A rail transit-oriented digital plan generation method and system

Abstract

The application relates to a rail transit-oriented digital plan generation method and system, which comprises the following steps: receiving multi-source passenger flow data and generating a standard event stream with a unified timestamp; selecting candidate events through a sliding time window, generating multi-source evidence support degree according to time proximity, historical accuracy and data integrity coefficient; calculating the normalized measurement difference degree among data sources, obtaining a global conflict degree index through weighted average of the product of the evidence support degree; when the conflict index exceeds the threshold, correcting the evidence support degree by using a passenger flow conservation model, and normalizing the corrected or uncorrected support degree into a fusion weight; weighting and fusing the measurement values of each data source by using the weight to obtain a passenger flow state, and generating a digital plan accordingly. Thus, in the case that multi-source sensing data exist time synchronization deviation and semantic conflict, the fusion result can be corrected by physical law constraint, and a physically consistent passenger flow state and a safe evacuation plan can be generated.

Description

Technical Field

[0001] This application relates to the field of digital data processing and traffic control technology, specifically to a method and system for generating digital contingency plans for rail transit. Background Technology

[0002] In the integrated monitoring system of rail transit subway transfer stations, generating accurate passenger flow management plans requires the integration of multiple types of sensor data, such as video analysis, turnstile counting, and train weighing. However, existing systems face severe challenges due to data heterogeneity: there are time synchronization deviations ranging from hundreds of milliseconds to several seconds between sensor devices from different manufacturers, network transmission delays exhibit random jitter, and the sampling frequency coverage of each device varies. to Furthermore, data loss occurs. These factors cause the timestamps of the same passenger flow event to be scattered across uncertain intervals in multi-source records. Existing technologies typically employ precise timestamp pairing or interpolation alignment methods, which are prone to incorrectly associating delayed data with the current moment or losing valid observations, resulting in severe distortion of the fusion results.

[0003] Furthermore, when multiple high-reliability sources provide conflicting measurements of the same passenger flow status—for example, a significant discrepancy between video estimates of stranded passengers and weighing estimates of disembarkation—conventional fusion methods based on support-weighted averaging only provide a compromise value. This value may violate fundamental physical laws such as the conservation of passenger flow quality in platform areas, leading to a disconnect between the generated flow control or evacuation plans and the actual physical conditions on site, thus posing safety risks. The data alignment distortion under these non-ideal factors and the physical violations caused by semantic conflicts have become key bottlenecks restricting the accuracy and safety of contingency plans. Therefore, how to achieve reliable alignment and credible quantification of multi-source passenger flow data, and how to introduce physical constraints to correct the fusion results in cases of severe conflicts, has become an urgent technical problem to be solved. Summary of the Invention

[0004] To achieve reliable alignment and credible quantification of multi-source passenger flow data, and to introduce physical constraints to correct the fusion results in the event of severe conflicts, this application provides a digital contingency plan generation method and system for rail transit.

[0005] Firstly, this application provides a method for generating digital contingency plans for rail transit, including: Receive multi-source passenger flow data and perform time-series standardization processing on the multi-source passenger flow data to generate a standard event stream with a unified timestamp; Candidate events corresponding to the standard event stream are selected by sliding time windows, and a corresponding candidate alignment set is formed based on the candidate events; Based on the temporal proximity, historical accuracy, and data integrity coefficient of each data source in the candidate alignment set, the evidence support level corresponding to each data source is generated. Calculate the normalized measure difference between any two data sources in the candidate alignment set, and then perform a weighted average of the normalized measure difference using the product of the evidence support levels corresponding to the two data sources as the weight to obtain a global conflict index. If the global conflict level index does not exceed the preset threshold, the evidence support level corresponding to the data source is retained; If the global conflict level index exceeds the preset threshold, the evidence support level corresponding to the data source is corrected using the passenger flow quality conservation model, and a corresponding corrected evidence support level is generated. The evidence support and the modified evidence support are normalized to obtain the fusion weights corresponding to each data source. The measurement values ​​from each data source are weighted and merged using the fusion weights to obtain the passenger flow status, and a digital contingency plan is generated based on the passenger flow status.

[0006] Optionally, generating a standard event stream with a unified timestamp includes: converting the event records output by each data source in the multi-source passenger flow data into a standard event record structure containing event type, spatial location identifier, data source identifier, unified timestamp, and measurement value, wherein the unified timestamp is a timestamp after clock deviation compensation, with an accuracy of milliseconds.

[0007] Optionally, candidate events corresponding to the standard event stream are selected by a sliding time window, and a corresponding candidate alignment set is formed based on the candidate events. This includes: setting a sliding step size and a window half-width; for each time to be processed, creating a time interval centered on the time to be processed and defined by the window half-width as a sliding window; forming a candidate alignment set by all candidate events whose timestamps from all data sources fall within the time interval; and setting the window half-width based on the larger of the maximum end-to-end transmission delay and the tolerable fusion lag.

[0008] Optionally, based on the temporal proximity, historical accuracy, and data integrity coefficient of each data source in the candidate alignment set, an evidence support score is generated for each data source, including: obtaining the device historical accuracy for each data source, which is obtained by long-term verification and statistical analysis of the frequency of historical device measurements matching the benchmark measurement; obtaining the data integrity coefficient for each data source, which is calculated by the proportion of successfully received data packets within a recent time window; for each data source in the candidate alignment set, calculating the absolute value of the time difference between the timestamp recorded by each data source and the current time, and calculating the temporal proximity using a negative exponential decay function; and multiplying the temporal proximity, the device historical accuracy, and the data integrity coefficient to obtain the evidence support score for each data source at the current time.

[0009] Optionally, if multiple records exist within the sliding window from the same data source, the maximum value of the temporal proximity among them represents the temporal proximity of the same data source at the current moment.

[0010] Optionally, calculating the normalized measurement difference between any two data sources in the candidate alignment set includes: obtaining the measurement values ​​of any two different data sources at the current time, calculating the absolute value of the difference between the measurement values, dividing it by the larger of the maximum value of the two measurement values ​​and a preset zero-prevention constant, to obtain the normalized measurement difference.

[0011] Optionally, a weighted average of the normalized measure discrepancies is performed using the product of the evidence support levels corresponding to any two data sources as the weight to obtain a global conflict level index. This includes: multiplying the product of the evidence support levels corresponding to any two different data sources by the corresponding normalized measure discrepancies and summing the results, then dividing by the sum of the products of all evidence support levels to obtain the global conflict level index; if the number of valid data sources is less than two, the global conflict level index is directly set to zero.

[0012] Optionally, if the global conflict level index exceeds the preset threshold, the evidence support corresponding to the data source is corrected using the passenger flow quality conservation model to generate a corresponding corrected evidence support. This includes: calculating the source measurement deviation between the measurement value of each data source and the initial fusion value; selecting a credible anchor subset from each data source based on the evidence support threshold and consistency tolerance conditions; if the credible anchor subset contains at least two data sources, combining the state estimate provided by the credible anchor subset with the passenger flow quality conservation model, and using the least squares criterion to solve for the physical consistency reference state; if the credible anchor subset cannot be obtained, predicting historical trend values ​​based on historical fusion state values ​​from several moments before the current moment as the physical consistency reference state; calculating the physical penalty coefficient based on the degree to which the measurement value of each data source deviates from the physical consistency reference state, multiplying the original evidence support by the physical penalty coefficient, and normalizing to obtain the corrected evidence support.

[0013] Optionally, the evidence support level corresponding to the data source is corrected using the passenger flow quality conservation model, and further includes: if the evidence support level of the corrected abnormal data source is lower than a preset threshold, an equipment maintenance alarm is generated.

[0014] Secondly, this application provides a digital contingency plan generation system for rail transit, comprising: The standard event stream generation module is used to receive multi-source passenger flow data and perform time-series standardization processing on the multi-source passenger flow data to generate a standard event stream with a unified timestamp. The candidate alignment set module is used to select candidate events corresponding to the standard event stream through a sliding time window, and to form a corresponding candidate alignment set based on the candidate events; The evidence support metric module is used to generate the evidence support degree corresponding to each data source based on the temporal proximity, historical accuracy and data integrity coefficient of each data source in the candidate alignment set; The conflict level assessment module calculates the normalized measurement difference between any two data sources in the candidate alignment set, and performs a weighted average of the normalized measurement difference using the product of the evidence support levels corresponding to the two data sources as the weight, to obtain a global conflict level index. The conflict handling module is used to retain the evidence support level corresponding to the data source if the global conflict level index does not exceed the preset threshold, and to correct the evidence support level corresponding to the data source by using the passenger flow quality conservation model to generate the corresponding corrected evidence support level. The fusion weight module is used to normalize the evidence support and the modified evidence support to obtain the fusion weight corresponding to each data source. The contingency plan generation module is used to use the fusion weight to weight and fuse the measurement values ​​of each data source to obtain the passenger flow status, and generate a digital contingency plan based on the passenger flow status.

[0015] This application has the following technical advantages: 1. By using a sliding time window to extract candidate events and fusing time proximity, historical accuracy, and data integrity coefficients to generate multi-source evidence support, the mechanism effectively mitigates the timing misalignment problem caused by time synchronization deviations and transmission delay jitter. The negative exponential decay of time proximity ensures that observations closer to the current moment contribute more, while historical accuracy and data integrity coefficients automatically suppress interference from disconnection or high-latency sources, constraining credibility based on long-term reliability and communication quality, respectively. This mechanism provides an objective and dynamic weighting basis for subsequent fusion, avoiding data miscorrelation and loss of effective observations caused by traditional hard timestamp pairing.

[0016] 2. A global conflict level index, weighted by the product of evidence support, is employed. This amplifies measurement contradictions between high-support sources while suppressing noise interference from outliers in low-support sources, thus accurately quantifying the level of multi-source semantic conflict. A tiered processing approach is implemented by comparing the index with a preset threshold. When the index is normal, it is directly and quickly weighted and fused; physical constraint reconstruction is only invoked when a substantial conflict is detected. This strategy, while maintaining the system's real-time response capability, helps to reliably identify potentially fatal contradictions and avoids resource waste caused by frequent deep computation triggered by minor noise.

[0017] 3. When severe semantic conflicts occur between high-confidence sources, a passenger flow quality conservation model is introduced as a hard physical constraint. A physically consistent reference state is obtained through anchored subset selection and least squares estimation. The support of the original evidence is then redistributed based on the physical violation of each source's measurements using a negative exponential penalty coefficient. This mechanism forces the fusion results to revert to the physically feasible region, which helps resolve contradictions, ensures physical self-consistency in passenger flow state estimation, and reduces the risk of generating high-risk contingency plans. Simultaneously, it automatically identifies and alerts sensors with abnormally weakened support, contributing to improved overall system operation and maintenance efficiency and security. Attached Figure Description

[0018] Figure 1 This is a flowchart of steps S1-S8 in a digital contingency plan generation method for rail transit according to this application. Detailed Implementation

[0019] This application discloses a method for generating digital contingency plans for rail transit, referring to... Figure 1 ,include: S1. Receive multi-source passenger flow data and perform time-series standardization processing on the multi-source passenger flow data to generate a standard event stream with a unified timestamp.

[0020] It should be noted that during the multi-source passenger flow data access phase, raw passenger flow event records from sensing devices of different manufacturers and using different communication protocols need to be received. These sensing devices include at least video passenger flow analysis cameras, gate counting systems, and train weighing sensing units. Due to time synchronization deviations ranging from hundreds of milliseconds to several seconds between the hardware clocks of each device and the system reference clock, and the random jitter caused by differences in network transmission paths, the timestamps of passenger flow changes occurring at the same physical moment recorded at the system access layer are scattered within an uncertain range. If these raw records with inconsistent time coordinate systems and formats are directly sent to the subsequent alignment and fusion stages, the same passenger flow event will be incorrectly associated with different times or different events will be incorrectly merged, thus introducing structural errors at the fusion source. Therefore, this step first standardizes the raw event records output by each data source, generating a standard event stream with a unified timestamp.

[0021] In one embodiment, generating a standard event stream with a unified timestamp includes: converting event records output from various data sources in multi-source passenger flow data into a standard event record structure containing event type, spatial location identifier, data source identifier, unified timestamp, and measurement value. The unified timestamp is a timestamp compensated for clock skew, with a precision in milliseconds.

[0022] As an example, when standardizing the output of various data sources, a unified standard event record structure is first defined. This standard event record structure will serve as the sole data carrier for all subsequent processing. Each standard event record contains the following fields: an event type field, used to identify the type of passenger flow measurement described by the record, such as "instantaneous number of people remaining on the platform," "number of people exiting the gate per unit time," and "estimated number of people disembarking from the train"; a spatial location identifier field, expressed using platform numbers or pre-defined grid area codes, used to anchor the record to a specific physical area; a data source identifier field, recording the specific sensor device that generated the record in the form of a unique device serial number or system registration number; a unified timestamp field, accurate to milliseconds (ms), recording the time of occurrence of the event under the system's reference clock; a measurement value field, carrying the passenger flow-related measurement value given by the device in the current event; and a source built-in reliability field, which is optional and only filled in if the sensor device itself has the capability to output measurement reliability; otherwise, it is left blank. The above field definitions shield the differences in message format, field naming, and semantic expression between different brands of devices, enabling data from heterogeneous sources to be integrated into the processing pipeline in a unified form.

[0023] Next, clock skew compensation is applied to the original timestamps to generate uniform timestamps. For each sensor device registered in the operation and maintenance system, the system maintains the deviation correction value between it and the reference clock. This deviation correction value is obtained through a periodic or event-driven clock synchronization protocol, such as calculated under the Network Time Protocol (NTP) or Precision Time Protocol (PTP) framework, and is expressed in milliseconds (ms). Let the... A device-side timestamp carried by a certain original record from a data source is Then its unified timestamp Calculate as follows: In the formula, For the first The deviation correction value of the data source relative to the system reference clock, in milliseconds (ms), is continuously updated by clock synchronization. After this compensation, records generated by different data sources at the same physical moment will obtain a consistent timestamp under the system reference clock, thereby eliminating the fixed time offset between the sources.

[0024] Finally, after completing time deviation compensation and field normalization, the system assigns a monotonically increasing event sequence number to each newly arrived standard event record and appends the record to the standard event stream. The standard event stream is a sequence of events arranged in ascending order according to a uniform timestamp, managed by a message queue or stream processing framework, and serves as the direct input for subsequent sliding time window operations and evidence support calculations.

[0025] At this point, this step transforms the heterogeneous, asynchronous, and differently formatted raw passenger flow data from various sensing devices such as video analytics, gate counting, and train weighing into a standard event stream expressed in a standard event log structure, with timestamps unified to the system's reference clock and millisecond precision.

[0026] S2. Select candidate events corresponding to the standard event stream by sliding time windows, and form a corresponding candidate alignment set based on the candidate events.

[0027] In one embodiment, candidate events corresponding to a standard event stream are selected by a sliding time window, including: setting a sliding step size and a window half-width; for each time to be processed, creating a time interval centered on the time to be processed and defined by the window half-width as a sliding window; forming a candidate alignment set by all candidate events whose timestamps from all data sources fall within the time interval; and setting the window half-width based on the larger of the maximum end-to-end transmission delay and the tolerable fusion lag.

[0028] S3. Generate the evidence support level for each data source based on the temporal proximity, historical accuracy, and data integrity coefficient of each data source in the candidate alignment set.

[0029] It should be noted that after the standard event stream is generated, due to differences in network transmission paths and communication quality among different data sources, the standard event records corresponding to passenger flow events occurring at the same physical moment may still be scattered across the timeline due to random delays and jitter. Furthermore, the sampling frequency and packet loss conditions of each device are also different. Therefore, pairing based solely on the principle of closest timestamps can easily lead to erroneous associations. This application expands the tolerance range in the time dimension by constructing a sliding time window, and comprehensively calculates the time proximity, historical accuracy, and data integrity coefficient for each data source to generate multi-source evidence support, i.e., the evidence support corresponding to each data source, to objectively evaluate the data credibility of each source at the current moment.

[0030] The process involves generating the evidence support level for each data source based on its temporal proximity, historical accuracy, and data integrity coefficient in the candidate alignment set. This includes: obtaining the device's historical accuracy for each data source, calculated by long-term statistical verification of the frequency of historical device measurements matching the baseline measurement; obtaining the data integrity coefficient for each data source, calculated by the proportion of successfully received data packets within a recent time window; for each data source in the candidate alignment set, calculating the absolute value of the time difference between the timestamp of each data source record and the current moment, and using a negative exponential decay function to calculate the temporal proximity; and multiplying the temporal proximity, device historical accuracy, and data integrity coefficient to obtain the evidence support level for each data source at the current moment. Furthermore, if multiple records from the same data source exist within a sliding window, the maximum temporal proximity among them represents the temporal proximity of the same data source at the current moment.

[0031] In one embodiment, the system presets a sliding step size. The default value is 1 second, which can be adjusted according to the system's processing capacity. The maximum end-to-end transmission delay can also be set. Take 4 seconds of experience and tolerated fusion lag The window is set to an empirical value of 5 seconds. This empirical value is a preset parameter and can be reconfigured by the implementer according to the specific network environment. (Window width is half-width.) Pick and The larger value in this embodiment The time limit is set at 5 seconds. Specifically, for each pending moment that is processed in increments... The system is based on Construct time intervals around the center As the current sliding window, the standard event stream is then scanned, and records with uniform timestamps within this interval from all data sources are collected to form a candidate alignment set. This candidate alignment set contains all observation records that may be related to the passenger flow status at that moment. Through this sliding window design, records that arrive late due to network congestion but actually occur... Nearby records are included in the candidate set at that moment, thus compensating for the temporal misalignment caused by random delays.

[0032] Secondly, regarding the candidate alignment set, the first... For each record in the data source, obtain its unique timestamp. Calculate the current time absolute value of time difference And adopt a negative exponential decay function As the temporal proximity of the record, The time proximity decay coefficient is taken as an empirical value. The unit is per second, which is a preset parameter and can also be calibrated according to the statistical characteristics of the on-site delay; if the first... If multiple records exist within the same window from the same data source, the maximum temporal proximity is taken as the temporal proximity of that source at the current moment. This is to reflect the reliability of the most recent observations.

[0033] At the same time, the system reads the first data from the device health status database. Device historical accuracy of data sources This parameter is derived from long-term calibration statistics of the sensors by the operation and maintenance platform. It represents the frequency at which the historical measurement values ​​of the device match the baseline measurement values, and its value range is [not specified]. to Dimensionless, reflecting the inherent measurement reliability of the device; in addition, the average packet loss rate of the device over the past 5 minutes can be obtained through the network management interface. The 5-minute time window is the statistical window length, which is taken as an empirical value of 5 minutes. It can also be adjusted according to the data update frequency, and then the data integrity coefficient is calculated. , Similarly, the value is taken as follows to Dimensionless, used to measure the stability and data integrity of the current communication link.

[0034] Then, the time proximity obtained above Historical accuracy of equipment With data integrity coefficient Multiplying the three together, we get the first... Data sources at time The support of multiple sources of evidence, i.e. The product value is dimensionless and lies between arrive Between, the closer their values ​​are The stronger the evidence that the measurement value from that source at this moment supports the actual state of the passenger flow event, the better; when a data source has no records within the sliding window, its... Take directly This leads to for This source is considered invalid at this moment and is automatically excluded. Finally, the system outputs the measurement values ​​from all valid data sources. and the corresponding multi-source evidence support This serves as the input for subsequent calculations of the global conflict level.

[0035] S4. Calculate the normalized measure dissimilarity between any two data sources in the candidate alignment set, and then perform a weighted average of the normalized measure dissimilarity using the product of the evidence support levels of the two data sources as the weight to obtain the global conflict level index.

[0036] It should be noted that when obtaining data from various data sources at time... Measurement values Support from multiple sources of evidence Subsequently, although the evidence support level comprehensively characterizes the quality of time alignment, the accuracy of equipment history, and the integrity of real-time communication, due to the differences in the working principles, on-site obstructions, and sensing ranges of different sensors in passenger flow observation, multiple sources with high support may still output contradictory measurement values. For example, the number of people stranded on the platform estimated by video analysis may be much higher than the number of people disembarking estimated by train weighing. If such contradictory measurement values ​​are directly weighted and fused, a compromise will be obtained, but it may deviate significantly from the physical reality, posing a safety risk to the subsequent contingency plan generation. Therefore, it is necessary to first objectively quantify the overall semantic inconsistency between the multi-source measurement values ​​at the current moment, that is, to calculate the global conflict degree index.

[0037] The calculation of the normalized measurement dissimilarity between any two data sources in the candidate alignment set includes: obtaining the measurement values ​​of any two different data sources at the current time, calculating the absolute value of the difference between the measurement values, and dividing by the larger of the maximum value of the two measurement values ​​and a preset zero-prevention constant to obtain the normalized measurement dissimilarity. A weighted average of the normalized measurement dissimilarity is then performed using the product of the evidence support levels corresponding to any two data sources as the weight to obtain the global conflict level index. This includes: multiplying the product of the evidence support levels corresponding to any two different data sources by the corresponding normalized measurement dissimilarity and summing the results, then dividing by the sum of all evidence support products to obtain the global conflict level index; if the number of valid data sources is less than two, the global conflict level index is directly set to zero.

[0038] In one embodiment, during implementation, the calculation is first performed on any two different data sources. and Normalized measure difference between Its expression is: In the formula, and The first and the Data sources at the current moment The unit of measurement depends on the type of event; for example, the number of people on the platform is measured in people. The denominator is the maximum of the two measured values ​​and a preset zero-prevention constant. The larger the value, the more empirical the zero constant should be. Dimensionless, used to prevent errors in division by zero calculations when all measured values ​​are zero or extremely small. This is the basis for... Normalized in to Between, dimensionless, when the two source measurements are completely identical The greater the measurement difference, the closer the value is to the mean. This eliminates the influence of the absolute magnitude of the measured value on the measure of difference.

[0039] Next, using the product of the support levels of the two sources as weights, a weighted average of the normalized measure discrepancies among all valid source pairs is calculated to obtain the global conflict index. The calculation method is as follows: In the formula, For support greater than The total number of valid data sources; and The first and the Data sources at time Multi-source evidence support, dimensionless, range of values The output is calculated from the step "selecting candidate events through a sliding time window, and generating evidence support for each data source based on its temporal proximity, historical accuracy, and data integrity coefficient". If the number of valid data sources... If a source pair cannot be formed, then directly... Set as This indicates that the conditions for evaluating multi-source conflicts are not met at the current moment. This is because the weights are calculated using the support product. Significant measurement differences among sources with high support will have an impact on This produces a strong amplification effect, while the contribution of low-support sources, even if they carry outliers, is rapidly attenuated by the product, making... It can more accurately highlight the substantive semantic contradictions that exist between trusted sources.

[0040] Furthermore, the results obtained through the above calculations Also normalized in arrive Between, dimensionless, their values ​​are closer to This indicates a more severe semantic conflict between multi-source measurements at the current moment. In this step, a global conflict level index is calculated. Then, it was compared with the measurement values ​​from each data source. and the degree of support of the original evidence The results are output together and compared with the preset conflict triggering threshold in the subsequent conflict resolution process to determine whether to directly perform weighted fusion or initiate the physical constraint reconstruction process based on the passenger flow quality conservation model.

[0041] S5. If the global conflict level index does not exceed the preset threshold, the evidence support level corresponding to the data source is retained.

[0042] S6. If the global conflict level index exceeds the preset threshold, the evidence support level corresponding to the data source is corrected using the passenger flow quality conservation model, and the corresponding corrected evidence support level is generated.

[0043] It should be noted that when the calculated global conflict level index If the threshold is not exceeded, the evidence support level corresponding to the current data source is retained, and the calculated global conflict level index is... Exceeding the preset conflict trigger threshold In this embodiment If an empirical value of 0.6 is taken as the preset parameter, which is dimensionless, it can also be adjusted by the implementer based on historical data playback simulation test. If it is determined that there is a serious semantic conflict between the current multi-source measurement values, the passenger flow estimate obtained by direct weighted fusion may seriously violate the physical laws such as the conservation of passenger flow quality at the platform. Therefore, it is necessary to use the passenger flow quality conservation model to correct the evidence support.

[0044] Specifically, if the global conflict level index exceeds a preset threshold, the evidence support corresponding to the data source is corrected using the passenger flow quality conservation model to generate a corresponding corrected evidence support. This includes: calculating the source measurement deviation between the measurement value of each data source and the initial fusion value; selecting a credible anchor subset from each data source based on the evidence support threshold and consistency tolerance conditions; if the credible anchor subset contains at least two data sources, combining the state estimate provided by the credible anchor subset with the passenger flow quality conservation model, and using the least squares criterion to solve for the physical consistency reference state; if a credible anchor subset cannot be obtained, the historical trend prediction value is predicted based on the historical fusion state values ​​several times prior to the current time as the physical consistency reference state; calculating the physical penalty coefficient based on the degree of deviation of the measurement value of each data source from the physical consistency reference state, multiplying the original evidence support by the physical penalty coefficient, and normalizing to obtain the corrected evidence support. Furthermore, correcting the evidence support corresponding to the data source using the passenger flow quality conservation model also includes: generating an equipment maintenance alarm if the corrected evidence support is lower than a preset threshold for an abnormal data source.

[0045] In one embodiment, the existing original evidence support is first utilized. Calculate an initial weighted fusion value The calculation method is as follows: In the formula, This represents the total number of currently valid data sources. For the first The degree of support from the original evidence of the source, dimensionless. For the first The measurement value of the source depends on the type of passenger flow; for example, the number of people on the platform is [number missing]. While it cannot be directly taken as the final state in cases of severe conflict, it provides a baseline reference for preliminary analysis of the relative positions of the various sources.

[0046] In obtaining Subsequently, in order to extract an internally consistent and highly reliable subset of observations from the contradictory data, the source measurement deviation of each data source is calculated. And set the anchor source support threshold. With consistency tolerance In this embodiment Take an empirical value of 0.7, which is dimensionless. The empirical value of 0.3 is used; it is dimensionless and represents preset parameters. These can be adjusted by the implementer based on sensor characteristics, only when a certain data source meets the requirements. Furthermore, the normalized measure of difference between the source and other sources in the candidate set that meet the support criteria. Only then will the source be included in the trusted anchor subset. This condition ensures that the selected subset is highly consistent and of reliable origin.

[0047] like If at least two data sources are included, the physically consistent reference state can be solved by combining the passenger flow quality conservation model. For the platform area, during the time period (This embodiment) (Taking 60 seconds), changes in the number of people on the platform It should equal the net increase in the number of turnstiles. Net increase in train boarding and alighting The sum of these values ​​yields the conserved recurrence relation: In the formula, for Historical figures for the number of people on the front platform. and Separately, the turnstile and the weighing system are located at... The cumulative values ​​within the time period are provided. The measurements of each anchor subset are recorded as observations, and a least-squares objective function is constructed: Minimize the function to obtain the analytical solution: In the formula, The number of data sources within the trusted anchor subset. This refers to an optimized reference state that integrates internal consistency and physical conservation. When a reliable anchor subset containing at least two data sources cannot be obtained, it indicates that there is no sufficiently consistent and reliable subset in the current contradictory data to provide anchorage. In this case, historical trends are used as the physical reference, and the system saves the previous data. The final merged passenger flow status at that moment, The empirical value of 10 is used as the preset parameter, which can also be adjusted. The historical trend forecast value is calculated by using an exponentially weighted moving average. The recurrence relation is: in, This represents the passenger flow status value output by the system at the previous moment. The smoothing coefficient is taken as an empirical value of 0.3, which is dimensionless. The initial value is the historical prediction value calculated in the previous round. The first valid state value can be taken at the initial moment. At this time, directly set... .regardless Regardless of the method used to obtain the data, the support level of the original evidence needs to be adjusted based on the degree of deviation of each source measurement. The definition of the first... Physical violation of each data source : In the formula, To prevent zero infinitesimal constants, take The unit is consistent with the measured value. A smaller value indicates that the source output conforms more closely to the laws of physics, thus allowing for the calculation of the physical penalty coefficient. : Among them, attenuation coefficient Using an empirical value of 2.0 (dimensionless), and controlling the sharpness of the penalty, the original evidence support is multiplied by this value to obtain the revised evidence support. : Source regions with high physical violation The sharp decrease leads to a significant reduction in the support level for the correction.

[0048] Finally, all valid sources are normalized according to the revised evidence support level, i.e.: Normalized The dimensionless value with a sum of 1 represents the final evidence support output from this step, which will be used for weighted fusion later. Data sources with values ​​below the preset abnormal alarm threshold (an empirical value of 0.2, dimensionless) are identified as suspected faulty equipment. The system automatically generates an equipment maintenance alarm and pushes it to the monitoring system.

[0049] S7. Normalize the evidence support and the revised evidence support to obtain the fusion weights for each data source.

[0050] In one embodiment, after completing conflict resolution and possible physical penalty correction, the system has obtained the current time. The degree of evidence support from each valid data source. When the global conflict index in the preceding steps... Not exceeding the preset threshold If the evidence support is not adjusted, then the evidence support at this time is the evidence support corresponding to the data source output in step S3. ;when When the threshold is exceeded, the evidence support level has been adjusted by the physical penalty coefficient in step S6. In either case, the sum of the supporting evidence from all valid data sources is usually not equal to... Weighted fusion requires that the weights of each source satisfy a normalization condition. Therefore, it is necessary to uniformly transform the evidence support of all current valid data sources into a sum. The fusion weight.

[0051] As an example, the specific method of normalization is: for the current The first of the valid data sources For each source, divide its evidence support by the sum of the evidence support of all valid sources to obtain the source's position at time [time]. Fusion weights : In the formula, For the first The degree of actual evidence supporting the source at the current moment, when the conflict is not serious. When the conflict is serious, take All are dimensionless; This represents the total number of currently valid data sources. For the first The fusion weights of the sources are dimensionless and satisfy the following conditions: After normalization This refers to the fusion weights used in the weighted fusion calculation of passenger flow status.

[0052] S8. Use fusion weights to weight and fuse the measurement values ​​from each data source to obtain passenger flow status, and generate digital contingency plans based on passenger flow status.

[0053] It should be noted that, at the time of acquisition The fusion weights of the following valid data sources and corresponding measurement values Subsequently, passenger flow status estimates Weighted fusion Given, in the formula, For the current moment Estimated passenger flow status, in person. For the first The fusion weights of the data sources are dimensionless. For the first Measurements from one data source, in units of one person. The weighted average result, representing the total number of valid data sources, is directly used as the current passenger flow status output value. The system has preset multi-level passenger flow density thresholds, with the first level threshold... Get experience points Human, second threshold Get experience points Human, third threshold Get experience points The number of people is preset and can be adjusted by the operator according to the actual capacity of the platform. If the contingency plan is not activated, Then the mild evacuation plan will be triggered. Then a moderate flow restriction plan will be triggered. This will trigger the highest level of emergency evacuation plan.

[0054] In one embodiment, the contingency plan generation unit extracts corresponding measure templates from the contingency plan template library according to the matched contingency plan level. The mild diversion template includes instructions such as opening two more exit gates and switching the guide screens to diversion guidance mode, while the moderate flow restriction template further includes issuing instructions to the entrance gates. Instructions such as limiting entry flow and switching some staircases to one-way exit were issued. The highest-level evacuation mode triggered a station-wide evacuation announcement, forced all turnstiles to switch to exit direction, and closed entry channels. The system then displayed the current platform location and passenger flow status. The template is filled with timestamps to generate a digital contingency plan message in JSON format, which is then pushed to the gate controller, directional sign controller and field terminal in real time through the message middleware, and automatically parsed and executed by each execution unit.

[0055] This application also discloses a digital contingency plan generation system for rail transit, including: The standard event stream generation module is used to receive multi-source passenger flow data and perform time-series standardization processing on the multi-source passenger flow data to generate a standard event stream with a unified timestamp. The candidate alignment set module is used to select candidate events corresponding to the standard event stream through a sliding time window, and to form a corresponding candidate alignment set based on the candidate events; The evidence support metric module is used to generate the evidence support level for each data source based on the temporal proximity, historical accuracy, and data integrity coefficient of each data source in the candidate alignment set. The conflict assessment module calculates the normalized measure difference between any two data sources in the candidate alignment set, and performs a weighted average of the normalized measure difference using the product of the evidence support levels of the two data sources as the weight, to obtain the global conflict level index. The conflict resolution module retains the evidence support level corresponding to the data source if the global conflict level index does not exceed the preset threshold. If the global conflict level index exceeds the preset threshold, the conflict resolution module uses the passenger flow quality conservation model to correct the evidence support level corresponding to the data source and generates the corresponding corrected evidence support level. The fusion weight module is used to normalize the evidence support and the revised evidence support to obtain the fusion weight corresponding to each data source. The contingency plan generation module is used to weight and merge the measurement values ​​from various data sources using fusion weights to obtain passenger flow status and generate digital contingency plans based on passenger flow status.

[0056] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for generating digital contingency plans for rail transit, characterized in that, include: Receive multi-source passenger flow data and perform time-series standardization processing on the multi-source passenger flow data to generate a standard event stream with a unified timestamp; Candidate events corresponding to the standard event stream are selected by sliding time windows, and a corresponding candidate alignment set is formed based on the candidate events; Based on the temporal proximity, historical accuracy, and data integrity coefficient of each data source in the candidate alignment set, the evidence support level corresponding to each data source is generated. Calculate the normalized measure difference between any two data sources in the candidate alignment set, and then perform a weighted average of the normalized measure difference using the product of the evidence support levels corresponding to the two data sources as the weight to obtain a global conflict index. If the global conflict level index does not exceed the preset threshold, the evidence support level corresponding to the data source is retained; If the global conflict level index exceeds the preset threshold, the evidence support level corresponding to the data source is corrected using the passenger flow quality conservation model, and a corresponding corrected evidence support level is generated. The evidence support and the modified evidence support are normalized to obtain the fusion weights corresponding to each data source. The measurement values ​​from each data source are weighted and merged using the fusion weights to obtain the passenger flow status, and a digital contingency plan is generated based on the passenger flow status.

2. The method for generating digital contingency plans for rail transit according to claim 1, characterized in that, Generating a standard event stream with a unified timestamp includes: converting the event records output by each data source in the multi-source passenger flow data into a standard event record structure containing event type, spatial location identifier, data source identifier, unified timestamp and measurement value, wherein the unified timestamp is a timestamp after clock deviation compensation and has a precision of milliseconds.

3. The method for generating digital contingency plans for rail transit according to claim 2, characterized in that, Candidate events corresponding to the standard event stream are selected by sliding a time window, and a corresponding candidate alignment set is formed based on the candidate events, including: Set the sliding step size and window half-width. For each time to be processed, create a time interval centered on the time to be processed and defined by the window half-width as a sliding window. Form a candidate alignment set for all candidate events whose timestamps from all data sources fall within the time interval. The window half-width is set according to the larger of the maximum end-to-end transmission delay and the tolerable fusion lag.

4. The method for generating digital contingency plans for rail transit according to claim 3, characterized in that, Based on the temporal proximity, historical accuracy, and data integrity coefficient of each data source in the candidate alignment set, the evidence support level corresponding to each data source is generated, including: The historical accuracy of the device corresponding to each data source is obtained by long-term verification and statistical analysis of the frequency of the historical measurement value of the device matching the benchmark measurement value. Obtain the data integrity coefficient for each data source, which is calculated by the proportion of data packets successfully received within a recent time window; For each data source in the candidate alignment set, calculate the absolute value of the time difference between the timestamp of each data source record and the current time, and use the negative exponential decay function to calculate the time proximity. Multiplying the time proximity, device historical accuracy, and data integrity coefficient yields the evidence support level for each data source at the current moment.

5. The method for generating digital contingency plans for rail transit according to claim 4, characterized in that, If multiple records exist within the sliding window from the same data source, the maximum value of the temporal proximity among them represents the temporal proximity of the same data source at the current moment.

6. The method for generating digital contingency plans for rail transit according to claim 1, characterized in that, Calculating the normalized measurement dissimilarity between any two data sources in the candidate alignment set includes: Obtain the measurement values ​​from any two different data sources at the current moment, calculate the absolute value of the difference between the measurement values, divide it by the larger of the maximum value of the two measurement values ​​and the preset zero-prevention constant, and obtain the normalized measurement difference degree.

7. The method for generating digital contingency plans for rail transit according to claim 1, characterized in that, The normalized measure discrepancies are weighted and averaged using the product of the evidence support levels corresponding to any two data sources to obtain a global conflict index, including: The global conflict index is obtained by multiplying the product of the evidence support for any two different data sources by the corresponding normalized measure of dissimilarity and summing the results, and then dividing by the sum of the products of all evidence support. If there are fewer than two valid data sources, the global conflict level indicator will be set to zero.

8. The method for generating digital contingency plans for rail transit according to claim 1, characterized in that, If the global conflict level index exceeds the preset threshold, the evidence support level corresponding to the data source is corrected using the passenger flow quality conservation model, and a corresponding corrected evidence support level is generated, including: Calculate the source measurement deviation between the measurement value from each data source and the initial fused value; Based on the evidence support threshold and consistency tolerance conditions, a credible anchor subset is selected from each data source; If the trusted anchoring subset contains at least two data sources, then the state estimate provided by the trusted anchoring subset is combined with the passenger flow quality conservation model, and the least squares criterion is used to solve for the physically consistent reference state. If the trusted anchor subset cannot be obtained, the historical trend prediction value is obtained based on the historical fusion state value of several times before the current time and is used as the physical consistency reference state. The physical penalty coefficient is calculated based on the degree to which the measurement values ​​from each data source deviate from the physical consistency reference state. The original evidence support is then multiplied by the physical penalty coefficient and normalized to obtain the corrected evidence support.

9. A method for generating digital contingency plans for rail transit according to claim 8, characterized in that, The method of correcting the evidence support level corresponding to the data source using the passenger flow quality conservation model also includes: if the evidence support level of the corrected data source is lower than a preset threshold, then an equipment maintenance alarm is generated.

10. A digital contingency plan generation system for rail transit, characterized in that, include: The standard event stream generation module is used to receive multi-source passenger flow data and perform time-series standardization processing on the multi-source passenger flow data to generate a standard event stream with a unified timestamp. The candidate alignment set module is used to select candidate events corresponding to the standard event stream through a sliding time window, and to form a corresponding candidate alignment set based on the candidate events; The evidence support metric module is used to generate the evidence support degree corresponding to each data source based on the temporal proximity, historical accuracy and data integrity coefficient of each data source in the candidate alignment set; The conflict level assessment module calculates the normalized measurement difference between any two data sources in the candidate alignment set, and performs a weighted average of the normalized measurement difference using the product of the evidence support levels corresponding to the two data sources as the weight, to obtain a global conflict level index. The conflict handling module is used to retain the evidence support level corresponding to the data source if the global conflict level index does not exceed the preset threshold, and to correct the evidence support level corresponding to the data source by using the passenger flow quality conservation model to generate the corresponding corrected evidence support level. The fusion weight module is used to normalize the evidence support and the modified evidence support to obtain the fusion weight corresponding to each data source. The contingency plan generation module is used to use the fusion weight to weight and fuse the measurement values ​​of each data source to obtain the passenger flow status, and generate a digital contingency plan based on the passenger flow status.

Images