Subway ventilation demand dynamic decision-making method and system fused with people flow prediction

Abstract

The invention discloses a subway ventilation demand dynamic decision-making method and system fused with people flow prediction, and belongs to the technical field of intelligent traffic and environment control, and the method comprises the steps: collecting real-time people flow data, historical passenger flow records, station layout information and external environment parameters of each region in a subway station; constructing a deep learning prediction model through multi-source data fusion, and generating short-term and medium-term people flow distribution prediction results; based on the prediction result, combining the ventilation equipment operation parameters, the air quality index and the energy consumption cost to construct a multi-objective optimization model; dynamic decision making is carried out on a ventilation system by utilizing the model, and an optimal ventilation strategy in different regions and different time periods is output; through a real-time feedback mechanism, prediction and control parameters are continuously corrected according to actual people flow changes, closed-loop regulation and control are formed, the response precision and the energy efficiency level of a subway ventilation system can be remarkably improved, and passenger comfort and environment safety are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation and environmental control, and in particular to a subway ventilation demand dynamic decision-making method and system integrating passenger flow prediction. Background Art

[0002] With the rapid development of urban rail transit, the subway system has become one of the main ways for people to travel in their daily lives. In order to ensure the comfort and safety of passengers, the intelligent control of the subway ventilation system is particularly important. Traditional subway ventilation control is mainly set according to time periods, seasons or basic passenger flow prediction models, and lacks the ability to sensitively respond to real-time passenger flow dynamics and sudden passenger flow fluctuations. In recent years, with the development of the Internet of Things, big data and artificial intelligence technologies, dynamic ventilation decision-making methods that integrate passenger flow predictions have gradually become a research hotspot.

[0003] The prior art has the following deficiencies: In large-scale transfer hubs, due to the intersection of multiple lines and the interweaving of passenger flows, nonlinear peak behavior is formed. When the ventilation system responds to sudden extreme changes in passenger flow (such as temporary evacuation and sudden peaks on holidays), it may cause a sudden drop in air quality in local areas or even heat and humidity accumulation due to algorithm delays or insufficient data granularity, seriously affecting passenger safety and comfort. Summary of the invention

[0004] The purpose of the present invention is to provide a subway ventilation demand dynamic decision-making method and system integrating passenger flow prediction to solve the shortcomings of the background technology.

[0005] In order to achieve the above object, the present invention provides the following technical solution: a dynamic decision method for subway ventilation demand integrating passenger flow prediction, comprising: Collect real-time passenger flow data, historical passenger flow records, station layout information and external environmental parameters in multiple areas of the subway station; Perform multi-source fusion processing on the collected data, use the deep learning model to build a dynamic crowd flow prediction model, and generate short-term and medium-term crowd flow distribution prediction results; Based on the prediction results, a multi-objective optimization model is constructed by combining the station ventilation equipment operating parameters, current air quality indicators, and energy consumption cost model; The optimization model is used to dynamically decide the operation strategy of the subway ventilation system and output the optimal ventilation plan for each time period and each area; The ventilation plan is fed back to the ventilation control system in real time, and the prediction results and control strategies are continuously revised according to actual changes in passenger flow.

[0006] Preferably, collecting real-time passenger flow data of multiple areas in a subway station includes: Video surveillance equipment, infrared counters, Wi-Fi or Bluetooth signal detection devices are deployed at entrances and exits, security checkpoints, escalator entrances, platform edges, and transfer passages to capture the flow of people in and out and the density of gatherings per unit time, and perform preliminary structured processing through edge computing modules.

[0007] Preferably, the collected real-time passenger flow data, historical passenger flow records, station structure information and external environmental parameters are cleaned, standardized and missing values ​​are supplemented; Slice the crowd flow data according to spatial regions and time granularity to construct a multi-dimensional time series; Encode the station layout information, extract the spatial topological relationship of each area, and construct a graph structure or regional feature vector; Integrate external environmental parameters with contextual information such as holidays, weather types, and major events as auxiliary input features; All features are normalized to adapt to neural network training.

[0008] Preferably, a hybrid structure of a graph convolutional neural network and a long short-term memory network is selected; The GCN module is used to process the spatial dependencies between regions within the station and construct a graph structure through the adjacency matrix and regional features; The LSTM module receives the flow time series and context features of each area to capture short-term and long-term temporal dependencies; Introducing the attention mechanism to enhance the model’s prediction sensitivity to different regions and at different times; The output layer performs regression prediction on the flow of people in each future time period, and the results are expressed in the form of a region-time matrix.

[0009] Preferably, historical data is used to divide the training set, validation set and test set, and the prediction accuracy is evaluated using MSE and MAE indicators; Improve the ability to perceive short-term mutations through sliding window training; Use early stopping strategies and learning rate decay mechanisms to prevent overfitting; The prediction results are output in real time according to the prediction period set by the user; a mapping relationship is established between the predicted passenger flow and the regional ventilation load as the input of the ventilation decision model; if abnormal fluctuations are detected, an emergency ventilation response strategy is triggered.

[0010] Preferably, the station is divided into multiple functional areas, including: station hall area, platform area, transfer channel and equipment area, each area is equipped with independent or independently controllable ventilation equipment, and is given a zone identification code Zon; provide the optimization model with: the predicted passenger density of each zone, the current air quality index, the status of the ventilation equipment in the previous cycle; the real-time operating capacity and limitation of the ventilation system, and the energy consumption cost model; the optimization engine runs and starts in each control cycle: calculates the candidate control strategy using the multi-objective function; adjusts the weight of the objective function according to the current strategy priority; outputs the ventilation equipment configuration parameters corresponding to the optimal solution; The optimal ventilation strategy for each area is sent to the corresponding fan controller, frequency converter or centralized control unit through the SCADA system to perform real-time start and stop and wind speed adjustment; the execution status and air quality monitoring values ​​are fed back in real time to compare the predicted target with the actual effect to determine whether there is a deviation; if a deviation is detected, the rapid re-optimization process is immediately triggered and the original ventilation strategy is replaced.

[0011] Preferably, the continuous correction of the prediction results and control strategies according to the actual crowd flow changes includes: the real-time crowd flow error is defined as: ; represents the prediction error in region z and time period t, is the actual observed flow of people, The flow of people output by the prediction model; Set the correction factor to adjust the predicted value for the next cycle, and set the ventilation control parameter for the optimized output to be wind speed , then the corrected control parameters are: ; is the adjusted wind speed control parameter, and β is the control sensitivity coefficient, which is set according to the response capability of the ventilation equipment.

[0012] Preferably, the model weights are updated based on a time sliding window: ; is the parameter of the prediction model at time t, η is the learning rate, L is the prediction loss function, ∇θL is the gradient of the loss function with respect to the model parameters, which is used to adjust the weights.

[0013] The present invention also provides a subway ventilation demand dynamic decision system integrating passenger flow prediction, including a multi-source data acquisition module, a passenger flow prediction and analysis module, a multi-objective optimization decision module, a ventilation strategy generation module and an adaptive correction module; Multi-source data collection module: collects real-time passenger flow data, historical passenger flow records, station layout information and external environmental parameters in multiple areas of the subway station; Crowd flow prediction and analysis module: It performs multi-source fusion processing on the collected data, builds a dynamic crowd flow prediction model using a deep learning model, and generates short-term and medium-term crowd flow distribution prediction results; Multi-objective optimization decision module: Based on the prediction results, combined with the station ventilation equipment operating parameters, current air quality indicators, and energy consumption cost model, a multi-objective optimization model is constructed; Ventilation strategy generation module: using the optimization model to dynamically decide on the operation strategy of the subway ventilation system, and output the optimal ventilation plan for each time period and each area; Adaptive correction module: Feedback the ventilation plan to the ventilation control system in real time, and continuously correct the prediction results and control strategies according to the actual changes in passenger flow.

[0014] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. This invention significantly improves the subway ventilation system's perception and prediction accuracy of dynamic changes in passenger flow by introducing multi-source data fusion and deep learning prediction models. Compared with the traditional method that relies on fixed time period rules or empirical value control, this method can achieve minute-level prediction granularity and spatial regional analysis, and flexibly formulate ventilation strategies for different station structures and operating conditions, thereby more effectively avoiding insufficient ventilation or excessive energy consumption, and improving the system's initiative and foresight.

[0015] 2. The multi-objective optimization and closed-loop adaptive control mechanism constructed by the present invention enables the ventilation strategy to not only comprehensively consider air quality, energy consumption cost and equipment stability, but also have the ability to self-learn and quickly respond to emergencies. By adjusting the prediction and control parameters through real-time feedback, the system can continuously optimize the operating status, improve passenger comfort and air safety level in the station, and reduce the long-term operating cost of the ventilation system, with significant energy-saving benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0017] Figure 1 The figure is a mind map of the method of the present invention.

[0018] Figure 2 This is a mind map of the system modules of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Example 1, please refer to Figure 1 As shown, the dynamic decision-making method for subway ventilation demand integrating passenger flow prediction described in this embodiment includes: Collect real-time passenger flow data, historical passenger flow records, station layout information and external environmental parameters in multiple areas of the subway station; Perform multi-source fusion processing on the collected data, use the deep learning model to build a dynamic crowd flow prediction model, and generate short-term and medium-term crowd flow distribution prediction results; Based on the prediction results, a multi-objective optimization model is constructed by combining the station ventilation equipment operating parameters, current air quality indicators, and energy consumption cost model; The optimization model is used to dynamically decide the operation strategy of the subway ventilation system and output the optimal ventilation plan for each time period and each area; The ventilation plan is fed back to the ventilation control system in real time, and the prediction results and control strategies are continuously revised according to actual changes in passenger flow.

[0021] In order to realize intelligent dynamic control of subway station ventilation demand, it is necessary to first build a comprehensive data perception system to collect multi-dimensional data on the flow of people and environmental factors that affect the ventilation load, including: Real-time crowd flow data collection: Deploy video recognition equipment, infrared counters, Wi-Fi / Bluetooth signal detectors and other sensors at key locations of subway stations (such as entrances and exits, security checkpoints, escalators, platform edges, transfer channels, etc.) to record the inflow, outflow and stay density of people in real time per unit time. The edge computing module preliminarily processes the trajectory and aggregation of crowds and outputs structured data for subsequent prediction.

[0022] Historical passenger flow records: By accessing the urban rail transit data center, long-term passenger flow data of the station and its adjacent stations can be obtained, including daily, weekly, holiday and emergency traffic statistics. The data granularity can be refined to a time window of every 5 minutes, which is convenient for training time series prediction models.

[0023] Station layout information: Integrate the station layout information in CAD / BIM format, such as the building structure diagram, ventilation system distribution diagram, and air circulation path of the subway station. Focus on extracting spatial parameters such as the volume of each area, opening location, and transfer channel length and width to provide physical boundary conditions for ventilation load modeling.

[0024] External environmental parameters: Access the city meteorological data platform to obtain the temperature, humidity, wind speed, and air quality index (PM2.5, CO 2 At the same time, microclimate data at the entrance and exit of the station, such as the temperature difference between the ground and underground, humidity gradient, etc., are collected to build a heat exchange model.

[0025] The above multi-source data will be transmitted to the central processing system through a unified data bus and stored in the data lake to provide support for crowd flow prediction and optimization control models. All sensing devices support automatic calibration and fault tolerance mechanisms to ensure data continuity and reliability.

[0026] In order to achieve high-precision passenger flow prediction and support the dynamic optimization of subway ventilation strategies, this paper builds a dynamic passenger flow prediction model that integrates multi-source data based on deep learning technology. The model can output short-term (such as 15 minutes to 1 hour) and medium-term (1 hour to 6 hours) passenger flow distribution trends in different areas of the station. The specific steps are as follows: Perform data cleaning, time alignment, missing value filling and other operations on the collected real-time and historical passenger flow data; Slice the crowd flow data according to spatial regions and time granularity to construct a multi-dimensional time series; Encode the station layout information, extract the spatial topological relationship of each area, and construct a graph structure or regional feature vector; Integrate external environmental parameters (temperature, humidity, AQI, etc.) with contextual information such as holidays, weather types, and major events as auxiliary input features; All features are normalized to adapt to neural network training.

[0027] The hybrid structure of graph convolutional neural network (GCN) and long short-term memory network (LSTM) is used to model spatial topological associations and capture time series changes; The GCN module is used to process the spatial dependencies between regions within the station and construct a graph structure through the adjacency matrix and regional features; The LSTM module receives the flow time series and context features of each area to capture short-term and long-term temporal dependencies; Optionally introduce the attention mechanism to enhance the model's prediction sensitivity to key areas and key moments; The output layer performs regression prediction on the flow of people in each future time period, and the results are expressed in the form of a region-time matrix.

[0028] Use historical data to divide the training set, validation set and test set, and use indicators such as MSE (mean square error) and MAE (mean absolute error) to evaluate the prediction accuracy; Improve the ability to perceive short-term mutations through sliding window training; Use early stopping strategies and learning rate decay mechanisms to prevent overfitting; The model supports periodic online updates and continuous fine-tuning using the latest data to adapt to changing passenger flow patterns in actual operations.

[0029] The prediction results are output in real time according to the prediction period set by the user (such as 15 minutes, 1 hour, 4 hours); a mapping relationship is established between the predicted passenger flow and the regional ventilation load as the key input of the ventilation decision model; if abnormal fluctuations are detected (such as an abnormal surge in passenger flow), the model can link the early warning module to trigger an emergency ventilation response strategy.

[0030] In order to realize intelligent and efficient control of the ventilation system in subway stations, the present invention, based on the results of passenger flow prediction, comprehensively considers the operation capacity of the ventilation system, air quality indicators and energy consumption costs, and formulates a ventilation strategy through a multi-objective optimization method. Specifically: Predicting crowd flow distribution: Introduce the output results of the aforementioned dynamic crowd flow prediction model to obtain the expected crowd flow density and fluctuation trend in different time periods and different areas.

[0031] Ventilation equipment operating parameters: including fan type, air volume level, ventilation frequency range, response delay, equipment power limit, switching frequency limit, etc., which are used to define the controllable parameter space.

[0032] Current air quality indicators: data from sensors on site, including CO 2 Concentration, PM2.5, temperature, humidity, etc. are important constraints for the optimization objectives.

[0033] Energy consumption and cost model: Considering factors such as the energy consumption per unit operating time of the equipment, time-of-use electricity prices, ventilation operating time periods, and equipment startup costs, a time-power consumption-cost mapping model is formed.

[0034] Construct a multi-objective function covering the following core objectives: Objective 1: Optimize air quality (environmental comfort): Make the CO 2 Concentration and PM2.5 levels remain within health standards or converge toward ideal target values; Goal 2: Minimize energy consumption (operating costs): Reduce the overall power consumption of ventilation equipment and lower the electricity cost per unit time period, especially adopting load shifting strategies during the peak period of time-of-use electricity prices; Goal 3: Minimize response time and dynamics (system stability): Minimize the number of fan starts and stops and the frequency of mode switching, and reduce equipment mechanical wear and control jitter.

[0035] The air quality in all areas must meet the minimum safety standards; the fan operating parameters must not exceed the rated upper limit; the ventilation strategy adjustment frequency is subject to the execution delay of the control system; the optimization results must be real-time and meet the computing efficiency of the scheduling cycle (such as 5 minutes).

[0036] Solve using multi-objective evolutionary algorithms (such as NSGA-II or MOEA / D) or reinforcement learning methods (such as policy gradient-based optimizers); The predicted passenger flow and air quality are input into the optimizer to iteratively search for the Pareto optimal solution set; Sort multiple solutions according to the scheduling strategy priority (e.g. comfort first, followed by minimum energy consumption); Output the optimal ventilation equipment parameter configuration table for each scheduling cycle.

[0037] The optimization results are output as a control instruction set and sent to the subway ventilation control system; Monitor the execution effect in real time, and feed back air quality data and actual changes in passenger flow to the prediction and optimization module to form a closed-loop control mechanism; If a large prediction error or an abnormal event (such as sudden congestion) is detected, the optimization process is immediately retriggered.

[0038] Based on the construction of a multi-objective optimization model, the present invention provides a dynamic decision-making mechanism that can analyze the current and predicted passenger flow, air quality and energy consumption status in real time, and formulate a ventilation operation strategy that is highly targeted, regionally refined and time-accurate. This mechanism can adapt to the rapidly changing subway operating environment, specifically including: The subway operating hours are divided into peak, off-peak, night-time low-load and other time segments (such as every 5 minutes or 15 minutes as a control cycle) to adapt to the output frequency of the prediction model.

[0039] The station is divided into multiple functional areas, including: station hall area, platform area, transfer passage, equipment area, etc. Each area is equipped with independent or independently controllable ventilation equipment and is assigned a zone identification code (Zone-ID).

[0040] The data input module provides the following to the optimization model: Predicted crowd density in each zone; current air quality indicators; ventilation equipment status in the previous cycle; real-time operating capacity and limitations of the ventilation system; energy consumption cost model.

[0041] The optimization engine runs and is started in each control cycle: candidate control strategies are calculated using multi-objective functions (comfort, energy efficiency, stability); objective function weights are adjusted according to the current strategy priority (such as giving priority to ventilation efficiency during peak hours); ventilation equipment configuration parameters corresponding to the optimal solution are output (such as wind speed level, start / stop status, ventilation frequency, etc.); if there are multiple equivalent optimal solutions, historical decision consistency or equipment wear cost is introduced as a secondary judgment basis.

[0042] The optimal ventilation strategy for each area is sent to the corresponding fan controller, inverter or centralized control unit through the SCADA system to perform real-time start and stop and wind speed adjustment.

[0043] The execution status and air quality monitoring values ​​are fed back in real time to compare the predicted target with the actual effect to determine whether there is any deviation.

[0044] If a significant deviation is detected (such as the actual passenger flow is more than 30% higher than the predicted value), the rapid re-optimization process will be triggered immediately and the original ventilation strategy will be replaced to achieve emergency response.

[0045] The system records historical strategies and execution results to form a knowledge base and improve the accuracy of future strategy predictions.

[0046] Use reinforcement learning methods (such as Q-Learning or DDPG) to establish a reward and punishment mechanism for strategy effectiveness and optimize the model's intelligent decision-making capabilities in multi-period scheduling.

[0047] If it is an interconnected transfer station or a multi-station collaborative system, a regional collaborative optimizer can be introduced to achieve coordinated adjustment of air volume and balance of energy consumption load.

[0048] In order to achieve continuous optimization of the prediction model and control strategy, the system has designed a closed-loop adaptive correction mechanism. This mechanism analyzes the prediction deviation and control effect based on the real-time feedback of the flow observation value, and realizes the self-iterative optimization of the model through weight update and strategy adjustment.

[0049] The real-time flow error is defined as: ; represents the prediction error in region z and time period t, is the actual observed flow of people (obtained by sensors), This is the flow of people output by the prediction model.

[0050] Set the correction factor to , used to adjust the forecast value for the next period: ; In the formula, The adjusted passenger flow forecast value is used as the control optimization input for the next cycle. To correct the scaling factor, the range is [0,1], which can be dynamically learned based on historical errors or set to a constant.

[0051] Assume the ventilation control parameter of the optimized output is wind speed , then the corrected control parameters are: ; is the adjusted wind speed control parameter, β is the control sensitivity coefficient, which is set according to the response capacity of the ventilation equipment, and the unit is: wind speed / number of people; If air quality feedback (such as carbon dioxide concentration) is considered, a composite correction term can be introduced: ; To monitor the carbon dioxide concentration in real time, The target control concentration (e.g. 1000ppm); They represent the response coefficients to crowd flow error and air quality deviation respectively.

[0052] The system can update the model weights based on the time sliding window: ; is the parameter of the prediction model at time t, η is the learning rate, L is the prediction loss function, such as MSE, ∇θL is the gradient of the loss function with respect to the model parameters, and is used to adjust the weights. Through the above mechanism, a complete closed-loop process from real-time feedback, short-term prediction adjustment to long-term model optimization can be achieved.

[0053] In this embodiment, a dynamic decision-making method for subway ventilation demand that integrates passenger flow prediction is proposed. By building an intelligent control system for complex operation scenarios of subway stations, multiple key technologies such as perception, prediction, optimization and feedback are integrated to achieve refined, dynamic and efficient operation of the ventilation system. Specifically, in this embodiment, passenger flow sensing devices are first deployed in multiple key areas of the subway station, and the historical passenger flow database, station layout information and external environmental parameters are connected to perform unified perception and structured processing through a multi-source data acquisition module. Subsequently, a deep learning model combining a graph convolutional neural network (GCN) and a long short-term memory network (LSTM) is used to predict the passenger flow in each area in the future short-term (such as 15 minutes to 1 hour) and medium-term (1 hour to 6 hours), fully exploring the spatial topological association and time evolution characteristics, and improving the accuracy and timeliness of the prediction. Based on the prediction results, the system further introduces the operating parameters of the ventilation equipment, the current air quality indicators (such as CO 2, PM2.5, etc.) and energy cost models, and build a multi-objective optimization model with "air quality compliance", "minimum energy consumption" and "minimum response dynamics" as the main optimization goals. The optimization function is solved by evolutionary algorithms or reinforcement learning methods, and the system dynamically outputs the optimal ventilation plan for each area in each scheduling cycle, including control parameters such as fan start and stop status, wind speed gear and ventilation frequency.

[0054] In order to enhance the adaptive capability of the system, this embodiment also builds a closed-loop feedback and correction mechanism to collect actual human flow and environmental data in real time, calculate the prediction deviation, and dynamically adjust the prediction value and control strategy for the next cycle based on the correction factor. At the same time, it supports online updating of model parameters based on gradient descent, realizes long-term autonomous learning and strategy evolution, and ensures that the system always maintains efficient and accurate operation.

[0055] In summary, this embodiment not only improves the perception and response capabilities of the subway ventilation system in a complex passenger flow environment, but also significantly reduces energy consumption costs and equipment wear risks, taking into account both passenger comfort and air safety, and has good application prospects and promotion value.

[0056] Example 2, please refer to Figure 2 As shown, the subway ventilation demand dynamic decision-making system integrating passenger flow prediction described in this embodiment includes a multi-source data acquisition module, a passenger flow prediction and analysis module, a multi-objective optimization decision-making module, a ventilation strategy generation module and an adaptive correction module; Multi-source data collection module: collects real-time passenger flow data, historical passenger flow records, station layout information and external environmental parameters in multiple areas of the subway station; Crowd flow prediction and analysis module: It performs multi-source fusion processing on the collected data, builds a dynamic crowd flow prediction model using a deep learning model, and generates short-term and medium-term crowd flow distribution prediction results; Multi-objective optimization decision module: Based on the prediction results, combined with the station ventilation equipment operating parameters, current air quality indicators, and energy consumption cost model, a multi-objective optimization model is constructed; Ventilation strategy generation module: using the optimization model to dynamically decide on the operation strategy of the subway ventilation system, and output the optimal ventilation plan for each time period and each area; Adaptive correction module: Feedback the ventilation plan to the ventilation control system in real time, and continuously correct the prediction results and control strategies according to the actual changes in passenger flow.

[0057] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0058] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0059] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0060] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A dynamic decision-making method for subway ventilation demand integrating passenger flow prediction, characterized by: include: Collect real-time passenger flow data, historical passenger flow records, station layout information and external environmental parameters in multiple areas of the subway station; Perform multi-source fusion processing on the collected data, use the deep learning model to build a dynamic crowd flow prediction model, and generate short-term and medium-term crowd flow distribution prediction results; Based on the prediction results, a multi-objective optimization model is constructed by combining the station ventilation equipment operating parameters, current air quality indicators, and energy consumption cost model; The optimization model is used to dynamically decide the operation strategy of the subway ventilation system and output the optimal ventilation plan for each time period and each area; The ventilation plan is fed back to the ventilation control system in real time, and the prediction results and control strategies are continuously revised according to actual changes in passenger flow.

2. The dynamic decision-making method for subway ventilation demand integrating passenger flow prediction according to claim 1 is characterized by: The real-time passenger flow data of multiple areas in the subway station is collected, including: Video surveillance equipment, infrared counters, Wi-Fi or Bluetooth signal detection devices are deployed at entrances and exits, security checkpoints, escalator entrances, platform edges, and transfer passages to capture the flow of people in and out and the density of gatherings per unit time, and perform preliminary structured processing through edge computing modules.

3. The dynamic decision-making method for subway ventilation demand integrating passenger flow prediction according to claim 2 is characterized by: Clean, standardize, and fill in missing values ​​for the collected real-time passenger flow data, historical passenger flow records, station structure information, and external environmental parameters; Slice the crowd flow data according to spatial regions and time granularity to construct a multi-dimensional time series; Encode the station layout information, extract the spatial topological relationship of each area, and construct a graph structure or regional feature vector; Integrate external environmental parameters with contextual information such as holidays, weather types, and major events as auxiliary input features; All features are normalized to adapt to neural network training.

4. The dynamic decision-making method for subway ventilation demand integrating passenger flow prediction according to claim 3 is characterized by: A hybrid structure of graph convolutional neural network and long short-term memory network is selected; The GCN module is used to process the spatial dependencies between regions within the station and construct a graph structure through the adjacency matrix and regional features; The LSTM module receives the flow time series and context features of each area to capture short-term and long-term temporal dependencies; Introducing the attention mechanism to enhance the model’s prediction sensitivity to different regions and at different times; The output layer performs regression prediction on the flow of people in each future time period, and the results are expressed in the form of a region-time matrix.

5. The dynamic decision-making method for subway ventilation demand integrating passenger flow prediction according to claim 4 is characterized in that: Use historical data to divide the training set, validation set and test set, and use MSE and MAE indicators to evaluate the prediction accuracy; Improve the ability to perceive short-term mutations through sliding window training; Use early stopping strategies and learning rate decay mechanisms to prevent overfitting; The prediction results are output in real time according to the prediction period set by the user; a mapping relationship is established between the predicted passenger flow and the regional ventilation load as the input of the ventilation decision model; if abnormal fluctuations are detected, an emergency ventilation response strategy is triggered.

6. The dynamic decision-making method for subway ventilation demand integrating passenger flow prediction according to claim 5 is characterized by: The station is divided into multiple functional areas, including: station hall area, platform area, transfer channel and equipment area. Each area is equipped with independent or independently controllable ventilation equipment and is assigned a zone identification code Zon; the optimization model is provided with: the predicted passenger density of each zone, the current air quality index, the status of the ventilation equipment in the previous cycle; the real-time operating capacity and limitations of the ventilation system, and the energy consumption cost model; the optimization engine runs and starts in each control cycle: the candidate control strategy is calculated using the multi-objective function; the objective function weight is adjusted according to the current strategy priority; the ventilation equipment configuration parameters corresponding to the optimal solution are output; The optimal ventilation strategy for each area is sent to the corresponding fan controller, frequency converter or centralized control unit through the SCADA system to perform real-time start and stop and wind speed adjustment; the execution status and air quality monitoring values ​​are fed back in real time to compare the predicted target with the actual effect to determine whether there is a deviation; if a deviation is detected, the rapid re-optimization process is immediately triggered and the original ventilation strategy is replaced.

7. The dynamic decision-making method for subway ventilation demand integrating passenger flow prediction according to claim 1 is characterized by: The continuous correction of the prediction results and control strategies according to the actual crowd flow changes includes: the real-time crowd flow error is defined as: ; represents the prediction error in region z and time period t, is the actual observed flow of people, The flow of people output by the prediction model; Set the correction factor to adjust the predicted value for the next cycle, and set the ventilation control parameter for the optimized output to be wind speed , then the corrected control parameters are: ; is the adjusted wind speed control parameter, and β is the control sensitivity coefficient, which is set according to the response capability of the ventilation equipment.

8. The dynamic decision-making method for subway ventilation demand integrating passenger flow prediction according to claim 7 is characterized in that: Based on the time sliding window, the model weights are updated: ; is the parameter of the prediction model at time t, η is the learning rate, L is the prediction loss function, ∇θL is the gradient of the loss function with respect to the model parameters, which is used to adjust the weights.

9. A subway ventilation demand dynamic decision system integrating passenger flow prediction, used to implement the subway ventilation demand dynamic decision method integrating passenger flow prediction as described in any one of claims 1 to 8, characterized in that: It includes multi-source data acquisition module, crowd flow prediction and analysis module, multi-objective optimization decision module, ventilation strategy generation module and adaptive correction module; Multi-source data collection module: collects real-time passenger flow data, historical passenger flow records, station layout information and external environmental parameters in multiple areas of the subway station; Crowd flow prediction and analysis module: It performs multi-source fusion processing on the collected data, builds a dynamic crowd flow prediction model using a deep learning model, and generates short-term and medium-term crowd flow distribution prediction results; Multi-objective optimization decision module: Based on the prediction results, combined with the station ventilation equipment operating parameters, current air quality indicators, and energy consumption cost model, a multi-objective optimization model is constructed; Ventilation strategy generation module: using the optimization model to dynamically decide on the operation strategy of the subway ventilation system, and output the optimal ventilation plan for each time period and each area; Adaptive correction module: Feedback the ventilation plan to the ventilation control system in real time, and continuously correct the prediction results and control strategies according to the actual changes in passenger flow.

Images