A railway safety processing method, device and system based on multi-agent cooperation

Abstract

The present application relates to the technical field of railway safety transportation, and discloses a railway safety processing method, device and system based on multi-agent cooperation. According to the received railway safety related task, the meta-cognition arranger dynamically forms a virtual cooperation group including multiple railway task agents, and generates an initial workflow. The meta-cognition arranger drives each railway task agent to cooperatively operate based on the initial workflow and a communication bus, and performs interim reasoning and path optimization during the cooperative operation of each railway task agent. The meta-cognition arranger generates a final decision report according to the output data of each railway task agent. The present application can encapsulate different applications as agents, drive multiple agents to actively cooperate and negotiate to process railway safety related tasks, and generate a final decision report according to the output data of the multiple agents, effectively realizing dynamic cooperation and adaptive global optimization of multiple specialties and multiple tasks.

Description

Technical Field

[0001] This invention relates to the field of railway transportation safety technology, and in particular to a railway safety processing method, apparatus and system based on multi-agent cooperation. Background Technology

[0002] Currently, the railway transportation system is developing rapidly towards high density, high speed and intelligence, and the requirements for railway operation safety and transportation efficiency are also getting higher and higher.

[0003] Railway safety systems in related technologies can perform multimodal and multidimensional data acquisition and event analysis, such as video analytics and sensor diagnostics, to identify risks and respond to them.

[0004] However, while the various applications in the related technologies (such as video analytics and sensor diagnostics) can detect problems, they cannot proactively collaborate, discuss, and generate comprehensive response strategies, making it difficult to form a globally optimized security solution. Summary of the Invention

[0005] This invention provides a railway safety handling method, device, and system based on multi-agent collaboration, which addresses the shortcomings of related technologies where applications cannot proactively collaborate, discuss, and generate comprehensive response strategies, making it difficult to form a globally optimized safety handling plan. By encapsulating different applications as agents, the invention drives multiple agents to proactively collaborate and discuss to handle railway safety-related tasks, and generates a final decision report based on the output data of the multiple agents, effectively achieving dynamic collaboration and adaptive global optimization across multiple disciplines and tasks.

[0006] In a first aspect, the present invention provides a railway safety processing method based on multi-agent cooperation, applied to a metacognitive orchestrator, the method comprising:

[0007] The metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on the received railway safety-related tasks, and generates an initial workflow; wherein, the initial workflow includes the execution order, data dependencies and / or interaction protocols between each railway task agent; The metacognitive orchestrator drives each railway task agent to operate collaboratively based on the initial workflow and communication bus, and performs mid-way reasoning and path optimization during the collaborative operation of each railway task agent. The metacognitive orchestrator generates a final decision report based on the output data of each railway task agent.

[0008] Optionally, the metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on the received railway safety-related tasks, including: The metacognitive orchestrator acquires the role and capability data of each registered railway task agent; The metacognitive orchestrator selects multiple registered railway task agents that match the railway safety-related tasks based on the railway safety-related tasks and the role and capability data of each registered railway task agent, and as a whole, forms the virtual collaborative group.

[0009] Optionally, the output data includes the analysis results, intermediate conclusions, and discussion records between different railway task agents for each agent. The metacognitive orchestrator generates a final decision report based on the output data of each railway task agent, including: The metacognitive orchestrator generates a structured analysis report, including root cause inference, multi-evidence chain support, associated risk warnings, and suggested handling measures, based on the analysis results, intermediate conclusions, and discussion records between different railway task agents, and serves as the final decision report.

[0010] Secondly, the present invention provides a railway safety processing device based on multi-agent cooperation, applied to a metacognitive orchestrator, the device comprising: The response unit is used to dynamically form a virtual collaborative group comprising multiple railway task agents based on the received railway safety-related tasks, and generate an initial workflow; wherein the initial workflow includes the execution order, data dependencies and / or interaction protocols between each railway task agent; The driving unit is used to drive each of the railway task agents to operate collaboratively based on the initial workflow and communication bus; The optimization unit is used to perform mid-course reasoning and path optimization during the collaborative operation of each railway task agent; The generation unit is used to generate a final decision report based on the output data of each railway task agent.

[0011] Thirdly, the present invention provides a railway safety processing system based on multi-agent cooperation, comprising: A communication bus and multiple railway task agents; wherein each railway task agent encapsulates specific railway safety analysis capabilities, and each railway task agent registers and communicates through the communication bus; The metacognitive orchestrator is used to receive railway safety-related tasks, dynamically organize multiple related railway task agents into a virtual collaborative group, and orchestrate the execution logic of each railway task agent in the virtual collaborative group to complete the railway safety-related tasks.

[0012] Optionally, the metacognitive orchestrator includes: The task parsing module is used to understand the task intent of the railway safety-related tasks; The agent scheduling module is used to dynamically form the virtual cooperative group based on the capability description of each railway task agent; The workflow generation module is used to generate dynamic workflow diagrams that include execution order and data dependencies; The conflict resolution module is used to dynamically adjust the execution path or introduce new railway task agents based on intermediate results during workflow execution.

[0013] Optionally, the railway task intelligent agent includes a role declaration module, a capability description module, and a state management module; The communication bus uses an enhanced Model Context Protocol (MCP) to support the railway task agent in broadcasting and subscribing to events based on semantic topics.

[0014] Optionally, the system further includes: A capability repository is used to store workflow templates abstracted from historical success cases; A cloud-based simulation platform is used to build digital twin models with railway environments, as well as to test and optimize new railway task agents or workflow templates.

[0015] Optionally, the system further includes: The interface adapter module is used to connect to the underlying artificial intelligence (AI) model or business processing layer. The role capability modeling module is used to declare the professional roles and structured capability descriptions of the railway task agent; The state machine management module is used to manage the working state of the railway task agent; The communication module is used for event-driven message interaction with external systems via the enhanced Model Context Protocol (MCP).

[0016] Optionally, the system further includes: A digital twin model is used to connect multiple railway task agents, simulate preset complex fault or risk scenarios, record and analyze the interaction data and decision results of each railway task agent in the simulated scenario, and optimize the collaboration rules or workflow templates between agents based on the analysis results.

[0017] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the railway safety processing method based on multi-agent cooperation described in the first aspect or any corresponding embodiment thereof.

[0018] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the railway safety processing method based on multi-agent cooperation described in the first aspect or any corresponding embodiment thereof.

[0019] This invention provides a railway safety processing method, apparatus, and system based on multi-agent collaboration. A metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on received railway safety-related tasks, and generates an initial workflow. The metacognitive orchestrator drives each railway task agent to collaboratively operate based on the initial workflow and communication bus, performing mid-process inference and path optimization during the collaborative operation of each agent. The metacognitive orchestrator generates a final decision report based on the output data of each railway task agent. This invention encapsulates different applications as agents, driving multiple agents to actively collaborate and deliberate to handle railway safety-related tasks, and generates a final decision report based on the output data of multiple agents, effectively achieving dynamic collaboration and adaptive global optimization across multiple disciplines and tasks.

[0020] This invention can also achieve the following technical effects: Improved early warning accuracy: Multimodal data fusion reduces the false alarm rate by more than 70% (actual data), especially in low visibility environments such as fog and night; Reduced response time: Edge-side local decision-making enables emergency braking within 500ms, which is 6 times faster than traditional systems; Reduce operation and maintenance costs: Through the MCP standardized interface, the time to access the new model has been shortened from an average of 3 weeks to less than 1 day; Enhance system scalability: Support rapid integration of third-party AI capabilities (such as third-party geological landslide prediction models) and build an open ecosystem; Improve decision-making transparency: Large model outputs include a chain of evidence (such as which regulations were cited or which databases were used), facilitating auditing and accountability; Achieve closed-loop management throughout the entire process: from "identifying problems" to "generating solutions" and then to "implementing verification," the entire process is digitized, automated, and traceable. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1A flowchart illustrating a railway safety processing method based on multi-agent cooperation, provided as an embodiment of the present invention; Figure 2 A content data table for a three-tier decision-making architecture of edge-region-cloud provided in an embodiment of the present invention; Figure 3 A schematic diagram of a railway safety processing device based on multi-agent cooperation provided in an embodiment of the present invention; Figure 4 A schematic diagram of a railway safety processing system based on multi-agent cooperation provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] It should be noted that the railway safety system based on relevant technologies is not only a data silo, but also a decision-making silo and a response silo. Various applications (such as video analytics and sensor diagnostics) are dormant experts; they can identify problems, but they cannot proactively collaborate, discuss, and generate comprehensive response strategies. The railway safety early warning system faces three fundamental bottlenecks: (1) Data silos are severe, and heterogeneous systems are difficult to coordinate. Subsystems such as video surveillance, track sensors, weather stations, and positioning operate independently, using different protocols and lacking a unified data context description mechanism. Multimodal data (images, text, time series) cannot be aligned, resulting in "seeing but not understanding".

[0025] (2) Intelligent algorithms operate in a closed system, resulting in high model call costs. Single-point artificial intelligence (AI) applications are widespread (e.g., only used for video intrusion detection), but models cannot be linked together. Model deployment relies on dedicated hardware and customized software stacks, and redevelopment is required when changing scenarios, forming "model silos". There are no standard interfaces between business systems (such as scheduling, customer service, and maintenance) and AI models, making it difficult to implement AI outputs.

[0026] (3) The decision-making chain is broken, resulting in large response delays. The system chain of "collection → upload → central analysis → command issuance" for related technologies is long, and the end-to-end delay generally exceeds 3 seconds, making it difficult to meet the emergency braking (<500ms) requirement. The lack of a hierarchical decision-making mechanism means that all data is sent to the cloud, causing network congestion and resource waste.

[0027] (4) Communication protocols are not suitable for large-scale model requirements. Message Queuing Telemetry Transport (MQTT) is suitable for lightweight telemetry, but does not support the transmission of complex structured contexts. Some protocols focus on industrial control and lack support for natural language and semantic understanding. Some protocols are flexible but highly coupled, making it difficult to support dynamic tool discovery and invocation.

[0028] Therefore, a new protocol framework is urgently needed that can both support the understanding capabilities of large models and connect real-world sensors, actuators, and business systems to solve the following problems: how to break down the fragmented state of "data → model → decision → execution" in the railway system; how to achieve efficient alignment and fusion of multi-source heterogeneous data at the spatiotemporal and semantic levels; how to support low-latency, high-reliability, and interpretable dynamic decision-making of large models in railway safety scenarios; and how to build an open and scalable intelligent railway ecosystem to avoid reinventing the wheel.

[0029] The following is combined with Figures 1-2 This invention describes a railway safety processing method based on multi-agent cooperation.

[0030] like Figure 1 As shown, this embodiment proposes a first railway safety processing method based on multi-agent cooperation, which may include the following steps: S101. The metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on the received railway safety-related tasks, and generates an initial workflow. This initial workflow includes the execution order, data dependencies, and / or interaction protocols between each railway task agent.

[0031] Among them, the metacognitive orchestrator is responsible for handling complex tasks across agents. For complex non-urgent tasks that require multimodal collaborative analysis, the metacognitive orchestrator coordinates the process at the regional / cloud center. For defined emergency events (such as obstacles), the edge agents directly trigger braking according to preset rules to ensure real-time performance.

[0032] Specifically, this embodiment can receive railway safety-related tasks, form a virtual collaboration group based on railway safety-related task components, and generate an initial workflow.

[0033] Optionally, step S101 includes: The metacognitive orchestrator acquires the role and capability data of each registered railway task agent; The metacognitive orchestrator selects multiple registered railway task agents that match railway safety-related tasks based on the role and capability data of each registered railway task agent, and treats them as a virtual collaborative group.

[0034] Optionally, the initial workflow may include the execution order, data dependencies, and / or interaction protocols between each railway task agent.

[0035] The interaction protocol can be the Model Context Protocol (MCP).

[0036] It should be noted that this embodiment does not simply allow large models to mechanically call tools, but rather endows each specialized functional module (such as "track inspector," "weather analyst," and "dispatch advisor") with an intelligent agent personality. Each AI model or business system connected to the system (such as a video analytics server, wind speed sensor interface, and regulations database) is no longer exposed as a simple "function interface," but is encapsulated as an "intelligent agent wrapper based on a role-capability-state model." This wrapper includes: The character description reads: "I am an expert in track deformation monitoring, specializing in analyzing fiber optic grating waveform data and assessing track health indices." Capability List: Declare the types of events that can be processed (such as event_type: track_deformation), input and output formats, and confidence calculation methods in a structured description (based on MCP Schema enhancement).

[0037] Internal state machine: The agent has states such as "idle", "monitoring", "diagnosing", and "cooperating". For example, when an anomaly is detected, the state changes from "monitoring" to "diagnosing" and triggers autonomous in-depth analysis.

[0038] This embodiment endows the cold algorithm module with "identity" and "proactivity," laying the foundation for subsequent anthropomorphic collaboration. This differs from microservices or function calls in related technologies; it is an entity abstraction oriented towards task collaboration.

[0039] This embodiment allows for the establishment of proactive notification and subscription mechanisms among different intelligent agents. Based on the MCP protocol, this embodiment can design an "event broadcasting" and "interest subscription" mechanism. For example, when the "track deformation agent" diagnoses a "level 2 risk," it not only reports the result but also broadcasts a semantic event via MCP: "Topic: Track health alarm; Location: K125+300; Severity: Level 2; Possible associations: geological subsidence, abnormal train load." Other intelligent agents, such as the "geological risk assessment agent" and the "train operation monitoring agent," if they have subscribed to "track health" or related topics in advance, will proactively receive this notification and trigger their own correlation analysis.

[0040] This embodiment can transform "central scheduling" into "proactive collaboration", reducing the decision-making pressure on the central node, improving the system's speed of detecting associated risks, and realizing the application of the MCP protocol in distributed sensing scenarios.

[0041] Regarding the initial workload, the metacognitive orchestrator (which is itself a special intelligent agent driven by a large model) can be activated when it receives a high-level task or complex alarm (such as a dispatcher's voice command: "Inspect why there are multiple train swaying incidents at K125+300 section"). First, it decomposes the task intent, then scans the "role-capability" list of all currently registered agents, dynamically assembling a temporary "virtual expert group" like a "director casting." For example, the group assembled for the above task might include: a "track deformation agent," a "video image analysis agent," a "historical fault retrieval agent," and a "real-time train data agent." The metacognitive orchestrator also generates a dynamic directed acyclic graph workflow, defining the execution order, data dependencies, and interaction protocols between agents (e.g., first, synchronously analyze video and track data, using the results as keywords to trigger historical fault retrieval).

[0042] S102, the metacognitive orchestrator drives each railway task agent to operate collaboratively based on the initial workflow and communication bus.

[0043] Specifically, the metacognitive orchestrator can mobilize each railway task agent to operate collaboratively based on the initial workflow and communication bus.

[0044] S103, the metacognitive orchestrator performs mid-course reasoning and path optimization during the collaborative operation of each railway task agent.

[0045] This embodiment addresses mid-process reasoning and path optimization within the workflow, which is not statically executed. During execution, the output of any agent may be evaluated midway. For example, if the "video analysis agent" reports "no visible foreign object detected," but the "orbit deformation agent" insists "deformation continues to expand," the metacognitive orchestrator will activate a conflict resolution mechanism. The metacognitive orchestrator may dynamically insert a "high-precision lidar verification agent" into the workflow, or command the two agents to exchange raw data for qualitative analysis. The metacognitive orchestrator in this embodiment is a workflow system with online reasoning and self-adjustment capabilities, rather than a predefined fixed process. It can leverage the logical reasoning capabilities of a large model to manage the collaborative process of lower-level agents in real time, addressing uncertainties.

[0046] S104. The metacognitive orchestrator generates a final decision report based on the output data of each railway task agent.

[0047] Optionally, the output data includes the analysis results, intermediate conclusions, and discussion records between different railway task agents for each agent. Step S104 includes: The metacognitive orchestrator generates a structured analysis report based on the analysis results, intermediate conclusions, and discussion records between different railway task agents. This report includes root cause inferences, multi-evidence chain support, associated risk warnings, and suggested handling measures, and serves as the final decision report.

[0048] Specifically, the comprehensive report on workflow results, including the analysis results, intermediate conclusions, and even discussion records among all participating agents (retained in the form of MCP messages), will be submitted to the metacognitive orchestrator. The metacognitive orchestrator can then synthesize this information to generate a structured analysis report, including: root cause inferences, multi-chain evidence support, and associated risk warnings (e.g., "The main cause of the train swaying is local track subsidence (confidence level 85%), which may be related to recent continuous rainfall. It is recommended to trigger the ground-penetrating radar re-examination workflow and notify the maintenance section to limit speed during this period."). It should be noted that the final output is not simply data, but a logical, evidence-based, and traceable "expert consultation conclusion," greatly improving the transparency and credibility of decision-making.

[0049] It should be noted that this embodiment can realize a fully intelligent railway system architecture based on large-scale model technology and MCP, especially focusing on applying MCP to data fusion, real-time decision-making, and cross-modal collaborative control in railway safety early warning systems, achieving closed-loop intelligent management from perception to execution. It integrates large-scale pre-trained models (large language model + multimodal large model), industrial communication protocol optimization, edge computing and cloud computing collaboration, and real-time response mechanisms for safety-critical systems, making it suitable for intelligent safety monitoring and dynamic scheduling systems in complex operating environments such as high-speed railways, urban rail transit, and heavy-haul freight railways.

[0050] This embodiment executes... Figure 1 The method shown establishes a railway full-service intelligent agent system with the MCP protocol as the nervous system and a large model as the brain, realizing the leap from passive alarm to active prediction and collaborative response.

[0051] Optionally, this embodiment can train the agent. Regarding the accumulation and reuse of agent capabilities, the workflow (including agent combination, interaction logic, and decision results) of each successful resolution of complex events is desensitized and abstracted, and accumulated into a "tactical workflow template" and stored in a "capability repository". When similar scenarios reappear, the metacognitive orchestrator can prioritize recommending or directly reuse the validated template, significantly improving response efficiency. For example, "responding to the risk of slope landslides during the rainy season" can become a standard template.

[0052] Optionally, in other railway safety processing methods based on multi-agent collaboration proposed in this embodiment, cloud-based digital twin simulation and training can also be performed. In the cloud, a digital twin environment for the railway line is constructed using historical data and physical models. New "agents" (new algorithm models) or new "workflow templates" are periodically injected into this environment to simulate massive complex scenarios (such as extreme weather and cascading equipment failures) for stress testing and collaborative drills. Through simulation, the impact of adding new agents on the overall system performance can be evaluated, the collaboration strategies between agents can be optimized, and the overall intelligence level of the system can be "evolved offline." This embodiment expands the system's evolution from a single "model parameter update" to "collaboration strategy optimization" and "organizational structure learning," achieving the core guarantee of "full-service intelligence."

[0053] This embodiment realizes the role-based encapsulation and state management of professional intelligent agents in the railway field. It encapsulates the artificial intelligence (AI) functional modules into proactive collaborative entities with role declarations, capability lists, and internal state machines, and performs identity registration and event notification through the enhanced MCP protocol.

[0054] This embodiment also implements metacognitive dynamic workflow orchestration based on a large language model. It designs a top-level intelligent agent that can analyze intents in real time according to complex tasks, dynamically form virtual groups of intelligent agents, generate and execute workflow diagrams that can be adjusted midway, and integrate the outputs of each intelligent agent to generate a final decision report.

[0055] This embodiment employs an active subscription-broadcast collaborative mechanism among intelligent agents. An event-driven communication layer is built on the MCP protocol, allowing intelligent agents to subscribe to events based on semantic interests, thereby enabling proactive and rapid discovery of risk associations and forming a distributed collaborative perception network.

[0056] This embodiment of the railway safety intelligent system's capability repository and simulation evolution system condenses successful workflow cases into reusable templates, and utilizes a cloud-based digital twin environment to perform simulation training and collaborative strategy optimization on intelligent agent clusters, thereby achieving continuous evolution of system-level capabilities.

[0057] The railway safety processing method based on multi-agent collaboration proposed in this embodiment can be applied to a metacognitive orchestrator. The metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on received railway safety-related tasks and generates an initial workflow. The metacognitive orchestrator drives each railway task agent to collaborate based on the initial workflow and communication bus, performing mid-way inference and path optimization during the collaborative operation of each agent. Based on the output data of each railway task agent, the metacognitive orchestrator generates a final decision report. This embodiment can encapsulate different applications as agents, driving multiple agents to actively collaborate and discuss to handle railway safety-related tasks, and generating a final decision report based on the output data of multiple agents, effectively achieving dynamic collaboration and adaptive global optimization across multiple disciplines and tasks.

[0058] The railway safety processing method based on multi-agent cooperation proposed in this embodiment can also achieve the following technical effects: Improved early warning accuracy: Multimodal data fusion reduces the false alarm rate by more than 70% (actual data), especially in low visibility environments such as fog and night; Reduced response time: Edge-side local decision-making enables emergency braking within 500ms, which is 6 times faster than traditional systems; Reduce operation and maintenance costs: Through the MCP standardized interface, the time to access the new model has been shortened from an average of 3 weeks to less than 1 day; Enhance system scalability: Support rapid integration of third-party AI capabilities (such as third-party geological landslide prediction models) and build an open ecosystem; Improve decision-making transparency: Large model outputs include a chain of evidence (such as which regulations were cited or which databases were used), facilitating auditing and accountability; Achieve closed-loop management throughout the entire process: from "identifying problems" to "generating solutions" and then to "implementing verification," the entire process is digitized, automated, and traceable.

[0059] based on Figure 1 In the second railway safety processing method based on multi-agent cooperation proposed in this embodiment, this embodiment can realize a railway safety early warning system architecture based on the MCP protocol, specifically including: (1) Perception Layer: Multimodal Data Acquisition and MCP Encapsulation. Deploy BeiDou / Global Navigation Satellite System (GNSS) high-precision positioning terminals, fiber Bragg grating strain sensors, 4K 60fps video surveillance, lidar, weather stations, and other equipment; all perception nodes encapsulate data through a lightweight MCP client, adding spatiotemporal tags, device identity, confidence level, and metadata description; data is transmitted to the decision layer via the MCP protocol, supporting hierarchical transmission (such as prioritizing bandwidth for emergency events). For example, when a small deformation occurs in a certain section of the track, the fiber Bragg grating sensor detects a wavelength shift, and the MCP client automatically adds position coordinates, sampling time, threshold deviation degree, and triggers an "orbit health status change" event.

[0060] (2) Decision-making level: Large-model-driven cross-modal fusion and hierarchical decision-making. For example... Figure 2 As shown, a three-tier decision-making architecture of "edge-region-cloud" is adopted. Key mechanisms include: all models are accessed through the MCP server, exposing their capabilities in the form of intelligent agents; large models dynamically select and call multiple intelligent agents according to the context (e.g., first call the geographic information system to obtain terrain, then call the meteorological application port to obtain wind speed, and finally generate scheduling suggestions); it supports retrieval-enhanced generation, and accesses historical accident databases, regulatory documents, and maintenance records to improve the compliance and interpretability of output.

[0061] (3) Execution layer: closed-loop control and MCP feedback. Decision instructions are sent to the vehicle system, dispatch terminal, engineering management system, etc. through the MCP protocol; execution results (such as braking completion, speed adjustment) are returned through MCP to form a closed loop; abnormal non-execution situations trigger secondary alarms and manual takeover procedures.

[0062] (4) Core Enhancement Technologies of MCP Protocol. In response to the special needs of railways, the original MCP protocol is optimized as follows: spatiotemporal alignment engine; introduction of Precision Time Protocol (PTP) synchronization mechanism to achieve a clock deviation of <1ms for all network devices; embedding WGS84 coordinates and UTC timestamps in data packets to support cross-modal data alignment; dynamic knowledge evolution mechanism; the model receives an incremental training signal every minute (small sample fine-tuning task based on newly collected data); push parameter differential updates through MCP to achieve online learning closed loop; security and encryption; support for 5G slicing to ensure that the transmission delay of key instructions is <10ms; use of national cryptographic encryption algorithm + two-way certificate authentication to prevent man-in-the-middle attacks and data tampering.

[0063] It should be noted that this embodiment does not simply register the model as an MCP tool, but rather upgrades it into a professional intelligent agent with "perception-thinking-action-collaboration" capabilities. These intelligent agents are then dynamically organized through a "metacognitive workflow orchestrator" to resolve complex, cross-modal, and multifaceted security incidents.

[0064] It should also be noted that the following technical solutions in related technologies have obvious defects compared with this embodiment. Specifically: Pure rule-based engine systems in related technologies have major drawbacks: they cannot handle ambiguous scenarios (such as "suspected foreign object intrusion"); they have extremely high maintenance costs; and they struggle to cope with new types of risks. They are only suitable for simple logical judgments and cannot handle complex uncertainties.

[0065] The main drawbacks of artificial intelligence platforms in related technologies (such as TensorFlow Serving + Kafka) are: lack of a unified semantic interface; independent deployment of each model; inability to support natural language interaction and dynamic orchestration; and fragmented toolchains that still require a lot of custom development and cannot achieve "plug and play".

[0066] The main drawbacks of microservice architectures based on API gateways in related technologies are: cumbersome interface definitions; complex version management; difficulty in supporting context-aware calls; high coupling; lack of large-model friendliness; and inability to achieve "intent-driven" service composition.

[0067] Other AI proxy protocols used in related technologies (such as OpenAI Plugins and JSON Schema-based Tools) have major drawbacks: closed ecosystems (e.g., limited to the OpenAI ecosystem); non-open protocols; and lack of support for industrial-grade reliability requirements. This does not align with the railway industry's strategic needs for independent control, security, and reliability.

[0068] The relevant technologies rely entirely on 5G+ multi-access edge computing for localized processing. The main drawback is that they only solve edge computing problems and do not address the essence of model collaboration and knowledge evolution. They remain "capability silos" and cannot achieve global optimization and cross-regional collaboration.

[0069] This embodiment, through the combination of the MCP protocol and a large model, can simultaneously meet the four core requirements of "openness, intelligence, low latency, and evolvability," and can effectively support the intelligent upgrade of the entire railway business chain.

[0070] like Figure 3 As shown, this embodiment proposes a railway safety processing device based on multi-agent cooperation, applied to a metacognitive orchestrator. The device includes: The response unit 101 is used to dynamically form a virtual collaborative group including multiple railway task agents according to the received railway safety-related tasks, and generate an initial workflow; wherein, the initial workflow includes the execution order, data dependencies and / or interaction protocols between each railway task agent; Drive unit 102 is used to drive each railway task agent to operate collaboratively based on the initial workflow and communication bus; The optimization unit 103 is used to perform mid-way reasoning and path optimization during the collaborative operation of each railway task agent; The generation unit 104 is used to generate a final decision report based on the output data of each railway task agent.

[0071] It should be noted that the processing procedures of the response unit 101, the driving unit 102, the optimization unit 103, and the generation unit 104, and their beneficial effects, can be referred to respectively. Figure 1 Steps S101 to S104 in the process will not be described again.

[0072] Optionally, response unit 101 is also used for: Obtain the role and capability data of each registered railway task agent; Based on railway safety-related tasks and the role and capability data of each registered railway task agent, multiple registered railway task agents that match the railway safety-related tasks are selected and collectively formed as a virtual collaborative group.

[0073] Optionally, the output data may include the analysis results, intermediate conclusions, and discussion records between different railway task agents for each railway task agent. The generating unit 104 is also used for: Based on the analysis results, intermediate conclusions, and discussion records between different railway task agents, a structured analysis report is generated, including root cause inferences, multi-evidence chain support, associated risk warnings, and suggested handling measures, and serves as the final decision report.

[0074] The railway safety processing device based on multi-agent collaboration proposed in this embodiment can be applied to a metacognitive orchestrator. The metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on received railway safety-related tasks and generates an initial workflow. The metacognitive orchestrator drives each railway task agent to collaborate based on the initial workflow and communication bus, performing mid-process inference and path optimization during the collaborative operation of each agent. Based on the output data of each railway task agent, the metacognitive orchestrator generates a final decision report. This embodiment can encapsulate different applications as agents, driving multiple agents to actively collaborate and discuss to process railway safety-related tasks, and generating a final decision report based on the output data of multiple agents, effectively achieving dynamic collaboration and adaptive global optimization across multiple disciplines and tasks.

[0075] The railway safety processing device based on multi-agent cooperation in this embodiment is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0076] like Figure 4 As shown, this embodiment also proposes a railway safety processing system based on multi-agent cooperation, including: A communication bus and multiple railway task agents; each railway task agent encapsulates specific railway safety analysis capabilities, and each railway task agent registers and communicates through the communication bus; The metacognitive orchestrator is used to receive railway safety-related tasks, dynamically organize multiple railway task agents into a virtual collaborative group, and orchestrate the execution logic of each railway task agent in the virtual collaborative group to complete the railway safety-related tasks.

[0077] Optional metacognitive orchestrators include: The task parsing module is used to understand the task intent of railway safety-related tasks; The agent scheduling module is used to dynamically form virtual collaborative groups based on the capability description of each railway task agent; The workflow generation module is used to generate dynamic workflow diagrams that include execution order and data dependencies; The conflict resolution module is used to dynamically adjust the execution path or introduce new railway task agents based on intermediate results during workflow execution.

[0078] Optionally, the railway task intelligent agent includes a role declaration module, a capability description module, and a state management module; The communication bus uses an enhanced Model Context Protocol (MCP) to support railway task agents in broadcasting and subscribing to events based on semantic topics.

[0079] Optionally, the system may also include: A capability repository is used to store workflow templates abstracted from historical success cases; A cloud-based simulation platform is used to build digital twin models with railway environments, as well as to test and optimize new railway task agents or workflow templates.

[0080] Optionally, the system may also include: The interface adapter module is used to connect to the underlying artificial intelligence (AI) model or business processing layer. The role capability modeling module is used to declare the professional roles and structured capability descriptions of railway task agents; The state machine management module is used to manage the working state of the railway task intelligent agent; The communication module is used for event-driven message interaction with external systems via the enhanced Model Context Protocol (MCP).

[0081] Optionally, the system may also include: Digital twin models are used to connect multiple railway task agents, simulate pre-set complex fault or risk scenarios, record and analyze the interaction data and decision results of each railway task agent in the simulated scenario, and optimize the collaboration rules or workflow templates between agents based on the analysis results.

[0082] The railway safety processing system based on multi-agent collaboration proposed in this embodiment can be applied to a metacognitive orchestrator. The metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on received railway safety-related tasks and generates an initial workflow. The metacognitive orchestrator drives each railway task agent to collaborate based on the initial workflow and communication bus, performing mid-process inference and path optimization during the collaborative operation of each agent. Based on the output data of each railway task agent, the metacognitive orchestrator generates a final decision report. This embodiment can encapsulate different applications as agents, driving multiple agents to actively collaborate and discuss to handle railway safety-related tasks, and generating a final decision report based on the output data of multiple agents, effectively achieving dynamic collaboration and adaptive global optimization across multiple disciplines and tasks.

[0083] This invention also provides a computer device having the above-described features. Figure 3 The diagram shows a railway safety processing device based on multi-agent cooperation.

[0084] Please see Figure 5The present invention provides a schematic diagram of the structure of a computer device according to an optional embodiment. The computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, multiple processors and / or multiple buses can be used with multiple memories, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 5 Take a processor 10 as an example.

[0085] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0086] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0087] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function. The data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0088] Memory 20 may include volatile memory, such as random access memory. Memory may also include non-volatile memory, such as flash memory, hard disk, or solid-state drive. Memory 20 may also include combinations of the above types of memory.

[0089] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0090] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A railway safety processing method based on multi-agent cooperation, characterized in that, Applied to metacognitive orchestrators, the method includes: The metacognitive orchestrator dynamically assembles a virtual collaborative group comprising multiple railway task agents based on the received railway safety-related tasks, and generates an initial workflow; wherein, the initial workflow includes the execution order, data dependencies and / or interaction protocols between each railway task agent; The metacognitive orchestrator drives each railway task agent to operate collaboratively based on the initial workflow and communication bus, and performs mid-way reasoning and path optimization during the collaborative operation of each railway task agent. The metacognitive orchestrator generates a final decision report based on the output data of each railway task agent.

2. The method according to claim 1, characterized in that, The metacognitive orchestrator dynamically assembles virtual collaborative groups comprising multiple railway task agents based on received railway safety-related tasks, including: The metacognitive orchestrator acquires the role and capability data of each registered railway task agent; The metacognitive orchestrator selects multiple registered railway task agents that match the railway safety-related tasks based on the railway safety-related tasks and the role and capability data of each registered railway task agent, and as a whole, forms the virtual collaborative group.

3. The method according to claim 1, characterized in that, The output data includes the analysis results, intermediate conclusions, and discussion records between different railway task agents for each of the railway task agents. The metacognitive orchestrator generates a final decision report based on the output data of each railway task agent, including: The metacognitive orchestrator generates a structured analysis report, including root cause inference, multi-evidence chain support, associated risk warnings, and suggested handling measures, based on the analysis results, intermediate conclusions, and discussion records between different railway task agents, and serves as the final decision report.

4. A railway safety processing device based on multi-agent cooperation, characterized in that, Applied to a metacognitive orchestrator, the device includes: The response unit is used to dynamically form a virtual collaborative group comprising multiple railway task agents based on the received railway safety-related tasks, and generate an initial workflow; wherein the initial workflow includes the execution order, data dependencies and / or interaction protocols between each railway task agent; The driving unit is used to drive each of the railway task agents to operate collaboratively based on the initial workflow and communication bus; The optimization unit is used to perform mid-course reasoning and path optimization during the collaborative operation of each railway task agent; The generation unit is used to generate a final decision report based on the output data of each railway task agent.

5. A railway safety processing system based on multi-agent cooperation, characterized in that, include: A communication bus and multiple railway task agents; wherein each railway task agent encapsulates specific railway safety analysis capabilities, and each railway task agent registers and communicates through the communication bus; The metacognitive orchestrator is used to receive railway safety-related tasks, dynamically organize multiple related railway task agents into a virtual collaborative group, and orchestrate the execution logic of each railway task agent in the virtual collaborative group to complete the railway safety-related tasks.

6. The system according to claim 5, characterized in that, The metacognitive orchestrator includes: The task parsing module is used to understand the task intent of the railway safety-related tasks; The agent scheduling module is used to dynamically form the virtual cooperative group based on the capability description of each railway task agent; The workflow generation module is used to generate dynamic workflow diagrams that include execution order and data dependencies; The conflict resolution module is used to dynamically adjust the execution path or introduce new railway task agents based on intermediate results during workflow execution.

7. The system according to claim 5, characterized in that, The railway task intelligent agent includes a role declaration module, a capability description module, and a state management module; The communication bus uses an enhanced Model Context Protocol (MCP) to support the railway task agent in broadcasting and subscribing to events based on semantic topics.

8. The system according to claim 5, characterized in that, The system also includes: A capability repository is used to store workflow templates abstracted from historical success cases; A cloud-based simulation platform is used to build digital twin models with railway environments, as well as to test and optimize new railway task agents or workflow templates.

9. The system according to claim 5, characterized in that, The system also includes: The interface adapter module is used to connect to the underlying artificial intelligence (AI) model or business processing layer. The role capability modeling module is used to declare the professional roles and structured capability descriptions of the railway task agent; The state machine management module is used to manage the working state of the railway task agent; The communication module is used for event-driven message interaction with external systems via the enhanced Model Context Protocol (MCP).

10. The system according to claim 5, characterized in that, The system also includes: A digital twin model is used to connect multiple railway task agents, simulate preset complex fault or risk scenarios, record and analyze the interaction data and decision results of each railway task agent in the simulated scenario, and optimize the collaboration rules or workflow templates between agents based on the analysis results.

Images