Rail transit scheduling analysis method and system based on artificial intelligence

Abstract

The invention provides a rail transit dispatching analysis method and system based on artificial intelligence, and belongs to the technical field of rail transit dispatching.The method comprises the steps that firstly, a dynamic dispatching baseline of a rail transit system is established, the dynamic dispatching baseline comprises reference operation features, rule constraint features and scene association features, and then real-time dispatching data is collected; the method comprises the following steps: performing multi-dimensional comparative analysis on a dynamic scheduling baseline to generate a baseline deviation feature set, calling a pre-trained scheduling collaborative evaluation model to perform dynamic correlation analysis on the baseline deviation feature set to generate a scheduling collaborative evaluation result, and constructing a multi-target scheduling optimization model based on the scheduling collaborative evaluation result to perform collaborative optimization processing. According to the method, an initial scheduling adjustment scheme set is generated, finally, dynamic constraint verification processing is performed on the initial scheduling adjustment scheme set, and a final rail transit scheduling instruction is generated and used for triggering a scheduling control system to execute scheduling parameter updating operation, so that the running efficiency, safety and stability of a rail transit system can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit scheduling, and in particular to an artificial intelligence-based rail transit scheduling analysis method and system. Background Art

[0002] In the field of rail transit scheduling, traditional scheduling analysis methods have numerous limitations. For one thing, some methods rely heavily on historical scheduling data and pre-set fixed rules to make scheduling decisions. However, historical data only reflects past operations and fails to fully account for real-time changing factors, such as sudden passenger flow and equipment failures. Furthermore, pre-set fixed rules struggle to adapt to complex and changing operational scenarios. When faced with special circumstances, these rules may not provide effective scheduling solutions, resulting in a lack of flexibility and adaptability in scheduling decisions.

[0003] On the other hand, existing scheduling analysis methods often focus on scheduling information in a single dimension, such as only considering scheduling in the time or space dimensions, while ignoring the correlations and mutual influences between different dimensions. Rail transit systems are complex and integrated, with multiple dimensions such as time, space, and resources being closely intertwined. Single-dimensional analysis cannot fully and accurately grasp the scheduling situation, easily leading to deviations in scheduling decisions and affecting the normal operation of the entire rail transit system. Summary of the Invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a rail transit scheduling analysis method based on artificial intelligence, the method comprising: Establishing a dynamic scheduling baseline for the rail transit system, which includes benchmark operating characteristics constructed from historical normal scheduling data, rule constraint characteristics from a preset scheduling rule library, and scenario-related characteristics from typical operating scenarios; Collecting real-time scheduling data of the rail transit system, performing a multi-dimensional comparison analysis on the real-time scheduling data and the dynamic scheduling baseline to generate a baseline deviation feature set, wherein the baseline deviation feature set includes a time dimension deviation feature, a space dimension deviation feature, and a resource dimension deviation feature; Calling a pre-trained scheduling collaborative evaluation model to perform dynamic correlation analysis on the baseline deviation feature set to generate a scheduling collaborative evaluation result, wherein the scheduling collaborative evaluation result includes a deviation conduction path parameter, a collaborative conflict probability parameter, and a resource adaptability parameter; Building a multi-objective scheduling optimization model based on the scheduling collaborative evaluation results, and using the multi-objective scheduling optimization model to collaboratively optimize the deviation conduction path parameters, the collaborative conflict probability parameters, and the resource adaptability parameters to generate an initial scheduling adjustment solution set; Dynamic constraint verification processing is performed on the initial scheduling adjustment plan set, and a final rail transit scheduling instruction is generated after verification. The final rail transit scheduling instruction is used to trigger the rail transit scheduling control system to perform a scheduling parameter update operation.

[0005] On the other hand, an embodiment of the present invention also provides an artificial intelligence-based rail transit scheduling analysis system, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0006] Based on the above aspects, an embodiment of the present invention establishes a dynamic scheduling baseline that includes benchmark operation characteristics constructed from historical normal scheduling data, rule constraint characteristics of a preset scheduling rule library, and scenario association characteristics of typical operating scenarios. It collects real-time scheduling data and performs multi-dimensional comparison analysis with the dynamic scheduling baseline to generate a baseline deviation feature set including time dimension deviation characteristics, space dimension deviation characteristics, and resource dimension deviation characteristics. This can accurately capture abnormal situations in the scheduling process and provide detailed data support for subsequent analysis. It calls a pre-trained scheduling collaborative evaluation model to perform dynamic correlation analysis on the baseline deviation feature set, and generates a scheduling collaborative evaluation result including deviation conduction path parameters, collaborative conflict probability parameters, and resource adaptability parameters. It deeply analyzes the inherent connections and impacts between deviations, and constructs a multi-objective scheduling optimization model based on the scheduling collaborative evaluation results for collaborative optimization processing. It generates an initial scheduling adjustment plan set, which can comprehensively consider multiple optimization goals and achieve overall optimization of the scheduling plan. Finally, it performs dynamic constraint verification processing on the initial scheduling adjustment plan set to generate a final rail transit scheduling instruction, ensuring the feasibility and effectiveness of the scheduling plan, and effectively improving the efficiency, safety, and stability of the rail transit system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a schematic diagram of the execution flow of the rail transit scheduling analysis method based on artificial intelligence provided by an embodiment of the present invention.

[0008] Figure 2 Schematic diagram of exemplary hardware and software components of an artificial intelligence-based rail transit scheduling and analysis system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of an artificial intelligence-based rail transit scheduling analysis method provided by an embodiment of the present invention. The artificial intelligence-based rail transit scheduling analysis method is introduced in detail below.

[0010] Step S110: establishing a dynamic scheduling baseline for the rail transit system, which includes benchmark operation characteristics constructed from historical normal scheduling data, rule constraint characteristics of a preset scheduling rule library, and scenario association characteristics of typical operating scenarios.

[0011] This example uses an urban rail transit network as an example. This network encompasses multiple interwoven lines connecting major urban areas, involving dozens of stations, hundreds of trains, and a centralized dispatching control center. Establishing a dynamic dispatch baseline is fundamental to the entire dispatch analysis method. Its core purpose is to create a benchmark framework that reflects the system's normal operating state, thereby accurately identifying deviations from this baseline state during real-time operations.

[0012] Step S111: collecting historical normal dispatching data of the rail transit system, wherein the historical normal dispatching data includes historical train operation data, historical line status data, historical passenger flow data and historical dispatching instruction data.

[0013] In this urban rail transit network, each train is equipped with a global positioning device, speed sensors, and in-car passenger counters. These devices record the train's trajectory in real time, including its position coordinates every second; the exact time of each stop, accurate to the second; the departure time; and the continuous speed curve during travel. These data together constitute the core content of historical train operation data. Track condition monitoring sensors are deployed along the line to detect track smoothness and wear. The signal system itself records its operating status, such as the display status of signal lights and the response time of signal switching. The power supply system has voltage and current monitoring devices. These data are aggregated to form historical line status data. Gate counters are installed at each station entrance and exit, recording the number of people entering and leaving the station every hour. Video analysis equipment is installed within the station, using image recognition technology to count the number of people staying in the station at different times. This data constitutes historical passenger flow data. The command system of the dispatching control center automatically archives all issued dispatch instructions, including the specific departure time, the station and duration of each stop, and the specific speed adjustment values, forming historical dispatch instruction data.

[0014] Step S112: Perform scene division processing on the historical normal scheduling data, and divide the historical normal scheduling data into multiple typical operation scene data units according to the operation period characteristics, weather condition characteristics and holiday attribute characteristics.

[0015] After completing data collection, the scenario division phase begins. The division of operating period characteristics is based on the travel patterns of urban residents. The morning peak period is set as the concentrated travel period for people going to work in the morning on weekdays, the evening peak period is the concentrated travel period for people leaving work in the evening on weekdays, the off-peak period is the normal daytime period outside the peak period, and the night period is the period from late evening to early morning. Weather condition characteristics are obtained by connecting to the historical meteorological data of the city's meteorological department, clearly distinguishing different weather conditions such as sunny, cloudy, rainy, and snowy days. Rainy days are further subdivided into light rain, moderate rain, and heavy rain, and snowy days are similarly subdivided according to the amount of snowfall. Holiday attribute characteristics are determined based on the national statutory holidays, weekends, and ordinary working days.

[0016] Scenario segmentation uses a multi-dimensional cross-combination approach, combining different values ​​of the three characteristics mentioned above to form multiple typical operational scenario data units. For example, the "Morning Peak - Light Rain - Weekday" scenario data unit contains all historical train operation data, historical line status data, historical passenger flow data, and historical dispatch instruction data during the morning peak period on weekdays with light rain; the "Evening Peak - Moderate Snow - Weekend" scenario data unit contains all corresponding data during the weekend evening peak period with moderate snow.

[0017] Step S113: Extract train punctuality characteristics, interval operation time characteristics, station stay time characteristics and resource occupancy rate characteristics from each typical operation scenario data unit, and combine the extracted characteristics to generate benchmark operation characteristics.

[0018] For each data unit of a typical operational scenario, feature extraction is performed using specialized data analysis tools. Train punctuality features are extracted by analyzing the operational status of all trains at all stops within that scenario. The difference between the actual and planned arrival times of each train at each station is calculated. Trains considered on-time when the difference is within a preset allowable range are then considered on-time. The proportion of on-time arrivals to the total number of arrivals is then calculated. The weighted average of the on-time rates for all trains and stations is then taken to obtain the train punctuality feature for that scenario. This feature is presented as a multidimensional array, containing the distribution of punctuality rates for different trains at different stations.

[0019] To extract interval running time features, the line is divided into multiple continuous intervals. For each interval, the travel time of all trains passing through the interval in that scenario is counted to form a time distribution sequence, including the minimum, maximum, average, and frequency of occurrence of intervals of different lengths. These data together constitute the interval running time features.

[0020] To extract the characteristics of the length of stay at the station, for each station, the stop time of all trains in the scenario is counted, and a feature sequence containing the minimum value, maximum value, average value and distribution frequency is also formed to reflect the characteristics of the stay time of different stations in the scenario.

[0021] The resource occupancy rate feature involves line resources and station platform resources. The line resource occupancy rate is the ratio of the time that the line section is occupied by trains to the total time per unit time; the station platform resource occupancy rate is the ratio of the time that the platform is occupied by trains per unit time and the number of people stranded in the station to the maximum capacity of the platform, etc. These data are integrated to form the resource occupancy rate feature.

[0022] Finally, the extracted train punctuality characteristics, interval operation time characteristics, station stay time characteristics and resource occupancy rate characteristics are combined in a set order to form a high-dimensional feature vector, which is the benchmark operation characteristic corresponding to the typical operation scenario data unit.

[0023] Step S114: Obtain a preset dispatching rule library, extract line capacity constraint rules, train minimum interval rules, station passenger flow carrying rules, and emergency dispatching priority rules from the preset dispatching rule library, and convert the extracted rules into rule constraint features.

[0024] The pre-set dispatching rules library, stored in the dispatching control center's database, represents a digital representation of a series of regulations and rules developed based on rail transit operational safety and efficiency requirements. Line capacity constraints clearly define the maximum number of trains that can safely operate per hour on each line during different time periods. This number is determined based on factors such as track strength, signal system response speed, and platform length. For example, a particular line may have a maximum number of trains per hour during peak hours and a smaller number during off-peak hours.

[0025] The minimum train interval rule stipulates the minimum time interval that must be maintained between two trains running on the same line. This minimum time interval is determined based on factors such as the braking distance of the train and the reaction time of the signal system. Different lines and time periods may have different values.

[0026] The station passenger flow carrying rules determine the maximum number of passengers that each station can carry at different times based on the station's platform area, aisle width, number of entrances and exits and other facility conditions. When the number of people in the station exceeds this number, flow control measures can be initiated.

[0027] The emergency dispatch priority rules list in detail the execution order of various dispatch instructions when different types of emergency situations occur. For example, in the event of a fire, the priority of the train emergency evacuation instruction is higher than the normal departure instruction; in the event of an equipment failure, the dispatch instruction of the maintenance train has a higher priority than the dispatch instruction of an ordinary passenger train.

[0028] When converting these rules into rule constraint features, feature encoding is used. For line capacity constraint rules, the maximum number of trains for each line in each time period is converted into a numerical matrix, where the rows represent the lines, the columns represent the time periods, and the matrix elements are the corresponding maximum number of trains. The minimum train interval rule is converted into a list containing line identifiers and corresponding minimum interval values. The station passenger flow carrying rule is converted into a dictionary structure indexed by the station identifier and containing the maximum number of passengers in different time periods. The emergency dispatch priority rule is converted into a priority sorting table that clearly defines the priority value of each instruction type.

[0029] Step S115: Analyze the correlation between the benchmark operation characteristics and the rule constraint characteristics in each typical operation scenario data unit, calculate the matching parameters of the correlation relationship, and based on the matching parameters, fuse the benchmark operation characteristics, rule constraint characteristics and scenario identification characteristics of the typical operation scenario to generate a dynamic scheduling baseline.

[0030] Analyzing the correlation between baseline operating characteristics and rule-constraint characteristics within each typical operational scenario data unit is a meticulous, multi-step process. Taking the "Morning Peak - Sunny Day - Weekday" scenario data unit as an example, the first step is to analyze whether the train running intervals in the baseline operating characteristics are within the minimum train interval requirements specified in the rule-constraint characteristics. The percentage of actual train running intervals that meet the minimum interval requirements in this scenario is calculated. Secondly, the number of trains in the baseline operating characteristics is analyzed to determine whether they fall within the maximum number of trains specified by the line capacity constraint rules, calculating the distribution of the ratio of the actual number of trains to the maximum permitted number. Furthermore, the station passenger flow data is analyzed to determine whether they fall within the range of the station passenger flow carrying rules. Through these analyses, a matching parameter is calculated to reflect the closeness of the correlation between the two. Based on this parameter, the baseline operating characteristics, rule-constraint characteristics, and corresponding scenario identification features are then organically integrated to form the dynamic scheduling baseline components for this scenario. Finally, the baseline components of all scenarios are integrated to construct a dynamic scheduling baseline for the entire rail transit system.

[0031] Step S1151: Extract the feature vector of the benchmark operation feature and the feature vector of the rule constraint feature from the typical operation scenario data unit, convert them into vector representations of the same dimension, calculate the cosine similarity between the benchmark operation feature vector and the rule constraint feature vector, and use the cosine similarity value as the preliminary matching parameter.

[0032] For each typical operational scenario data unit, a feature vector is first extracted from the baseline operational characteristics. This feature vector contains the specific values ​​of train punctuality in different dimensions, various statistical indicators of interval running time, various statistical indicators of station dwell time, various data on resource utilization, etc. Each feature is considered a dimension of the vector. Similarly, a feature vector is extracted from the rule constraint characteristics, including various values ​​of line capacity constraints, specific values ​​of minimum train intervals, various values ​​of station passenger flow, and values ​​of emergency dispatch priority.

[0033] When two vectors have different dimensions, feature expansion is performed on the vector with fewer dimensions, for example, by adding new dimensions through interpolation so that it has the same number of dimensions as the vector with more dimensions; or feature selection is performed on the vector with more dimensions, retaining the dimensions related to the vector with fewer dimensions to achieve the unification of the dimensions of the two.

[0034] After dimensionality unification, the cosine similarity of the two vectors is calculated. The calculation process involves first taking the dot product of the two vectors (multiplying the values ​​of the corresponding dimensions and then summing them). Next, the modulus of the two vectors is calculated (the square root of the sum of the squares of the values ​​of each dimension). Finally, the dot product is divided by the product of the two moduli to obtain the cosine similarity value. This cosine similarity value ranges from -1 to 1, with values ​​closer to 1 indicating a higher degree of similarity between the two vectors. This cosine similarity value is used as the initial matching parameter.

[0035] Step S1152: Analyze the correlation strength between the train punctuality feature in the benchmark operation feature and the train minimum interval rule in the rule constraint feature, and calculate the compliance parameter of the punctuality rate to the interval rule.

[0036] To analyze the correlation between train punctuality and the minimum interval rule, we first screened the baseline train punctuality data for different punctuality intervals, such as those above 90%, 80%-90%, and 70%-80%. For each punctuality interval, we counted the number of train intervals that met the minimum interval rule, as well as the total number of train runs within that interval. We then calculated the ratio of the number of such compliances to the total number of such compliances to obtain the compliance ratio for that punctuality interval.

[0037] The compliance ratios of all punctuality intervals are weighted averaged according to the level of punctuality, with the weight being the ratio of the number of runs in each interval to the total number of runs. The result is the compliance parameter of punctuality to the interval rule, which reflects the degree of correlation between train punctuality and compliance with the minimum train interval rule.

[0038] Step S1153: analyzing the correlation strength between the resource occupancy rate feature in the benchmark operation feature and the line capacity constraint rule in the rule constraint feature, and calculating the compliance parameter of the resource occupancy rate with the capacity rule.

[0039] When analyzing the correlation between resource utilization characteristics and line capacity constraints, divide the line resource utilization within the resource utilization characteristics into multiple intervals, such as below 60%, 60%-80%, and above 80%. For each resource utilization interval, count the number of trains actually running on the line within that interval that meet the maximum train count requirement in the line capacity constraint, as well as the total number of train runs. Calculate the ratio of these compliance times to the total number of runs to obtain the compliance ratio for that resource utilization interval.

[0040] Similarly, the compliance ratio of all resource occupancy intervals is weighted averaged according to the ratio of the number of operations within the resource occupancy interval to the total number of operations, and the compliance parameter of the resource occupancy to the capacity rule is obtained. This compliance parameter reflects the degree of correlation between the resource occupancy and compliance with the line capacity constraint rule.

[0041] Step S1154: performing weighted summation on the preliminary matching parameter, the compliance parameter of the punctuality rate to the interval rule, and the compliance parameter of the resource occupancy rate to the capacity rule to generate a comprehensive matching parameter.

[0042] To comprehensively assess the degree of match between baseline operating characteristics and rule constraints, a weighted summation of the preliminary matching parameters, the punctuality rate's compliance with spacing rules, and the resource utilization rate's compliance with capacity rules is required. Weights are determined based on expert experience and statistical analysis of historical data. Because minimum train spacing rules and line capacity constraints are directly related to operational safety, the punctuality rate's compliance with spacing rules and the resource utilization rate's compliance with capacity rules are weighted higher, while the preliminary matching parameters are weighted lower.

[0043] Specifically, each parameter is multiplied by its corresponding weight, and then the three products are added together. The resulting sum is the comprehensive matching parameter. This comprehensive matching parameter integrates matching information from multiple aspects and can more comprehensively reflect the matching between the baseline operating characteristics and the rule constraint characteristics.

[0044] Step S1155: If the comprehensive matching degree parameter is higher than the preset matching threshold, the benchmark operation characteristics, rule constraint characteristics and corresponding typical operation scenario identification characteristics are spliced ​​to generate a scenario baseline unit of the typical operation scenario.

[0045] The preset matching threshold is determined based on the historical safe operation data and dispatch quality requirements of the urban rail transit network. This threshold has been tested and adjusted repeatedly to ensure that only highly matched baseline operating characteristics and rule-constrained characteristics are used to construct the scenario baseline unit. When the calculated comprehensive matching parameter is higher than the preset matching threshold, it indicates that the baseline operating characteristics and rule-constrained characteristics have good consistency and adaptability in this typical operating scenario.

[0046] At this time, after converting the feature vector of the benchmark operation feature, the feature vector of the rule constraint feature, and the identification feature of the typical operation scenario (such as "morning rush hour-sunny day-weekday") into a unified feature format, they are spliced ​​in the order of the benchmark operation feature vector, the rule constraint feature vector, and the scenario identification feature to form a longer feature vector, which is the scenario baseline unit of the typical operation scenario.

[0047] Step S1156: Classify and store the scenario baseline units of all typical operation scenarios according to the scenario identification features to build a baseline library of dynamic scheduling baselines.

[0048] After generating scenario baseline units for all typical operating scenarios, they were categorized according to the various dimensions of the scenario identification features. First, they were categorized according to operating time periods into morning peak, evening peak, off-peak, and nighttime categories. Within each time period category, weather conditions were further categorized into subcategories such as sunny, cloudy, rainy, and snowy. Within each weather subcategory, holiday attributes were further categorized into subcategories such as weekdays, weekends, and statutory holidays.

[0049] After classification, the scene baseline units for each category are stored in a corresponding table in the database. The database uses a distributed storage architecture to improve data access efficiency. Simultaneously, an index based on scene identification features is established for the database. This allows subsequent queries to quickly locate the corresponding scene baseline unit based on the current scene information, thereby building a baseline library for dynamic scheduling baselines.

[0050] Step S1157: setting a dynamic update cycle for the baseline library, recalculating the comprehensive matching parameters of each scene baseline unit in each update cycle, and replacing scene baseline units whose comprehensive matching parameters are lower than a preset threshold.

[0051] Considering that the operational status of the urban rail transit network changes over time, such as the addition of new lines, changes in the number of trains, and changes in passenger travel habits, the baseline library is dynamically updated once a week. During each update cycle, historical normal scheduling data from the previous week is first collected. Then, according to the method of steps S111 to S1154, the comprehensive matching parameters of each typical operation scenario data unit are recalculated.

[0052] For existing scenario baseline units, if the recalculated comprehensive matching parameters are lower than the preset matching threshold, it means that the scenario baseline unit can no longer accurately reflect the current operating conditions and needs to be replaced with a newly calculated scenario baseline unit with a comprehensive matching parameter higher than the preset threshold. For newly added typical operation scenario data units, the comprehensive matching parameters are calculated according to the same process. If they meet the requirements, the corresponding scenario baseline unit is added to the baseline library, thus ensuring that the baseline library can always accurately reflect the current operating conditions and rule requirements.

[0053] Step S120: Collect real-time scheduling data of the rail transit system, perform multi-dimensional comparison analysis on the real-time scheduling data and the dynamic scheduling baseline, and generate a baseline deviation feature set, which includes time dimension deviation features, space dimension deviation features, and resource dimension deviation features.

[0054] During the real-time operation of the rail transit system, the dispatch monitoring system continuously collects various types of real-time dispatch data, which is transmitted in real time to the dispatch analysis center via a high-speed data transmission network. At the dispatch analysis center, this real-time dispatch data is meticulously compared with the corresponding scenario baseline units from the dynamic dispatch baseline library, analyzing the differences between the two across three key dimensions: time, space, and resources. Deviations are identified, and a baseline deviation feature set is generated that contains information on these three dimensions.

[0055] Step S121: real-time dispatching data is collected through the rail transit dispatching monitoring system. The real-time dispatching data includes real-time train position data, real-time line occupancy status data, real-time station passenger flow data and real-time dispatching instruction execution data.

[0056] The rail transit dispatching and monitoring system is a comprehensive system that integrates multiple monitoring devices and data transmission protocols. Real-time train location data is collected every second by high-precision positioning equipment installed on trains. This data includes the train's real-time latitude and longitude coordinates, line identifier, and section identifier. Simultaneously, the train's onboard computer packages this location information with the train's number and current timestamp, and transmits it to the dispatching center via a dedicated wireless communication network.

[0057] Real-time line occupancy status data is collected by sensors installed at the entrances and exits of line sections. Each sensor can detect the time when a train enters and leaves the section. This time information can be used to calculate whether each line section is currently occupied by a train and the duration of the occupancy. The signal system will also upload its judgment results on the line section occupancy status in real time as a supplement.

[0058] Real-time station passenger flow data is collected collaboratively by a variety of equipment within the station. At each entrance and exit of the station, the gate system records every opening and closing action in real time and links it to the corresponding ticket information, thereby counting the number of people entering and exiting the station every minute. High-definition cameras are installed in the station hall and platform areas. These cameras are connected to the image analysis server, which processes the video stream in real time through human body recognition algorithms to count the number of people stranded in different areas at different times, including the total number of stranded people in the station hall, the number of stranded people on each platform, and the number of people moving in the passage. In addition, some stations also install passenger flow sensors on escalators and stairwells to further assist in counting the direction and density changes of passenger flow. These data together constitute the real-time station passenger flow data, which is transmitted to the dispatching center in real time through a dedicated data channel.

[0059] Real-time dispatch instruction execution data is collected by the dispatch center's instruction execution feedback system. When the dispatch center issues a dispatch instruction, such as adjusting a train's departure time or changing a train's platform stop, the relevant train onboard systems and station control systems provide real-time feedback on the instruction's reception status, execution start time, execution progress, and execution results. For example, after receiving a departure time adjustment instruction, a train can provide feedback on whether the instruction was successfully received, whether the expected departure time was adjusted according to the instruction, and the deviation between the actual departure time and the instruction requirements. After receiving a passenger flow diversion instruction, a station can provide feedback on the activation of diversion measures, the broadcast status of the station broadcast, and the dispatch of staff. This feedback information is aggregated to form real-time dispatch instruction execution data.

[0060] Step S122: Based on the timestamp information of the real-time scheduling data, the corresponding typical operation scenario data unit is matched from the dynamic scheduling baseline to obtain the benchmark operation characteristics and rule constraint characteristics corresponding to the typical operation scenario data unit.

[0061] Each record in real-time dispatch data contains precise timestamp information, accurate to the millisecond, recording the exact moment the data was collected. The dispatch analysis system first parses this timestamp information to determine the current date and time. Based on the date information, it queries the preset holiday database to determine whether the current date is a weekday, weekend, or statutory holiday, thereby determining the holiday attributes. Based on the specific time, the system compares the preset operating period classification standards to determine whether the current period is morning peak, evening peak, off-peak, or nighttime, thereby determining the operating period characteristics. Simultaneously, the dispatch analysis system maintains a real-time connection with the city's meteorological service system to obtain current weather conditions, including weather type (sunny, cloudy, rainy, snowy, etc.) and precipitation or snowfall intensity, to determine weather characteristics.

[0062] After determining the three characteristics described above, the scheduling analysis system combines the operating period, weather conditions, and holiday attributes into a current scenario identifier, such as "morning rush hour - moderate rain - weekday." Based on this scenario identifier, the system then searches the baseline library of the dynamic scheduling baseline for the corresponding typical operating scenario data unit. The search process leverages the index established within the baseline library based on the scenario identifier to quickly locate the matching unit. Once the corresponding typical operating scenario data unit is found, the corresponding baseline operating characteristics and rule constraint characteristics are extracted from the unit, serving as a baseline reference for subsequent comparative analysis.

[0063] Step S123: perform time axis alignment processing on the real-time train position data and the historical train position data in the benchmark operation characteristics, and calculate the difference between the real-time arrival time and the benchmark arrival time in the same line section as the time dimension deviation feature.

[0064] To ensure comparability between real-time and historical train location data, timeline alignment is required. First, the train's route information and the estimated transit order for each section are extracted from the real-time train location data. Then, within the historical train location data for the baseline operating characteristics, historical travel records for the same train under similar dates and weather conditions are identified and their corresponding transit order is extracted. The transit orders of the two sections are then compared to ensure identical routes. If there are any temporary reroutes, the matching historical data must be re-screened.

[0065] After identifying the matching historical data, the timeline is aligned, using the starting point of the track section as the reference point. The time when the real-time train enters each track section is aligned with the time when the historical train entered the same track section. For example, using the time when the train enters the first track section as the reference zero point, the offsets of the entry times of the real-time train and the historical train in each subsequent track section relative to this reference zero point are calculated. This method achieves timeline alignment.

[0066] After alignment is complete, for each line section, the real-time train arrival time and the benchmark time of the historical train arrival at the same section are found, and the difference between these two times is calculated. If the real-time arrival time is later than the benchmark arrival time, the difference is positive; if the real-time arrival time is earlier than the benchmark arrival time, the difference is negative. These differences for all line sections are arranged in order of the line sections to form a time dimension deviation feature, which reflects the train's deviation from the benchmark operating state in the time dimension.

[0067] Step S1231: extracting real-time train location data from real-time scheduling data, and parsing the train identification information, line section identification information and arrival timestamp information in the real-time train location data.

[0068] From the overall stream of real-time dispatch data, data filtering rules are used to select all records marked as real-time train location data. Each record contains a header that clearly identifies the data type as train location data. Each train location data record is parsed to extract the train identification information. The train identification is a unique string consisting of the train's line code and train number, which uniquely identifies a specific train.

[0069] The line section identification information is parsed. This information consists of a line code and a section number. The line code represents the line the train is currently on, and the section number represents the specific section on that line. For example, the section numbers for a line increase sequentially from the starting point, with each number corresponding to a continuous section of track. The arrival timestamp information is also extracted. This timestamp, accurate to the second, records the moment when the train fully enters the end of the line section—that is, the time when both the front and rear of the train pass the signal marking at the end of the section.

[0070] Step S1232: Based on the train identification information and the line section identification information, the corresponding historical train position data is matched from the benchmark operation characteristics, and the historical arrival timestamp information is extracted.

[0071] Using the parsed train identification information, we search the historical train location data for the benchmark operation characteristics to find all the historical operation records of the train over the past period. Then, based on the line section identification information, we filter out the historical records containing the same line section identification, that is, the records in which the train has traveled the same line section.

[0072] The selected historical records are further matched against the current scenario characteristics, prioritizing those with the same or similar characteristics as the current operating period, weather conditions, and holiday attributes to ensure their reference value. From the highest-matching historical records, the train's historical arrival timestamp at the end of the corresponding route section is extracted. This timestamp is also accurate to the second and maintains the same format as the real-time arrival timestamp.

[0073] Step S1233: Convert the real-time arrival timestamp information and the historical arrival timestamp information into time values ​​in the same time coordinate system.

[0074] Both real-time and historical arrival timestamps contain time elements such as year, month, day, hour, minute, and second. To calculate the time difference, they must be converted to time values ​​in the same time coordinate system. This conversion method converts the timestamp information into the total number of seconds starting at 00:00:00 on the current day.

[0075] For example, if the real-time arrival timestamp is 8:10:05 AM on a weekday, the converted time value is the total number of seconds from midnight to that moment. This is calculated by multiplying eight hours by 3,600 seconds per hour, adding ten minutes by 60 seconds per minute, and adding five seconds. Historical arrival timestamps are converted similarly to obtain the corresponding total number of seconds. This conversion places the real-time and historical arrival times in the same time coordinate system, facilitating subsequent difference calculations.

[0076] Step S1234: Calculate the difference between the real-time arrival time value and the historical arrival time value to obtain the time deviation value of a single line section.

[0077] In the same time coordinate system, the time deviation value for a single line section is obtained by subtracting the historical arrival time value from the real-time arrival time value. If the deviation value is a positive number, it means that the real-time train arrives at the end of the line section later than the historical benchmark time, indicating a delay. If the deviation value is a negative number, it means that the real-time train arrives earlier than the historical benchmark time, indicating an advance. If the deviation value is zero or close to zero, the real-time arrival time is basically consistent with the historical benchmark time.

[0078] For each line section, the corresponding time deviation value is calculated according to the above method. These time deviation values ​​respectively reflect the time deviation of the train in different sections.

[0079] Step S1235: performing sequence analysis on the time deviation values ​​of the same train in consecutive line sections, identifying the changing trend of the time deviation values, and calculating the trend change rate parameter.

[0080] The time deviation values ​​for the same train on consecutive route sections are arranged according to the order of their travel, forming a time deviation sequence. This sequence is visualized, plotting a curve showing the deviation values ​​as they change over the route sections. Trends can be identified by observing the curve's direction. For example, a continuously rising curve indicates increasing train delays; a continuously declining curve indicates increasing train early arrivals; and fluctuations without a clear pattern indicate unstable time deviations.

[0081] To quantify these changing trends, a trend change rate parameter is calculated. Specifically, a continuous line section is divided into multiple adjacent section pairs. For each section pair, the difference between the time deviation value of the latter section and the time deviation value of the previous section is calculated. This difference reflects the change in the deviation value. The average of these changes is then used as the core indicator of the trend change rate parameter. The standard deviation of the changes is also calculated to reflect the stability of the trend. The combination of the average and standard deviation forms the trend change rate parameter, which comprehensively reflects the changing trend and fluctuation of the time deviation value.

[0082] Step S1236: The time deviation value and trend change rate parameter of a single line section are combined to generate a time dimension deviation feature, which includes an interval time deviation sequence and a deviation trend change sequence.

[0083] The time deviation values ​​of all individual line sections are arranged in the order of the line sections that the train travels to form an interval time deviation sequence. Each element in the interval time deviation sequence corresponds to the time deviation value of a line section, which can intuitively show the time deviation distribution of the train on the entire route.

[0084] At the same time, the trend change rate parameters are arranged in order according to the corresponding interval pairs to form a deviation trend change sequence. For example, for n line sections, trend change rate parameters for n-1 interval pairs can be formed. These parameters are arranged in the order of the interval pairs to form a deviation trend change sequence, which reflects the changing trend of time deviation between different intervals.

[0085] The interval time deviation sequence and the deviation trend change sequence are combined, that is, the two sequences are spliced ​​together in order to form the time dimension deviation feature, which comprehensively reflects the deviation of the train in the time dimension and its changing trend.

[0086] Step S124: compare the real-time line occupancy status data with the line capacity constraint rule in the rule constraint feature, and identify the deviation value between the current line section occupancy rate and the maximum occupancy rate allowed by the rule as the spatial dimension deviation feature.

[0087] First, occupancy information for each section is extracted from real-time line occupancy status data, including the number of trains currently operating in the section, the start time of each train occupying the section, and the expected departure time. Based on this information, the occupancy rate of each section at the current moment is calculated. This is done by taking the ratio of the total length of trains currently occupying the section to the total length of the section, and then converting this into the occupancy ratio per unit time, taking into account the train speed and section length. For example, if the total length of a section is a certain value, and there are two trains currently occupying the section, the ratio of their total length to the total length of the section is a certain ratio. This ratio, combined with the average time it takes for trains to pass through the section, is used to calculate the proportion of the section occupied per unit time, i.e., the current section occupancy rate.

[0088] The maximum occupancy rate allowed for the corresponding line section in the line capacity constraint rule is extracted from the rule constraint features. This maximum occupancy rate is determined based on factors such as the line design standards, the capabilities of the signal system, and safety redundancy. Different line sections and different time periods may have different values. For example, in order to improve transportation efficiency during peak hours, the maximum occupancy rate may be set higher, while it may be relatively lower during off-peak hours.

[0089] Compare the current route section occupancy rate with the maximum allowed occupancy rate and calculate the difference between the two: the current route section occupancy rate minus the maximum allowed occupancy rate. If the difference is positive, it indicates that the current occupancy rate exceeds the maximum allowed by the rules, indicating an over-occupancy deviation. If the difference is negative, it indicates that the current occupancy rate is within the allowed range and has some excess space. If the difference is zero, it indicates that the current occupancy rate has exactly reached the maximum allowed. Arrange the deviation values ​​for all route sections in order of route sections to form a spatial dimension deviation feature.

[0090] Step S125: Match the real-time station passenger flow data with the historical passenger flow data in the benchmark operation characteristics, combine the resource allocation information in the real-time scheduling instruction execution data, and calculate the adaptation deviation value between the current resource configuration and passenger flow demand as the resource dimension deviation feature.

[0091] First, real-time station passenger flow data is segmented into time periods, dividing a day into multiple, consecutive, one-hour periods. The number of passengers entering, exiting, and averaging the number of passengers stranded within each period is calculated to generate real-time passenger flow period data. From the historical passenger flow data in the baseline operating characteristics, historical passenger flow data with the same characteristics as the current scenario (operating period, weather conditions, and holiday attributes) is found. Using the same time period segmentation, the historical number of passengers entering, exiting, and averaging the number of passengers stranded within each period is calculated to generate historical passenger flow period data.

[0092] Match the real-time passenger flow period data with the historical passenger flow period data, that is, compare the passenger flow data of the same period, calculate the ratio of the real-time number of people entering the station to the historical number of people entering the station, the ratio of the real-time number of people leaving the station to the historical number of people leaving the station, and the ratio of the real-time average number of people stranded to the historical average number of people stranded. These ratios reflect the degree of difference between the current passenger flow and the historical benchmark passenger flow, and constitute the passenger flow demand characteristics.

[0093] Resource allocation information is extracted from real-time dispatch instruction execution data, including the number of staff members at each station in each time period, the number of open gates, the number of escalators in operation, the number of platforms the train stops at the station, and the duration of the stops. This resource allocation information is compared with historical baseline resource allocation information (extracted from the baseline operation characteristics). The ratio of the current resource allocation to the historical baseline resource allocation is calculated, such as the ratio of the current number of staff members to the historical baseline number of staff members, and the ratio of the current number of open gates to the historical baseline number of open gates, etc., to form the resource allocation characteristics.

[0094] To calculate the adaptation deviation between the current resource configuration and passenger flow demand, each ratio in the resource configuration feature is compared with the corresponding ratio in the passenger flow demand feature. The difference between each corresponding item is calculated: the resource configuration ratio minus the passenger flow demand ratio. For example, the adaptation deviation between the staff configuration and the incoming passenger flow demand is calculated by subtracting the real-time number of people entering the station from the current staff ratio; the adaptation deviation between the gate configuration and the outgoing passenger flow demand is calculated by subtracting the real-time number of people leaving the station from the current number of gates open. All these adaptation deviation values ​​are combined in a set order to form the resource dimension deviation feature.

[0095] Step S126: Correlate and integrate the time dimension deviation features, space dimension deviation features, and resource dimension deviation features according to preset dimension weights to generate a baseline deviation feature set.

[0096] Preset dimension weights are determined based on the actual needs of rail transit scheduling and historical experience, and are set through a combination of expert evaluation and data analysis. Time dimension deviation characteristics directly impact train punctuality and operating efficiency, and are weighted higher during peak hours. Spatial dimension deviation characteristics are related to line safety and capacity utilization, and are weighted relatively higher when line loads are high. Resource dimension deviation characteristics affect passenger flow management and passenger experience, and are weighted higher at stations with high passenger flow.

[0097] The correlation and integration process first converts the deviation features of the time dimension, space dimension, and resource dimension into standardized feature vectors, where each element of the feature vector is a dimensionless ratio or deviation value. Next, each element in each feature vector is multiplied by the corresponding dimension weight to obtain a weighted feature vector. Finally, the three weighted feature vectors are concatenated in the order of time, space, and resource dimensions to form a baseline deviation feature set that contains the deviation information of the three dimensions and their weighted impact.

[0098] Step S130: calling the pre-trained scheduling collaborative evaluation model to perform dynamic correlation analysis on the baseline deviation feature set to generate a scheduling collaborative evaluation result, which includes deviation conduction path parameters, collaborative conflict probability parameters, and resource adaptation parameters.

[0099] The Dispatching Collaborative Evaluation Model is a deep learning-based neural network model specifically designed to analyze the correlations between various deviation features in rail transit scheduling. Trained with a large amount of historical baseline deviation feature data and corresponding dispatch evaluation results, the Dispatching Collaborative Evaluation Model automatically learns the hidden correlations and transmission patterns between deviation features. When the model is called, it takes the baseline deviation feature set as input. The Dispatching Collaborative Evaluation Model uses its internal multi-layer neural network to perform feature extraction and correlation analysis, outputting a dispatching collaborative evaluation result that includes deviation transmission path parameters, collaborative conflict probability parameters, and resource adaptability parameters.

[0100] Step S131: Input the baseline deviation feature set into the feature preprocessing layer of the scheduling collaborative evaluation model, standardize the time dimension deviation feature, space dimension deviation feature, and resource dimension deviation feature, and generate a standard deviation feature vector.

[0101] The baseline deviation feature set first enters the feature preprocessing layer of the scheduling collaborative evaluation model. This layer's primary function is to normalize the input features to eliminate dimensional differences and numerical range differences between features of different dimensions, thereby improving the model's analytical accuracy. For each element in the time dimension deviation feature, a mean-variance normalization method is applied. This method subtracts the mean of the time dimension deviation feature from each element and then divides it by its standard deviation.

[0102] Similarly, each element in the spatial and resource dimension deviation features is normalized by subtracting their respective mean and dividing by their respective standard deviations. After normalization, the deviation features in all three dimensions are converted into standard feature vectors with a mean of zero and a standard deviation of one. These three standard feature vectors are concatenated in the order in which they were input to form a unified standard deviation feature vector, which serves as the input for the next layer of the model.

[0103] Step S132: By scheduling the graph structure construction module of the collaborative evaluation model, the deviation association graph structure is constructed with the standard deviation feature vector as the node and the correlation strength between the deviation features of each dimension as the edge weight.

[0104] After the standard deviation feature vector enters the graph structure construction module, the graph structure construction module first regards each element in the standard deviation feature vector as an independent node. Each node represents a specific deviation feature element, such as the time deviation value of a certain line section, the resource adaptation deviation value of a certain station, etc.

[0105] Next, the strength of the association between any two nodes is calculated as the weight of the edge connecting them. This strength is calculated based on the co-occurrence frequency and similarity of the change trends of the deviation feature elements represented by the two nodes in the historical data. Specifically, the co-occurrence frequency is calculated as the proportion of the total number of records in which two deviation feature elements simultaneously exhibit non-zero deviations. The correlation coefficient between the change curves of the two deviation feature elements in the historical data is calculated as the similarity of the change trends. The co-occurrence frequency and the change trend similarity are weighted and summed according to a set ratio to obtain the strength of the association between the two nodes, i.e., the edge weight.

[0106] Based on the node and edge weights, an undirected deviation association graph structure is constructed. Each node in the graph corresponds to a deviation feature element, and the weight of each edge corresponds to the strength of the association between the two deviation feature elements. The above deviation association graph structure can intuitively show the mutual connection between different deviation feature elements. For example, the element of train arrival delay in the time dimension deviation feature may have a strong correlation with the element of line section congestion in the space dimension deviation feature. In this case, there will be a high-weight edge between the nodes representing these two elements in the graph; and the element of insufficient station staff in the resource dimension deviation feature may be closely related to the element of extended train stop time caused by slow passengers getting on and off in the time dimension deviation feature. Accordingly, the edge weight between them will also be high.

[0107] Step S133: Use the path mining algorithm of the scheduling collaborative evaluation model to perform path search processing on the deviation association graph structure, identify the key conduction path from the initial deviation node to the influencing node, and extract the path length parameter, node influence intensity parameter and path conduction efficiency parameter as the deviation conduction path parameter.

[0108] The path mining algorithm in the scheduling collaborative evaluation model is a tool specifically designed to identify deviation transmission paths within the deviation association graph structure. Initial deviation nodes are those whose deviation values ​​exceed a preset deviation threshold; these nodes are the starting points of deviation transmission. Influencing nodes are those that have a significant impact on the overall operational efficiency or safety of the rail transit system, such as passenger flow nodes at key transfer stations and section occupancy nodes on trunk lines. The path mining algorithm, starting from the initial deviation node, explores all possible paths that can reach the influencing node. It then selects key transmission paths and extracts relevant parameters.

[0109] Step S1331: Identify all initial deviation nodes from the deviation association graph structure. The initial deviation nodes are nodes whose deviation values ​​in the time dimension deviation feature, space dimension deviation feature, or resource dimension deviation feature exceed a preset deviation threshold.

[0110] The preset deviation threshold is determined based on the operating standards and historical data of the rail transit system. Different types of deviation feature nodes will have different preset deviation thresholds. For example, for the train arrival time deviation node in the time dimension deviation feature, the preset deviation threshold may be set to a time length. When the difference between the actual train arrival time and the benchmark arrival time exceeds this length, the node is identified as an initial deviation node. For the line section occupancy deviation node in the spatial dimension deviation feature, the preset deviation threshold may be a percentage. When the difference between the actual occupancy and the benchmark occupancy exceeds this percentage, the node becomes an initial deviation node. By comparing the deviation value of each deviation feature node with the corresponding preset deviation threshold, all initial deviation nodes are screened out.

[0111] Step S1332: Taking each initial deviation node as a starting point, a depth-first search algorithm is used to perform path search in the deviation association graph structure, and the conduction paths from the initial deviation node to all other nodes are recorded.

[0112] The depth-first search algorithm is an algorithm that traverses the nodes of a graph in the depth-first order. It starts from the initial deviation node and searches as deeply as possible along a path until it can no longer go forward. It then backtracks to the previous node and selects another unexplored path to continue searching. During the search process, every conductive path that starts from the initial deviation node and can reach all other nodes is recorded in detail, including each node passed by the path and the order of connections between nodes. For example, starting from the initial deviation node "Train A is delayed in section X", there may be a path that is "Train A is delayed in section X → the subsequent train B in section X is forced to slow down → Train B arrives at station Y late → passengers at station Y are stranded → transfer to train C at station Y is delayed", and the algorithm will record the above path in full.

[0113] Step S1333: Calculate the sum of the edge weights of each conduction path, and use the sum of the edge weights as a path influence strength parameter. A larger path influence strength parameter indicates a more significant conduction influence of the conduction path.

[0114] Each transmission path consists of a series of nodes and edges. The edge weight reflects the strength of the connection between two adjacent nodes, that is, the degree of influence of the bias transmitted from one node to another. For each recorded transmission path, the weights of all edges along the path are added together. The resulting sum is the path influence strength parameter for that path. The larger the value of this parameter, the more significant the influence of the bias transmission along this path. For example, a large sum of edge weights along a path means that the bias transmission along this path will have a greater impact on subsequent nodes.

[0115] Step S1334: Count the number of nodes included in each conductive path, and use the number of nodes as a path length parameter.

[0116] The path length parameter (PLP) is a measure of the length of a conduction path. It is determined by counting the number of nodes in each conduction path. The greater the number of nodes, the larger the PLP, indicating that the deviation must pass through more nodes to complete the conduction. For example, a conduction path with 5 nodes has a PLP of 5, while a conduction path with 3 nodes has a PLP of 3.

[0117] Step S1335: Calculate the ratio of the path influence intensity parameter to the path length parameter, and use the ratio as the path conduction efficiency parameter. A higher path conduction efficiency parameter indicates a higher influence transfer efficiency per unit path length.

[0118] The path conduction efficiency parameter reflects the efficiency of deviation transmission along the conduction path. To calculate this, the path influence strength parameter for each conduction path is divided by the path length parameter for that path. If a path has a larger path influence strength parameter and a smaller path length parameter, its path conduction efficiency parameter will be higher, indicating a stronger deviation transmission per unit path length. For example, if a path influence strength parameter is a certain value and a path length parameter is 3, its path conduction efficiency parameter will be that value divided by 3. If another path has the same path influence strength parameter and a path length parameter of 5, its path conduction efficiency parameter will be that value divided by 5. Clearly, the former is more efficient.

[0119] Step S1336: sorting all conduction paths from high to low according to the path influence strength parameters, and selecting a preset number of conduction paths before sorting as key conduction paths.

[0120] The preset number of paths is determined based on the complexity of the rail transit system and the analysis requirements. This ensures that the selected critical transmission paths include those deviation transmission paths with the most significant impact on the system. After all transmission paths are ranked according to the path impact strength parameter, the top-ranked transmission paths have a more significant impact on the system. The preset number of transmission paths before ranking are selected and designated as critical transmission paths, which serve as the focus of subsequent analysis and optimization.

[0121] Step S1337: extracting the path length parameter, node influence strength parameter, and path conduction efficiency parameter of each path from the key conduction path, and combining them to generate the deviation conduction path parameter.

[0122] For each key conduction path, its path length parameter, node influence strength parameter (i.e., path influence strength parameter) and path conduction efficiency parameter are extracted respectively. Then, these parameters are combined in a set order to form a deviation conduction path parameter containing the parameters of multiple key conduction paths. It comprehensively reflects the information such as the length, influence strength and conduction efficiency of the key deviation conduction path.

[0123] Step S134: Calculate the cumulative impact value of the deviation characteristics on each key conduction path based on the deviation conduction path parameters, input the cumulative impact value into the coordination conflict probability calculation module, and generate the coordination conflict probability parameter in combination with the rule constraint characteristics of the preset scheduling rule library.

[0124] First, for each key transmission path, the cumulative impact value of the deviation characteristics on the path is calculated based on the deviation value of each node on the path and the weight of the edge between the nodes. During the calculation process, the deviation value of each node will be transferred and accumulated according to the weight of the edge between it and the next node, and finally the cumulative impact value of the entire path is obtained. Then, these cumulative impact values ​​are input into the collaborative conflict probability calculation module. The collaborative conflict probability calculation module will combine the rule constraint characteristics in the preset scheduling rule library to analyze the conflicts between different scheduling links that may be caused by these cumulative impacts. For example, if the cumulative impact value of a key transmission path is large, it may cause a conflict between the train departure plan and the line capacity. The module will calculate the probability of the above conflict based on historical data and rule constraints. The collection of these probability values ​​is the collaborative conflict probability parameter.

[0125] Step S135: Call the resource adaptability analysis module of the scheduling collaborative evaluation model, match and analyze the resource dimension deviation characteristics with the resource configuration information in the real-time scheduling data, and calculate the matching ratio parameter between the resource supply and the deviation repair demand as the resource adaptability parameter.

[0126] The resource adaptability analysis module is specifically designed to analyze the degree of compatibility between resource configuration and deviation remediation requirements. Resource dimension deviation characteristics reflect the deviation between the current resource configuration and the baseline resource configuration. Resource configuration information in real-time scheduling data includes the number of trains, station staff, platform space, and security equipment. The resource dimension deviation characteristics are matched and analyzed with this resource configuration information to determine the demand for each resource type required to remediate the deviation. This is then compared with the current actual resource supply and the ratio between the supply and demand of each resource is calculated. The collection of these ratio parameters is the resource adaptability parameter. The closer the ratio is to 1, the higher the resource adaptability.

[0127] Step S136: Integrate the deviation conduction path parameter, the coordination conflict probability parameter, and the resource adaptability parameter to generate a scheduling coordination evaluation result.

[0128] The deviation transmission path parameter obtained in step S1337, the coordination conflict probability parameter obtained in step S134, and the resource adaptation parameter obtained in step S135 are integrated and combined according to the specified format to form a complete scheduling coordination evaluation result. This scheduling coordination evaluation result comprehensively reflects the deviation transmission path information, the possible coordination conflict probability, and the resource adaptation status.

[0129] Step S140: Construct a multi-objective scheduling optimization model based on the scheduling collaborative evaluation results, and perform collaborative optimization processing on the deviation conduction path parameters, collaborative conflict probability parameters, and resource adaptability parameters through the multi-objective scheduling optimization model to generate an initial scheduling adjustment plan set.

[0130] The multi-objective scheduling optimization model is a mathematical model that can handle multiple optimization objectives at the same time. It is based on the scheduling collaborative evaluation results, takes the deviation conduction path parameters, collaborative conflict probability parameters and resource adaptation parameters as optimization objects, sets corresponding optimization objectives and constraints, and obtains multiple possible scheduling adjustment schemes through solution to form an initial scheduling adjustment scheme set.

[0131] Step S141: Using the deviation conduction path parameters, collaborative conflict probability parameters and resource adaptability parameters in the scheduling collaborative evaluation results as optimization targets, construct an objective function of the multi-objective scheduling optimization model. The objective function includes the deviation conduction path shortening target, the collaborative conflict probability reduction target and the resource adaptability improvement target.

[0132] The objective function defines the areas to be optimized. The goal of shortening the deviation transmission path aims to shorten the critical transmission path and reduce the scope of the deviation by adjusting the scheduling plan. The goal of reducing the probability of coordinated conflicts aims to reduce the probability of conflicts between different scheduling links and improve system stability. The goal of improving resource adaptability aims to improve the adaptability between resource supply and deviation correction requirements to ensure efficient resource utilization. These three objectives together constitute the objective function of the multi-objective scheduling optimization model.

[0133] Step S142: extracting line capacity constraint rules, train minimum interval rules and station passenger flow carrying rules from the preset scheduling rule library as constraint conditions of the multi-objective scheduling optimization model.

[0134] Constraints are restrictions that must be adhered to in a multi-objective scheduling optimization model to ensure that the optimized scheduling solution is feasible in actual operations. Line capacity constraints limit the maximum number of trains that can run on each line per unit time, preventing line overload; minimum train separation rules ensure sufficient safe distance and time interval between consecutive trains on the same line; and station passenger capacity rules ensure that the number of passengers within a station does not exceed its carrying capacity, avoiding safety accidents. Incorporating these rules as constraints into the model ensures that the optimization process proceeds within a reasonable range.

[0135] Step S143: Using the path length parameter and the node influence strength parameter in the deviation conduction path parameter as input variables of the deviation conduction path shortening target, and setting the path length shortening weight coefficient.

[0136] The path length parameter directly reflects the length of the conduction path, while the node impact strength parameter reflects the impact of deviation conduction on the path. Using these two parameters as input variables for the deviation conduction path shortening objective means that during the optimization process, we can prioritize shortening paths that are longer and have greater impact. The path length shortening weighting factor is set based on the impact of different paths on the overall system operation. Paths with greater impact are given higher weighting factors to ensure they receive more attention during the optimization process.

[0137] Step S144: using the collaborative conflict probability parameter as an input variable of the collaborative conflict probability reduction target, and setting a conflict probability reduction weight coefficient.

[0138] The coordination conflict probability parameter contains the probability values ​​of various possible scheduling conflicts. It serves as an input variable for the coordination conflict probability reduction objective, aiming to minimize these probabilities during the optimization process. The conflict probability reduction weighting factor is determined based on the severity of the conflict. Conflicts that could potentially lead to serious consequences, such as those associated with the risk of rear-end collisions, are given a higher weighting factor to prioritize reducing their probability.

[0139] Step S145: Using the resource adaptability parameter as an input variable of the resource adaptability improvement target, and setting the adaptability improvement weight coefficient.

[0140] Resource fit parameters reflect the matching ratio between supply and demand for various resources. They serve as input variables for resource fit improvement objectives, aiming to optimize these ratios by optimizing scheduling. Fit improvement weights are set based on resource importance. Critical resources, such as peak-hour train capacity, receive higher weights to prioritize their fit.

[0141] Step S146: Solve the objective function under the constraints through a multi-objective optimization algorithm to generate multiple sets of candidate optimization solutions including train timetable adjustment parameters, line resource allocation adjustment parameters, and station passenger flow management adjustment parameters.

[0142] The multi-objective optimization algorithm utilizes advanced algorithms such as the non-dominated sorting genetic algorithm to solve the objective function while satisfying constraints. By simulating the selection, crossover, and mutation processes of biological evolution, the algorithm continuously iterates and optimizes, generating multiple sets of candidate solutions. Each candidate solution includes specific parameters for adjusting the train timetable, such as adjusting train departure times and changing stop stations; parameters for adjusting line resource allocation, such as adjusting the number of trains allocated to different lines; and parameters for adjusting station passenger flow management, such as increasing the number of station staff and adjusting the accessibility of entrances and exits.

[0143] For example, step S1461: initialize the population parameters of the multi-objective optimization algorithm, set the population size, maximum number of iterations, and crossover mutation probability.

[0144] The setting of population parameters has a significant impact on the performance of multi-objective optimization algorithms. The population size refers to the number of candidate solutions initially generated. A population size that is too large may increase the computational effort, while a population size that is too small may result in an inadequate search. An appropriate population size should be set based on the complexity of the problem. The maximum number of iterations is the maximum number of steps the algorithm can run, ensuring that the algorithm completes the search within a reasonable time. The crossover probability refers to the probability that two parent individuals will produce offspring individuals during the algorithm's crossover operation; the mutation probability refers to the probability that an individual's genes will mutate. The setting of these two probabilities requires balancing the algorithm's exploration capability and convergence speed, and appropriate values ​​are usually determined through multiple experiments.

[0145] Step S1462: The train operation diagram adjustment parameters, line resource allocation adjustment parameters, and station passenger flow management adjustment parameters are used as optimization variables, and the optimization variables are encoded to generate initial population individuals.

[0146] Optimization variables are the parameters to be optimized by the multi-objective optimization algorithm. Train schedule adjustment parameters, line resource allocation adjustment parameters, and station passenger flow management adjustment parameters are converted into a coding format that the algorithm can process, such as binary or real number encoding. For example, the adjustment amount for train departure time can be encoded as a binary number of a set length, with each binary bit representing a different adjustment amount. This coding process generates an initial population of individuals, each of which corresponds to a set of possible optimization variable combinations.

[0147] Step S1463: Calculate the objective function value corresponding to each individual in the initial population. The objective function value includes the shortening value of the deviation conduction path, the reduction value of the probability of collaborative conflict, and the improvement value of resource adaptability.

[0148] For each individual in the initial population, its corresponding optimization variable is substituted into the objective function of the multi-objective scheduling optimization model. The shortened deviation conduction path (i.e., the shortened length compared to the original path); the reduced probability of collaborative conflict (i.e., the magnitude of the reduction in conflict probability); and the improved resource adaptability (i.e., the degree of improvement in resource adaptability) are calculated. These objective function values ​​reflect the performance of the scheduling solution corresponding to that individual in terms of the optimization objectives.

[0149] Step S1464: Perform non-dominated sorting on the individuals in the initial population based on the objective function value, determine the Pareto rank of each individual, and calculate the crowding distance of the individual.

[0150] Non-dominated sorting is a method used to distinguish the superiority of individuals in multi-objective optimization. An individual is said to dominate another individual if it is not inferior to the other individual on all objective functions and is superior to the other individual on at least one objective function. Using non-dominated sorting, individuals in the initial population are divided into different Pareto ranks, with individuals at lower ranks being superior. The crowding distance measures the density of individuals within the same Pareto rank. A larger distance indicates a sparser distribution of individuals in the solution space and greater diversity. To calculate the crowding distance, for each objective function, individuals within that rank are sorted according to the objective function value. The crowding distances of individuals at the two ends are set to infinity, while the crowding distance of the middle individual is the sum of the differences between the objective function values ​​of its two adjacent individuals.

[0151] Step S1465: Select excellent individuals to enter the next generation population based on the Pareto rank and crowding distance, and generate new population individuals through crossover and mutation operations.

[0152] During the selection process, individuals with low Pareto ranks are prioritized. Within the same rank, individuals with large crowding distances are selected to ensure the quality and diversity of the population. A crossover operation swaps the codes of two selected individuals according to a set method to generate a new individual. This is similar to a single-point crossover in binary coding, where a random intersection point is selected and the codes of the two individuals after the intersection are swapped. A mutation operation randomly changes the codes of an individual, such as a bit flip in binary coding, where a 0 is changed to a 1 or a 1 is changed to a 0. This increases population diversity and prevents the algorithm from falling into local optima.

[0153] Step S1466: Check the constraints of the new population individuals and remove individuals that do not meet the line capacity constraint rules, the minimum train interval rules, and the station passenger flow carrying rules.

[0154] Newly generated individuals may not meet the constraints and require verification. The dispatch plan corresponding to each new individual is compared with the line capacity constraints, the minimum train spacing rules, and the station passenger flow carrying capacity rules to check for any violations. For example, check whether the adjusted number of trains exceeds the line capacity limit, whether the intervals between trains are less than the minimum spacing requirements, and whether the station passenger flow exceeds the carrying capacity. Individuals that do not meet the constraints are eliminated, and only those that meet the requirements are retained for the next iteration.

[0155] Step S1467: Repeat the objective function value calculation, non-dominated sorting, selection, crossover mutation, and constraint checking steps until the maximum number of iterations is reached.

[0156] Following the above steps, the population is continuously updated and optimized, generating better individuals with each iteration. When the number of iterations reaches the preset maximum number of iterations, the algorithm stops running. At this point, the population contains multiple sets of candidate optimization solutions that are relatively good under the current conditions.

[0157] Step S1468: extract all non-dominated solutions from the final population as multiple groups of candidate optimization solutions, each group of candidate optimization solutions includes corresponding train timetable adjustment parameters, line resource allocation adjustment parameters, and station passenger flow management adjustment parameters.

[0158] The non-dominated solutions in the final population are those that are not dominated by other solutions on all objective functions. They constitute the Pareto optimal solution set. These non-dominated solutions are extracted from the final population. Each set of solutions corresponds to a specific set of parameters for adjusting the train schedule, line resource allocation, and station passenger flow management. Together, these parameters form a complete scheduling adjustment plan that can optimize rail transit system operations from different perspectives.

[0159] Train timetable adjustment parameters involve details such as train departure times, arrival times, and speeds within various sections. For example, to address congestion on a particular line during the morning rush hour, the departure times of some trains may be adjusted to create a more reasonable interval between trains, avoiding overcrowding or excessively long gaps. Furthermore, based on the actual conditions of the line, train speeds within certain sections may be fine-tuned to ensure that trains can travel more efficiently and safely through these sections.

[0160] Line resource allocation adjustment parameters primarily include the allocation of train numbers and track access rights. For example, on a line with high passenger volume, the number of trains during peak hours might be increased to improve capacity. On branch lines or lines with lower passenger volume, the number of trains might be reduced to avoid wasted resources. Furthermore, track access rights are rationally allocated based on the operating schedules of different trains to ensure that track conflicts do not occur between trains.

[0161] Parameters for adjusting passenger flow at stations include the placement of signage, staff deployment, and the number of gates opened. For example, at a station experiencing a sudden increase in passenger flow, temporary signage might be added to guide passengers in and out quickly. At the same time, additional staff might be deployed at key locations like platforms and stairwells to guide passengers and avoid congestion. The number of gates opened could be adjusted based on real-time passenger flow to expedite entry and exit.

[0162] Step S150: Perform dynamic constraint verification processing on the initial scheduling adjustment plan set, and generate a final rail transit scheduling instruction after verification. The final rail transit scheduling instruction is used to trigger the rail transit scheduling control system to perform a scheduling parameter update operation.

[0163] While the initial set of scheduling adjustment solutions performs well in the optimization algorithm, they still need to be verified against actual constraints to ensure their feasibility in actual operations. Through dynamic constraint verification, solutions that meet all constraints are screened out, and the optimal solution is selected to generate the final rail transit scheduling instructions, guiding the scheduling control system to update parameters.

[0164] Step S151: selecting an initial scheduling adjustment plan from the initial scheduling adjustment plan set, and extracting the train operation diagram adjustment parameters, line resource allocation adjustment parameters, and station passenger flow diversion adjustment parameters contained in the initial scheduling adjustment plan.

[0165] From the set of initial scheduling adjustment plans, the first one is selected in a predetermined order. This plan is then parsed to extract the train schedule adjustment parameters, line resource allocation adjustment parameters, and station passenger flow management adjustment parameters. These parameters serve as input to the subsequent verification process to verify that the plan meets the various constraints.

[0166] Step S152: Input the train diagram adjustment parameters into the line simulation module, simulate the adjusted train running path and stop time, and generate a line simulation result.

[0167] The Line Simulation Module is a computer-based simulation system that simulates train operation on track. After inputting the extracted train diagram adjustment parameters into the module, it constructs a virtual train operation scenario based on these parameters, simulating the entire process of a train departing from the starting station, traveling through various sections, and stopping at various stations according to the adjusted departure time, speed, and other parameters.

[0168] During the simulation, the system records the train's route in real time, including each track section the train passes through and the travel time in each section. It also records the train's stops at each station, including detailed information such as arrival and departure times. This information forms the line simulation results, visually demonstrating the adjusted train operation.

[0169] Step S153: Compare the line resource allocation adjustment parameters with the real-time line occupancy status data to verify whether the adjusted resource allocation meets the line capacity constraint rules and the train minimum interval rule.

[0170] Real-time line occupancy data reflects the current train distribution and track usage on the line. Line resource allocation adjustment parameters are compared with this real-time data to first check whether the adjusted number of trains is within the line's capacity constraints. This ensures that the line is not overloaded due to an excessive number of trains, thus ensuring safe operation.

[0171] Secondly, the adjusted intervals between trains are verified to ensure they comply with the minimum interval regulations. By comparing the operation plans of the preceding and following trains, the time interval between adjacent trains is calculated to ensure that the interval is no less than the prescribed minimum interval, thus preventing safety accidents such as collisions between trains.

[0172] Step S154: input the station passenger flow diversion adjustment parameters into the station passenger flow simulation module, simulate the passenger flow change trend after adjustment, and verify whether the passenger flow diversion effect meets the station passenger flow carrying rules.

[0173] The Station Passenger Flow Simulation module simulates passenger flow within a station based on input parameters. After inputting station flow adjustment parameters into the module, it constructs a virtual station scene based on these parameters, simulating passenger movements within the station, including entry, ticket purchase, waiting, ticket inspection, and exit.

[0174] During the simulation, the system tracks passenger flow trends in real time, including the number of passengers in different areas of the station, passenger flow speed, and passenger retention time. This data is used to analyze whether the adjusted passenger flow control measures can effectively alleviate the station's passenger pressure, ensure that the number of passengers in the station does not exceed the station's passenger capacity, and verify whether the passenger flow control measures meet the station's passenger capacity requirements.

[0175] Step S155: Calculate the comprehensive constraint satisfaction parameters of the initial scheduling adjustment plan based on the line simulation results, resource allocation verification results, and passenger flow diversion verification results.

[0176] The comprehensive constraint satisfaction parameter is an indicator that comprehensively reflects the degree to which the initial scheduling adjustment plan satisfies various constraints. To calculate this parameter, the route simulation results, resource allocation verification results, and passenger flow management verification results are first quantitatively scored.

[0177] For line simulation results, scores are given based on whether the trains can run smoothly according to the adjusted plan and whether delays occur; for resource allocation verification results, scores are given based on whether the number of trains meets capacity constraints and whether the train intervals meet minimum interval rules; for passenger flow management verification results, scores are given based on whether the number of passengers in the station is within the carrying capacity and whether the passenger flow is smooth.

[0178] Then, weights are assigned to each score based on the importance of each constraint. For example, line capacity constraints and minimum train separation rules, which are related to operational safety, may be given higher weights; while passenger flow management effectiveness may have a relatively lower weight. Finally, each score is multiplied by its corresponding weight and added together to obtain the comprehensive constraint satisfaction parameter for the initial scheduling adjustment plan.

[0179] Step S156: If the comprehensive constraint satisfaction parameter reaches a preset threshold, the initial scheduling adjustment plan is marked as a valid plan.

[0180] The preset threshold is determined based on the operational requirements and safety standards of the rail transit system and represents the minimum level of satisfaction required for a plan to be accepted. When the calculated comprehensive constraint satisfaction parameter reaches this threshold, it indicates that the initial scheduling adjustment plan meets the specified requirements for all constraints and can ensure the safe and efficient operation of the rail transit system. Therefore, it is marked as a valid plan.

[0181] Step S157: If the comprehensive constraint satisfaction parameter does not reach the preset threshold, the next initial scheduling adjustment plan is selected from the initial scheduling adjustment plan set and re-verified.

[0182] If the comprehensive constraint satisfaction parameter does not reach the preset threshold, it indicates that the initial scheduling adjustment plan does not meet the constraint conditions in some aspects, which may bring risks to the operation of the rail transit system or affect its efficiency. In this case, it is necessary to abandon the plan and select the next plan from the initial scheduling adjustment plan set. The process of steps S152 to S155 is repeated until a valid plan that meets the conditions is found.

[0183] Step S158: sort all valid solutions from high to low according to the comprehensive constraint satisfaction parameter, and select the valid solution ranked first as the final scheduling adjustment solution.

[0184] After verifying all the options in the initial set of scheduling adjustment options, all options marked as valid are collected. These options are then sorted from high to low based on their comprehensive constraint satisfaction parameters. The higher the comprehensive constraint satisfaction parameter, the better the option performs in satisfying the constraints and the more it can adapt to the actual operational needs of the rail transit system.

[0185] The first-ranked effective plan is selected as the final scheduling adjustment plan. The final scheduling adjustment plan has the best overall performance among all effective plans and can optimize the operating status of the rail transit system to the greatest extent.

[0186] Step S159: Convert the final scheduling adjustment plan into the final rail transit scheduling instruction.

[0187] The final dispatch adjustment plan includes a series of parameters and measures that need to be converted into an instruction format that the rail transit dispatch control system can recognize and execute. This conversion process requires refining and standardizing each parameter and measure in the plan. For example, train departure times can be converted into specific time instructions, line resource allocation plans can be converted into clear train dispatch instructions, and station passenger flow control measures can be converted into specific staff deployment instructions and equipment operation instructions.

[0188] The final rail transit dispatching instructions after conversion will be sent to the rail transit dispatching control system. Based on these instructions, the dispatching control system will make real-time adjustments to train operation, line resource allocation, passenger flow management at stations, etc., to ensure that the rail transit system can operate efficiently and safely according to the optimized plan.

[0189] Figure 2A schematic diagram illustrating exemplary hardware and software components of an artificial intelligence-based rail transit scheduling analysis system 100 that can implement the concepts of the present application, as provided in some embodiments of the present application, is shown. For example, a processor 120 can be used in the artificial intelligence-based rail transit scheduling analysis system 100 to perform the functions described in the present application.

[0190] For example, the artificial intelligence-based rail transit scheduling analysis system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the artificial intelligence-based rail transit scheduling analysis system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The artificial intelligence-based rail transit scheduling analysis system 100 also includes an I / O interface 150 between the computer and other input and output devices.

[0191] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the above-mentioned artificial intelligence-based rail transit scheduling analysis method is implemented.

[0192] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A rail transit scheduling analysis method based on artificial intelligence, characterized in that: The method comprises: Establishing a dynamic scheduling baseline for the rail transit system, which includes benchmark operating characteristics constructed from historical normal scheduling data, rule constraint characteristics from a preset scheduling rule library, and scenario-related characteristics from typical operating scenarios; Collecting real-time scheduling data of the rail transit system, performing a multi-dimensional comparison analysis on the real-time scheduling data and the dynamic scheduling baseline to generate a baseline deviation feature set, wherein the baseline deviation feature set includes a time dimension deviation feature, a space dimension deviation feature, and a resource dimension deviation feature; Calling a pre-trained scheduling collaborative evaluation model to perform dynamic correlation analysis on the baseline deviation feature set to generate a scheduling collaborative evaluation result, wherein the scheduling collaborative evaluation result includes a deviation conduction path parameter, a collaborative conflict probability parameter, and a resource adaptability parameter; Building a multi-objective scheduling optimization model based on the scheduling collaborative evaluation results, and using the multi-objective scheduling optimization model to collaboratively optimize the deviation conduction path parameters, the collaborative conflict probability parameters, and the resource adaptability parameters to generate an initial scheduling adjustment solution set; Dynamic constraint verification processing is performed on the initial scheduling adjustment plan set, and a final rail transit scheduling instruction is generated after verification. The final rail transit scheduling instruction is used to trigger the rail transit scheduling control system to perform a scheduling parameter update operation.

2. The rail transit scheduling analysis method based on artificial intelligence according to claim 1 is characterized in that: The establishment of a dynamic scheduling baseline for the rail transit system includes: Collecting historical normal dispatching data of the rail transit system, wherein the historical normal dispatching data includes historical train operation data, historical line status data, historical passenger flow data, and historical dispatching instruction data; Performing scene division processing on the historical normal scheduling data, dividing the historical normal scheduling data into multiple typical operation scene data units according to operation period characteristics, weather condition characteristics, and holiday attribute characteristics; Extract train punctuality characteristics, interval operation time characteristics, station dwell time characteristics, and resource occupancy rate characteristics from each typical operation scenario data unit, and combine the extracted features to generate benchmark operation characteristics; Obtaining a preset dispatching rule library, extracting line capacity constraint rules, train minimum interval rules, station passenger flow carrying rules, and emergency dispatching priority rules from the preset dispatching rule library, and converting the extracted rules into rule constraint features; Analyze the correlation between the benchmark operation characteristics and the rule constraint characteristics in each typical operation scenario data unit, calculate the matching parameters of the correlation relationship, and based on the matching parameters, fuse the benchmark operation characteristics, rule constraint characteristics and scenario identification characteristics of the typical operation scenario to generate a dynamic scheduling baseline.

3. The rail transit scheduling analysis method based on artificial intelligence according to claim 2 is characterized in that: The analysis of the correlation between the benchmark operation characteristics and the rule constraint characteristics in each typical operation scenario data unit, calculating the matching degree parameter of the correlation relationship, and fusing the benchmark operation characteristics, the rule constraint characteristics, and the scenario identification characteristics of the typical operation scenario based on the matching degree parameter to generate a dynamic scheduling baseline includes: Extract the feature vectors of the baseline operation characteristics and the feature vectors of the rule constraint characteristics from the typical operation scenario data unit, convert them into vector representations of the same dimension, calculate the cosine similarity of the baseline operation feature vector and the rule constraint feature vector, and use the cosine similarity value as the preliminary matching parameter; Analyze the correlation strength between the train punctuality characteristic in the benchmark operation characteristics and the minimum train interval rule in the rule constraint characteristics, and calculate the compliance parameter of the punctuality rate with the interval rule; Analyze the correlation strength between the resource occupancy rate characteristics in the benchmark operation characteristics and the line capacity constraint rules in the rule constraint characteristics, and calculate the compliance parameter of the resource occupancy rate with the capacity rule; The preliminary matching parameter, the compliance parameter of the punctuality rate to the interval rule, and the compliance parameter of the resource occupancy rate to the capacity rule are weighted and summed to generate a comprehensive matching parameter; If the comprehensive matching degree parameter is higher than the preset matching threshold, the benchmark operation characteristics, rule constraint characteristics and corresponding typical operation scenario identification characteristics are spliced ​​to generate the scenario baseline unit of the typical operation scenario; Classify and store the scenario baseline units of all typical operating scenarios according to scenario identification features to build a baseline library for dynamic scheduling baselines; Set a dynamic update cycle for the baseline library, recalculate the comprehensive matching parameters of each scene baseline unit in each update cycle, and replace the scene baseline units whose comprehensive matching parameters are lower than the preset threshold.

4. The rail transit scheduling analysis method based on artificial intelligence according to claim 1 is characterized in that: The collecting of real-time dispatching data of the rail transit system, performing a multi-dimensional comparison analysis on the real-time dispatching data and the dynamic dispatching baseline to generate a baseline deviation feature set includes: Collecting real-time dispatch data through the rail transit dispatch monitoring system, the real-time dispatch data including real-time train position data, real-time line occupancy status data, real-time station passenger flow data and real-time dispatch instruction execution data; Based on the timestamp information of the real-time scheduling data, matching the corresponding typical operation scenario data unit from the dynamic scheduling baseline to obtain the benchmark operation characteristics and rule constraint characteristics corresponding to the typical operation scenario data unit; The real-time train position data is aligned with the historical train position data in the benchmark operation characteristics, and the difference between the real-time arrival time and the benchmark arrival time in the same line section is calculated as the time dimension deviation feature; Compare the real-time line occupancy status data with the line capacity constraint rules in the rule constraint features, and identify the deviation between the current line section occupancy rate and the maximum occupancy rate allowed by the rules as the spatial dimension deviation feature; The real-time station passenger flow data is matched with the historical passenger flow data in the benchmark operation characteristics for a certain period of time. Combined with the resource allocation information in the real-time dispatch instruction execution data, the adaptation deviation value between the current resource configuration and passenger flow demand is calculated as the resource dimension deviation feature. The time dimension deviation features, space dimension deviation features and resource dimension deviation features are associated and integrated according to the preset dimension weights to generate a baseline deviation feature set.

5. The rail transit scheduling analysis method based on artificial intelligence according to claim 4 is characterized in that: The method of performing time axis alignment processing on the real-time train position data and the historical train position data in the benchmark operation feature, and calculating the difference between the real-time arrival time and the benchmark arrival time in the same line section as the time dimension deviation feature, includes: Extracting real-time train location data from real-time dispatch data, parsing the train identification information, line section identification information, and arrival timestamp information in the real-time train location data; Based on the train identification information and line section identification information, the corresponding historical train position data is matched from the benchmark operation characteristics to extract the historical arrival timestamp information; Convert the real-time arrival timestamp information and the historical arrival timestamp information into time values ​​in the same time coordinate system; Calculate the difference between the real-time arrival time and the historical arrival time to obtain the time deviation value of a single route section; Perform sequence analysis on the time deviation values ​​of the same train in consecutive line sections, identify the changing trend of the time deviation values, and calculate the trend change rate parameter; The time deviation value and trend change rate parameter of a single line section are combined to generate a time dimension deviation feature, which includes an interval time deviation sequence and a deviation trend change sequence.

6. The rail transit scheduling analysis method based on artificial intelligence according to claim 1 is characterized in that: The calling of the pre-trained scheduling collaborative evaluation model to perform dynamic correlation analysis on the baseline deviation feature set to generate a scheduling collaborative evaluation result includes: Input the baseline deviation feature set into the feature preprocessing layer of the scheduling collaborative evaluation model, perform standardization processing on the time dimension deviation feature, space dimension deviation feature and resource dimension deviation feature, and generate a standard deviation feature vector; Through the graph structure construction module of the scheduling collaborative evaluation model, the deviation correlation graph structure is constructed with the standard deviation feature vector as the node and the correlation strength between the deviation features of each dimension as the edge weight; Utilizing the path mining algorithm of the scheduling collaborative evaluation model to perform path search processing on the deviation association graph structure, identifying the key conduction path from the initial deviation node to the influencing node, and extracting the path length parameter, the node influence intensity parameter, and the path conduction efficiency parameter as the deviation conduction path parameters; Calculate the cumulative impact value of the deviation characteristics on each key conduction path based on the deviation conduction path parameters, input the cumulative impact value into the coordination conflict probability calculation module, and generate the coordination conflict probability parameter in combination with the rule constraint characteristics of the preset scheduling rule library; The resource adaptability analysis module of the scheduling collaborative evaluation model is called to match and analyze the resource dimension deviation characteristics with the resource configuration information in the real-time scheduling data. The matching ratio parameter between the resource supply and the deviation repair demand is calculated as the resource adaptability parameter. The deviation conduction path parameters, collaborative conflict probability parameters and resource adaptability parameters are integrated to generate scheduling collaborative evaluation results.

7. The rail transit scheduling analysis method based on artificial intelligence according to claim 6 is characterized in that: The path mining algorithm of the scheduling collaborative evaluation model is used to perform path search processing on the deviation association graph structure, identify the key conduction path from the initial deviation node to the influencing node, and extract the path length parameter, the node influence intensity parameter, and the path conduction efficiency parameter as the deviation conduction path parameters, including: Identify all initial deviation nodes from the deviation association graph structure, wherein the initial deviation nodes are nodes whose deviation values ​​in the time dimension deviation feature, the space dimension deviation feature, or the resource dimension deviation feature exceed a preset deviation threshold; Taking each initial deviation node as the starting point, a depth-first search algorithm is used to search for paths in the deviation association graph structure, and the conduction paths from the initial deviation node to all other nodes are recorded; Calculate the sum of the edge weights of each conduction path and use the sum of the edge weights as the path influence strength parameter. The larger the path influence strength parameter, the more significant the conduction influence of the conduction path. Count the number of nodes contained in each conduction path and use the number of nodes as the path length parameter; Calculate the ratio of the path influence intensity parameter to the path length parameter, and use this ratio as the path conduction efficiency parameter. The higher the path conduction efficiency parameter, the higher the influence transfer efficiency per unit path length. All conduction paths are sorted from high to low according to the path influence strength parameter, and a preset number of conduction paths before sorting are selected as key conduction paths; The path length parameter, node influence strength parameter and path conduction efficiency parameter of each path are extracted from the key conduction path and combined to generate the deviation conduction path parameter.

8. The rail transit scheduling analysis method based on artificial intelligence according to claim 1 is characterized in that: The multi-objective scheduling optimization model is constructed based on the scheduling collaborative evaluation result, and the deviation conduction path parameters, the collaborative conflict probability parameters, and the resource adaptability parameters are collaboratively optimized by the multi-objective scheduling optimization model to generate an initial scheduling adjustment solution set, including: The deviation transmission path parameters, coordination conflict probability parameters, and resource adaptability parameters in the scheduling coordination evaluation results are used as optimization targets to construct the objective function of the multi-objective scheduling optimization model. The objective function includes the objectives of shortening the deviation transmission path, reducing the coordination conflict probability, and improving the resource adaptability. Extract line capacity constraint rules, train minimum interval rules, and station passenger flow carrying rules from the preset scheduling rule library as constraints for the multi-objective scheduling optimization model; The path length parameter and node influence strength parameter in the deviation conduction path parameters are used as input variables of the deviation conduction path shortening target, and the path length shortening weight coefficient is set; The collaborative conflict probability parameter is used as the input variable of the collaborative conflict probability reduction target, and the conflict probability reduction weight coefficient is set; Use resource adaptability parameters as input variables for resource adaptability improvement targets and set adaptability improvement weight coefficients; The objective function is solved under constraints using a multi-objective optimization algorithm to generate multiple sets of candidate optimization solutions, including train timetable adjustment parameters, line resource allocation adjustment parameters, and station passenger flow management adjustment parameters. Perform non-dominated sorting on multiple groups of candidate optimization solutions, and select candidate optimization solutions that meet the optimization requirements of all objective functions as the initial scheduling adjustment solution set.

9. The rail transit scheduling analysis method based on artificial intelligence according to claim 1 is characterized in that: The dynamic constraint verification process is performed on the initial scheduling adjustment plan set, and the final rail transit scheduling instruction is generated after the verification, including: Select an initial scheduling adjustment plan from the initial scheduling adjustment plan set, and extract the train operation diagram adjustment parameters, line resource allocation adjustment parameters, and station passenger flow diversion adjustment parameters contained in the initial scheduling adjustment plan; Input the train operation diagram adjustment parameters into the line simulation module, simulate the adjusted train operation path and stop time, and generate the line simulation results; Compare the line resource allocation adjustment parameters with the real-time line occupancy status data to verify whether the adjusted resource allocation meets the line capacity constraint rules and the minimum train interval rules; Input the station passenger flow diversion adjustment parameters into the station passenger flow simulation module, simulate the passenger flow change trend after adjustment, and verify whether the passenger flow diversion effect meets the station passenger flow carrying rules; Based on the route simulation results, resource allocation verification results, and passenger flow management verification results, the comprehensive constraint satisfaction parameters of the initial scheduling adjustment plan are calculated; If the comprehensive constraint satisfaction parameter reaches the preset threshold, the initial scheduling adjustment plan is marked as a valid plan; If the comprehensive constraint satisfaction parameter does not reach the preset threshold, the next initial scheduling adjustment plan is selected from the initial scheduling adjustment plan set for re-verification; All valid plans are sorted from high to low according to the comprehensive constraint satisfaction parameter, and the first valid plan is selected as the final scheduling adjustment plan; The final dispatch adjustment plan is converted into the final rail transit dispatch instruction.

10. A rail transit dispatching analysis system based on artificial intelligence, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the artificial intelligence-based rail transit scheduling analysis method described in any one of claims 1 to 9.

Images