Rail transit intelligent scheduling management method and system

Abstract

The invention relates to the technical field of rail transit intelligence, and discloses a rail transit intelligent scheduling management method and system, and the method comprises the steps: collecting the entering and exiting data of passengers, the number of people in a waiting area, the train load factor and the platform congestion degree through an automatic fare collection system, a video monitor and a sensor of each station of rail transit; the collected passenger flow data are preprocessed, a box plot about passenger flow distribution is constructed, and sudden passenger flow fluctuation areas are identified in different time windows; a K-means clustering algorithm is adopted to classify passenger flow modes in peak periods, and the distribution type of passenger flow fluctuation is analyzed; a time sequence prediction model is constructed by adopting a Transform model in combination with weather, holidays and festivals and emergencies, the future short-term and medium-and-long-term passenger flow trend is predicted, and the train departure interval is optimized; and based on the predicted passenger flow distribution, a scheduling optimization objective function is constructed, and a reinforcement learning algorithm is combined. The method has the advantage of improving the passenger flow prediction precision in the peak period.

Description

Technical Field

[0001] This invention relates to the field of intelligent rail transit technology, specifically to intelligent scheduling and management methods and systems for rail transit. Background Technology

[0002] In modern urban rail transit systems, peak-hour passenger flow forecasting is a key factor in ensuring efficient system operation. Traditional dispatching systems typically rely on historical data, fixed timetables, or simple statistical models for passenger flow forecasting, but these methods often lack accuracy in dynamic and complex peak-hour environments. This is mainly reflected in the following aspects: First, historical data only reflects past travel patterns and cannot accurately predict sudden changes in passenger flow, such as the impact of holidays, special events, or weather changes. Second, traditional forecasting methods neglect the dynamic adjustment of real-time data and cannot react promptly to the current passenger flow status, resulting in delayed forecasts and an inability to effectively warn of excessively high or low passenger flows. Due to the lack of accurate forecasts, dispatching systems often cannot adjust train departure frequencies in a timely manner during peak hours, leading to platform congestion or excessively high train empty rates, thereby affecting passenger travel experience and increasing operating costs. Therefore, it is essential to design intelligent dispatching and management methods and systems for rail transit that improve the accuracy of peak-hour passenger flow forecasting. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an intelligent scheduling and management method and system for rail transit, which has the advantage of improving the accuracy of passenger flow prediction during peak hours and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving the accuracy of peak-hour passenger flow forecasting, this invention provides the following technical solution: an intelligent dispatching and management method for rail transit, comprising the following steps:

[0005] S1: Collect passenger entry and exit data, waiting area numbers, train occupancy rates, and platform congestion through automatic fare collection systems, video surveillance, and sensors at various rail transit stations, and combine this data with historical data to build basic data.

[0006] Preferably, S1 further includes real-time monitoring and analysis of passenger entry and exit, transfers, platform density, and train occupancy rate through automatic ticketing systems, video surveillance, sensors, and computer vision and deep learning technologies; and passenger flow data analysis, pattern recognition, and time-series prediction through time-series databases and edge computing devices, using statistical methods and machine learning algorithms.

[0007] S2: Preprocess the collected passenger flow data to construct a box plot of passenger flow distribution and identify areas of sudden passenger flow fluctuations within different time windows.

[0008] Preferably, step S2 further includes visualizing passenger flow distribution using box plots by calculating quartiles and interquartile ranges, identifying abnormal fluctuation areas, detecting and filtering out abnormal data and densely distributed abnormal areas using Z-score, sliding window, and DBSCAN algorithms, filling in missing data with high precision using linear interpolation, ARIMA model, or LSTM neural network, and using data normalization and standardization methods to analyze passenger flow fluctuations through moving average, CUSUM algorithm, and K-means clustering to identify sudden changes and peak areas.

[0009] S3: The K-means clustering algorithm is used to classify passenger flow patterns during peak hours, analyze the distribution types of passenger flow fluctuations, and match corresponding scheduling strategies based on different types.

[0010] Preferably, step S3 further includes evaluating the clustering effect under different K values ​​using the elbow method and silhouette coefficient, selecting the optimal K value, initializing cluster centers using the K-means++ algorithm, iteratively updating cluster centers and assigning data points to the most suitable clusters until clustering converges, identifying passenger flow patterns in different time periods by analyzing the average passenger flow, fluctuation range, and congestion level of each cluster, and optimizing train departure frequency and capacity, and adjusting train schedules and formations by adopting fixed-interval departures, dynamic additional trains, differentiated scheduling, and flexible formations based on passenger flow distribution characteristics.

[0011] Preferably, the average passenger flow for each cluster is obtained by calculating the mean of all data points in that cluster, using the following formula: In the formula, It is the average passenger flow of cluster k. It refers to all data points in cluster k. It is the number of data points in cluster k. These are data points; the range of fluctuation within each cluster can be represented by calculating the variance of the data points within that cluster, using the following formula: In the formula, represents the fluctuation range of cluster k; the degree of crowding is measured by the cohesion of the cluster, and the root mean square error within the cluster is used to represent the degree of crowding, as shown in the formula: In the formula, It represents the degree of crowding of cluster k; By analyzing the average passenger flow, fluctuation range, and congestion level of each cluster, passenger flow patterns in different time periods can be identified:

[0012] During peak hours, the cluster experiences high and fluctuating passenger flow, resulting in significant crowding.

[0013] During off-peak hours, the number of passengers in the cluster is lower and fluctuates less, resulting in less crowding.

[0014] S4: Using the Transformer model, combined with weather, holidays, and emergencies, a time-series forecasting model is constructed to predict future short-term and medium-to-long-term passenger flow trends and optimize train departure intervals.

[0015] Preferably, S4 further includes using the classic Encoder-Decoder architecture of the Transformer model for passenger flow prediction, combining historical data, external features and prediction results to optimize train departure intervals and scheduling strategies, dynamically adjusting departure frequencies through a combination of real-time monitoring and prediction, evaluating the model's generalization ability using cross-validation, and periodically updating external data and performing transfer learning.

[0016] S5: Based on the predicted passenger flow distribution, construct a scheduling optimization objective function, and combine reinforcement learning algorithms to dynamically adjust train departure intervals, increase or decrease capacity, or adjust train formations.

[0017] Preferably, step S5 further includes predicting future passenger flow distribution through a time series analysis model, dynamically adjusting train departure intervals, train formations, and increasing or decreasing capacity based on the prediction results, using reinforcement learning algorithms combined with Q-learning and deep learning, training the model to make optimal scheduling decisions based on real-time passenger flow information and the current state, maximizing the minimization of passenger waiting time and reducing train empty load rate, and simulating the impact of different scheduling strategies by constructing an environment model and reward function to optimize the train scheduling scheme.

[0018] The intelligent dispatching and management system for rail transit includes:

[0019] Data acquisition module: responsible for collecting various real-time data from the rail transit system and building a basic data model by combining historical data;

[0020] Data preprocessing module: Cleans, processes, and analyzes the collected data;

[0021] Passenger flow analysis module: Uses machine learning algorithms to analyze and classify passenger flow patterns during peak hours;

[0022] Time series forecasting module: Uses time series forecasting models to predict future passenger flow trends and optimizes train departure intervals based on the forecast results;

[0023] Scheduling optimization module: Combining reinforcement learning algorithms, the scheduling strategy is dynamically adjusted based on passenger flow prediction results to optimize train departure intervals, increase or decrease capacity, or adjust train formations.

[0024] Compared with the prior art, the present invention provides a method and system for intelligent scheduling and management of rail transit, which has the following beneficial effects:

[0025] This invention achieves efficient scheduling decisions by comprehensively collecting and analyzing passenger data from various stations, combined with advanced algorithms and models. Through precise passenger flow data preprocessing and pattern recognition, the system can promptly detect passenger flow fluctuations and respond effectively. The combination of K-means clustering and the Transformer time-series prediction model accurately predicts passenger flow trends at different times, ensuring flexible adjustments to train departure intervals and capacity configuration during peak hours, optimizing train formations, and improving transportation efficiency. The introduction of reinforcement learning algorithms enables real-time dynamic adjustments to scheduling strategies, continuously optimizing system operation based on actual passenger flow changes, minimizing passenger waiting time, improving system stability and safety, reducing resource waste, and enhancing overall transportation efficiency. The implementation of this method will significantly improve the service level and operational efficiency of the rail transit system, enhancing its ability to respond to emergencies and changes. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the method of the present invention;

[0027] Figure 2 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Example 1

[0030] like Figure 1 The intelligent scheduling and management method for rail transit described in this embodiment of the invention includes the following steps:

[0031] S1: Collect passenger entry and exit data, waiting area numbers, train occupancy rates, and platform congestion through automatic fare collection systems, video surveillance, and sensors at various rail transit stations, and combine this data with historical data to build basic data.

[0032] Collect passenger card swipe data for entry and exit, including entry and exit times and ticket types. Statistically analyze passenger flow at different times of day for each station, identifying peak periods. Combine passenger transfer data to analyze passenger flow direction and provide a reference for scheduling optimization. Obtain real-time passenger flow data through surveillance cameras on rail transit platforms and passageways. Use computer vision algorithms (object detection and tracking) to analyze platform passenger density, queuing conditions, and flow direction. Combine historical video data to identify abnormal passenger flow (such as sudden large-scale passenger influx). Deploy infrared, lidar, or pressure sensors on platforms and inside trains to monitor platform passenger density in real time. Install pressure sensors at train car doors to measure train occupancy rates and determine passenger congestion levels. Monitor the number of people in waiting areas using ultrasonic or lidar sensors to identify high-load conditions on platforms. Match AFC data, video surveillance data, and sensor data by timestamp and geographic location to ensure data synchronization. Use data interpolation or time window matching algorithms to supplement missing time-series data. Statistical methods (such as box plots) are employed to detect outlier data points, such as sudden increases or decreases in passenger flow. Historical and real-time data are compared to eliminate erroneous or duplicate records, improving data quality. A time-series database is used to store passenger flow data, with indexes categorized by time, station, and route to improve query efficiency. Long-term historical data is combined with short-term real-time data to form a complete passenger flow information database. Deep learning models (such as YOLO and ResNet) are used to detect passenger targets in video surveillance and calculate passenger flow. Optical flow analysis is used to predict passenger flow trends and identify platform congestion points. K-means clustering is used to cluster passenger flow data across different time periods, identifying passenger flow characteristics for different patterns, such as weekdays and holidays. Decision tree or random forest algorithms are used to analyze the impact of weather, holidays, and unforeseen events on passenger flow changes. Long Short-Term Memory (LSTM) networks or Transformer models are used for time-series passenger flow prediction, improving the accuracy of future passenger flow predictions. The adaptability of predictions is improved by incorporating influencing factors such as weather, holidays, and historical passenger flow patterns. Edge computing devices are deployed at the station to preprocess the collected data locally, such as filtering out abnormal data and extracting key features, thereby improving data processing efficiency. Only preprocessed key data is uploaded to the cloud to reduce network transmission latency. A cloud-based distributed storage architecture is adopted to store long-term passenger flow data, supporting real-time querying and analysis. Streaming computing frameworks (such as Apache Flink) are used to process large-scale passenger flow data, enabling real-time data updates and historical data storage and management.

[0033] By combining AFC (Automatic Facing), video surveillance, and sensors, comprehensive and multi-angle data collection is achieved, avoiding the limitations of single data sources. Spatiotemporal data fusion ensures data consistency and improves data integrity and accuracy. The integration of computer vision and deep learning enhances the automation of passenger flow identification and analysis, enabling platform congestion early warning. Pattern recognition allows for advance prediction of passenger flow trends, optimizing peak-hour scheduling plans. Deep learning time-series prediction models improve the accuracy of short-term and medium-to-long-term passenger flow forecasts. The impact of weather, holidays, and unforeseen events is considered to improve the adaptability of passenger flow forecasts and avoid scheduling imbalances caused by prediction errors. Based on predicted passenger flow changes, train departure intervals are dynamically adjusted and train formations are optimized to improve capacity matching. A smart dispatching system monitors passenger flow in real time and dynamically adjusts train scheduling, reducing passenger waiting time and improving the passenger travel experience. Precise scheduling reduces empty train operation, increases vehicle utilization, and lowers energy consumption and operating costs. Optimized passenger flow management reduces platform congestion and improves the overall operational efficiency of the rail transit system.

[0034] S2: Preprocess the collected passenger flow data to construct a box plot of passenger flow distribution and identify areas of sudden passenger flow fluctuations within different time windows.

[0035] Outlier detection is performed using statistical methods, calculating the quartiles (Q1, Q3) and interquartile range (IQR = Q3 - Q1) of passenger flow data. Thresholds are set (typically Q1 - 1.5 × IQR and Q3 + 1.5 × IQR); points exceeding these ranges are considered outliers. Box plots visualize the passenger flow data distribution, quickly identifying areas of abnormal fluctuation. A sliding window method is used to calculate average passenger flow over different time periods to determine if there are sudden increases or decreases. The Z-score method is used to calculate the deviation of each data point; if the Z-value exceeds a set threshold (e.g., the 3σ principle), it is considered an outlier. The DBSCAN (density-based clustering) method is combined to detect densely distributed abnormal passenger flow areas and eliminate isolated points. For small-scale missing data points, linear interpolation is performed based on passenger flow at adjacent time points to fill in the missing values. The average value of several time points before and after the missing point (e.g., a 3-hour moving average) is calculated to smoothly fill in the data. An autoregressive integral moving average (ARIMA) model is used to predict the passenger flow value for missing points. A Long Short-Term Memory (LSTM) neural network model is trained to infer missing data based on historical passenger flow trends, improving the accuracy of data imputation. The minimum and maximum values ​​of the data are calculated and transformed to the [0,1] range to improve the comparability of data across different time periods. The mean and standard deviation of the data are calculated, converting the data to a standard normal distribution to improve the robustness of the model. Median and interquartile range are used for normalization to reduce the impact of outliers and improve data stability. Passenger flow statistics (median and interquartile range) are calculated for different time windows (e.g., 10 minutes, 30 minutes, 1 hour). Extremely high (upper quartile) and extremely low (lower quartile) values ​​are identified to determine abnormal passenger flow fluctuation points. A moving average window method is used to calculate the rate of change of passenger flow over a short period to identify sudden fluctuation points. The Cumulative Sum (CUSUM) algorithm is used to detect sudden changes in passenger flow data and provide early warning of areas of surge in passenger flow. The K-means clustering method is used to group passenger flow data by time and space to identify concentrated areas of passenger flow fluctuation. By combining hotspot analysis, we can identify peak passenger flow areas and predict future passenger flow trends.

[0036] Outlier detection removes erroneous data caused by sensor errors and system malfunctions, improving data reliability. Intelligent imputation methods reduce analytical bias caused by missing data, enhancing data integrity. Normalization reduces data volume differences, improving comparability across different stations and time periods. Cluster analysis and box plot analysis improve the sensitivity of detecting sudden passenger flow fluctuations. Time series modeling accurately predicts future passenger peaks, optimizing train departure intervals. Combined with sudden passenger flow early warning, it improves the responsiveness of the dispatching system, reducing congestion and empty trains. Early warning of peak passenger flow allows for proactive passenger diversion, improving platform and carriage utilization efficiency. This enhances the stability of the rail transit system and reduces passenger waiting time caused by prediction errors.

[0037] S3: The K-means clustering algorithm is used to classify passenger flow patterns during peak hours, analyze the distribution types of passenger flow fluctuations, and match corresponding scheduling strategies based on different types.

[0038] Clustering performance under different K values ​​was evaluated using the elbow method and silhouette coefficient. The optimal K value was selected, and the K-means++ algorithm was used to initialize cluster centers. Cluster centers were iteratively updated, and data points were assigned to the most suitable clusters until clustering converged. By analyzing the average passenger flow, fluctuation range, and congestion level of each cluster, passenger flow patterns in different time periods were identified. Based on passenger flow distribution characteristics, fixed-interval departures, dynamic train additions, differentiated scheduling, and flexible train formations were adopted to optimize train departure frequency and capacity, and adjust train schedules and formations. The average passenger flow of each cluster was obtained by calculating the mean of all data points in that cluster, using the following formula: In the formula, It is the average passenger flow of cluster k. It refers to all data points in cluster k. It is the number of data points in cluster k. These are data points; the range of fluctuation within each cluster can be represented by calculating the variance of the data points within that cluster, using the following formula: In the formula, represents the fluctuation range of cluster k; the degree of crowding is measured by the cohesion of the cluster, and the root mean square error within the cluster is used to represent the degree of crowding, as shown in the formula: In the formula, It represents the degree of crowding of cluster k;

[0039] By analyzing the average passenger flow, fluctuation range, and congestion level of each cluster, passenger flow patterns in different time periods can be identified:

[0040] During peak hours, the number of passengers in the cluster is high and fluctuates greatly, resulting in a high degree of congestion.

[0041] During off-peak hours, the number of passengers in the cluster is lower and fluctuates less, resulting in less crowding.

[0042] Calculate the average passenger flow, fluctuation range, and congestion level for each cluster to analyze passenger flow distribution characteristics. Combine time information to identify different passenger flow patterns during morning peak, evening peak, and off-peak hours. Passenger flow remains consistently high during morning and evening peak hours, but fluctuations are small, and the passenger flow trend is relatively stable. This is suitable for fixed-interval departure patterns to maintain high capacity supply. Due to sudden events (weather, events, activities, etc.), passenger flow surges, leading to severe platform congestion in a short period. This is suitable for strategies such as dynamically adding trains and temporarily adjusting operating intervals. Station passenger flow is unevenly distributed, with some stations severely congested and others relatively empty. This is suitable for differentiated scheduling, such as adding express trains to specific stations or optimizing short-distance train formations. Using fixed timetable scheduling, setting stable departure intervals based on historical passenger flow data improves train occupancy rates. Using dynamic train addition strategies, combined with real-time passenger flow monitoring, train schedules are adjusted as needed. An emergency response mechanism is activated to increase train density for a short period during peak hours. Flexible formation schemes are adopted, adding larger train formations to high-load stations and reducing capacity at low-load stations. At stations with high passenger flow, priority will be given to direct express trains or local trains to reduce passenger congestion.

[0043] K-means clustering is used to classify different types of passenger flow patterns, improving the understanding of passenger flow changes during peak hours. Data standardization and feature optimization are employed to reduce noise and improve clustering results. Optimal scheduling schemes are matched based on different passenger flow types to improve the accuracy of train departures. Real-time scheduling adjustments are achieved through sudden passenger flow identification, improving system responsiveness. Intelligent scheduling optimizes passenger waiting times, reduces congestion, and improves train utilization efficiency. Combining express and local trains and dynamic scheduling enhances passenger flow management and reduces platform pressure during peak hours. Optimized scheduling reduces unnecessary train operations, lowering energy consumption. Increased train occupancy rates reduce empty runs, improving operational efficiency.

[0044] S4: Using the Transformer model, combined with weather, holidays, and emergencies, a time-series forecasting model is constructed to predict future short-term and medium-to-long-term passenger flow trends and optimize train departure intervals.

[0045] The classic Encoder-Decoder architecture of Transformer is adopted, where the Encoder processes the input time-series data (including historical passenger flow data and external features), and the Decoder generates the prediction results (future passenger flow predictions). The Encoder uses a self-attention mechanism to capture long-distance dependencies in the time series and leverages a multi-head attention mechanism to enhance the model's expressive power. The Decoder generates passenger flow predictions for future periods and processes the time-series information through positional encoding. MSE is used as the loss function to measure the error between the predicted and actual values ​​for model optimization. Batch gradient descent (such as the Adam optimizer) is used to train the model, and hyperparameters (such as learning rate and batch size) are tuned. The input time window size is optimized by adjusting the time window length according to the characteristics of passenger flow fluctuations to better capture short-term and medium-term passenger flow trends. The trained Transformer model is used to predict passenger flow in the short term (e.g., 1 hour, 2 hours) and medium to long term (e.g., 1 day, 3 days). Based on the forecast results, the passenger flow trends in different time periods are analyzed to predict future peak and off-peak periods, helping to optimize scheduling strategies. Based on the passenger flow forecast results, combined with the system's actual capacity and station capacity, train departure intervals are optimized. For short-term forecasts, flexible scheduling can be implemented based on sudden passenger flow fluctuations, such as reducing departure intervals for short periods to increase capacity. For medium- and long-term forecasts, train frequency and formations are adjusted to ensure train operation efficiency during peak hours and avoid excessive consumption of trains during off-peak hours. By combining real-time passenger flow monitoring results with forecast results, departure intervals are dynamically adjusted, and a dynamic scheduling system is used to improve the flexibility of train scheduling. During peak hours, train departure frequency can be increased to reduce passenger waiting time; during off-peak hours, departure frequency can be appropriately reduced to save operating costs. Cross-validation techniques are used to evaluate the model's generalization ability and prevent overfitting. Prediction accuracy is evaluated using metrics such as RMSE and MAE. Data on external factors (weather, holidays, emergencies) are regularly updated, and the model is retrained to adapt to constantly changing passenger flow patterns. By using transfer learning, the model is fine-tuned on a new dataset to adapt to changes in passenger flow patterns in different time periods and geographical areas.

[0046] The powerful time-series modeling capabilities of the Transformer model effectively capture long-term dependencies and complex nonlinear relationships, improving the accuracy of short-term and medium-to-long-term passenger flow forecasts. The introduction of external factors (weather, holidays, emergencies) makes the forecasts more flexible and adaptable to actual conditions, enhancing their comprehensiveness. Accurate passenger flow forecasts help the scheduling system identify peak and off-peak periods in advance, allowing for reasonable adjustments to train intervals. Dynamically adjusting train intervals not only improves the capacity utilization efficiency of the rail transit system but also avoids train congestion during peak hours, reducing passenger waiting times and improving passenger experience. Optimizing train intervals based on forecast results reduces empty train operations, saves energy, and lowers operating costs. Appropriately reducing train intervals during off-peak hours avoids resource waste and increases the overall operational efficiency of the system. Combining real-time forecasts with historical data effectively addresses the impact of various emergencies by dynamically adjusting train departure frequencies. Real-time scheduling optimization strategies enhance the system's emergency response capabilities, improving service quality and operational efficiency in special circumstances. By regularly training and updating the model, the Transformer model can continuously adapt to changes in passenger flow. Over time, the system will become increasingly accurate in predicting passenger flow trends and adapt to changes during holidays and emergencies.

[0047] S5: Based on the predicted passenger flow distribution, construct a scheduling optimization objective function, and combine reinforcement learning algorithms to dynamically adjust train departure intervals, increase or decrease capacity, or adjust train formations.

[0048] This system integrates multi-dimensional data such as historical passenger flow, weather information, holiday schedules, and special events. It predicts passenger flow distribution for different time periods using a time-series analysis model. Based on the prediction results, it obtains passenger flow distribution for different time periods (e.g., hourly, daily passenger flow). The predicted passenger flow results are used as input features for scheduling optimization. By dynamically adjusting train departure intervals, it minimizes the average waiting time for passengers. During off-peak hours, it reduces train departure frequency to avoid empty train operation and save resources. During peak hours, it increases train departure frequency or adds larger train formations to improve train capacity. Scheduling decisions balance train capacity and passenger demand to avoid system overcrowding or resource waste. This includes train departure intervals, train formations (e.g., short or long trains), and capacity adjustments. The maximum passenger capacity of each train is considered; exceeding this limit may cause overloading. During short peak periods, excessively short departure intervals may lead to operational risks, necessitating consideration of train scheduling safety. It ensures platform capacity to avoid safety hazards caused by passenger overloading. The maximum carrying capacity of each line in the system is also considered to prevent excessive congestion. By constructing an environmental model of the rail transit system, the impact of train scheduling decisions is simulated. The environmental model includes states (passenger flow distribution, train position, current departure interval, etc.), actions (train departure interval, train formation, capacity increase / decrease, etc.), and rewards (passenger waiting time, train empty rate, etc.). The state at each moment is composed of multi-dimensional features such as passenger flow, train capacity, and departure interval. Adjustments include changing departure intervals, increasing or decreasing the number of trains, and adjusting train formation. A table-based Q-learning algorithm maximizes long-term rewards by selecting different scheduling strategies (departure interval, capacity increase / decrease, formation adjustment). Q-learning is combined with deep learning, and a neural network is used to approximate the Q function to solve the large-scale state space and action space problem. The model is trained using a reinforcement learning algorithm to make appropriate scheduling decisions based on the current state under different passenger flow conditions. Through continuous iterative training, the system's scheduling strategy optimization under different scenarios is strengthened, gradually improving the overall system efficiency. The reinforcement learning model dynamically generates scheduling decisions based on real-time passenger flow predictions, the current state, and the strategies obtained during training. The train scheduling plan is adjusted in real time based on the actions output by the model (such as adjusting departure intervals, adding or removing trains, and train formations). During system operation, the actual situation (such as passenger waiting time and train empty load rate) is fed back in real time, continuously updating the status and guiding subsequent scheduling decisions. The model continuously optimizes itself based on new passenger flow data and scheduling feedback to ensure long-term stability. A reasonable reward function is designed, incorporating factors such as passenger waiting time, train empty load rate, and system resource utilization efficiency as criteria for evaluating the effectiveness of the scheduling strategy. The reward function should comprehensively consider multiple indicators to avoid the optimization of a single factor affecting the overall system efficiency. The scheduling strategy is validated by simulating different passenger flow scenarios (such as peak holiday periods and severe weather) to evaluate the model's adaptability and stability.Based on the actual scheduling results, the reward function and model parameters are optimized periodically to improve system performance.

[0049] Reinforcement learning-based scheduling optimization algorithms can flexibly adjust train departure intervals, add or remove trains, and adjust train formations based on real-time passenger flow predictions, ensuring sufficient capacity during peak hours and avoiding excessive resource waste during off-peak hours. Through intelligent scheduling, train departure frequencies and carriage formations are optimized, improving train occupancy rates while ensuring passenger comfort and avoiding empty trains and platform congestion. During peak periods such as holidays and special events, the system can intelligently increase capacity to balance passenger demand, avoiding excessive empty trains and energy waste. By reducing passenger waiting time, train punctuality and comfort are improved, significantly enhancing the passenger travel experience. The system can flexibly respond to emergencies, such as sudden increases in passenger flow or weather changes, by promptly adjusting departure intervals and capacity allocation to avoid overcrowding during peak periods. Through precise capacity adjustments and train formation optimization, unnecessary train operations and empty trains are reduced, significantly decreasing energy consumption and labor costs. This improves the overall operational efficiency of the rail transit system and reduces operational waste. The dynamic scheduling model based on reinforcement learning can self-optimize through continuous training and feedback, gradually adapting to different passenger flow patterns and external changes over time. The system can automatically adjust its strategies to better cope with ever-changing passenger flow demands, improving scheduling efficiency and accuracy.

[0050] Example 2

[0051] like Figure 2 The intelligent dispatching and management system for rail transit described in this embodiment of the invention includes:

[0052] Data acquisition module: responsible for collecting various real-time data from the rail transit system and building a basic data model by combining historical data;

[0053] Data preprocessing module: Cleans, processes, and analyzes the collected data;

[0054] Passenger flow analysis module: Uses machine learning algorithms to analyze and classify passenger flow patterns during peak hours;

[0055] Time series forecasting module: Uses time series forecasting models to predict future passenger flow trends and optimizes train departure intervals based on the forecast results;

[0056] Scheduling optimization module: Combining reinforcement learning algorithms, the scheduling strategy is dynamically adjusted based on passenger flow prediction results to optimize train departure intervals, increase or decrease capacity, or adjust train formations.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent dispatching management of rail transit, characterized in that, Comprise the following steps: S1: Collect passenger entry and exit data, waiting area passenger number, train full load rate and platform congestion degree through automatic fare collection system, video monitoring and sensors of rail transit stations, and combine historical data to build basic data; S2: Preprocess the collected passenger flow data, construct a box plot about passenger flow distribution, and identify the sudden passenger flow fluctuation area in different time windows; S3: Classify the peak period passenger flow mode using K-means clustering algorithm, analyze the distribution type of passenger flow fluctuation, and match the corresponding scheduling strategy according to different types; S4: Use the Transformer model to combine weather, holidays, and unexpected events to build a time series prediction model to predict short-term and medium-term passenger flow trends and optimize train departure interval; S5: Based on the predicted passenger flow distribution, build a scheduling optimization objective function, combine reinforcement learning algorithm, and dynamically adjust train departure interval, increase or decrease transport capacity, or adjust train formation. 2.The rail transit intelligent dispatching management method according to claim 1, characterized in that, The S1 further comprises real-time monitoring and analyzing passenger entry and exit, transfer, platform density, and train full load rate through automatic fare collection system, video monitoring, sensors, computer vision and deep learning technology, applying statistical methods and machine learning algorithms for passenger flow data analysis, pattern recognition and time series prediction through time series database and edge computing devices. 3.The rail transit intelligent dispatching management method of claim 1, wherein, The S2 further comprises calculating quartiles and interquartile ranges, visualizing passenger flow distribution using box plots, identifying abnormal fluctuation areas, detecting and filtering out abnormal data and densely distributed abnormal areas using Z-score method, sliding window method and DBSCAN algorithm, filling in missing data with high precision using linear interpolation, ARIMA model or LSTM neural network, and analyzing passenger flow fluctuations using data normalization and standardization methods, sliding average, CUSUM algorithm and K-means clustering to identify sudden changes and peak areas.

4. The rail transit intelligent dispatching management method of claim 1, wherein, The S3 further comprises evaluating clustering effects under different K values through elbow method and contour coefficient, selecting the optimal K value, initializing cluster centers using K-means++ algorithm, iteratively updating cluster centers and assigning data points to the most suitable cluster until clustering converges, identifying passenger flow patterns in different time periods by analyzing the average passenger flow, fluctuation range and congestion degree of each cluster, and optimizing train departure frequency and transport capacity, adjusting train schedule and formation by using fixed interval departure, dynamically adding trains, differentiated scheduling and flexible formation according to passenger flow distribution characteristics. 5.The rail transit intelligent dispatching management method of claim 4, wherein, The average passenger flow of each cluster is obtained by calculating the mean value of all data points in the cluster, and the formula is: In the formula, is the average passenger flow of cluster k, is all data points in cluster k, is the number of data points in cluster k, is a data point; the fluctuation range within each cluster can be represented by calculating the variance of the data points in the cluster, and the formula is: In the formula, is the fluctuation range of cluster k; the congestion degree is measured by the cohesion of the cluster, and the root mean square error within the cluster is used to represent the congestion degree, and the formula is: In the formula, is the congestion degree of cluster k; by analyzing the average passenger flow, fluctuation range and congestion degree of each cluster, the passenger flow mode of different time periods is identified: in the peak period, the passenger flow of the cluster is high and the fluctuation range is large, and the congestion degree is high; In non-peak periods, the passenger flow of the cluster is low and the fluctuation range is small, and the congestion degree is low. 6.The rail transit intelligent dispatching management method according to claim 1, characterized in that, The S4 further comprises using the classic Encoder-Decoder architecture of the Transformer model for passenger flow prediction, combining historical data, external features and prediction results to optimize train departure interval and scheduling strategy, dynamically adjusting departure frequency through real-time monitoring and prediction, evaluating model generalization ability using cross-validation, and regularly updating external data and performing transfer learning.

7. The rail transit intelligent dispatching management method of claim 1, wherein, The S5 further comprises predicting future passenger flow distribution through a time series analysis model and dynamically adjusting train departure intervals, train formation and increasing or decreasing transport capacity based on the prediction results. The reinforcement learning algorithm is combined with Q learning and deep learning to train the model to make the best scheduling decision based on real-time passenger flow information and the current state, maximize passenger waiting time minimization and reduce train empty load rate. By constructing an environment model and a reward function, the impact of different scheduling strategies is simulated to optimize the train scheduling scheme.

8. The intelligent dispatching management system for rail transit, applied to the intelligent dispatching management method for rail transit according to any one of claims 1-7, characterized in that, Comprise: Data acquisition module: responsible for collecting various real-time data in the rail transit system and establishing a basic data model combined with historical data; Data preprocessing module: clean, process and analyze the collected data; Passenger flow analysis module: use machine learning algorithms to analyze and classify peak period passenger flow patterns; Time series prediction module: use time series prediction models to predict future passenger flow trends and optimize train departure intervals based on prediction results; Scheduling optimization module: combine reinforcement learning algorithms to dynamically adjust scheduling strategies based on passenger flow prediction results to optimize train departure intervals, increase or decrease transport capacity or adjust train formation.

Images