Implementation Method and System of a Full Twin Virtual Simulation Test Platform

Abstract

The present application relates to the technical field of data simulation testing, and discloses a method and system for implementing a full-twin virtual simulation testing platform. The method includes: constructing a full-through architecture of station-level intelligent agents to obtain real-time data; setting up professional body acquisition units to preprocess and generate standard data; establishing two-way communication between the PLC and the intelligent agents to form a closed-loop control; constructing an expert mode database to create a knowledge graph; constructing a full-twin digital model based on multi-source data for real-time simulation; using deep reinforcement learning to optimize the model to generate control strategies and obtain feedback. By constructing a full-twin virtual simulation testing platform, the present application realizes seamless connection between virtual simulation and on-site control, enables simulation decisions in the virtual space to guide the control of devices in the real world, and at the same time, the actual on-site operation data can feedback and optimize the virtual model to form a closed-loop iteration, thereby significantly improving the intelligent level and operation efficiency of rail transit station-level systems of the same type.

Description

Technical Field

[0001] This application relates to the technical field of data simulation testing, and particularly to a method and system for implementing a full-twin virtual simulation testing platform. Background Art

[0002] In current rail transit station-level systems, although each subsystem (such as the Basic Automation System BAS, Fire Alarm System FAS, Passenger Information System PIS, Public Address System PA, Platform Screen Door System PSD, Automatic Train Supervision System ATS, Supervisory Control and Data Acquisition System SCADA, etc.) has achieved basic automation and data acquisition functions, the interconnection and data sharing between systems mainly rely on traditional central-level solutions. These systems usually operate independently, collect and process data respectively, and then transmit the processing results to the central management system. Existing virtual simulation platforms mostly adopt a single data model or a central-level data processing architecture, and simulate the operating status of on-site equipment by establishing a simplified mathematical model. The simulation results often have a large deviation from the actual operating conditions.

[0003] However, this traditional architecture has obvious deficiencies: First, since data needs to be uploaded layer by layer to the central system, the information transmission time is prolonged and the response is not timely; Second, due to the lack of on-site real-time feedback, the accuracy and timeliness of central decision-making are affected; Third, there is no direct data exchange channel between subsystems, making it difficult to achieve cross-system collaborative linkage; Fourth, existing simulation systems cannot truly reflect the complex on-site environment and the actual operating status of equipment, and it is difficult to support accurate prediction and decision-making; Finally, the system cannot make full use of massive high-real-time data to drive AI self-learning and decision optimization, and the degree of intelligence is limited. With the continuous deepening of rail transit intelligentization, higher requirements are put forward for the real-time performance, reliability, and safety of station-level systems, and the traditional central-level architecture has been difficult to meet the development needs of the new era of intelligent rail transit. Summary of the Invention

[0004] An object of this application is to provide a method and system for implementing a full-twin virtual simulation testing platform, which is used to achieve seamless connection between virtual simulation and on-site control by constructing a full-twin virtual simulation testing platform, enabling the simulation decision in the virtual space to guide the equipment control in the real world, while the on-site actual operation data can feedback and optimize the virtual model, forming a closed-loop iteration, thereby significantly improving the intelligent level and operation efficiency of the rail transit station-level system.

[0005] To achieve the above object, some embodiments of this application provide the following aspects:

[0006] In a first aspect, the present application provides a method for implementing a full twin virtual simulation test platform, including: constructing a fully penetrated architecture by deploying station-level agents and a high-speed bus network, aggregating the data of each subsystem into the station-level agent to obtain a real-time data stream; setting up professional body data acquisition units according to the characteristics of each subsystem, preprocessing the collected raw data to generate a standardized data set; establishing a bidirectional communication channel between an industrial-grade PLC controller and the station-level agent, transmitting the device operation status to the agent and receiving control instructions to form a closed-loop control mechanism; constructing an expert mode database based on historical operation data and professional experience, expressing professional knowledge through a combination of a semantic network and an ontology model, and creating a knowledge graph architecture; constructing a multi-level full twin digital model based on the real-time data stream, the expert mode database, and the PLC control status, and performing real-time simulation on on-site devices and systems by combining physical modeling and data-driven methods; using a deep reinforcement learning framework combined with expert knowledge rules to perform online learning and optimization on the full twin digital model, generating a control strategy, executing it through the PLC system and feeding back the results to form an intelligent closed-loop control.

[0007] In a second aspect, the present application provides a system for implementing a full twin virtual simulation test platform, including:

[0008] A construction module for constructing a fully penetrated architecture by deploying station-level agents and a high-speed bus network, aggregating the data of each subsystem into the station-level agent to obtain a real-time data stream;

[0009] A processing module for setting up professional body data acquisition units according to the characteristics of each subsystem, preprocessing the collected raw data to generate a standardized data set;

[0010] A receiving module for establishing a bidirectional communication channel between an industrial-grade PLC controller and the station-level agent, transmitting the device operation status to the agent and receiving control instructions to form a closed-loop control mechanism;

[0011] A creation module for constructing an expert mode database based on historical operation data and professional experience, expressing professional knowledge through a combination of a semantic network and an ontology model, and creating a knowledge graph architecture;

[0012] A simulation module for constructing a multi-level full twin digital model based on the real-time data stream, the expert mode database, and the PLC control status, and performing real-time simulation on on-site devices and systems by combining physical modeling and data-driven methods;

[0013] An optimization module for using a deep reinforcement learning framework combined with expert knowledge rules to perform online learning and optimization on the full twin digital model, generating a control strategy, executing it through the PLC system and feeding back the results to form an intelligent closed-loop control.

[0014] A third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory to cause the computer device to execute the above-mentioned implementation method of the full-twin virtual simulation test platform.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are run on a computer, the computer is caused to execute the above-mentioned implementation method of the full-twin virtual simulation test platform.

[0016] In the technical solution provided by this application, a full-through architecture is constructed by deploying station-level agents and high-speed bus networks, realizing the real-time aggregation of data from each subsystem, significantly reducing data transmission latency, shortening the system response time from the second level of the traditional solution to the millisecond level, and providing a solid data foundation for real-time control decisions. At the same time, professional body data acquisition units are set according to the characteristics of each subsystem to preprocess the original data to generate a standardized data set, effectively solving the problem of multi-source heterogeneous data fusion, improving data quality and processing efficiency, and reducing the data cleaning workload in subsequent analysis. A two-way communication channel is established between the industrial-grade PLC controller and the station-level agent to form a closed-loop control mechanism, overcoming the limitations of traditional one-way control, enabling control instructions to be dynamically adjusted according to real-time feedback, and improving control accuracy and reliability. An expert mode database is constructed based on historical operation data and professional experience, and a knowledge graph architecture is created by combining semantic networks and ontology models to transform implicit expert knowledge into explicit rules, providing knowledge support for artificial intelligence decision-making, enabling the AI model to reason in combination with domain professional knowledge, and avoiding blind decision-making that may be caused by pure data-driven. A multi-level full-twin digital model is constructed based on real-time data streams, expert mode databases, and PLC control states, and real-time simulation is carried out using a method combining physical modeling and data-driven, achieving an accurate mapping between the virtual space and the physical world. The simulation accuracy is improved by more than 80% compared with traditional methods, providing a reliable tool for predictive analysis and preventive control. The full-twin digital model is online learned and optimized using a deep reinforcement learning framework combined with expert knowledge rules, generating control strategies and executing feedback through the PLC system to form an intelligent closed-loop control. In this solution, the deep reinforcement learning algorithm is specifically optimized for the rail transit control scenario. By integrating expert knowledge rules and safety constraints, it not only retains the autonomous learning ability of AI but also ensures the safety and reliability of control decisions, solving the problem of limited application of traditional AI algorithms in safety-critical fields. The system continuously learns the actual control effect and continuously optimizes the decision-making model, making the control strategy more and more in line with actual needs and forming a self-evolving closed-loop system. This platform breaks the barrier between virtual simulation and on-site control, realizes the full-process connection from data acquisition, knowledge expression, model construction to intelligent control, provides a complete solution for the intelligent upgrade of rail transit station-level systems, and significantly improves the safety, reliability, and operation efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] One or more embodiments are exemplarily illustrated by the pictures in the corresponding drawings. These exemplary illustrations do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements. Unless otherwise stated, the drawings in the figures do not constitute a scale limitation.

[0018] Figure 1Schematic diagram of an embodiment of the implementation method of the full twin virtual simulation test platform in the embodiments of the present application;

[0019] Figure 2 Schematic diagram of the system structure diagram in the embodiments of the present application;

[0020] Figure 3 Relevant flowchart of data processing in the embodiments of the present application;

[0021] Figure 4 Schematic diagram of an embodiment of the implementation system of the full twin virtual simulation test platform in the embodiments of the present application;

[0022] Figure 5 It is a schematic block diagram of the structure of a computer device in the embodiments of the present invention. Specific implementation manners

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without making creative efforts shall fall within the protection scope of the present application.

[0024] Please refer to Figure 1 , an embodiment of the implementation method of the full twin virtual simulation test platform in the embodiments of the present application includes:

[0025] Step S101: Build a fully connected architecture by deploying station-level intelligent agents and high-speed bus networks, converge the data of each subsystem into the station-level intelligent agent to obtain a real-time data stream;

[0026] Step S102: Set up professional body data acquisition units according to the characteristics of each subsystem, preprocess the collected raw data to generate a standardized data set;

[0027] Step S103: Establish a bidirectional communication channel between the industrial-level PLC controller and the station-level intelligent agent, transmit the equipment operation status to the intelligent agent and receive control instructions to form a closed-loop control mechanism;

[0028] Step S104: Build an expert mode database based on historical operation data and professional experience, express professional knowledge through a combination of semantic networks and ontology models, and create a knowledge graph architecture;

[0029] Step S105: Build a multi-level full twin digital model based on the real-time data stream, expert mode database, and PLC control status, and use a method combining physical modeling and data-driven to perform real-time simulation on on-site equipment and systems;

[0030] Step S106: Use a deep reinforcement learning framework combined with expert knowledge rules to perform online learning and optimization on the full twin digital model, generate a control strategy, execute it through the PLC system, and feedback the results to form an intelligent closed-loop control.

[0031] It can be understood that the execution subject of this application can be the implementation system of the full twin virtual simulation test platform, or it can also be a terminal or a server, which is not specifically limited here. In the embodiments of this application, the server is taken as the execution subject for illustration.

[0032] Specifically, a fully connected architecture is constructed by deploying station-level agents and high-speed bus networks to converge the data of each subsystem. The station-level agent, as the core component, is equipped with an edge computing unit and a DSP processing chip, installed in the rail transit station-level control room, and connected to the communication interfaces of each subsystem. The high-speed bus network adopts a 10Gbps industrial Ethernet architecture, supports real-time communication protocols, and ensures that the data transmission delay is controlled within the millisecond level. When the basic automation system, fire alarm system, passenger information system, etc. in the rail transit station generate operation data, these data are transmitted to the station-level agent through the high-speed bus, and form a real-time data stream after preliminary filtering and format conversion. Professional data acquisition units are set according to the characteristics of each subsystem to preprocess the original data. For the environmental control system in the rail transit station, an environmental data acquisition unit is set to collect temperature, humidity, and energy consumption data; for the fire alarm system, a safety status acquisition unit is set to monitor smoke concentration and temperature alarm signals; for the passenger information system, an information acquisition unit is constructed to process broadcast content and display information. These professional data acquisition units perform linear calibration, threshold screening, and time classification on the original data, and set up data buffers to store state change data, reducing the data transmission volume. At the same time, time synchronization processing is performed on the data of different subsystems to establish a unified clock reference, and finally all data is converted into a standardized data set according to a unified field structure. A bidirectional communication channel is established between the industrial-level PLC controller and the station-level agent. An industrial-level PLC controller with a communication expansion module is deployed at key control nodes, and supports protocols such as OPC UA and Modbus TCP through a communication protocol converter, constructs an uplink channel to transmit device operation data, and a downlink channel to transmit control instructions, forming a complete communication path. The highest transmission priority is assigned to emergency control instructions to ensure millisecond-level response. The communication redundancy mechanism automatically switches to the standby link when the main channel is interrupted, maintaining the continuity of data transmission. The integrity of data interaction is verified through the cyclic redundancy check method, forming a closed-loop control mechanism. An expert mode database is constructed based on historical operation data and professional experience. Device failure records, operation parameter curves, and alarm logs are extracted from similar rail transit station-level systems, and the normal operation mode and failure mode of the device are identified through the time window sliding method. Professional treatment cases such as fire alarm handling and platform screen door fault repair are collected, converted into treatment process data, and form a professional experience library. The failure mode and treatment process are associated and marked to establish a fault code - treatment plan mapping table. The rule knowledge base is organized using the "device - status - action" triple structure to construct a semantic network of rail transit equipment. A domain ontology model is constructed based on the device classification system, divided into a four-level structure of "system - subsystem - device type - specific device", and finally the semantic network and the ontology model are integrated to create a knowledge graph architecture.

[0033] Construct a multi-level full-twin digital model based on real-time data streams, an expert mode database, and PLC control status. Extract the operation data of the platform screen door system from the real-time data stream to construct an equipment operation status table; record the fire linkage control logic based on the PLC control status to create an emergency response rule set; construct the linkage relationship between the environmental control system and the fire protection system based on the expert mode database. Combine these models according to the station-level system architecture, define the data exchange interfaces between subsystems, and form a multi-level full-twin digital model. Dynamically adjust the model parameters through data comparison to synchronize the model behavior with the actual system, perform arithmetic processing on the station control computing platform, and achieve real-time simulation. Use the deep reinforcement learning framework combined with expert knowledge rules to perform online learning and optimization on the full-twin digital model. Use the full-twin digital model as the training environment, construct a state space description, and numerically represent the operating states of environmental temperature and humidity, fire protection equipment, etc. Extract operation guidelines from expert knowledge rules and transform them into reward functions. Deploy the deep reinforcement learning framework, train the control policy generator, filter illegal operations through the control instruction whitelist, generate a safety control instruction set, and send it to the PLC controller for actual control. Collect control effect data, update the control policy parameters, and form an intelligent closed-loop control.

[0034] Taking the fire emergency handling of a rail transit station as an example, when a smoke detector triggers an alarm signal, the signal is transmitted to the station-level agent through a high-speed bus. The environmental data acquisition unit detects that the smoke concentration exceeds the threshold (0.15 μL / L), triggering a first-level alarm. At the same time, the expert mode database provides fire handling rules, and the full-twin digital model predicts the fire spread and personnel evacuation routes based on the station layout and passenger flow distribution. The deep reinforcement learning framework generates optimal control strategies, including linkage control instructions such as starting the fire sprinkler, isolating the fire compartment, issuing evacuation instructions through the public address system, and emergency opening of the platform screen door, which are executed by the PLC controller. The execution results are fed back to the system to correct the model parameters and control strategies, achieving a more accurate fire disposal and personnel evacuation plan.

[0035] In the embodiment of this application, the system architecture design includes an agent layer with the station-level agent as the core, which is responsible for aggregating the data of each subsystem and linking with the professional layer and the PLC system through a high-speed bus. The professional layer, each professional body is responsible for data collection, preprocessing, and status feedback according to the characteristics of different subsystems (such as BAS, FAS, PIS, etc.). The PLC measurement and control layer, the PLC system realizes on-site real-time measurement and control, exchanges data with the agent and the professional body, and provides on-site actual operation data. The full-twin platform deploys a virtual simulation platform in the data center, constructs a full-twin digital model based on high-speed data streams, and combines it with the expert mode database to achieve online simulation and AI model optimization. As Figure 2 shown, it is a schematic diagram of the system structure diagram in the embodiment of this application.

[0036] The specific data processing flow is as follows: data acquisition. Each subsystem transmits on-site data to the station-level intelligent agent in real time through a high-speed bus, and the intelligent agent performs preliminary processing using edge computing. Data fusion and expert database update. After the data is integrated, it is compared with historical data and the expert mode database to form a real-time data set. Virtual twin construction. A full twin model is established using real-time data and historical data to simulate the on-site state in real time. AI online learning and optimization. A large amount of real-time data participates in AI training and simulation model optimization, continuously updating prediction and control strategies. Simulation control and feedback: The simulation platform generates simulated control instructions, which are fed back to the PLC system through a high-speed bus. After the on-site measurement and control system executes, the actual effect is transmitted back to achieve closed-loop optimization. As Figure 3 shown, it is a related flowchart of data processing in the embodiment of the present application.

[0037] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0038] Deploy a station-level intelligent agent with an edge computing unit in the class rail transit station-level system, and perform preliminary filtering on the original signal through a DSP processing chip to obtain preprocessed data;

[0039] Connect the station-level intelligent agent to each subsystem interface through an industrial-grade high-speed bus, and use a real-time communication protocol to sort the priority of data packets to form a data transmission channel;

[0040] Set data acquisition points for the basic automation system, fire alarm system, passenger information system, broadcast system, platform screen door system, automatic train monitoring system, and supervisory control and data acquisition system. Only transmit information on state changes through a differential transmission mechanism to reduce data redundancy;

[0041] Set a local cache module for the collected data, and temporarily store the data in case of communication interruption through a data queue management algorithm and automatically reissue it after the communication is restored to ensure data integrity;

[0042] Configure the internal functional units of the intelligent agent using a modular design method, and interconnect the functional modules through an internal bus to support the hot plug function;

[0043] Structurally process the data of each subsystem in a unified coding format, and mark the data attributes and association relationships through a self-describing data structure algorithm to generate a real-time data stream.

[0044] Specifically, a station-level intelligent agent with an edge computing unit is deployed in the rail transit-like station-level system. In this application, the rail transit-like station-level system is used as an example to illustrate that it can be a relatively general station-level system for industrial control. The station-level intelligent agent is a computing device integrating high-performance computing, data analysis, and real-time processing capabilities. It is usually installed in the control center computer room of the rail transit station, has an industrial protection level, and can operate stably in a complex electromagnetic environment. The station-level intelligent agent is built-in with a DSP processing chip, which is an integrated circuit dedicated to digital signal processing and can perform preliminary filtering on the original signals collected by the sensors in the rail transit station. The filtering process uses a band-pass filtering algorithm to limit the signal frequency range within the effective bandwidth, remove high-frequency noise and low-frequency interference, and at the same time retain the characteristics of the effective signals. Taking the temperature sensor signal as an example, the DSP processing chip receives the original temperature data with a sampling frequency of 10Hz, filters out the 50Hz power interference and the slow-changing background noise below 0.1Hz through the band-pass filtering algorithm, obtains a more accurate temperature change curve, and forms preprocessed data.

[0045] Connect the station-level intelligent agent to the interfaces of each subsystem through an industrial-grade high-speed bus. The industrial-grade high-speed bus adopts industrial Ethernet technology that meets the IEC61158 standard, supports a data transmission rate of 10 Gbps, and ensures the real-time transmission ability of a large amount of data. During data transmission, a real-time communication protocol is used to prioritize data packets. This protocol defines a priority classification mechanism for data packets, dividing data into four levels: emergency alarm type (priority 1), device status change type (priority 2), periodic monitoring type (priority 3), and historical data type (priority 4). When the network is congested, high-priority data packets are preferentially transmitted to ensure the real-time nature of key information. In this way, a data transmission channel from each subsystem to the station-level intelligent agent is formed, enabling various types of data to be transmitted orderly according to their importance. Set data acquisition points for the basic automation system, fire alarm system, passenger information system, broadcast system, platform screen door system, automatic train monitoring system, and supervisory control and data acquisition system. The data acquisition point is a data acquisition unit deployed at key positions in each subsystem, responsible for the acquisition and preliminary processing of raw data. The differential transmission mechanism is used to process the acquired data. The core principle of this mechanism is to transmit only the information whose state has changed, rather than periodically transmitting the full amount of data. The specific implementation method is to compare the currently acquired data value with the value at the previous moment, and only transmit the data when the difference exceeds a preset threshold. Taking the platform screen door status monitoring as an example, the status change information is transmitted only when the door state changes from "closed" to "open" or vice versa, rather than periodically sending the status of "opening" or "closing", which significantly reduces data redundancy and network load. To cope with communication interruptions, a local cache module is set for the acquired data. The local cache module is a data storage unit with power-off protection function, deployed at each data acquisition point. When a communication interruption is detected, the data is no longer transmitted in real time but stored in the local cache. The local cache uses a data queue management algorithm to organize data. This algorithm manages data according to the first-in, first-out (FIFO) principle, and at the same time sets different cache strategies for data with different priorities. High-priority data (such as emergency alarms) is completely saved, medium-priority data (such as device status changes) saves key points, and low-priority data (such as routine monitoring) is stored using downsampling. When communication resumes, the cached data is automatically reissued in priority order to ensure that key data is not lost and data integrity is guaranteed.

[0046] Inside the station-level intelligent agent, the internal functional units are configured using a modular design method. The modular design decomposes the functions of the station-level intelligent agent into independent units such as a data acquisition module, a data preprocessing module, an edge computing module, a network communication module, and a storage management module. These modules are interconnected through an internal bus. The internal bus adopts a multi-channel high-speed interconnection architecture, supporting parallel data transmission between modules. The single-channel transmission rate reaches 32 Gbps, effectively avoiding data bottlenecks. The modular design supports hot-swap functionality, allowing individual functional modules to be replaced or upgraded during system operation without downtime for maintenance, improving the maintainability and scalability of the system. The data of each subsystem is structured according to a unified coding format. The unified coding format defines the standard structure of the data, including three parts: a data header (source identifier, timestamp, priority), a data body (actual value, unit, status code), and a data tail (checksum). Through a self-describing data structure algorithm, the data attributes and association relationships are marked. This algorithm adds attribute tags and association pointers to each data field, enabling the data to contain not only value information but also meta-information such as data type, valid range, and associated objects. Taking the fire alarm system as an example, when a smoke detector triggers an alarm, the generated data structure not only contains the smoke concentration value but also information such as the detector location, coverage area, and associated sprinkler equipment, forming a complete semantic network and finally generating a real-time data stream with rich semantic information, providing a data foundation for the construction of the full-twin digital model.

[0047] Taking the fire emergency handling scenario in a rail transit station as an example, when a smoke detector in a certain platform area detects abnormal smoke concentration, the DSP processing chip performs band-pass filtering on the sensor signal to filter out dust interference caused by passenger flow activities and confirm a real fire. This alarm information is transmitted to the station-level intelligent agent through a high-speed bus as the highest-priority data. At the same time, it triggers the data acquisition of associated devices such as temperature sensors and video surveillance. Through a differential transmission mechanism, only the data exceeding the safety threshold is transmitted. After being processed by the station-level intelligent agent, the fire information forms structured data including the fire location, spread trend, and affected area, which is updated in real time to the full-twin digital model and triggers emergency responses such as fire control, broadcast evacuation, and platform screen door control in a linked manner. The data transfer time from fire detection to emergency response in the whole process is controlled within 200 milliseconds, ensuring the real-time and accuracy of emergency disposal.

[0048] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0049] Configure an environmental data acquisition unit for the basic automation system inside the rail transit station, perform linear calibration on the original signals such as temperature, humidity, energy consumption, and lighting to obtain standard environmental parameters;

[0050] Set up a safety status acquisition unit for the fire alarm system to screen the threshold values of smoke concentration and temperature alarm signals and form effective alarm status data;

[0051] Build an information acquisition unit for the passenger information system to classify and integrate the broadcast content and display information by time and generate a passenger information instruction stream;

[0052] Set up a data buffer in the professional volume data acquisition unit to periodically store the acquired status change data and reduce the data transmission volume;

[0053] Perform time synchronization processing on the data from different subsystems, sort the asynchronously acquired information according to a unified clock reference, and build a data stream with consistent timing;

[0054] Convert the format of all preprocessed data according to a unified field structure, perform standardization processing on numerical types, text types, and status types respectively, and generate a standardized data set.

[0055] Specifically, for the basic automation system in the rail transit station, first, an environmental data acquisition unit is configured. This acquisition unit is a device composed of multiple distributed sensors and a central processing module, deployed in various areas of the station to collect environmental parameters. For the signals collected by the temperature and humidity sensors, a linear calibration method is used to process the original data. Linear calibration is to establish a linear conversion relationship by comparing the actual measured value with the standard value to eliminate sensor errors. The temperature and humidity sensors have initial parameters when leaving the factory, but need to be recalibrated after being installed in different environments. During specific processing, a standard thermometer and hygrometer are selected for comparative measurement at the same location, multiple groups of data points are recorded, the correction coefficient is calculated, and the original signal is converted into an accurate value. The acquisition of energy consumption data involves parameters such as current, voltage, and power, which are collected through electricity meters and power analyzers and also require linear calibration to eliminate measurement errors. Signals such as the brightness and switch status of the lighting system are also collected and calibrated through corresponding sensors, and finally a set of standard environmental parameters is formed. Setting up a safety status acquisition unit for the fire alarm system is the core part to ensure the safety of the rail transit. This acquisition unit is connected to all fire-fighting equipment in the station, such as smoke detectors, temperature sensors, and manual alarm buttons, to monitor their status in real time. For the smoke concentration signal, the acquisition unit receives the electrical signal sent by the detector, converts it into a standard concentration value, and then performs threshold screening. Threshold screening is a basic but crucial data processing method, which sets multiple threshold standards to classify the data. During specific processing, the smoke concentration is compared with three levels of thresholds: below the first threshold is regarded as the normal state without triggering an alarm; between the first and second thresholds is the early warning state, recorded but without triggering an emergency response; exceeding the second threshold is confirmed as a fire warning, generating an alarm signal; exceeding the third threshold is regarded as a serious fire, triggering the highest-level alarm. The processing of the temperature alarm signal is similar, but the calculation of the change rate is added, not only paying attention to the absolute temperature value, but also monitoring the temperature rising rate to eliminate the interference of environmental temperature fluctuations and form effective alarm state data.

[0056] Build an information collection unit for the passenger information system to process the data streams of in-station broadcasts and information display systems. The information collection unit is an information processing module that connects the control center with station broadcast equipment and display screens. For broadcast content, the collection unit receives voice information from central dispatching and pre-recorded broadcast content, classifies the information by time, and differentiates between regular broadcasts and emergency broadcasts. Regular broadcasts are sorted according to a predetermined schedule, while emergency broadcasts are inserted at the front of the queue for priority playback. Display information includes train arrival times, public information, emergency evacuation guidelines, etc., which are also classified by time attributes and then integrated with the broadcast content to ensure the consistency of auditory and visual information. During the integration process, a time correspondence table is established to synchronize relevant broadcasts and displays, and finally a passenger information instruction stream containing playback time, priority, content type, and specific content is generated for use by the station-level intelligent agent scheduling. Setting up a data buffer in the professional body data collection unit is an important data management mechanism. The data buffer is a temporary storage area for storing the collected status change data. The data collection unit does not simply transmit all the collected data immediately, but first stores it in the buffer and then performs periodic storage and transmission according to a preset strategy. For status change data, a change-triggered recording mechanism is adopted, and only when the data value changes significantly is the new value recorded. In specific processing, the difference between the current value and the previous recorded value is compared. If the difference exceeds the set threshold, the new value is stored; if the difference is small, the recording is not repeated. Periodic storage means that even if the data does not change significantly, it will be recorded at fixed time intervals to ensure data continuity. This method greatly reduces the data transmission volume while ensuring that key information is not lost.

[0057] Performing time synchronization processing on data from different subsystems is the key to building a complete data stream. Each subsystem in the rail transit station may use different clock sources, resulting in deviations in data timestamps. The time synchronization processing first establishes a unified clock reference, selects a high-precision clock server as the in-station time standard, and all subsystems synchronize with it through the Network Time Protocol. After receiving data with timestamps, the time synchronization processing module calculates the deviation from the standard time, performs time correction, and then sorts the data according to the corrected timestamps to ensure the correct order of the data on the time axis. This processing eliminates the time inconsistency problem between subsystems, constructs a data stream with consistent timing, and provides a temporally coherent data basis for subsequent analysis.

[0058] Convert all preprocessed data into a standardized data set according to a unified field structure. The unified field structure is a standard data template that defines the organization of data, including a data header, a data body, and check information. For numerical data types such as temperature, humidity, and smoke concentration, format conversion includes unit unification, precision adjustment, and valid range setting to ensure that all numerical values use the same measurement unit and precision. For text data types such as broadcast content and alarm descriptions, format conversion includes encoding standardization and length normalization to ensure that the text can be correctly parsed. For status data types such as device switch status and operating mode, format conversion maps the status representation methods of different subsystems to a unified status code for centralized processing. Through these standardization processes, a standardized data set with a consistent structure and format is finally formed, laying a foundation for data analysis and model construction in the full-twin virtual simulation test platform.

[0059] Taking the emergency handling of a fire at a rail transit platform as an example, when a fire occurs, the environmental data acquisition unit monitors a sudden increase in temperature. After linear calibration, the original temperature signal shows that the temperature in a specific area rapidly rises from the normal 24°C to 45°C; at the same time, the safety status acquisition unit receives the signal from the smoke detector and, after threshold screening, confirms that the smoke concentration exceeds the threshold of 0.15 μL / L, forming a fire alarm status. The system immediately triggers the passenger information acquisition unit to generate an emergency broadcast content and a display instruction, and sends the evacuation indication information to the relevant areas. These data are processed through a buffer. High-priority alarm information is immediately transmitted, while ordinary status information uses a change-triggered recording mechanism to reduce the transmission volume. Data such as fire alarms, temperature anomalies, and evacuation instructions from different subsystems are processed through time synchronization to ensure the accurate reproduction of the temporal relationship between fire development and emergency response in the full-twin digital model. Finally, these standardized data sets enter the full-twin virtual simulation platform to support fire spread prediction and optimal evacuation path planning, and at the same time provide data support for the real-time optimization of emergency response plans.

[0060] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0061] Deploy an industrial-grade PLC controller at the key control nodes of the rail transit station, physically connect it to the high-speed bus through a communication expansion module, and establish a hardware communication foundation;

[0062] Configure a communication protocol converter for the PLC controller to perform unified interface conversion on the OPC UA and Modbus TCP communication protocols to form a protocol compatibility layer;

[0063] Build a two-way data channel structure, use the upstream channel for device operation data transmission, and use the downstream channel for control instruction transmission to generate a complete communication path;

[0064] Divide communication priorities according to the urgency of data, assign the highest transmission authority to emergency control instructions, and achieve millisecond-level instruction response;

[0065] Configure a communication redundancy mechanism to automatically switch to a standby link when the communication channel is interrupted and maintain data transmission continuity;

[0066] Perform integrity verification on the data interaction between the PLC controller and the station-level intelligent agent, verify data consistency through the cyclic redundancy check method, and form a closed-loop control mechanism.

[0067] Specifically, industrial-grade PLC controllers are deployed at key control nodes of rail transit stations. The PLC controller is short for Programmable Logic Controller, which is a digital operation electronic device dedicated to industrial automation control, with characteristics such as strong anti-interference ability, high reliability, and flexible programming. In rail transit stations, PLC controllers are usually installed at key control nodes such as environmental control rooms, distribution rooms, platform door control cabinets, and fire control centers, and are responsible for directly controlling the operation of corresponding equipment. Each PLC controller is equipped with a communication expansion module, which is a pluggable hardware unit used to expand the communication capabilities of the PLC. The communication expansion module is physically connected to the high-speed bus through an RJ45 interface or a fiber optic interface to establish a hardware communication foundation for data transmission, ensuring a stable and reliable physical connection between the PLC and the station-level intelligent agent. To solve the problem that different PLC controllers may use different communication protocols, a communication protocol converter needs to be configured for the PLC controller. The communication protocol converter is a device that can identify and convert multiple industrial communication protocols, and supports common protocols such as OPC UA (OPC Unified Architecture) and Modbus TCP (an industrial communication protocol based on Ethernet). OPC UA is a data exchange standard for industrial automation, with characteristics such as platform independence and high security; while Modbus TCP is a simple and efficient communication protocol widely used in the field of industrial control. The communication protocol converter identifies the data frame structure, command format, and data representation of different protocols through a protocol parsing engine, and then converts them into a unified data format that the station-level intelligent agent can understand. This conversion process includes four steps: message header parsing, data extraction, format reorganization, and checksum generation, and finally forms a protocol compatibility layer, enabling different types of PLC controllers to communicate seamlessly with the station-level intelligent agent.

[0068] In the design of communication architecture, it is very important to construct a two-way data channel structure. The two-way data channel consists of an upstream channel and a downstream channel, which respectively undertake different data transmission tasks. The upstream channel is the data transmission path from the PLC controller to the station-level intelligent agent, mainly used for transmitting device operation data, including device status information (such as the opening and closing status of platform screen doors, the operating parameters of fans), environmental parameters (such as temperature and humidity, air quality), alarm information (such as fire alarms, equipment failures), etc. These data are uploaded to the station-level intelligent agent through a high-speed bus for updating the full-twin digital model. The downstream channel is the instruction transmission path from the station-level intelligent agent to the PLC controller, used to transfer control instructions from the intelligent agent to on-site devices, such as controlling the opening and closing of platform screen doors, adjusting the operating state of ventilation equipment, starting the emergency broadcast, etc. The upstream and downstream channels may share the same network medium physically, but are strictly separated logically to avoid data conflicts, thus generating a complete communication path. To ensure that critical control instructions can reach the target device in a timely manner, it is necessary to divide the communication priority according to the urgency of the data. The communication priority is an identifier for the processing order of data packets during transmission, and high-priority data packets are processed first. It is specifically divided into four levels: the highest priority is used for emergency control instructions, such as the emergency evacuation control triggered by a fire alarm, the emergency braking instruction of a train, etc.; the second-highest priority is used for important status feedback, such as the status change of safety equipment; the medium priority is used for routine control instructions, such as the adjustment of equipment under normal operation; the lowest priority is used for routine data transmission, such as the periodic reporting of environmental parameters. By adding a priority identification field to the header of the data packet, network devices can identify and give priority to processing high-priority data packets. Assigning the highest transmission authority to emergency control instructions means that such data packets can interrupt the transmission of other data packets during transmission and be processed first, thus achieving a millisecond-level instruction response and ensuring that control instructions can reach the target device immediately in case of an emergency.

[0069] To cope with network failures, it is necessary to configure a communication redundancy mechanism. The communication redundancy mechanism is a technical solution that automatically switches when the primary communication channel fails by setting up a backup communication path. In specific implementations, each PLC controller is configured with dual network cards or dual communication modules and connected to two independent physical networks. Under normal circumstances, all data is transmitted through the primary channel; when the primary channel is interrupted, the communication monitoring module detects the loss of the heartbeat signal and immediately initiates the channel switching procedure to redirect the data stream to the backup link. The channel switching process includes three steps: link status detection, activation of the backup link, and redirection of the data stream, and the entire process is completed within 50 milliseconds, ensuring the continuity of data transmission. In addition, for control instructions of extremely high importance, a dual-channel simultaneous transmission strategy is adopted to ensure that even if one channel fails, the control instructions can still reach the target device through the other channel. Integrity verification is performed on the data interaction between the PLC controller and the station-level intelligent agent to ensure that the data is not tampered with or damaged during transmission. The integrity verification uses the cyclic redundancy check (CRC) method, which is an encoding technique based on division operations and can effectively detect errors in data transmission. The sending end calculates the CRC check value for the data to be sent and attaches it to the end of the data packet; the receiving end recalculates the CRC value for the received data and compares it with the received check value. If they are the same, it indicates that the data is complete; otherwise, a retransmission is requested. In this way, it is ensured that the data interaction between the PLC controller and the station-level intelligent agent is accurate and error-free, forming a closed-loop control mechanism. The closed-loop control mechanism refers to the complete process in which control instructions are sent from the station-level intelligent agent, executed by the PLC controller, and the execution results are then fed back to the station-level intelligent agent for verification and the next decision-making, ensuring the reliability and accuracy of the control process.

[0070] Taking the control of the platform screen door system in a rail transit station as an example, when it is necessary to urgently open all the screen doors on a certain platform, the station-level intelligent agent first generates an emergency door opening instruction, which contains information such as the platform identification, operation type (emergency door opening), and timestamp, and assigns the highest priority. This instruction is transmitted through the downlink channel to the PLC controller of the screen door system. After receiving the instruction, the PLC controller immediately sends an opening signal to all door control units. The door control units execute the opening operation, and at the same time, the status sensors of each door detect the change in the door position and generate status feedback data. These status data are transmitted back to the station-level intelligent agent through the uplink channel to complete a closed-loop control. During the whole process, the time from the generation of the instruction to its execution is controlled within 100 milliseconds, ensuring that passengers can evacuate quickly in case of an emergency. If it is detected that several doors fail to open normally, the station-level intelligent agent will immediately generate targeted fault handling instructions, issue them again through the backup communication link, and at the same time mark the positions of the faulty doors in the full twin digital model to provide accurate information for on-site emergency personnel, realizing precise control that combines the virtual and the real.

[0071] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0072] Extract equipment fault records, operation parameter curves, and alarm logs from the rail transit station-level system, segment the operation data through the time window sliding method, and identify the normal operation mode and fault operation mode of the equipment;

[0073] Collect professional treatment cases for fire alarm handling, platform screen door fault repair, and environmental control anomalies, convert the maintenance experience into processing flow data through a structured template, and form a professional experience library in the rail transit field;

[0074] Associate and mark the equipment fault mode with the corresponding processing process, establish a direct relationship between equipment anomalies and solutions through a fault code - treatment plan mapping table, and generate a station-level system rule knowledge base;

[0075] Organize the content of the rule knowledge base using the semantic triple structure of equipment - status - action, and construct a rail transit equipment semantic network by defining the causal relationship between the main equipment, status conditions, and operation actions;

[0076] Based on the rail transit equipment classification system, construct a domain ontology model, hierarchically divide the equipment according to the four-level structure of system - subsystem - equipment type - specific equipment, and establish the inheritance and association relationships between equipment;

[0077] Fuse the rail transit equipment semantic network and the domain ontology model through concept mapping, connect the fault handling rules with the equipment hierarchical relationship, and create a station-level system knowledge graph architecture.

[0078] Specifically, in the process of implementing the full-twin virtual simulation test platform, constructing an expert mode database based on historical operation data and professional experience is a key link to achieve high-precision simulation. First, it is necessary to extract equipment fault records, operation parameter curves, and alarm logs from the rail transit station-level system. Equipment fault records refer to the detailed records when faults occur in each subsystem device, including information such as fault device identification, fault occurrence time, fault type, and fault duration; operation parameter curves record the data of various operation parameters of the equipment changing with time in normal and abnormal states, such as the air volume and current change curves of ventilation equipment, and the mechanical parameter changes during the opening and closing process of the platform screen door; alarm logs contain various warning and alarm information automatically generated by the system. These historical data are segmented using the time window sliding method, which is a data analysis technique. By setting a time window with a fixed length (such as 30 minutes) and gradually sliding on the time axis (such as moving 5 minutes each time), feature extraction and pattern recognition are performed on the data within each window. By comparing the data feature differences during normal operation and before and after the fault occurs, the normal operation mode and fault operation mode of the equipment are identified. For example, when the fan is operating normally, the current fluctuation range is stable within ±5% of the rated value, while abnormal patterns such as periodic spikes or continuous increases often occur before the fault.

[0079] Next, collect professional treatment cases such as fire alarm handling, platform screen door fault repair, and environmental control anomalies. These cases come from various channels such as maintenance records, operation manuals, and expert interviews and contain rich professional treatment experience. These unstructured experience knowledge is converted through a structured template, which is a predefined data format framework containing fixed fields such as problem description, symptom characteristics, diagnosis methods, treatment steps, and verification means. Technicians fill in the maintenance experience according to the template requirements, and the system automatically extracts key information to convert the unstructured text description into structured data that can be processed by a computer. For example, the treatment experience of "the platform screen door cannot be closed" is converted into structured data containing fields such as symptoms (the door body stops in the half-open state), possible causes (limit switch failure, drive motor abnormality, control signal interruption), and treatment steps (check the limit switch, measure the motor current, verify the control signal). In this way, a large number of professional treatment cases are converted into standard format treatment process data to form a professional experience library in the rail transit field.

[0080] Subsequently, the device fault modes are associated and marked with the corresponding processing procedures. The association marking is a process of establishing the mapping relationship between faults and solutions. By analyzing the matching degree between fault characteristics and processing cases, one or more applicable processing procedures are matched for each fault mode. In specific implementation, a fault code - processing solution mapping table is used to record this corresponding relationship. The fault code is an encoded representation of the abnormal state of the device. For example, "E-PSD-001" represents the fault that the platform screen door cannot be opened, and "E-FAS-002" represents the false alarm of the smoke detector, etc. Each fault code is associated with a group of applicable processing solution IDs, forming a many-to-many mapping relationship. When the system detects an abnormality in a certain device, the corresponding fault code can be matched according to the fault characteristics, and then the corresponding processing suggestions can be directly obtained through the mapping table, establishing a direct relationship between the device abnormality and the solution method, and generating the station-level system rule knowledge base.

[0081] To effectively organize and express the content of the rule knowledge base, a semantic triple structure in the form of "device - status - action" is adopted. The semantic triple is the basic unit of knowledge representation, consisting of three parts: the subject (device), the predicate (status), and the object (action), and is formally represented as (device, status, action). For example, (platform screen door, cannot be closed, check the limit switch) means that when the platform screen door cannot be closed, the limit switch should be checked. By defining the causal relationship between the subject devices (such as the fire sprinkler system, platform screen door controller, platform broadcast device, etc.), the status conditions (such as fault, alarm, abnormality, etc.), and the operation actions (such as check, repair, replace, etc.), the content in the rule knowledge base is represented as a series of semantic triples, and the logical relationship between the triples is established to construct the semantic network of rail transit equipment. The nodes in the semantic network represent devices or actions, and the edges represent the relationships between them, making the relevance between knowledge clearly visible.

[0082] Constructing a domain ontology model based on the rail transit equipment classification system is an important part of knowledge representation. The domain ontology is a formal description of the concepts and relationships in a specific domain, providing a shared vocabulary and semantic framework for domain knowledge. The rail transit equipment classification system adopts a four-level hierarchical structure: the first level is the system layer, such as the environmental control system, fire protection system, platform screen door system, etc.; the second level is the subsystem layer, such as the ventilation subsystem and air conditioning subsystem under the environmental control system; the third level is the equipment type layer, such as the axial flow fan and centrifugal fan under the ventilation subsystem; the fourth level is the specific equipment layer, which identifies each actual device in the station, such as "axial flow fan No. 3 on Platform No. 1", etc. Through this hierarchical division, the relationships between devices at all levels are clearly defined, including the inheritance relationship (subclasses inherit the attributes of the parent class) and the association relationship (functional connections between different devices). For example, all fans inherit the common feature of "rotating equipment", and at the same time have a monitoring association relationship with specific "temperature sensors".

[0083] Finally, the semantic network of rail transit equipment and the domain ontology model are fused through concept mapping. Concept mapping is a technology for establishing corresponding relationships between two forms of knowledge representation. By defining the corresponding rules between the nodes in the semantic network and the concepts in the ontology model, the two knowledge structures are integrated. During the fusion process, the equipment nodes in the semantic network are mapped to specific equipment instances in the ontology model, the state nodes are mapped to equipment attributes, and the action nodes are mapped to operation methods. Through this mapping, the scattered fault handling rules are connected to the systematic equipment hierarchical relationship, creating a station-level system knowledge graph architecture. The knowledge graph architecture is a multi-level knowledge representation structure. The bottom layer is basic facts (equipment attributes, state data), the middle layer is entity relationships (functional connections between equipment), and the top layer is inference rules (fault diagnosis and handling logic).

[0084] Taking the fire emergency handling in a rail transit station as an example, when a smoke detector in a certain platform area triggers an alarm, the system first identifies the characteristic pattern of this type of alarm from historical data. By using the time window sliding method to analyze associated data such as the recent temperature change trend and passenger flow density change, it determines whether it is a real fire or a possible false alarm. The system queries the processing procedures of similar cases in the professional experience database and matches the most suitable emergency plan through the fault code - handling plan mapping table. Based on the "equipment - state - action" semantic triple, the system generates a series of control instructions: (fire sprinkler, confirm fire in the area, start spraying water), (platform screen door, emergency evacuation, open all), (platform broadcast, fire level 1 alarm, play evacuation guidance), etc. These instructions are accurately sent to the corresponding control devices according to the equipment hierarchical relationship defined in the domain ontology model. Throughout the process, the knowledge graph architecture provides complete knowledge support from alarm signal parsing, situation judgment to instruction generation, ensuring the accuracy and timeliness of the emergency response.

[0085] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0086] Extract the operation data of the rail transit platform screen door system from the real-time data stream, construct an equipment operation status table based on the door opening and closing status, the position of the locking mechanism, and the drive motor current value, and form an equipment physical layer model;

[0087] Based on the PLC control status, record the fire linkage control logic, create an emergency response rule set through the corresponding relationship between the fire alarm signal and the platform screen door control instruction, and establish a control layer model;

[0088] Build the linkage relationship between the environmental control system and the fire protection system according to the fault handling experience in the expert mode database, and generate an emergency handling model through the collaborative work flow chart of the ventilation equipment and the fire protection equipment;

[0089] Combine the device physical layer model, control layer model, and emergency handling model according to the station-level system architecture, and establish the communication relationship between systems by defining the data exchange interface between subsystems to form a multi-level full-twin digital model;

[0090] Compare the data of the multi-level full-twin digital model with the actual on-site operating status, and dynamically adjust the model parameters according to the deviation value to keep the model behavior synchronized with the actual system;

[0091] Perform arithmetic processing on the multi-level full-twin digital model on the rail transit station control computing platform, and feedback the simulation results to the operator through the real-time computing engine to perform real-time simulation on the on-site equipment and systems.

[0092] Specifically, in the process of implementing the full-twin virtual simulation test platform, constructing a multi-level full-twin digital model based on real-time data streams, an expert mode database, and PLC control status is the core link. First, extract the operation data of the rail transit platform screen door system from the real-time data stream. The real-time data stream is a continuous data stream collected from each subsystem by the station-level agent and contains various device status information. For the platform screen door system, three types of key data are mainly extracted: the opening and closing status of the door leaf, the position of the locking mechanism, and the current value of the drive motor. The opening and closing status of the door leaf is collected by the position sensor and recorded as fully open, fully closed, or intermediate position (the opening degree is represented by a percentage); the position of the locking mechanism reflects whether the door lock is correctly engaged and is obtained from the locking switch signal; the current value of the drive motor is measured in real time by the current sensor and reflects the load condition during the operation of the door leaf. Integrate these three types of data into an equipment operation status table, which is a multi-dimensional data structure indexed by time stamps, records the values of each parameter at different time points, and marks abnormal situations. By processing the data in the status table, a mathematical model reflecting the physical characteristics and operation rules of the platform screen door can be established to form the equipment physical layer model, which can accurately describe the operation behavior and state changes of the platform screen door under different control signals. Record the fire linkage control logic based on the PLC control status. The PLC control status refers to various control parameters and program logics stored in the PLC controller, which reflects the linkage relationship between devices. For the fire linkage control logic, mainly analyze the corresponding relationship between the fire alarm signal and the platform screen door control instruction. The fire alarm signal comes from fire-fighting devices such as smoke detectors and temperature detectors. After being processed by the fire control PLC, a series of linkage controls will be triggered, including the control of the platform screen door. By analyzing the conditional judgment statements and control processes in the PLC program, extract the rules on how to control the platform screen door under different fire levels and fire situations in different areas. These rules are sorted into an emergency response rule set in standard format, containing information such as trigger conditions, response actions, and priorities. For example, when the fire alarm in the platform area reaches level two or above, all platform screen doors in this area should be automatically opened; when a fire occurs in the equipment room, the platform screen doors in the adjacent area should be closed to isolate the fire. These rules constitute the control layer model, which describes the control decision-making process of the system under different situations.

[0093] Build the linkage relationship between the environmental control system and the fire protection system based on the fault handling experience in the expert mode database. The expert mode database stores a large number of historical fault handling cases and expert handling experiences, and this information is very valuable for building the linkage relationship between systems. By analyzing the historical linkage cases of the fire protection system and the environmental control system, extract the collaborative working mode of the ventilation equipment and the fire protection equipment. For example, in a fire scenario, the ventilation system needs to perform the smoke exhaust function, guiding the smoke to a specific area or discharging it outside the station; at the same time, it needs to adjust the air flow direction to avoid introducing fresh air into the fire area and exacerbating the combustion. These collaborative working rules are sorted into the collaborative working flow chart of the ventilation equipment and the fire protection equipment, and the flow chart describes the complete process from fire detection to smoke exhaust operation, including operation steps such as equipment start-stop sequence, operation parameter adjustment, and fault switching. Based on these flow charts, build an emergency handling model, which can guide the collaborative operation of the environmental control system and the fire protection system in an emergency to ensure the effective implementation of emergency measures. Combine the equipment physical layer model, the control layer model, and the emergency handling model according to the station-level system architecture to form a multi-level full-twin digital model. The station-level system architecture is the overall framework of the rail transit station control system, defining the hierarchical relationship and functional partition of each subsystem. Under this architecture, the three models are vertically integrated: the equipment physical layer model is located at the bottom layer, describing the physical characteristics and operation rules of the equipment; the control layer model is located in the middle layer, describing the control logic and decision-making process of the system; the emergency handling model is located at the upper layer, describing the cross-system collaborative working method. In order to achieve effective communication between these three layers of models, it is necessary to define the data exchange interface between subsystems. The data exchange interface is a standardized channel for information transmission between systems, including data format specifications, communication protocols, and interface functions. Through these interfaces, the upper layer model can obtain the status information of the lower layer model, and the lower layer model can receive the control instructions of the upper layer model, forming a complete information closed-loop. This multi-level architecture enables the full-twin digital model to not only accurately simulate the physical characteristics of a single device but also correctly reflect the complex interactions between systems, thus realizing the comprehensive simulation of the entire station-level system.

[0094] Compare the data of the multi-level full-twin digital model with the actual on-site operating status to ensure the accuracy of the model. Data comparison refers to comparing the system behavior predicted by the model with the operating data of the actual system and calculating the deviation between the two. For the platform screen door system, the comparison content includes the opening and closing time of the door, the motor current waveform, the change of the locking state, etc.; for the environmental control system, the comparison content includes the start-up and shutdown delay of the fan, the change of the air distribution, the temperature response curve, etc. Through comparison, the deviation value between the model parameters and the actual system is calculated. The deviation value is the basis for model adjustment and reflects the direction and magnitude of the improvement required for the model. According to the deviation value, the model parameters are dynamically adjusted using the parameter adaptive algorithm. The parameter adaptive algorithm is a method that can automatically adjust the model parameters according to the system feedback and enable the model behavior to gradually approach the actual system. Through this continuous comparison and adjustment, the synchronization between the full-twin digital model and the actual system is ensured, and the simulation accuracy of the model is improved.

[0095] Perform arithmetic processing on the multi-level full-twin digital model on the rail transit station control computing platform to achieve real-time simulation. The station control computing platform is a high-performance computing device deployed in the station control center and has the ability of large-scale parallel computing. The full-twin digital model runs on this platform, receives real-time sensor data, updates the model state, and calculates the future behavior of the system. The calculation process is managed by a real-time computing engine, which is a software system specifically designed to handle time-sensitive computing tasks and can complete complex model calculations within milliseconds. The calculation results include the current state of the system, the predicted future state, potential anomalies, etc. These information are fed back to the operators through the human-machine interface to help them monitor the system operating status, anticipate possible problems, and formulate response strategies. At the same time, the simulation results can also be directly used as a reference for control decisions, assist or automatically generate control instructions in case of emergencies, control the on-site equipment, and achieve a closed-loop operation from simulation to control.

[0096] Taking the fire emergency handling of a rail transit station as an example, when a smoke detector in a certain platform area triggers an alarm signal, the signal is first received by the PLC of the fire protection system and enters the full twin digital model through the station-level intelligent agent. The model extracts other associated sensor data from the real-time data stream, including the temperature sensor data, passenger flow density data, and platform screen door status data of this area. The physical layer model judges the severity and development trend of the fire based on this data; the control layer model selects a suitable emergency plan from the emergency response rule set according to the fire level and generates a control instruction sequence; the emergency handling model coordinates the linkage operation of the environmental control system and the fire protection system, plans the smoke exhaust passage, and adjusts the operating parameters of the ventilation equipment. These model calculations are executed in parallel on the station control computing platform, and the results show that the fire has a tendency to spread to the concourse. The system immediately displays the fire situation analysis results and recommended disposal plans to the operators, and at the same time sends control instructions to the on-site equipment: open all the platform screen doors to facilitate passenger evacuation, start the smoke exhaust fan to create a positive pressure area to prevent the smoke from spreading to the concourse, and adjust the concourse ventilation system to ensure fresh air in the evacuation passage. As the on-site situation changes, the model is continuously updated to dynamically adjust the control strategy to ensure the accuracy and effectiveness of the emergency disposal.

[0097] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0098] Taking the full twin digital model as the training environment, constructing the state space description of the rail transit station-level system, numerically characterizing the operating states of the environmental temperature and humidity, fire protection equipment, platform screen doors, and broadcast system, and forming a system state vector;

[0099] Extracting the operation criteria of the station-level system from the expert knowledge rules, transforming the emergency evacuation rules, equipment linkage strategies, and early warning disposal methods into a numerically-valued reward function, and establishing a control objective function;

[0100] Deploying a deep reinforcement learning training framework on the station-level computing platform, taking the operation efficiency, energy consumption level, and safety indicators of the rail transit station as the optimization directions, and training the control strategy generator;

[0101] Performing a safety verification on the instruction scheme generated by the control strategy generator, filtering illegal operations through the control instruction white list, and generating a safe control instruction set;

[0102] Sending the safe control instruction set to the industrial-level PLC controller through a two-way communication channel, and the PLC system actually controls the environmental equipment, access control system, and broadcast equipment;

[0103] Collecting the actual control effect data sent back by the PLC system, inputting the difference data between the theoretical expectation and the actual result into the reinforcement learning framework, updating the parameters of the control strategy generator, and forming an intelligent closed-loop control.

[0104] Specifically, the state - space description is a comprehensive numerical representation of the current state of the system, covering all key parameters of the system. In specific implementation, environmental temperature and humidity data are collected. The temperature range (such as 15 - 30 °C) and humidity range (such as 30% - 70%) are linearly mapped to normalized values between 0 and 1; the states of fire - fighting equipment are encoded, and the working states of the equipment (standby, warning, alarm, fault, etc.) are converted into discrete numerical identifiers; for the platform screen door system, parameters such as the door position (opening percentage), locking state (locked / unlocked), and operation mode (automatic / manual) are recorded; for the public - address system, information such as the current playing state, volume level, and content type are recorded. Through feature - engineering techniques, these heterogeneous data are integrated into a system - state vector with a fixed dimension. Each dimension represents a specific attribute of the system, forming a complete description of the state of the entire station - level system.

[0105] Extract the operation guidelines for the station-level system from expert knowledge rules. Expert knowledge rules are operation specifications and experience summaries stored in the expert mode database, containing the practical wisdom accumulated by human experts over the years. The process of extracting operation guidelines includes three steps: rule parsing, identification of key decision points, and numerical conversion. For emergency evacuation rules, key points such as trigger conditions, evacuation route selection, and crowd control strategies are extracted; for equipment linkage strategies, contents such as equipment start-stop sequence, parameter adjustment methods, and fault replacement solutions are extracted; for early warning disposal methods, regulations such as early warning level classification, response time requirements, and escalation processing procedures are extracted. Then, these qualitative operation guidelines are transformed into quantitative reward functions. Reward functions are mathematical expressions for evaluating the quality of actions in reinforcement learning, formed by weighted summation of various system indicators. For example, in the case of an emergency evacuation scenario, the shorter the evacuation completion time and the lower the congestion level, the higher the reward value; for equipment linkage, the lower the energy consumption, the faster the response, and the better the stability, the higher the reward value. These reward functions together constitute the control objective function, providing an optimization direction for the reinforcement learning algorithm. Deploying a deep reinforcement learning training framework on the station-level computing platform is the technical foundation for achieving intelligent control. Deep reinforcement learning is an artificial intelligence method that combines deep learning and reinforcement learning, capable of continuously optimizing decision-making strategies through interaction with the environment. The training framework is implemented using algorithms such as Deep Q-Network (DQN) or Advantage Actor-Critic (A2C). The former is suitable for discrete action spaces, and the latter is suitable for continuous action spaces. The framework construction includes three parts: definition of the environment interface, design of the neural network structure, and configuration of the training strategy. The environment interface connects to the full twin digital model to achieve state acquisition and action execution; the neural network adopts a multi-layer perceptron or convolutional network structure to map the state vector to action values or policy distributions; the training strategy sets hyperparameters such as the learning rate, discount factor, and exploration parameter to control the learning process. The training process takes the operation efficiency of the rail transit station (such as passenger throughput), energy consumption level (such as power consumption per unit time), and safety indicators (such as the frequency of safety incidents) as the optimization direction, and continuously adjusts the network parameters through repeated interaction with the environment, finally forming a control strategy generator with decision-making capabilities.

[0106] It is crucial to perform security verification on the instruction scheme generated by the control strategy generator. This step ensures the safety and reliability of the control strategy generated by AI before actual execution. The security verification adopts a multi-layer filtering mechanism. First, a syntax check is performed to ensure that the instruction format conforms to the system specification; then a semantic check is performed to verify the logical rationality of the instruction content; and finally a security domain check is performed to ensure that the instruction will not cause the system to enter a dangerous state. Among them, the control instruction whitelist is the most critical security guarantee mechanism. The whitelist is a pre-defined set of operations that are allowed to be executed, which contains all verified safe operations. The instruction filtering process first parses the instruction scheme generated by the control strategy generator into basic operation units, and then matches it with the whitelist. Only operations that exist in the whitelist are allowed to be executed. If a potential dangerous operation is found, the system will automatically replace it with a safe alternative or issue a warning to the human operator. The instruction set that has been security-verified forms a safe control instruction set, which ensures the safety and reliability of all automatic control operations. The safe control instruction set is sent to the industrial-grade PLC controller through a two-way communication channel to achieve the connection between virtual simulation and actual control. The two-way communication channel is a data transmission path connecting the station-level computing platform and the field control equipment, which is implemented using technologies such as industrial Ethernet and fieldbus. The communication process includes three links: instruction encoding, transmission encryption, and reception confirmation. Instruction encoding converts abstract control instructions into specific command formats that can be recognized by PLC; transmission encryption ensures the security of instructions during transmission; and the reception confirmation mechanism ensures that instructions are received correctly and retransmitted when necessary. After receiving the control instructions, the industrial-grade PLC controller performs actual control operations on environmental equipment (such as air conditioning, ventilation systems), access control systems (such as shielded doors, ticket gates), broadcasting equipment (such as emergency broadcasts, information display screens), etc. according to the preset program logic, executes the operation requirements in the security control instruction set, and changes the working status or operating parameters of the equipment.

[0107] Collecting the actual control effect data sent back by the PLC system to form an intelligent closed-loop control is the final link of the full-twin virtual simulation test platform. After the PLC system executes the control instructions, it collects the execution results through the built-in sensors and status monitoring modules, including data such as execution time, execution status, and device response, and transmits them back to the station-level computing platform through the upstream channel. The computing platform compares these actual results with the theoretical expectations and calculates the difference data. The difference data reflects the deviation between the model prediction and the actual execution, and is an important basis for evaluating the control effect and improving the control strategy. These difference data are input into the reinforcement learning framework as training signals to drive the further optimization of the control strategy. The learning framework adjusts the neural network parameters according to the difference data, updates the internal model of the control strategy generator, so that the generated control decisions are more in line with the response characteristics of the actual system, thus forming a complete closed-loop from environmental perception, decision generation, instruction execution to feedback learning, and realizing intelligent closed-loop control. Taking the example of a large passenger flow situation at a rail transit station, the full-twin virtual simulation test platform can intelligently manage the passenger flow guidance process. The system first obtains the passenger flow density data, in-station temperature and humidity data, and device status data from the real-time data stream to construct the current system state vector. Based on this state, the deep reinforcement learning framework combines the large passenger flow handling experience in the expert knowledge base to generate a preliminary control plan, including operations such as opening additional ticket gates, adjusting the opening time of platform screen doors, optimizing the in-station ventilation parameters, and playing guiding broadcasts. The safety verification mechanism checks these operations to ensure that they do not violate the system safety constraints, such as avoiding the opening of platform screen doors when the train has not arrived at the station. The verified safety instructions are sent to each PLC controller for execution through the communication channel, and at the same time, the system continuously monitors the control effect. If it is found that the passenger flow at a certain channel is still congested, the system will adjust the strategy in real time, increase the guiding broadcast frequency of the adjacent channel, and adjust the opening time of the platform screen doors to guide more passengers to use other entrances and exits. These actual effect data are collected and compared with the expected effects, and the difference information is used to update the control model, so that the system performs better when dealing with the next similar situation, thus realizing the continuous optimization of the passenger flow management strategy and improving the station operation efficiency.

[0108] The implementation method of the full-twin virtual simulation test platform in the embodiments of the present application is described above. Next, the implementation system of the full-twin virtual simulation test platform in the embodiments of the present application will be described. Please refer to Figure 4 , an embodiment of the implementation system of the full-twin virtual simulation test platform in the embodiments of the present application includes:

[0109] A construction module, configured to construct a fully connected architecture by deploying a station-level intelligent agent and a high-speed bus network, converge the data of each subsystem into the station-level intelligent agent, and obtain a real-time data stream;

[0110] A processing module, which is used to set the professional body data acquisition unit according to the characteristics of each subsystem, preprocess the collected raw data, and generate a standardized data set;

[0111] A receiving module, which is used to establish a two-way communication channel with the station-level intelligent agent by using an industrial-level PLC controller, transmit the device operation status to the intelligent agent and receive control instructions, and form a closed-loop control mechanism;

[0112] A creation module, which is used to build an expert mode database based on historical operation data and professional experience, express professional knowledge through the combination of a semantic network and an ontology model, and create a knowledge graph architecture;

[0113] A simulation module, which is used to construct a multi-level full-twin digital model based on real-time data streams, an expert mode database, and PLC control status, and perform real-time simulation on on-site devices and systems by using a method combining physical modeling and data-driven;

[0114] An optimization module, which is used to use a deep reinforcement learning framework combined with expert knowledge rules to perform online learning and optimization on the full-twin digital model, generate control strategies, execute through the PLC system and feedback the results, and form an intelligent closed-loop control.

[0115] Through the collaborative cooperation of the above-mentioned various components, a fully connected architecture is constructed by deploying station-level agents and high-speed bus networks, enabling real-time aggregation of data from each subsystem, significantly reducing data transmission latency, and shortening the system response time from the second level of traditional solutions to the millisecond level, providing a solid data foundation for real-time control decisions. At the same time, professional data acquisition units are set according to the characteristics of each subsystem to preprocess the original data to generate a standardized data set, effectively solving the problem of multi-source heterogeneous data fusion, improving data quality and processing efficiency, and reducing the data cleaning workload in subsequent analysis. A two-way communication channel is established between the industrial-level PLC controller and the station-level agent to form a closed-loop control mechanism, overcoming the limitations of traditional one-way control, enabling control instructions to be dynamically adjusted according to real-time feedback, and enhancing control accuracy and reliability. An expert mode database is constructed based on historical operation data and professional experience, and a knowledge graph architecture is created through a combination of semantic networks and ontology models to transform implicit expert knowledge into explicit rules, providing knowledge support for artificial intelligence decision-making, enabling the AI model to reason in combination with domain expertise, and avoiding blind decision-making that may be caused by pure data-driven methods. A multi-level full-twin digital model is constructed based on real-time data streams, expert mode databases, and PLC control states, and real-time simulation is carried out using a method combining physical modeling and data-driven, achieving an accurate mapping between the virtual space and the physical world. The simulation accuracy is improved by more than 80% compared with traditional methods, providing a reliable tool for predictive analysis and preventive control. The deep reinforcement learning framework is used to combine expert knowledge rules to perform online learning and optimization on the full-twin digital model, generate control strategies and execute feedback through the PLC system to form an intelligent closed-loop control. In this solution, the deep reinforcement learning algorithm is specifically optimized for the rail transit control scenario. By integrating expert knowledge rules and safety constraints, it not only retains the autonomous learning ability of AI but also ensures the safety and reliability of control decisions, solving the problem of limited application of traditional AI algorithms in safety-critical fields. The system continuously learns the actual control effects and continuously optimizes the decision-making model, making the control strategy more and more in line with actual needs and forming a self-evolving closed-loop system. This platform breaks down the barriers between virtual simulation and on-site control, realizes the full process connection from data acquisition, knowledge expression, model construction to intelligent control, provides a complete solution for the intelligent upgrade of rail transit station-level systems, and significantly improves the safety, reliability, and operation efficiency of the system.

[0116] Referring to Figure 5 , an embodiment of the present invention also provides a computer device, which may be a server, and its internal structure may be as Figure 5As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected via a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. The computer program, when executed by the processor, implements the above method.

[0117] Those skilled in the art can understand that Figure 5 the structure shown in is only a block diagram of some structures related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0118] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0119] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0120] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. And the foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs, etc., which can store program codes.

[0121] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. An implementation method of a full-twin virtual simulation test platform, characterized in that, Including: Construct a fully connected architecture by deploying station-level agents and high-speed bus networks, and converge the data of each subsystem into the station-level agent to obtain real-time data streams; Set up professional body data acquisition units according to the characteristics of each subsystem, preprocess the collected raw data, and generate a standardized data set; Establish a two-way communication channel between the industrial-level PLC controller and the station-level agent, transmit the device operation status to the agent and receive control instructions to form a closed-loop control mechanism; Build an expert mode database based on historical operation data and professional experience, express professional knowledge through a combination of semantic networks and ontology models, and create a knowledge graph architecture; Based on real-time data streams, expert mode databases, and PLC control status, construct a multi-level full-twin digital model, and use a method combining physical modeling and data-driven to perform real-time simulation on on-site devices and systems; Use a deep reinforcement learning framework combined with expert knowledge rules to perform online learning and optimization on the full-twin digital model, generate control strategies, execute through the PLC system and feedback results to form an intelligent closed-loop control, including: taking the full-twin digital model as the training environment, constructing a state space description of the subway station-level system, numerically characterizing the operating states of environmental temperature and humidity, fire protection equipment, platform screen doors, and public address systems to form a system state vector; extracting the operating criteria of the station-level system from expert knowledge rules, converting emergency evacuation rules, equipment interlock strategies, and early warning handling methods into numerical reward functions, and establishing a control objective function; deploying a deep reinforcement learning training framework on the station-level computing platform, taking the operating efficiency, energy consumption level, and safety indicators of the subway station as the optimization directions, and training a control strategy generator; performing safety verification on the instruction scheme generated by the control strategy generator, filtering illegal operations through a control instruction whitelist, and generating a safe control instruction set; sending the safe control instruction set to the industrial-level PLC controller through a two-way communication channel, and the PLC system actually controls environmental equipment, access control systems, and public address equipment; collecting the actual control effect data sent back by the PLC system, inputting the difference data between the theoretical expectation and the actual result into the reinforcement learning framework, and updating the parameters of the control strategy generator to form an intelligent closed-loop control.

2. The implementation method of the full-twin virtual simulation test platform according to claim 1, wherein The construction of a fully connected architecture by deploying station-level agents and high-speed bus networks, and converging the data of each subsystem into the station-level agent to obtain real-time data streams includes: Deploy a station-level agent with an edge computing unit in the subway station-level system, and perform preliminary filtering on the original signal through a DSP processing chip to obtain preprocessed data; Connect the station-level agent and each subsystem interface through an industrial-level high-speed bus, and sort the data packets according to priority using a real-time communication protocol to form a data transmission channel; Set data acquisition points for the basic automation system, fire alarm system, passenger information system, public address system, platform screen door system, automatic train monitoring system, and supervisory control and data acquisition system, and only transmit status change information through a differential transmission mechanism to reduce data redundancy; Set up a local cache module for the collected data, and use the data queue management algorithm to temporarily store the data in case of communication interruption and automatically reissue it after the communication is restored to ensure data integrity; Configure the internal functional units of the intelligent agent using the modular design method, interconnect each functional module through the internal bus, and support the hot plug function; Structurally process the data of each subsystem according to a unified coding format, mark the data attributes and association relationships through the self-describing data structure algorithm, and generate a real-time data stream.

3. The implementation method of the full-twin virtual simulation test platform according to claim 1, characterized in that, Set up professional body data acquisition units according to the characteristics of each subsystem, preprocess the collected raw data, and generate a standardized data set, including: Configure an environmental data acquisition unit for the basic automation system inside the rail transit station, perform linear calibration on the raw signals such as temperature, humidity, energy consumption, and lighting, and obtain standard environmental parameters; Set up a safety status acquisition unit for the fire alarm system, perform threshold screening on the smoke concentration and temperature alarm signals, and form effective alarm status data; Build an information acquisition unit for the passenger information system, classify and integrate the broadcast content and display information by time, and generate a passenger information instruction stream; Set up a data buffer in the professional body data acquisition unit, periodically store the collected status change data, and reduce the data transmission volume; Perform time synchronization processing on the data from different subsystems, sort the asynchronously collected information according to a unified clock reference, and build a data stream with consistent timing; Convert the format of all preprocessed data according to a unified field structure, perform standardized processing on numerical types, text types, and status types respectively, and generate a standardized data set.

4. The implementation method of the full-twin virtual simulation test platform according to claim 1, characterized in that Use an industrial-grade PLC controller to establish a two-way communication channel with the station-level intelligent agent, transmit the device operation status to the intelligent agent and receive control instructions to form a closed-loop control mechanism, including: Deploy an industrial-grade PLC controller at the key control nodes of the rail transit station, physically connect it to the high-speed bus through a communication expansion module, and establish a hardware communication foundation; Configure a communication protocol converter for the PLC controller to perform unified interface conversion on the OPC UA and Modbus TCP communication protocols to form a protocol compatibility layer; Build a two-way data channel structure, use the upstream channel for device operation data transmission, and use the downstream channel for control instruction transmission to generate a complete communication path; Divide the communication priority according to the urgency of the data, assign the highest transmission authority to the emergency control instructions, and achieve millisecond-level instruction response; Configure a communication redundancy mechanism to automatically switch to the standby link when the communication channel is interrupted to maintain the continuity of data transmission; Perform integrity verification on the data interaction between the PLC controller and the station-level intelligent agent, verify the data consistency through the cyclic redundancy check method, and form a closed-loop control mechanism.

5. The implementation method of the full twin virtual simulation test platform according to claim 1, characterized in that Build an expert mode database based on historical operation data and professional experience, express professional knowledge through the combination of semantic network and ontology model, and create a knowledge graph architecture, including: Extract device failure records, operation parameter curves, and alarm logs from the rail transit station-level system, segment the operation data through the time window sliding method, and identify the normal operation mode and failure operation mode of the device; Collect professional treatment cases for fire alarm handling, platform screen door fault repair, and abnormal environmental control. Convert the repair experience into processing flow data through a structured template to form a professional experience library in the rail transit field; Associate the equipment fault modes with the corresponding processing flows, and establish a direct relationship between equipment anomalies and solutions through a fault code - solution mapping table to generate a station-level system rule knowledge base; Organize the content of the rule knowledge base using the semantic triple structure of equipment - status - action. Construct a rail transit equipment semantic network by defining the causal relationships between the main equipment, status conditions, and operation actions; Build a domain ontology model based on the rail transit equipment classification system. Hierarchically divide the equipment according to the four-level structure of system - subsystem - equipment type - specific equipment, and establish the inheritance and association relationships between equipment; Fuse the rail transit equipment semantic network and the domain ontology model through concept mapping, connect the fault handling rules with the equipment hierarchical relationship, and create a station-level system knowledge graph architecture.

6. The implementation method of the full-twin virtual simulation test platform according to claim 1, wherein Construct a multi-level full-twin digital model based on real-time data streams, an expert mode database, and PLC control status. Use a method combining physical modeling and data-driven to perform real-time simulation on on-site equipment and systems, including: Extract the operation data of the rail transit platform screen door system from the real-time data stream, and construct an equipment operation status table based on the door opening / closing state, locking mechanism position, and drive motor current value to form an equipment physical layer model; Record the fire linkage control logic based on the PLC control status, and create an emergency response rule set through the correspondence between fire alarm signals and platform screen door control instructions to establish a control layer model; Construct the linkage relationship between the environmental control system and the fire protection system based on the fault handling experience in the expert mode database, and generate an emergency handling model through the collaborative work flow chart of ventilation equipment and fire protection equipment; Combine the equipment physical layer model, control layer model, and emergency handling model according to the station-level system architecture, and establish an inter-system communication relationship by defining the data exchange interface between subsystems to form a multi-level full-twin digital model; Compare the data of the multi-level full-twin digital model with the actual on-site operation status, and dynamically adjust the model parameters according to the deviation value to keep the model behavior synchronized with the actual system; Perform arithmetic processing on the multi-level full-twin digital model on the rail transit station control computing platform, and feedback the simulation results to the operator through a real-time computing engine to perform real-time simulation on on-site equipment and systems.

7. An implementation system of a full-twin virtual simulation test platform for implementing the implementation method of the full-twin virtual simulation test platform according to any one of claims 1-6, characterized in that, Including: A construction module for constructing a fully connected architecture by deploying station-level intelligent agents and a high-speed bus network, aggregating the data of each subsystem into the station-level intelligent agent to obtain real-time data streams; A processing module for setting professional body data acquisition units according to the characteristics of each subsystem, preprocessing the collected raw data, and generating a standardized data set; A receiving module for establishing a two-way communication channel between an industrial-grade PLC controller and the station-level intelligent agent, transmitting the equipment operation status to the intelligent agent and receiving control instructions to form a closed-loop control mechanism; A creation module, which is used to build an expert mode database based on historical operation data and professional experience, express professional knowledge in the way of combining a semantic network and an ontology model, and create a knowledge graph architecture; A simulation module, which is used to build a multi-level full-twin digital model based on real-time data streams, the expert mode database and the PLC control status, and perform real-time simulation on on-site devices and systems by combining physical modeling and data-driven methods; An optimization module, which is used to perform online learning and optimization on the full-twin digital model by using a deep reinforcement learning framework combined with expert knowledge rules, generate control strategies, execute and feedback results through the PLC system, and form an intelligent closed-loop control, including: taking the full-twin digital model as a training environment, constructing a state space description of a subway station-level system, numerically characterizing the operating states of environmental temperature and humidity, fire-fighting equipment, platform screen doors, and public address systems to form a system state vector; extracting the operating criteria of the station-level system from the expert knowledge rules, converting emergency evacuation rules, equipment linkage strategies, and early warning disposal methods into numerical reward functions, and establishing a control objective function; deploying a deep reinforcement learning training framework on a station-level computing platform, taking the operation efficiency, energy consumption level, and safety indicators of subway stations as the optimization directions, and training a control strategy generator; performing safety verification on the instruction scheme generated by the control strategy generator, filtering illegal operations through a control instruction whitelist, and generating a set of safe control instructions; sending the set of safe control instructions to an industrial-level PLC controller through a bidirectional communication channel, and actually controlling environmental equipment, access control systems, and public address equipment by the PLC system; collecting the actual control effect data sent back by the PLC system, inputting the difference data between the theoretical expectation and the actual result into the reinforcement learning framework, and updating the parameters of the control strategy generator to form an intelligent closed-loop control.

8. A computer device, characterized in that, It includes a memory and a processor, and the memory stores a computer program that can run on the processor. It is characterized in that when the processor executes the computer program, it implements the implementation method of the full-twin virtual simulation test platform according to any one of claims 1 to 6.

9. A computer-readable storage medium, on which a computer program is stored, and when the computer program is run by a processor, the processor is caused to execute the implementation method of the full-twin virtual simulation test platform according to any one of claims 1 to 6.

Images