attached Figure Description

[0019] Figure 1 This is a system architecture diagram of an integrated intelligent passenger service performance management platform according to the present invention; Figure 2 This is an architecture diagram of the grade evaluation module in an integrated intelligent passenger service performance management platform of the present invention; Figure 3 This is a schematic diagram of a performance calculation page provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see the appendix Figure 1 -Appendix Figure 3 This invention provides an integrated intelligent passenger service performance management platform, comprising: The interface control module is used to capture performance calculation trigger instructions through event listeners and submit them to the task queue, so that the background thread pool can distribute them to the background worker threads for asynchronous calculation. Furthermore, the interface control module includes: The foreground user interface thread unit is used to render the graphical user interface and submit the captured performance calculation trigger instructions to the task queue. The background thread pool scheduling unit is used to dynamically activate a corresponding number of concurrent worker threads based on the number of tasks in the queue and the number of processor cores. The color encoding processing unit is used to receive node status signals sent by the background worker thread in real time, and call the color encoding processor to render the running status data into the corresponding preset color according to the preset mapping relationship between the status flag bits and the color matrix. Specifically, the front-end user interface thread unit, serving as the entry point for human-computer interaction, performs real-time rendering of the graphical user interface based on an event-driven architecture. When a flight attendant triggers the performance calculation control on the front-end interface, the event listener within the front-end user interface thread unit captures the calculation request signal triggered by the hardware interrupt. The event listener then encapsulates the request signal into a data packet containing a timestamp and fleet identifier, serving as the performance calculation trigger instruction. To avoid the calculation blocking the front-end rendering, the front-end user interface thread unit uses an inter-thread communication mechanism to immediately submit the performance calculation trigger instruction to the kernel's first-in-first-out task queue for queuing and buffering, thereby releasing front-end control and ensuring interface smoothness. The background thread pool scheduling unit continuously monitors the backlog status of the aforementioned task queue. Internally, this unit incorporates resource allocation logic that continuously calculates the total number of backlogged task objects in the task queue. It also obtains the total number of currently available physical processor cores on the server through system call interfaces. The background thread pool scheduling unit dynamically determines the number of active concurrent worker threads based on the following concurrency calculation formula. : ; in, This indicates the total number of background concurrent worker threads that need to remain active. This represents the total number of available processor physical cores detected by the system. Subtracting one core is to reserve space for the foreground user interface thread unit to prevent the interface from freezing. This represents the total number of unprocessed performance evaluation tasks backlogged in the first-in-first-out (FIFO) task queue. This represents a preset single-threaded processing task load factor constant, used to balance the task concurrency density under single-core resources. Its value range is determined based on experimental data of the throughput of hardware single-core parallel processing of multiple file handles. In this embodiment, it is preferably a positive integer between 3 and 5. This allows the background thread pool scheduling unit to wake up the corresponding number of worker threads from the pre-initialized background thread pool and distribute the performance accounting trigger instructions in the task queue to these activated concurrent worker threads, thereby realizing the parallel asynchronous calculation of multi-team service datasets in the background and improving the utilization efficiency of hardware multi-core resources. During the background worker threads' computation, the color encoding processing unit is responsible for maintaining the transparency of the runtime state. Each active background worker thread, when performing data cleaning, standardization calculations, or grade evaluation, will asynchronously synchronize node status signals to the color encoding processing unit after the completion of key operators. These signals contain status flags representing the current processing progress. The color encoding processing unit maintains a status attribute mapping table, which establishes a linear correspondence between status flags and the three primary color value matrices. For example, when a flag representing normal computation is received, the corresponding matrix output is a high-saturation value for the green channel. When a flag representing triggering boundary fault tolerance or anomaly marking is received, the corresponding matrix output is a value for the yellow or red channel. The color encoding processing unit passes this value matrix as rendering parameters to the color encoding processor. The color encoding processor directly overwrites the pixel buffer of the corresponding vehicle node in the front-end graphical interface based on these values, converting the real-time computation state of the background multi-threaded process into the corresponding preset color on the front-end interface for rendering. The entire mechanism decouples the front and back ends through a task queue, optimizes computational efficiency through dynamic resource allocation, and improves system observability through closed-loop feedback of status signals.

[0022] The data cleaning module is used to convert the original table file into a two-dimensional data matrix, perform cell splitting and filling and regular expression matching of fleet names, and remove invalid rows through compliance scanning; Furthermore, the data cleaning module includes: The matrix filling unit is used to convert the original table file into a two-dimensional data matrix and write the stored value of the valid cell into each corresponding independent cell after splitting and merging the area; The name conversion unit is used to scan the text columns in the two-dimensional data matrix and use a regular expression matching engine to uniformly replace non-standard fleet text with standard fleet identifiers. The missing value filtering unit is used to calculate the missing value rate of a single row and mark the row data as removed when the missing value rate is greater than or equal to the judgment threshold, and when the content of a single row matches the preset blacklist. Specifically, the matrix filling unit first calls the underlying file stream parser to read the original table file, maps its row and column structure into memory, and constructs a two-dimensional data matrix interwoven with horizontal row objects and vertical column objects. In this process, for merged cells in the original table, the matrix filling unit parses the topological coordinate range of each merged area, extracts the actual storage value of the first valid cell in the upper left corner of the merged area, and then performs a demerging operation, splitting the originally merged area into a grid composed of independent cell objects. Using a double loop traversal algorithm, the extracted valid storage value is sequentially written into each of the split independent cell objects, thereby eliminating the local data loss caused by table layout and completing the basic filling of the two-dimensional data matrix. To eliminate textual ambiguity caused by inconsistent team naming in multiple source files, the name conversion unit performs a text column scan on the fully populated two-dimensional data matrix. The unit sequentially extracts unstructured text strings containing team names and inputs them as input vectors to a pre-configured regular expression matching engine. This engine internally loads a team name regular expression mapping library, which contains multiple non-standard regular expression capture groups built for various team aliases, abbreviations, and common spelling errors. Each capture group is linearly mapped to a globally unique standard team identifier. When the matching engine detects that a non-standard team text matches a specific capture group, it triggers a replacement operator, rewriting the non-standard text to be converted in the current cell of the two-dimensional data matrix into the corresponding standard team identifier. After name conversion, the two-dimensional data matrix is ​​then passed to the missing value filtering unit for compliance scanning to remove low-quality invalid rows. The missing value filtering unit employs a single-row scanning mechanism based on a traversal accumulator, checking for null values ​​in the cells of each row. The single-row missing value rate is then calculated. The calculation formula is: ; in, This indicates the single-row missing value rate of the currently scanned data row. This represents the total number of cells with empty values ​​found by the accumulator in the current row. This indicates the total number of physical columns in the current row. The missing value filtering unit calculates the single-row null value missing rate. Compared with the preset judgment threshold Quantitative comparison is performed, and this judgment threshold is determined. This is an empirical limit established based on robustness analysis of missing historical attendance samples. Its value range is typically limited to 0.4 to 0.6; in this embodiment, it is set to 0.5. This is used to characterize that when more than half of the key performance fields in a single row are missing, the data in that row loses its statistical significance for grade evaluation. Simultaneously, the missing value filtering unit also assigns the text content of the current row to a keyword blacklist and performs precise matching with a preset redundant keyword blacklist at the end of the table. Once the single row's null value missing rate is detected... Greater than or equal to the judgment threshold If the current row content successfully matches any item in the redundant keyword blacklist at the end of the table, the missing rate filtering unit will write a removal flag in the header status bit of the data object of that row in memory. Data rows with removal flags will be skipped directly in the subsequent memory cascading and score pool construction process, thereby achieving fully automatic interception of invalid data.

[0023] The data fusion module is used to cascade structured data with employee ID as the primary key, dynamically build an adaptive sliding window, and call compound regular expressions to extract deduction values ​​to update the original performance score. Furthermore, the data fusion module includes: The structured cascading unit is used to retrieve and cascade the regular assessment scores, safe driving mileage without failure, and clock-in timestamps of the crew drivers using the employee number as the primary key, and assigns the regular assessment scores to the original performance score field. An adaptive window unit is used to determine the start and end boundary coordinates of an adaptive sliding window centered on the target work number position by bidirectional scanning of regular expressions and iteration of boundary symbols. The unstructured extraction unit is used to extract the deduction scores within the adaptive sliding window covering semantic blocks using compound regular expressions, and deduct them from the original performance scores of the corresponding driver nodes. Specifically, the structured cascading unit first initializes a data buffer pool with a hash table as its underlying architecture in the system memory. The structured cascading unit extracts the cleaned shift table data and basic attendance table data, and performs high-speed retrieval in the memory space using the unique employee number of the train attendant as the index key. When a matching employee number record is found, the structured cascading unit extracts the heterogeneous attribute fields of the train attendant, such as the regular assessment score, safe driving mileage without failure, and clock-in timestamp, across tables and instantiates them in memory as independent driver data entity nodes. The regular assessment score obtained from the cascading is used as the baseline score and is directly written and assigned to the original performance score field initialized inside each driver data entity node as the initial state value for subsequent deduction calculations. To accurately extract penalty information associated with a specific driver from unstructured text notification files, the adaptive window unit initiates a two-stage localization processing flow. First, the adaptive window unit calls a regular expression scanning engine to retrieve the target employee ID from the notification text file, precisely locating its absolute character offset within the text stream. Then, using the physical location of the target employee ID as the center, the adaptive window unit performs a bidirectional iterative scan to the left and right of the text using its built-in boundary marker scanner. During this iterative scan, the boundary marker scanner uses a single character as a step size, constantly comparing whether the current character hits the preset forward and backward paragraph separators. The starting boundary coordinates of the adaptive sliding window... coordinates of the end boundary The calculation and determination formula is as follows: ; ; in, This represents the starting physical boundary coordinates of the adaptive sliding window. This represents the coordinates of the end physical boundary of the adaptive sliding window. This represents the absolute character offset coordinates of the target employee ID within the text stream. This represents the cumulative dynamic expansion character step size as the boundary scanner iterates to the left until it hits the first forward paragraph separator. This represents the cumulative dynamic expansion character step size as the boundary scanner iterates to the right until it hits the first backward paragraph separator. In actual operation, the forward and backward paragraph separators are usually specified as newline characters, paragraph breaks, or periods to ensure that the segmented semantic block is a relatively complete independent text sentence. The system dynamically isolates an adaptive sliding window containing the complete contextual semantic information of the driver's penalty for each target employee number, eliminating interference from other irrelevant paragraphs in the text. After defining the window boundaries, the unstructured extraction unit parses the text content within the complete semantic block covered by the adaptive sliding window. The unstructured extraction unit inputs the string within the semantic block into the compound regular expression parsing kernel. This compound regular expression uses specific numerical capture groups to perform pattern matching and numerical extraction on numerical fragments immediately following sensitive trigger words such as "violation," "point deduction," and "point reduction" within the semantic block. Once a specific point deduction value is successfully captured, the unstructured extraction unit converts the value into a standard floating-point number and uses the target employee number associated with the current adaptive sliding window as the key to address and locate the corresponding driver data entity node in the memory hash table. The unstructured extraction unit then calls the subtraction operator to dynamically deduct the captured point deduction value from the existing original performance score field of the node, and directly overwrites the latest result after subtraction back into the original performance score field of the node, thus completing the fusion of structured basic indicators and unstructured dynamic point deduction data.

[0024] The grade assessment module is used to extract the extreme values ​​of the team scores and apply the min-max normalization method to calculate the standardized scores, and then perform grade division and verify the grade boundary scores; Furthermore, the grading module includes: The extreme value extraction and boundary protection unit is used to scan the final raw performance scores of all drivers in a specific fleet to extract the maximum and minimum values, and when the two are equal, it forces the standardized score of all drivers in the fleet to be the preset maximum score and stops the subsequent calculation. The linear normalization unit is used to apply min-max normalization logic to uniformly map the original performance scores of the flight attendants to a preset standard range when the maximum value is greater than the minimum value, so as to obtain a standardized score with a preset number of decimal places. The quota allocation and verification unit is used to sort the standardized score sets of all drivers in descending order, divide the performance grades based on the preset quota ratio, and use the rule constraint to make conditional judgments on the grade boundary scores. Linear normalization units include: Using the min-max normalization method, the minimum value of the original performance scores of all crew drivers in the current fleet is subtracted from the original performance score of the current fleet driver, divided by the difference between the maximum and minimum values, and then multiplied by the upper limit of the preset standard interval to obtain the standardized score. Specifically, the extreme value extraction and boundary protection unit first establishes a data traversal processing flow. It uses a memory iterator to poll and access the driver data entity nodes of all passengers belonging to a specific fleet identifier. During the polling process, the extreme value extraction and boundary protection unit extracts the final raw performance score of each node in turn and maintains two intermediate variables in real time through a memory comparator, namely the highest raw score and the lowest raw score in the current fleet. After completing the traversal, the extreme value extraction and boundary protection unit sends the highest raw score and the lowest raw score to the equivalence checker. If the equivalence checker determines that the two satisfy the absolute numerical equality relationship, it means that all drivers in the specific fleet have the same score. At this time, the extreme value extraction and boundary protection unit immediately triggers the boundary protection fault tolerance mechanism, skips the subsequent high-risk division by zero algebra operation, directly forces the preset highest standard score to be overwritten to the score attributes of all driver and passenger nodes in the fleet, and sends a calculation stop signal. When the highest raw score is determined to be greater than the lowest raw score, the system activates the linear standardization unit. The linear standardization unit calls the floating-point algebra unit to perform a minimum-maximum normalized linear mapping on the raw performance score of each driver in the fleet, eliminating benchmark deviations caused by different assessment standards between different fleets, and standardizing the scores of each driver. The calculation formula is as follows: ; in, This represents the standardized score generated after the current flight attendant has undergone standardized processing. This represents the current driver's final raw performance score. This represents the minimum original performance score among all drivers in the current driver's fleet. This represents the maximum original performance score of all drivers within the fleet to which the current driver belongs. This represents the upper limit of the standard interval uniformly preset across the entire network. Based on the percentage-based evaluation criteria, this upper limit is used in this embodiment. Setting it to 100, the linear normalization unit will calculate the standardized score. The precision truncation operator is input, the preset number of decimal places are retained, and the truncated standardized score is updated to the corresponding hash table node attribute to form a network-wide standardized score pool for network-wide sorting. After constructing the entire network score pool, the quota allocation and verification unit takes over the data flow. It merges the standardized scores of all vehicle drivers and uses a quicksort algorithm in memory to sort the score set in descending order. Then, the unit extracts a preset quota ratio parameter, multiplies it by the total number of drivers participating in the sorting, and calculates the quota number corresponding to each performance level (Excellent, Good, Qualified, etc.). The unit slices the descending sequence based on the quota truncation number, thus assigning each driver a corresponding performance level label. After slicing, to prevent logical loopholes at level boundaries due to tied scores, the unit calls a mandatory logic verification rule constraint. The rule constraint extracts the level boundary scores at each level slice boundary point and substitutes them into a preset inequality chain for condition judgment. If the boundary score violates the inequality chain constraint, a conflict signal is output so that the subsequent adaptive rework module can perform local state correction.

[0025] The adaptive reconstruction module is used to send a secondary reconstruction callback instruction when the verification fails, extract multi-level objective technical indicators and apply a multi-level decay weight factor reconstruction model to calculate the fine-tuning score for local rearrangement. Furthermore, the adaptive refactoring module includes: The linkage response unit is used to receive the boundary score absolute equality conflict signal and send back the secondary reconstruction callback instruction when the boundary score does not meet the constraint. The feature multi-level retrieval unit is used to respond to the secondary reconstruction callback instruction, and retrieve the corresponding driver node based on the employee number of the passenger driver in the boundary parallel state, and extract the safe driving fault-free mileage as the first-level indicator and the check-in timestamp millisecond value as the second-level indicator across the table. The perturbation fine-tuning unit is used to combine the first-level and second-level perturbation weight factors to perform algebraic reconstruction on the extracted safe driving fault-free mileage and check-in timestamps, and obtain the fine-tuning score for local descending reordering of drivers with boundary parallel conflicts. Specifically, the linkage response unit continuously monitors the verification status output by the grade evaluation module. When the grade boundary score violates the preset inequality chain constraint, the linkage response unit captures the boundary score absolute equality conflict signal. The linkage response unit converts the conflict signal into an event object containing the parallel scores and the affected vehicle set information, and triggers the secondary reconstruction callback instruction in reverse to the underlying adaptive reconstruction control logic through the kernel's internal event bus, activating the subsequent feature retrieval and reconstruction data processing flow. In response to the aforementioned secondary reconstruction callback instruction, the feature multi-level retrieval unit first extracts the specific boundary score that caused the conflict, and then retrieves the set of crew driver IDs whose scores are absolutely equal to the boundary score from the standardized score pool of the entire network. The feature multi-level retrieval unit uses the IDs of these crew drivers in the boundary parallel state as the primary key to perform directional addressing and retrieval of the corresponding driver node in the memory hash table. After locating the node, the feature multi-level retrieval unit initiates a cross-table multi-level cascading mechanism to extract the safe driving fault-free mileage as the first-level indicator and the attendance timestamp in the system attendance check-in record as the second-level indicator from the corresponding driver node attributes. In order to meet the requirements of high-precision fine-tuning calculation, the extracted attendance timestamp is converted into a standard millisecond value calculated from Coordinated Universal Time in memory. After the fine-tuning unit takes over the data stream, it calls the algebraic reconstruction model to perform decay weight fusion on the extracted multi-level features, and calculates the fine-tuning score used to break up parallel localities. The algebraic reconstruction calculation logic is based on the following formula: ; in, This indicates the current driver's fine-tuning score. This represents the preset first-level micro-perturbation weighting factor, specifically designed to amplify the contribution of safe driving fault-free mileage. To ensure that the fine-tuning score does not exceed the grade difference of the main standardized score, its order of magnitude is limited to between 0.0001 and 0.001. This represents the extracted value of safe driving mileage without failures, which is used as the first-level indicator. This represents the preset second-level micro-perturbation weighting factor, specifically used to adjust the contribution of attendance clocking-in timeliness; its order of magnitude is limited to between 0.01 and 0.1. This represents the extracted check-in timestamp in milliseconds, which serves as the second-level indicator. This represents the preset baseline start timestamp in milliseconds for the current shift. This parameter is a constant value automatically injected into memory by the system dispatch console during initialization based on the shift schedule. It serves as the zero-point reference for attendance time. A reciprocal operator ensures that drivers who clock in earlier receive a higher increment for the disturbance term, and by setting a very small disturbance weight factor value, fine-tuning scores are ensured. The magnitude of the error is much smaller than the standardized score, thus achieving micro-discrimination without disrupting the original performance structure. The perturbation fine-tuning unit will calculate the generated fine-tuning score. The score is appended to the attribute field of the corresponding micro-node, and the local memory sorting operator is called. Based on the fine-tuning score, the flight attendants who were originally in a boundary and conflict state are reordered in a local descending order. Finally, the reordered non-conflict grade results are updated to the corresponding hash table node to complete the closed-loop correction of the grade evaluation.

[0026] The streaming protection and logging module is used to perform event-driven line-by-line reading and memory control of data, synchronously generate traceability log files, and add regular expression missing tags when extraction fails. Furthermore, the streaming protection and logging module includes: The event flow control unit is used to read the crew dataset line by line in an event stream, and suspend the reading and call the garbage collection operator when the dynamic memory safety factor calculated based on the memory monitor is less than or equal to the critical trigger threshold. The full-stack traceability log unit is used to run synchronously on the core computing nodes to capture the input parameters, intermediate computing status, key operator outputs and worker ID primary key mapping relationships of each node in real time, and generate structured traceability log files. The abnormal feature labeling unit is used to write the employee number and the original text fragment as abnormal metadata into the traceability log file and add a regular expression missing label when the responsible employee number is matched and the deduction value cannot be extracted within the adaptive sliding window. Specifically, the event flow control unit establishes a file pointer-driven data consumption stream, calls the underlying streaming text parser to perform line-by-line physical reading of the crew dataset. For each successfully read line, the event flow control unit extracts the field information of that line, deserializes it, and writes it to a memory hash table. Then, it immediately destroys the physical reference pointer of that line in the file system to release file system-level memory usage. During this process, the built-in memory monitor polls the operating system's process memory interface at fixed time intervals to obtain the number of allocated bytes and the maximum available bytes of the current virtual machine memory in real time, dynamically adjusting the memory safety factor. The quantitative calculation formula is as follows: ; in, This represents the dynamic memory safety factor during the current runtime of the system. This indicates the number of bytes of heap memory currently allocated and occupied by the process, as obtained by the memory monitor. This indicates the maximum number of bytes of physical heap memory that the current process is allowed to request. The event flow control unit will calculate the dynamic memory safety factor. The value is compared in real time with a preset critical trigger threshold, which serves as a red line for preventing memory overflow. Its safe empirical value range is typically set between 0.10 and 0.15; in this embodiment, it is preferably set to 0.12. That is, when the available remaining heap memory is less than 12%, the flow control blocking mechanism is forcibly activated. Once the dynamic memory safety factor is detected... If the value is less than or equal to the critical trigger threshold, it means the system faces a risk of memory overflow. The event flow control unit immediately performs a suspension operation, forcibly blocking the current file reading thread, stopping the ingestion of new data lines, and simultaneously invoking the garbage collection operator to explicitly trigger the reclamation of underlying objects. This continues until the memory monitor provides feedback on the dynamic memory safety factor. Once the temperature rises above the critical trigger threshold, the system unblocks and resumes line-by-line reading of the file stream, thus using a feedback control loop to protect the system from crashing. While the event stream runs smoothly, the full-stack traceability log unit runs synchronously with the main computing thread on the core computing node through the log persistence kernel. The full-stack traceability log unit captures the context data generated by the main computing thread during the runtime of each module in real time through the aspect interception mechanism. This includes the initial input parameters of each node, the intermediate computing state during the cleaning and standardization process, the temporary floating-point values ​​output by key operators, and the dynamic primary key mapping relationship between the worker number and the driver node. The full-stack traceability log unit assembles these captured semi-structured data packets into data rows with globally unique auto-incrementing sequence numbers and synchronously writes them to the server's hard disk high-speed buffer using asynchronous append write operators, generating a structured traceability log file with a full-word search index, providing the system with a basis for tracing the data flow throughout its entire lifecycle. When the system encounters irregular text during unstructured extraction, the anomaly feature labeling unit connects to the data stream to perform anomaly control. The unit monitors the output status of the adaptive sliding window in real time. When the regular expression scanning engine successfully matches a valid responsible crew member's employee number in the notification text, but the compound regular expression parsing kernel cannot capture and extract the specific deduction score within the semantic block covered by the corresponding adaptive sliding window, the anomaly feature labeling unit determines that the current notification text has experienced a format change anomaly due to a manual input error. At this point, the anomaly feature labeling unit issues a state-maintaining instruction to keep the current working thread and the entire pipeline running normally, preventing abnormal termination. Meanwhile, the anomaly feature labeling unit encapsulates the target employee number that cannot be parsed and the original text fragment within the entire adaptive sliding window corresponding to that employee number into anomaly metadata. It then calls the write interface of the full-stack traceability log unit to forcibly rewrite and input it into the structured traceability log file that is being generated. At the same time as writing, the anomaly feature labeling unit adds a globally unique regular expression missing label to the line end attribute field of the log record. This label then serves as the index target and participates in the reverse tracing and local incremental refresh of the subsequent state repair module.

[0027] The status repair module is used to hot update the rules in the dynamic configuration dictionary, and to perform targeted source retrieval based on the missing regular expression label to perform local status overwrite. It also uses the cascade control unit to schedule the target fleet set for cascade refresh. Furthermore, the status repair module includes: The hot update configuration unit is used to decouple the regular expression rules used by the core parsing from the underlying logic execution engine as an independent dynamic configuration dictionary, and to overwrite the configuration dictionary in real time based on the modification events sent by the user interface. The targeted tracing and retrieval unit is used to perform pattern matching in the traceability log file with the missing regular expression tag as the target after the dynamic configuration dictionary is hot-updated and an incremental repair instruction is received, and to extract the associated employee number and the original text fragment that failed to be parsed by reverse tracing. The local cascading refresh unit is used to re-input the original text fragment into the updated rule dictionary for secondary extraction and to overwrite the original performance score of the corresponding node. It also uses the cascading control unit to schedule the target fleet set to perform local recalculation state activation. The partial cascading refresh unit includes: The set of affected passenger drivers with missing regular expression labels is compared with the total set of passenger drivers in each fleet of the entire network. If a specific fleet contains a passenger driver whose ID is included in the set of affected passenger drivers, then that specific fleet is assigned to the target fleet set. Specifically, when the system starts, the hot update configuration unit decouples the regular expression rules used by the core parsing from the underlying kernel execution engine and instantiates them as an independent dynamic configuration dictionary in an independent storage space in external memory. The configuration listener inside the hot update configuration unit is continuously mounted on the background bus of the graphical user interface. When the operation and maintenance personnel modify the regular expression through the front-end interface and submit the modification event, the configuration listener captures the event and extracts the new regular expression string. The hot update configuration unit then calls the memory overwrite operator to directly rewrite the dynamic configuration dictionary in external memory without restarting the system service, thereby achieving dynamic and smooth replacement of parsing rules. After the external dynamic configuration dictionary completes a hot update and the system receives an incremental repair instruction issued manually, the targeted tracing and retrieval unit starts data recovery processing. The targeted tracing and retrieval unit calls the underlying log retrieval unit to open the previously generated structured traceability log file in read-only stream mode. The log retrieval unit configures the preset regular expression missing label as the target for matching in the full word index of the log file and performs multi-threaded concurrent mode matching. When the retrieval unit hits a log record line with the label, the targeted tracing and retrieval unit activates the field splitting operator and extracts the associated crew driver's employee number from the structured data body of that log line, as well as the complete original text fragments that failed to be parsed in historical calculations and were retained as abnormal metadata. These extracted employee numbers and corresponding original text fragments are collected and encapsulated in memory as incremental data sources to be processed and input into the subsequent refresh process. After the local cascading refresh unit takes over the above incremental data source, it first starts the secondary extraction processing flow, re-inputs the extracted complete original text fragment into the updated dynamic configuration dictionary, and performs secondary pattern matching on the text fragment using the updated regular expression rules to capture and extract the correct deduction score. The local cascading refresh unit then uses the associated employee number as the key to address and locate the corresponding driver data entity node in the memory hash table, calls the subtraction operator to overwrite and update the original performance score field inside the node with the newly extracted score, and completes the status overwrite at the individual level. To ensure that changes in individual scores are correctly cascaded to the macro-statistical layer, the local cascaded refresh unit executes the convergence determination of the affected fleet set through the scheduling logic matrix of the cascaded control unit. The cascaded control unit uses the set of affected driver IDs with regularized missing labels as input features and performs an intersection existence comparison with the overall set of IDs within each fleet of the entire network extracted from the in-memory database. The convergence determination formula is as follows: ; in, This represents the set of target vehicle fleets that need to be triggered for incremental updates, generated after the convergence determination. This represents the total set of employee IDs of all crew members within any specific fleet in the entire network. This represents the set of affected flight attendant employee IDs extracted from log records with missing regular expression tags through reverse tracing. The primary key represents the employee ID of any passenger driver. If at least one passenger driver's employee ID is included in the set of affected passenger driver employee IDs within a specific fleet, the cascade control unit will partition and write the identifier of that specific fleet into the target fleet set. After the convergence and partitioning of the convoy assembly is completed, the cascaded control unit moves towards the target convoy. All the fleet nodes included in the set uniformly and in parallel issue local recalculation state activation instructions, driving these affected fleets to re-call the evaluation module to perform local extreme value extraction and linear standardization calculations. Meanwhile, the memory states of other fleets not included in this set in the entire network are in a completely isolated and blocked state, thus completing the local cascading refresh.

[0028] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart passenger service performance management integrated platform, characterized in that, include: The interface control module is used to capture performance calculation trigger instructions through event listeners and submit them to the task queue, so that the background thread pool can distribute them to the background worker threads for asynchronous calculation. The data cleaning module is used to convert the original table file into a two-dimensional data matrix, perform cell splitting and filling and regular expression matching of fleet names, and remove invalid rows through compliance scanning; The data fusion module is used to cascade structured data with employee ID as the primary key, dynamically build an adaptive sliding window, and call compound regular expressions to extract deduction values ​​to update the original performance score. The grade assessment module is used to extract the extreme values ​​of the team scores and apply the min-max normalization method to calculate the standardized scores, and then perform grade division and verify the grade boundary scores; The adaptive reconstruction module is used to send a secondary reconstruction callback instruction when the verification fails, extract multi-level objective technical indicators and apply a multi-level decay weight factor reconstruction model to calculate the fine-tuning score for local rearrangement. The streaming protection and logging module is used to perform event-driven line-by-line reading and memory control of data, synchronously generate traceability log files, and add regular expression missing tags when extraction fails. The state repair module is used to hot update the rules in the dynamic configuration dictionary, and to perform targeted source retrieval based on the missing regular expression tags to perform local state overlay, and to use the cascade control unit to schedule the target fleet set for cascade refresh.

2. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The interface control module includes: The foreground user interface thread unit is used to render the graphical user interface and submit the captured performance calculation trigger instructions to the task queue. The background thread pool scheduling unit is used to dynamically activate a corresponding number of concurrent worker threads based on the number of tasks in the queue and the number of processor cores. The color encoding processing unit is used to receive node status signals sent by the background worker thread in real time, and according to the preset mapping relationship between the status flag and the color matrix, call the color encoding processor to render the running status data into the corresponding preset color.

3. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The data cleaning module includes: The matrix filling unit is used to convert the original table file into a two-dimensional data matrix and write the stored value of the valid cell into each corresponding independent cell after splitting and merging the area; The name conversion unit is used to scan the text columns in the two-dimensional data matrix and use a regular expression matching engine to uniformly replace non-standard fleet text with standard fleet identifiers. The missing value filtering unit is used to calculate the missing value rate of a single row, and to mark the row data as removed when the missing value rate is greater than or equal to the judgment threshold, and when the content of a single row matches a preset blacklist.

4. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The data fusion module includes: The structured cascading unit is used to retrieve and cascade the regular assessment scores, safe driving mileage without failure, and clock-in timestamps of the crew drivers using the employee number as the primary key, and assigns the regular assessment scores to the original performance score field. An adaptive window unit is used to determine the start and end boundary coordinates of an adaptive sliding window centered on the target work number position by bidirectional scanning of regular expressions and iteration of boundary symbols. The unstructured extraction unit is used to extract the deduction scores within the adaptive sliding window covering semantic block using compound regular expressions, and deduct them from the original performance score of the corresponding driver node.

5. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The grade assessment module includes: The extreme value extraction and boundary protection unit is used to scan the final raw performance scores of all drivers in a specific fleet to extract the maximum and minimum values, and when the two are equal, it forces the standardized score of all drivers in the fleet to be the preset maximum score and stops the subsequent calculation. The linear normalization unit is used to apply min-max normalization logic to uniformly map the original performance scores of the flight attendants to a preset standard range when the maximum value is greater than the minimum value, so as to obtain a standardized score with a preset number of decimal places. The quota allocation and verification unit is used to sort the standardized score sets of all drivers in descending order, divide the performance grades based on the preset quota ratio, and use the rule constraint to make conditional judgments on the grade boundary scores.

6. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The adaptive reconstruction module includes: The linkage response unit is used to receive the boundary score absolute equality conflict signal and send back the secondary reconstruction callback instruction when the boundary score does not meet the constraint. The feature multi-level retrieval unit is used to respond to the secondary reconstruction callback instruction, and retrieve the corresponding driver node based on the employee number of the passenger driver in the boundary parallel state, and extract the safe driving fault-free mileage as the first-level indicator and the check-in timestamp millisecond value as the second-level indicator across the table. The perturbation fine-tuning unit is used to perform algebraic reconstruction on the extracted safe driving fault-free mileage and check-in timestamps by combining the first-level and second-level perturbation weight factors, and obtain the fine-tuning score for local descending reordering of drivers with boundary parallel conflicts.

7. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The streaming protection and logging module includes: The event flow control unit is used to read the crew dataset line by line in an event stream, and suspend the reading and call the garbage collection operator when the dynamic memory safety factor calculated based on the memory monitor is less than or equal to the critical trigger threshold. The full-stack traceability log unit is used to run synchronously on the core computing nodes to capture the input parameters, intermediate computing status, key operator outputs and worker ID primary key mapping relationships of each node in real time, and generate structured traceability log files. The abnormal feature labeling unit is used to write the employee number and the original text fragment as abnormal metadata into the traceability log file and add a regular expression missing label when the responsible employee number is matched and the deduction value cannot be extracted within the adaptive sliding window.

8. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The status repair module includes: The hot update configuration unit is used to decouple the regular expression rules used by the core parsing from the underlying logic execution engine as an independent dynamic configuration dictionary, and to overwrite the configuration dictionary in real time based on the modification events sent by the user interface. The targeted tracing and retrieval unit is used to perform pattern matching in the traceability log file with the missing regular expression tag as the target after the dynamic configuration dictionary is hot-updated and an incremental repair instruction is received, and to extract the associated employee number and the original text fragment that failed to be parsed by reverse tracing. The local cascading refresh unit is used to re-input the original text fragment into the updated rule dictionary for secondary extraction and to overwrite the original performance score of the corresponding node. It also uses the cascading control unit to schedule the target fleet to perform local recalculation state activation.

9. The integrated intelligent passenger service performance management platform according to claim 5, characterized in that, The linear normalization unit includes: Using the min-max normalization method, the minimum original performance score of all crew drivers in the fleet is subtracted from the original performance score of the current fleet driver, divided by the difference between the maximum and minimum scores, and then multiplied by the upper limit of the preset standard interval to obtain the standardized score.

10. The integrated intelligent passenger service performance management platform according to claim 1, characterized in that, The local cascaded refresh unit includes: The set of affected passenger drivers with missing regular expression labels is compared with the total set of passenger drivers in each fleet of the entire network. If a passenger driver in a specific fleet is included in the set of affected passenger drivers, then that specific fleet is assigned to the target fleet set.

Images