Railway maintenance robot cooperative scheduling method and system based on AI agent

Abstract

This invention provides a collaborative scheduling method and system for railway maintenance robots based on AI agents. Each railway maintenance robot detects heterogeneous railway maintenance robots in its neighborhood, generating a real-time capability sharing map of the neighborhood. The original task-executing robot, based on the demand benchmark of the current pre-set maintenance task to be executed, simulates task deduction and predicts its own potential capability blind spots and operational risk factors, outputting multi-dimensional task adaptation gap prediction descriptions and associated risk warning reference information. The original task-executing robot performs cross-robot capability complementarity matching calculations, locates candidate maintenance robots with precise replacement capabilities, and sends data packets carrying collaborative replacement request packets to these robots, enabling them to output standardized replacement collaborative consensus results. Based on this, collaborative scheduling instructions are generated to complete the collaborative execution of maintenance operations. This invention improves the overall reliability and completion quality of railway maintenance operations.

Description

Technical Field

[0001] This invention relates to the field of robot scheduling, and more specifically, to a collaborative scheduling method and system for railway maintenance robots based on AI agents. Background Technology

[0002] With the increasing demands for railway transportation safety, collaborative scheduling technology for railway maintenance robots has become one of the core supporting technologies for ensuring the stable operation of railway equipment. This technology is mainly used in railway maintenance operation scenarios to coordinate multiple maintenance robots with different functions to complete various maintenance tasks, thereby improving the coverage and efficiency of maintenance operations. Currently, the common railway maintenance robot scheduling methods in the industry mostly adopt a centralized architecture. However, such scheduling methods often suffer from information transmission delays between the central control system and the robots, making it difficult to match the issued scheduling instructions with the real-time operating status of the robots. Support requests initiated solely based on local information cannot fully consider all feasible replacement resources in the vicinity, which can easily lead to untimely replacement or insufficient adaptability of replacement objects, thereby affecting the overall progress and quality of maintenance operations. Summary of the Invention

[0003] In view of this, the present invention provides a collaborative scheduling method and system for railway maintenance robots based on AI intelligent agents.

[0004] According to one aspect of the present invention, a collaborative scheduling method for railway maintenance robots based on AI agents is provided. The method includes: each railway maintenance robot detecting heterogeneous railway maintenance robots in its neighborhood, establishing a two-way capability perception communication network among multiple robots, and each AI agent synchronously controlling its own robot to upload real-time operating status data and historical work completion records to generate a neighborhood real-time capability sharing map that associates the capabilities of all networked robots; the AI ​​agent of the original task execution robot, based on the neighborhood real-time capability sharing map and combined with the demand benchmark of the current preset maintenance task to be executed, simulates task deduction and predicts the potential capability blind spots and operational risk factors of its own task execution, and outputs a multi-dimensional task adaptation gap prediction description and associated risk warning reference information; the AI ​​agent of the original task execution robot, based on the neighborhood real-time capability sharing map, the multi-dimensional task adaptation gap prediction description, and associated risk warning reference information, performs a multi-dimensional task adaptation gap prediction and associated risk warning reference information. Based on the risk warning reference information, cross-robot capability complementarity matching calculations are performed to locate candidate maintenance robots with precise replacement capabilities. The original task-executing railway maintenance robot is then dispatched to send a collaborative replacement request data packet carrying complete predictive information to the candidate maintenance robot. The AI ​​agent of the candidate maintenance robot receiving the collaborative replacement request combines its own robot's real-time capability description with the complete predictive information to initiate bidirectional capability complementarity negotiation with the AI ​​agent of the original task-executing robot. The negotiation covers task splitting boundaries, operation timing connection nodes, and cross-authorization scope, and outputs a standardized replacement collaborative consensus result. The AI ​​agents of the original task-executing robot and the candidate maintenance robot jointly generate a collaborative scheduling instruction based on the standardized replacement collaborative consensus result, which includes a dynamic operation timing table, cross-authorization rules, and dynamic risk avoidance strategies. The maintenance operation is then collaboratively executed according to the collaborative scheduling instruction.

[0005] According to another aspect of the present invention, a cooperative scheduling system is provided, comprising: a processor; and a memory, wherein the memory stores computer-readable code, which, when executed by the processor, causes the processor to perform the method as described above.

[0006] This invention establishes a two-way capability-aware communication network among various railway maintenance robots and generates a real-time capability-sharing map of the neighborhood, linking the capabilities of all networked robots. The AI ​​agent of the original task-executing robot simulates the task based on this map and the requirements of the current pre-set maintenance task. This allows it to identify potential capability blind spots and operational risks in advance, making the requests for replacements more targeted and eliminating the need for adjustments during operation, thus reducing the possibility of work interruptions. Based on the real-time capability-sharing map, multi-dimensional task adaptation gap prediction descriptions, and associated risk warning reference information, the invention performs cross-robot capability complementarity matching calculations. This accurately locates candidate maintenance robots with replacement capabilities, avoiding the time wasted searching for replacement targets and improving the efficiency of replacements. The accuracy and efficiency of position matching are ensured. The AI ​​agent of the candidate maintenance robot receiving the collaborative replacement request combines its own robot's real-time capability description and complete predictive information to initiate two-way capability complementarity negotiation with the AI ​​agent of the original task execution robot. This clarifies the task splitting boundaries, operation sequence connection nodes, and cross-authorization scope, making the operational responsibilities and collaboration boundaries of both parties clear and unambiguous. This avoids problems such as overlapping operations or poor connection, ensuring the smoothness of collaborative operations. The AI ​​agents of the original task execution robot and the candidate maintenance robot jointly generate collaborative scheduling instructions based on the standardized replacement collaboration consensus results and execute maintenance operations. This ensures that the scheduling instructions can simultaneously match the operational capabilities and task requirements of both parties, guaranteeing the efficient advancement of collaborative operations and improving the overall reliability and completion quality of railway maintenance operations. Attached Figure Description

[0007] Figure 1 This is a flowchart illustrating a collaborative scheduling method for railway maintenance robots based on AI agents provided by the present invention.

[0008] Figure 2 This is a schematic diagram of the structure of a collaborative scheduling system provided in an embodiment of the present invention. Detailed Implementation

[0009] The AI-based railway maintenance robot collaborative scheduling method provided in this embodiment of the invention is executed by a collaborative scheduling system. The collaborative scheduling system is embedded in each communicating railway maintenance robot to receive and analyze the corresponding railway maintenance robot's operating data and external environment data, thereby realizing collaborative task scheduling.

[0010] Please see Figure 1 This is a flowchart illustrating the collaborative scheduling method for railway maintenance robots based on AI agents provided in an embodiment of the present invention, including the following steps:

[0011] Step S100: Each railway maintenance robot detects heterogeneous railway maintenance robots in the neighborhood, establishes a two-way capability perception and communication network among multiple robots, and each AI agent synchronously controls its own robot to upload real-time operating status data and historical work completion records, generating a neighborhood real-time capability sharing map that associates the capabilities of all networked robots.

[0012] Heterogeneous railway maintenance robots within a neighborhood refer to railway maintenance robots of different types and functions existing within a certain range around each other. Two-way capability-aware communication networking is a network architecture that enables multiple robots to perceive each other's capabilities and communicate bidirectionally. Real-time operational status data reflects the current operating status of the railway maintenance robots, such as battery level, speed, and position. Historical task completion records are relevant records of past maintenance tasks performed by the robots, including task time, task content, and task results.

[0013] In one implementation, step S100 may specifically include the following steps S110 to S160:

[0014] Step S110: Each railway maintenance robot sends a detection signal with a unique identification code to the neighborhood space, receives response signals with unique identification codes returned by heterogeneous railway maintenance robots in the neighborhood, decodes and verifies the signal codes, filters invalid coded signals, and obtains a set of valid unique identification codes for heterogeneous railway maintenance robots in the neighborhood.

[0015] The neighborhood space refers to the physical space range defined around each railway maintenance robot, within which other railway maintenance robots may be detected. A unique identification code is assigned to each railway maintenance robot to accurately identify different robots during communication. For example, each railway maintenance robot sends a detection signal with its unique identification code to the neighborhood space via its own signal transmitter. For instance, at a railway maintenance site, robot A activates its signal transmitter function, sending a detection signal with its unique identification code to the surrounding space wirelessly. Heterogeneous railway maintenance robots within the neighborhood receive this detection signal, identify the unique identification code in the signal using their own signal receiving and processing modules, and return a response signal with its own unique identification code. Robot A, upon receiving these response signals, decodes and verifies the signal encoding. The decoding process can be based on a binary encoding decoding algorithm to convert the received encoded signal into readable information. The verification process checks whether the decoded information conforms to preset rules and formats, such as checking the code length and range. If an invalid encoded signal is found, such as one whose code length does not meet requirements or whose code exceeds the preset range, it is filtered out. After decoding, verification, and filtering invalid signals, a set of unique identity codes for valid heterogeneous railway maintenance robots within the neighborhood is finally obtained.

[0016] Step S120: Each AI agent initiates a two-way communication link handshake request with the corresponding coded robot based on the set of unique identity codes of the effective heterogeneous railway maintenance robots in the neighborhood, performs link encryption verification and connectivity status confirmation, and obtains a two-way capability perception communication network among multiple robots.

[0017] AI agents are intelligent entities with autonomous decision-making and control capabilities, capable of controlling and managing railway maintenance robots. For example, each AI agent, based on the set of unique identifiers for heterogeneous railway maintenance robots within its neighborhood obtained in step S110, initiates a two-way communication link handshake request to the robot with the corresponding identifier. For instance, AI agent A controls robot A to send a handshake request to robot B, which has a specific unique identifier. The handshake request contains necessary information, such as the communication protocol version and initial communication parameters. Upon receiving the handshake request, robot B parses and verifies it; if the verification passes, it returns an acknowledgment response. During this process, the AI ​​agent and the robot perform link encryption verification. Encryption verification can employ symmetric or asymmetric encryption algorithms, such as the AES symmetric encryption algorithm. Before communication begins, both parties negotiate an encryption key and then use this key to encrypt and decrypt communication information. Encryption verification ensures the security of communication information during transmission. Simultaneously, both parties also confirm connectivity, for example, by sending test data packets to check if data packet transmission is normal and if there are any issues such as packet loss or delay. If the connectivity is confirmed, the two-way communication link is successfully established, and a two-way capability-aware communication network is formed between multiple robots. In this network, each robot can communicate two-way with other robots to perceive their capabilities and status.

[0018] Step S130: Each AI agent retrieves real-time operating status data and historical work completion records from the local storage unit of the railway maintenance robot it controls, performs unified data format conversion, and generates standardized operation information.

[0019] Real-time operational status data reflects the robot's current operating condition, such as battery level, speed, and position. Historical task completion records are relevant records of past maintenance tasks performed by the robot, including task time, task content, and task results.

[0020] For example, each AI agent interacts with the local storage unit of the railway maintenance robot it controls to retrieve real-time operating status data and historical task completion records. For instance, an AI agent controls robot A and reads real-time operating status data and historical task completion records from robot A's local storage unit via the data interface. Since different robots' local storage units may use different data formats to store this data, a unified data format conversion is necessary. This conversion process can employ data mapping and transformation algorithms, such as mapping data of different formats to a unified field and structure based on a preset format template.

[0021] Step S140: Each AI agent uploads standardized operation information to the shared storage node of the two-way capability perception communication network, and simultaneously obtains standardized operation information uploaded by other railway maintenance robots in the network, thus obtaining a set of standardized operation information for the entire network.

[0022] For example, each AI agent uploads the standardized operational information generated in step S130 to the shared storage node via the communication link of the two-way capability-aware communication network. For instance, AI agent A controls robot A to send its standardized operational information to the shared storage node. Upon receiving this information, the shared storage node stores and manages it, for example, by classifying and storing it according to the robot's unique identification code. Simultaneously, each AI agent also synchronously acquires standardized operational information uploaded by other railway maintenance robots within the network. The synchronization process can employ periodic polling or event-triggered methods. For example, the AI ​​agent sends a request to the shared storage node at regular intervals to obtain the latest information uploaded by other robots. Alternatively, when the shared storage node uploads new information, a notification event is triggered, and upon receiving the notification, the AI ​​agent immediately retrieves the new information from the shared storage node.

[0023] Step S150: Based on the standardized operation information set of the entire network, perform field association mapping on the operation coverage, historical operation completion records, and real-time operation status data of each railway maintenance robot to generate a neighborhood real-time capability sharing map that associates the capabilities of all networked robots.

[0024] The operational coverage area refers to the region where railway maintenance robots can perform maintenance tasks. Historical task completion records reflect the robot's past performance in performing maintenance tasks, while real-time operational status data reflects the robot's current operational status. Field association mapping associates and maps different data fields to establish logical relationships between them. The neighborhood real-time capability sharing graph is a visual representation of the capabilities of all networked robots and the relationships between them.

[0025] In one implementation, step S150 may specifically include the following steps S151 to S156:

[0026] Step S151: Extract the operation coverage field, historical operation completion record field, and real-time operation status data field of each railway maintenance robot from the standardized operation operation information set of the whole network, remove redundant content in the fields, and obtain the basic field set of single robot operation capability.

[0027] For example, the operation coverage area, historical operation completion record, and real-time operation status data fields for each railway maintenance robot are first extracted from the standardized operation information set across the entire network. This can be achieved through data filtering and extraction algorithms, such as accurately extracting the required fields from the information set based on the field name and identifier. For instance, in a database containing information on multiple robots, SQL queries can be written to filter the operation coverage area, historical operation completion record, and real-time operation status data fields for each robot based on the field name. Then, redundant content is removed from the extracted fields. Redundant content removal can employ preset algorithms, such as algorithms based on text similarity. For the operation coverage area field, if there are duplicate descriptions of areas or information on areas with different expressions but the same actual meaning, they can be merged or removed. For the historical operation completion record field, if there are duplicate task records or unnecessary embellishments in the task description, they can be cleaned up. For the real-time operation status data field, if there is duplicate data or invalid data, it can be filtered out. After removing redundant content, a set of basic fields for the single robot's operational capabilities is obtained.

[0028] Step S152: Bind each field in the basic field set of single robot operation capabilities to the unique identification code of the corresponding railway maintenance robot to generate a set of operation capability fields with unique identification codes.

[0029] For example, first, define the unique identification code of the railway maintenance robot corresponding to the basic field set of each single robot's operational capabilities. This can be achieved through the association established in the previous data processing, such as recording the correspondence between each robot's operational capability data and its unique identification code during data acquisition and storage. Then, bind each field in the basic field set of single robot operational capabilities to its corresponding unique identification code. The binding method can be in the form of a data structure, such as creating a dictionary or list, storing the unique identification code as the key and the basic field set of single robot operational capabilities as the value in the dictionary; or storing the unique identification code and the corresponding operational capability field set as an element in the list. For example, if a railway maintenance robot R1 has a unique identification code of ID1, and its basic field set of single robot operational capabilities is {Operation Coverage A, Historical Operation Completion Record B, Real-time Operation Status Data C}, then it can be bound as {ID1:{Operation Coverage A, Historical Operation Completion Record B, Real-time Operation Status Data C}}. In this way, bind all robot operational capability fields to their unique identification codes, ultimately generating a set of operational capability fields with unique identification codes.

[0030] Step S153: Cross-compare the set of operational capability fields with unique identification codes to identify overlapping content in the operational coverage field of different railway maintenance robots, similar content in the historical operation completion record field, and complementary content in the real-time operation status data field, so as to obtain the set of capability association fields within the network.

[0031] For example, the set of job capability fields with unique identification codes is traversed, and the job coverage field, historical job completion record field, and real-time operation status data field of different robots are compared separately. For comparing the job coverage field, Geographic Information System (GIS) technology can be used. The job coverage of each robot is represented in the form of geographic coordinates or regional boundaries, and then the spatial analysis function in GIS software is used to find the overlapping areas between different job coverage areas. For example, if robot R1's job coverage area is region A and robot R2's job coverage area is region B, GIS analysis can determine the overlapping part of A and B. For comparing the historical job completion record field, text matching algorithms can be used. The historical job records of each robot are converted into text strings, and then string matching algorithms, such as the Levenshtein distance algorithm or the cosine similarity algorithm, are used to calculate the similarity between different records. If the similarity exceeds a preset threshold, similar content is considered to exist. For the identification of complementary content in the real-time operation status data field, analysis can be performed based on the data type and characteristics. By cross-referencing these fields, overlapping content in the job coverage field, similar content in the historical job completion record field, and complementary content in the real-time operation status data field are identified. These contents are then organized and recorded to obtain the final set of capability association fields within the network.

[0032] Step S154: Based on the set of capability association fields within the network, perform topological mapping on the operational capability fields of each railway maintenance robot according to the associated content, and generate underlying topological structure data with unique identity codes, operational coverage areas, and capability association links.

[0033] For example, the relationships between railway maintenance robots are first determined based on the information in the set of capability association fields within the network. For instance, if the set shows that robots R1 and R2 have overlapping work coverage areas and similar content in their historical work completion records, then a capability association is established between R1 and R2. Then, the operational capability fields of each railway maintenance robot (including work coverage area, historical work completion records, real-time operating status data, etc.) are topologically mapped according to these associations. Graph theory can be used, treating each robot as a node and the capability associations between them as edges, constructing a topology graph. In this topology graph, each node is labeled with a corresponding unique identification code and work coverage area, and each edge represents a capability association link. For example, different colors or line thicknesses can be used to represent different types of capability associations, such as red lines representing overlapping work coverage areas and blue lines representing similar historical work completion records. Through this mapping, the operational capability information of each robot is displayed graphically. Finally, this topology graph is converted into underlying topology structure data. This data can be stored in the form of a data matrix or a list. For example, an adjacency matrix can be used to represent the connections between nodes, with elements indicating whether capability associations exist and the type of association. Each node's unique identification code, operational coverage area information, and associated link information are integrated into this data structure, ultimately generating underlying topology data with unique identification codes, operational coverage areas, and capability-related links.

[0034] Step S155: Embed the real-time operating status data fields of each railway maintenance robot into the topology node area corresponding to the unique identity code, and embed similar content of the historical operation completion record fields into the middle area of ​​the corresponding capability association link to generate a neighborhood real-time capability sharing map.

[0035] For example, the real-time operating status data fields of each railway maintenance robot are first associated and embedded with the topology node regions corresponding to their unique identification codes. This can be done by finding the node corresponding to each unique identification code in the topology data and then storing the real-time operating status data as the node's attribute information. For example, for robot R1, its unique identification code corresponds to the topology node N1, and its real-time operating status data includes battery level E1, speed V1, and position P1. These data are embedded into the attribute region of node N1, so that node N1 not only contains robot R1's unique identification code and work coverage area information, but also its real-time operating status information. Then, similar content from historical work completion records is embedded into the middle region of the corresponding capability association links. For association links with similar historical work completion records, the similar content is extracted and embedded into the middle region of the link. Through these two embedding operations, the real-time operating status data and similar content from historical work completion records are integrated into the topology structure, ultimately generating a neighborhood real-time capability sharing graph.

[0036] Step S156: Perform integrity verification on the neighborhood real-time capability sharing graph, confirm that all fields and associated content have been embedded, and mark the verification of the generated neighborhood real-time capability sharing graph as passed.

[0037] For example, a field-by-field and link-by-link check is performed on the neighborhood real-time capability sharing graph. First, a list of all fields and links that should be included in the graph is determined. This list can be based on the previous design and generation process. For example, the list should include each robot's unique identification code, work coverage area, real-time operating status data fields, capability linking links between different robots, and similar content from historical job completion records. Then, the graph is checked against the list to see if each field and link exists and is correctly embedded. For field checks, the corresponding attribute information can be found in the graph's data structure. For example, check if each topology node contains its corresponding unique identification code, work coverage area, and real-time operating status data. For link checks, check if the capability linking links are complete and if similar content from historical job completion records is correctly embedded in the links. If a field or link is found to be missing or incorrect during the check, it needs to be corrected and supplemented accordingly. Once all fields and links are confirmed to be correctly embedded, the verification of the generated neighborhood real-time capability sharing graph is marked as passed. This tag can be represented by a Boolean value (such as True) or by a preset character or encoding.

[0038] Step S160: Synchronize the neighborhood real-time capability sharing map to the local storage unit of all AI agents in the bidirectional capability perception communication network, and update the synchronization timestamp of the neighborhood real-time capability sharing map.

[0039] For example, through the communication mechanism of a two-way capability-aware communication network, the real-time capability sharing graph of the neighborhood can be sent to the local storage units of all AI agents within the network. This can be achieved using broadcast or multicast communication methods, sending the graph data to all AI agents at once. For instance, a central node can be set up in the network to broadcast the graph data to each AI agent. After receiving the graph data, each AI agent stores it in its local storage unit, overwriting any previously stored graph data. Simultaneously, it updates the synchronization timestamp of the real-time capability sharing graph. The synchronization timestamp can be represented using the current system time, for example, using a time value accurate to milliseconds. This timestamp is associated with the real-time capability sharing graph and stored in the AI ​​agent's local storage unit for subsequent determination of the graph's timeliness. When the real-time capability sharing graph is needed, the synchronization timestamp can be checked to determine if the graph is the latest version. If the timestamp is old, the graph data may need to be resynchronized.

[0040] Step S200: The AI ​​agent of the original task execution robot, based on the real-time capability sharing map of the neighborhood, combined with the demand benchmark of the current preset maintenance task to be executed, simulates the task to predict the potential capability blind spots and operational risk factors of its own task execution, and outputs multi-dimensional task adaptation gap prediction description and related risk warning reference information.

[0041] In one implementation, step S200 may specifically include the following steps S210 to S260:

[0042] Step S210: The AI ​​agent of the original task execution robot retrieves the set of requirement benchmark fields for the current preset maintenance task to be executed from the local storage unit, and extracts the set of operation capability fields for controlling the railway maintenance robot from the real-time capability sharing map of the neighborhood.

[0043] The original task execution robot is the robot initially assigned to perform the preset maintenance task. Its AI agent retrieves the set of requirement benchmark fields for the current preset maintenance task through a data interface with the local storage unit. This can be achieved by searching a pre-stored requirement benchmark file or database table in the local storage unit. For example, in a database, there might be a table named "Preset Maintenance Task Requirements." The AI ​​agent can retrieve the set of requirement benchmark fields for the current task by executing an SQL query. Simultaneously, the AI ​​agent extracts the set of operational capability fields for its controlled railway maintenance robot from the neighborhood real-time capability sharing graph. In the neighborhood real-time capability sharing graph, each robot's operational capability information is associated with a corresponding unique identification code. The AI ​​agent searches for and extracts the corresponding operational capability fields from the graph based on the unique identification code of its controlled robot. For example, by traversing the node information in the graph, it finds the node that matches its own robot's unique identification code, and then extracts fields such as work coverage, historical work completion records, and real-time operating status from that node to form the set of operational capability fields.

[0044] Step S220: Compare the set of requirement benchmark fields for the current preset maintenance task to be executed with the set of operation capability fields of the railway maintenance robot controlled by itself, identify the fields in the requirement benchmark that are not covered by its own capabilities, and obtain a preliminary set of requirement difference fields.

[0045] For example, the set of requirement benchmark fields for the currently scheduled pre-defined maintenance task is compared one by one with the set of operational capability fields of the self-controlled railway maintenance robot. For each requirement benchmark field, it is checked whether there is a corresponding field in the set of operational capability fields of the self-controlled railway maintenance robot that can meet that requirement. For example, if the requirement benchmark field set includes a field that specifies maintenance work in a specific area A, but the operational coverage field of the self-controlled railway maintenance robot does not include area A, then this field of performing maintenance work in a specific area A is identified as a field not covered by its own capabilities. By comparing all requirement benchmark fields in this way, the fields not covered by its own capabilities are filtered out, forming a preliminary set of requirement difference fields.

[0046] Step S230: Based on the initial set of demand difference fields, map the data of each job node of the task to be executed, generate the underlying data flow scenario of the simulated job, and start the node-by-node flow of the simulated job nodes.

[0047] In one implementation, step S230 may specifically include the following steps S231 to S236:

[0048] Step S231: Based on each job node field in the preliminary requirement difference field set, generate the corresponding node's flow rule data, and define the node's input data requirements, output data format, and flow trigger conditions.

[0049] For example, each task node field in the initial requirement difference field set is analyzed. For instance, if a task node field specifies flaw detection on a particular type of railway track, then corresponding node flow rule data needs to be generated based on this field. Input data requirements might specify the need for railway track type, location, and other information in a predefined encoding and string format. Output data format might specify that flaw detection results should be output as a report, including detection conclusions and identified problems, in a predefined document template. Flow trigger conditions might be set so that the node starts running when complete input data is received and the system is idle.

[0050] Step S232: Input the set of operational capability fields of the self-controlled railway maintenance robot into the underlying data flow scenario of the simulated operation, drive the flow of each operation node according to the flow rules, record the input data, output data and flow duration of each node, and obtain the set of simulated operation node flow data.

[0051] For example, the set of operational capability fields for the self-controlled railway maintenance robot is input into the underlying data flow scenario of the simulated operation. For instance, information such as the operation coverage, historical operation completion status, and real-time operating status from the operational capability field set is input as initial data into the scenario. Then, the data drives the flow of each operation node according to the flow rules. Each operation node is processed sequentially according to the input data requirements, flow trigger conditions, and output data formats defined in the flow rules. When an operation node meets the flow trigger conditions, the input data is processed, and corresponding output data is generated. For example, in the equipment status inspection operation node, the input equipment status information is processed according to the preset inspection logic, and an inspection result report is output. During the flow of each node, the input data, output data, and flow duration are recorded in detail. Data storage structures, such as lists or database tables, can be used to store this information. The flow data of all operation nodes is integrated to obtain the simulated operation node flow data set.

[0052] Step S233: Perform real-time analysis on the simulated operation node flow data set, compare the matching of the output data of each node with the preset output format, identify nodes with mismatched output data, and mark them as flow abnormal nodes.

[0053] For example, real-time analysis of the simulated job node workflow data set can be performed. Data comparison algorithms can be used to compare the output data of each node with a preset output format. For instance, for a job node that outputs data in document format, the preset output format stipulates that the document must contain preset headings, paragraph structures, and content fields. A text comparison algorithm checks whether the output document meets these requirements. If the format, content, or structure of the output data is inconsistent with the preset output format, the output data of that node is considered mismatched. These nodes with mismatched output data are marked as workflow abnormal nodes. The marking method can be to add a status identifier field to each node in the simulated job node workflow data set, setting the status identifier of abnormal nodes to preset values, such as abnormal or mismatched. By identifying and marking workflow abnormal nodes, nodes that may cause problems during the simulated job process can be quickly located.

[0054] Step S234: For abnormal nodes in the flow, backtrack the matching of their input data and flow rule data, identify nodes whose input data does not meet the rule requirements and classify them as node stagnation, identify nodes not covered by the flow rules and classify them as node jump failure, and obtain a set of abnormal node type tags.

[0055] Node stalling refers to a situation where a job node cannot function properly because the input data does not meet the requirements of the workflow rules. Node jump failure refers to a situation where the workflow rules do not specify certain conditions for job nodes, causing the node to fail to complete the workflow as expected.

[0056] For example, for each abnormal workflow node, the matching of its input data with the workflow rule data is traced back. The input data is checked to see if it meets the input data requirements defined in the workflow rules, such as data type, format, and range. If the input data does not meet the rule requirements, such as an incorrect data type or data exceeding the specified range, then the node is classified as a stalled node. For example, in a job node that calculates equipment parameters, if the workflow rule requires the input data to be an integer, but the actual input data is a string, then this node will be determined to be a stalled node.

[0057] Simultaneously, check whether the workflow rules data covers all possible scenarios for the work nodes. If the workflow rules do not specify certain possible situations, such as the lack of corresponding handling rules when input data contains preset abnormal values, causing the node to be unable to continue workflow, then that node is classified as a node jump failure. For example, in a work node for equipment fault diagnosis, if the input fault code is a value not defined in the workflow rules, the node cannot proceed to the next step and will be judged as a node jump failure.

[0058] Step S235: Bind the abnormal node type tag set to the corresponding job node data to generate an abnormal node data set with type tags.

[0059] For example, each exception node type tag in the exception node type tag set can be bound to the corresponding job node data. This binding can be achieved using the job node's identifier. For instance, in the exception node type tag set, there is a node with the identifier N1 and its type tag is "Node Stagnant." Find the job node data with the identifier N1 in the job node data set and associate the "Node Stagnant" type tag with that job node data. This binding relationship can be represented using data structures such as dictionaries or lists. For example, create a dictionary with the job node identifier as the key and a composite object containing the type tag and job node data as the value.

[0060] Step S236: Synchronize the set of abnormal node data with type tags to the temporary storage unit of the original task execution robot AI agent, and update the record timestamp of the set of abnormal node data with type tags.

[0061] For example, a set of anomaly node data with type tags can be synchronized to the temporary storage unit of the original task-executing robot AI agent via a data transmission interface. Data can be stored in the temporary storage unit using data copying or sharing. For example, data can be copied from the current storage location to a designated area of ​​the temporary storage unit. Simultaneously, the record timestamp of the anomaly node data set with type tags is updated. The record timestamp can be represented using the current system time, for example, using a time value accurate to milliseconds. In the temporary storage unit, this timestamp is stored in association with the anomaly node data set with type tags for subsequent determination of data timeliness. When the anomaly node data set with type tags is needed, the record timestamp can be checked to determine if the data is the latest version. If the timestamp is old, data resynchronization may be necessary. By synchronizing the anomaly node data set with type tags to the temporary storage unit and updating the record timestamp, it can be ensured that the original task-executing robot AI agent always has the latest anomaly node data.

[0062] Step S240: Monitor the flow process of the simulated work nodes in real time, identify the node stagnation and node jump failure during the flow process, record the corresponding work node data, and obtain the simulation simulation abnormal node data set.

[0063] For example, real-time monitoring of the workflow of simulated job nodes can be achieved. Monitoring points can be set up within the underlying data flow of the simulated job to obtain the operational status information of each job node in real time. For instance, monitoring programs can be set up at the input and output stages of each job node to check whether the input data meets requirements and whether the output data conforms to a preset format. When a node is found to be stalled or fails to jump, the corresponding job node data is recorded. This data includes the node's identifier, input data, output data, workflow duration, and anomaly type. Data storage structures, such as lists or database tables, can be used to store this data. Finally, the data of all job nodes that experienced anomalies can be integrated to obtain a set of simulated abnormal node data.

[0064] Step S250: Classify and organize the simulation and simulation abnormal node data set, classify the work nodes corresponding to node stagnation into the potential capability blind spot field, classify the work nodes corresponding to node jump failure into the work risk cause field, and obtain the classified abnormal node field set.

[0065] The "Potential Capability Blind Spot" field refers to the field related to the job node where the robot may encounter problems due to insufficient capabilities during task execution. The "Job Risk Trigger" field refers to the field related to the job node that may lead to risks during the job execution process.

[0066] For example, a detailed analysis and organization of the simulation and deduction abnormal node data set is performed. For each task node data in the set, its abnormality type is determined as node stagnation or node jump failure. If the abnormality type is node stagnation, it means that the task node cannot operate normally because the input data does not meet the requirements of the flow rules, which may be due to the original task execution robot's insufficient capability in this aspect. This task node is classified as a potential capability blind spot. For example, in a task node for equipment fault repair, if the robot lacks the ability to repair a certain type of fault, and the node stagnates when the corresponding fault data is input, then this node will be classified as a potential capability blind spot.

[0067] If the exception type is "node jump failure," it indicates that the workflow rules do not specify certain situations for job nodes, causing the node to fail to complete the workflow as expected. This situation may pose risks during actual operations, and the job node will be classified as a job risk trigger. For example, in a data transmission job node, if the workflow rules do not have a corresponding handling mechanism when encountering network anomalies, causing the node jump to fail, this node will be classified as a job risk trigger.

[0068] Step S260: Based on the classified abnormal node field set, generate a multi-dimensional task adaptation gap prediction description that includes a list of potential capability blind spots in operation nodes and descriptions of operation risk trigger nodes, generate a set of response direction fields for each abnormal node, and obtain related risk warning reference information.

[0069] For example, potential capability blind spot task nodes and task risk trigger nodes are extracted from the categorized set of abnormal node fields. The potential capability blind spot task nodes are compiled into a list, detailing each node's identifier, task content, and corresponding capability deficiency type. Task risk trigger nodes are described, including the node's identifier, a detailed description of the abnormal situation, and potential risks. The list of potential capability blind spot task nodes and the descriptions of task risk trigger nodes are integrated to generate a multi-dimensional task adaptation gap prediction description. This description analyzes the potential problems and gaps that the original task execution robot may have when performing tasks from multiple perspectives, providing comprehensive information for subsequent task adjustments and risk response.

[0070] For each abnormal node, possible solutions and measures are analyzed, generating a set of response direction fields. For example, for a potential capability blind spot node, if it is due to the robot lacking a certain technical capability, other robots with the corresponding technical capabilities can be called in to assist; for a node causing operational risk, if it is due to imperfect workflow rules, the workflow rules can be optimized and supplemented. The generated set of response direction fields is then correlated with the multi-dimensional task adaptation gap prediction description to obtain related risk warning reference information.

[0071] Step S300: The AI ​​agent of the original task execution robot performs cross-robot capability complementarity matching calculation based on the real-time capability sharing map of the neighborhood, the multi-dimensional task adaptation gap prediction description and the associated risk warning reference information, locates the candidate maintenance robot with accurate replacement capability, and schedules the original task execution railway maintenance robot to send a collaborative replacement request data packet carrying complete prediction information to the candidate maintenance robot.

[0072] In one implementation, step S300 may specifically include the following steps S310 to S360:

[0073] Step S310: The AI ​​agent of the original task execution robot extracts the set of operational capability fields of all other railway maintenance robots in the network from the real-time capability sharing map of the neighborhood, extracts the list of potential capability blind spot operation nodes and the description of operation risk inducement nodes from the multi-dimensional task adaptation gap prediction description, and extracts the response direction guidance from the associated risk warning reference information to obtain the basic data set for filling gap matching.

[0074] The set of operational capability fields for all other railway maintenance robots within the network is extracted from the real-time capability sharing graph of the neighborhood, and includes the operational capability fields of robots other than the original task-executing robot. The list of potential capability blind spots for operational nodes is a list of operational nodes where the original task-executing robot may have insufficient capabilities, as described in the multi-dimensional task adaptation gap prediction description. The description of operational risk-inducing nodes is a detailed description of nodes that may lead to operational risks, as described in the multi-dimensional task adaptation gap prediction description. The response guidance provides possible solutions and measures for each abnormal node in the associated risk warning reference information.

[0075] For example, the AI ​​agent of the original task-executing robot extracts the set of operational capability fields for all other railway maintenance robots in the network from the real-time capability sharing graph of the neighborhood through a data interface with the local storage unit. In the real-time capability sharing graph of the neighborhood, the operational capability information of each robot is associated with a corresponding unique identity code. The AI ​​agent finds the operational capability fields of other robots besides the original task-executing robot by traversing the node information in the graph, forming a set of operational capability fields. At the same time, it extracts a list of potential capability blind spot operation nodes and descriptions of operation risk inducing nodes from the multi-dimensional task adaptation gap prediction description. The relevant list and description content can be extracted from the text information of the multi-dimensional task adaptation gap prediction description using text parsing algorithms. In addition, it extracts response guidance from the associated risk warning reference information. The content of response guidance can also be obtained from the associated risk warning reference information using text parsing and data extraction methods. The extracted set of operational capability fields for all other railway maintenance robots in the network, the list of potential capability blind spot operation nodes, the descriptions of operation risk inducing nodes, and the response guidance are integrated together to obtain the basic data set for gap matching.

[0076] Step S320: Compare the potential capability blind spot operation node list and operation risk cause node description in the basic data set of the fill-in matching with the operation capability field set of other railway maintenance robots one by one, identify the railway maintenance robots that can cover the corresponding node content, and obtain the initial matching robot exclusive identity code set.

[0077] In one implementation, step S320 may specifically include the following steps S321 to S326:

[0078] Step S321: Set a unique requirement identifier field for each node in the potential capability blind spot operation node list and operation risk cause node description in the basic data set for filling in the gaps and matching.

[0079] For example, the potential capacity blind spot operation node list and operation risk cause node descriptions in the basic dataset for fill-in-the-blank matching are traversed. For each operation node in the potential capacity blind spot operation node list, a unique requirement identifier field is assigned. This identifier field can be generated using a preset encoding rule, such as a combination of letters and numbers, and it is guaranteed that each identifier field is unique throughout the entire list. For example, for an operation node performing flaw detection on a specific type A railway track, a requirement identifier field of REQ-001 can be assigned. Similarly, a unique requirement identifier field is also set for each node in the operation risk cause node description.

[0080] Step S322: Extract the operation coverage node field of each robot from the set of operation capability fields of other railway maintenance robots in the network, and set a unique capability identifier field for each operation coverage node.

[0081] For example, the set of operational capability fields for other railway maintenance robots within the network is traversed. For each robot's operational capability field, the operational coverage node field is extracted. For instance, a robot's operational capability field might include operational coverage node fields such as inspecting railway bridges and maintaining railway tunnels. For each extracted operational coverage node, a unique capability identifier field is assigned. The generation rule for the capability identifier field can be similar to that of the requirement identifier field, using a combination of letters and numbers, and ensuring that each identifier field is unique within the entire set of operational coverage nodes. For example, for the operational coverage node of inspecting railway bridges, a capability identifier field of CAP-001 can be assigned.

[0082] Step S323: Compare the demand identifier field and the capability identifier field to identify identifier pairs with the same job node content and generate a set of demand-capability identifier association fields.

[0083] In one implementation, step S323 may specifically include the following steps S3231 to S3236:

[0084] Step S3231: Perform numerical conversion on the job node content corresponding to the requirement identifier field to generate requirement node values; perform numerical conversion on the job coverage node content corresponding to the capability identifier field to generate capability node values.

[0085] For example, for the work node content corresponding to the demand identifier field, numerical conversion is performed based on its specific characteristics and attributes. If the work node content involves quantitative parameters, such as the number of equipment to be repaired or the duration of the operation, these parameters can be directly used as part of the numerical value. For example, if the work node content is to repair 5 railway signaling devices, then 5 can be used as a component of the work node value. For some qualitative descriptions, such as the type of operation (e.g., flaw detection, fault repair, etc.), numerical conversion can be performed using encoding. For example, flaw detection can be encoded as 1, fault repair as 2, etc. The numerical values ​​of each part are combined according to certain rules to form the demand node value.

[0086] Step S3232: Compare the numerical values ​​of the demand node with the numerical values ​​of the capability node, identify the identifier pairs with exactly the same values, and generate a set of numerical matching identifier pairs.

[0087] For example, all demand node values ​​and capability node values ​​are compared one by one. This can be done by iterating through each demand node value and comparing it with all capability node values. During the comparison, a precise numerical comparison algorithm is used to ensure that the two identifier fields are considered a numerical matching identifier pair only when the demand node value and capability node value are exactly the same. For example, if both the demand node value and the capability node value are 123, the corresponding demand identifier field and capability identifier field are recorded as an identifier pair. During the comparison process, attention must be paid to the precision and format of the values. If the values ​​have been rounded or otherwise processed, these factors must be taken into account during the comparison to avoid misjudgments due to precision issues. All identifier pairs with exactly the same value are collected to form a numerical matching identifier pair set.

[0088] Step S3233: Perform a second verification on the content of the job node corresponding to the numerical matching identifier pair set, confirm that the text description and job requirements of the node content are completely consistent, and generate a set of identifier pairs that have passed the verification.

[0089] For example, for each identifier pair in the numerical matching identifier pair set, re-examine the content of its corresponding job node. Perform a detailed text comparison between the job node content corresponding to the requirement identifier field and the job coverage node content corresponding to the capability identifier field. Check whether the text descriptions are completely identical, including the specific object of the job, the operation method of the job, and the quality requirements of the job. For example, the requirement node content is to conduct comprehensive flaw detection on railway bridges with a detection accuracy requirement of 0.1mm, and the capability node content is to have the capability to conduct flaw detection on railway bridges. Although the numerical values ​​may match, the text description and job requirements are not completely consistent. In this case, the identifier pair cannot pass the verification.

[0090] In addition to the textual description, it is also necessary to check whether the job requirements are consistent. Job requirements may include safety standards, time limits, tools and materials used, etc. If any inconsistencies are found, the corresponding identifier pair is removed from the set of numerical matching identifier pairs. After a second verification of all identifier pairs, the identifier pairs that completely match the textual description and job requirements are collected to generate a set of verified identifier pairs.

[0091] Step S3234: Store the verified set of identifier pairs to the shared storage node of the bidirectional capability-aware communication network, and update the generation timestamp of the verified set of identifier pairs.

[0092] For example, through the communication interface of a two-way capability-aware communication network, the set of verified identifier pairs is sent to a shared storage node. A data transmission protocol, such as TCP / IP, can be used to ensure reliable data transmission. In the shared storage node, the set of verified identifier pairs is stored in a pre-defined database table or file for subsequent querying and use.

[0093] The generation timestamp of the verified identifier pair set is updated. The generation timestamp can use the current system's precise time, such as a value accurate to milliseconds. This timestamp is then associated with the verified identifier pair set and stored in the shared storage node. When other AI agents need to use this identifier pair set, they can check the generation timestamp to determine if the data is up-to-date. If the timestamp is outdated, it may be necessary to retrieve the latest identifier pair set. By storing the verified identifier pair set in the shared storage node and updating the generation timestamp, data sharing and timeliness are ensured, providing accurate and up-to-date demand-capability association information for AI agents within the network.

[0094] Step S3235: Retrieve the historical demand-capability identifier association field set from the shared storage node, compare it with the currently verified identifier pair set, update duplicate association relationships, add new association relationships, and obtain the updated demand-capability identifier association field set.

[0095] For example, retrieve the historical demand-capability identifier association field set from the shared storage node of the two-way capability-aware communication network. Through the data interface with the shared storage node, execute the corresponding query operation to obtain the historical set data. Compare the historical demand-capability identifier association field set with the currently validated set of identifier pairs. For each identifier pair, check if it already exists in the historical set. If it exists, it indicates a duplicate association relationship, requiring updating according to the current situation. For example, if the job requirement corresponding to a certain association in the historical set has changed, it needs to be updated to the latest job requirement.

[0096] If an identifier pair does not exist in the historical set, it indicates a new association, and it is added to the historical set. By iterating through the set of currently validated identifier pairs, all duplicate associations are updated, and new associations are added, ultimately resulting in the updated set of requirement-capability identifier association fields.

[0097] Step S3236: Synchronize the updated set of requirement-capability identifier association fields to the local storage unit of all AI agents in the bidirectional capability-aware communication network.

[0098] For example, through the communication mechanism of a two-way capability-aware communication network, the updated set of requirement-capability identifier association fields is sent to all AI agents within the network. Broadcast or multicast communication methods can be used to send the set of data to all AI agents at once. Each AI agent, upon receiving the set of data, stores it in its local storage unit. During storage, previously stored requirement-capability identifier association field sets (if any) can be overwritten to ensure that the locally stored data is the latest.

[0099] Step S324: Traverse the capability identifier field of each other railway maintenance robot in the network, count the number of identifiers that appear in the demand-capability identifier association field set, and obtain the matching identifier count for each robot.

[0100] Match ID count is the number of times each robot's capability ID field appears in the demand-capability ID association field set, reflecting the degree of match between each robot and the demand.

[0101] For example, the capability identifier fields of each railway maintenance robot in the network, excluding the original task execution robot, are iterated through. For each capability identifier field of each robot, a search is performed in the demand-capability identifier association field set. It is checked whether the capability identifier field exists in a certain identifier pair in the set. If it exists, it means that the capability identifier field matches a demand identifier field, and the matching identifier count of the robot is incremented by 1.

[0102] The traversal process can be implemented using a loop. For example, an outer loop can be used to iterate through each robot, and for each robot, an inner loop can be used to iterate through all its capability identifier fields. In each inner loop iteration, a lookup and count operation is performed. In this way, the number of times each robot's capability identifier field appears in the demand-capability identifier association field set is counted, resulting in a matching identifier count for each robot.

[0103] Step S325: Bind the matching identifier count of each robot to the corresponding unique identity code to generate a set of matching count fields with unique identity codes.

[0104] For example, we can bind the match identifier count and unique identity code for each robot. This binding can be achieved using data structures such as dictionaries or lists. If using a dictionary, the unique identity code is stored as the key, and the match identifier count as the value. For example, robot R1's unique identity code is ID1, and its match identifier count is 5, so it can be stored as {ID1:5}. If using a list, the unique identity code and match identifier count can be stored as a single element, such as [ID1, 5]. By iterating through the match identifier count and unique identity code for each robot and binding them one by one, a set of match count fields with unique identity codes is ultimately generated.

[0105] Step S326: Select robot-specific identification codes that can cover all required identification fields to obtain a preliminary set of matching robot-specific identification codes.

[0106] For example, the matching status of each robot is analyzed based on a set of matching count fields with unique identification codes. The range of requirement identification fields corresponding to each robot's capability identification field is examined. If a robot's capability identification field covers all requirement identification fields, it indicates that the robot has the capability to meet all requirements.

[0107] The selection operation can be achieved by iterating through the set of matching count fields with unique identification codes and combining it with the set of demand-capability identifier association fields. For each robot, check the demand identifier fields associated with its capability identifier field in the demand-capability identifier association field set, and count the number of demand identifier fields covered. When the number of demand identifier fields covered by a robot equals the total number of demand identifier fields, add the robot's unique identification code to the initial matching robot unique identification code set.

[0108] Step S330: Extract the real-time running status data field corresponding to the preliminary matching robot exclusive identity code set from the neighborhood real-time capability sharing graph, filter the robot exclusive identity codes whose current real-time running status is not idle, and obtain the filtered replacement robot candidate exclusive identity code set.

[0109] For example, based on the initial set of robot-specific identification codes, the corresponding real-time operational status data fields are extracted from the neighborhood real-time capability sharing graph. In the neighborhood real-time capability sharing graph, each robot's information is associated with its corresponding unique identification code. By traversing the node information in the graph, the nodes corresponding to each code in the initial set of robot-specific identification codes are found, and the real-time operational status data fields are extracted from these nodes.

[0110] The extracted real-time operation status data fields are analyzed to filter out robot-specific identification codes whose current real-time operation status is not idle. A real-time operation status of not idle may include the robot performing other tasks, being in a fault repair state, etc. Whether the robot is idle can be determined by checking specific identifiers or parameters in the real-time operation status data fields. For example, if the real-time operation status data field contains a task status parameter, and the value of this parameter is "executing," it indicates that the robot is not idle, and its unique identification code is removed from the initial set of matched robot-specific identification codes.

[0111] Step S340: Analyze the set of operational capability fields corresponding to the selected candidate identity codes of the replacement robots, and generate replacement adaptability analysis results containing the replacement coverage node description and the replacement timing connection description of each candidate robot.

[0112] For example, based on the selected candidate unique identity code set for supplementary robots, the corresponding set of operational capability fields is extracted from the neighborhood real-time capability sharing graph. A detailed analysis of the operational capability field set for each candidate robot is then performed. For the supplementary coverage node description, combined with the demand-capability identifier association field set, the potential capability blind spot operational nodes and operational risk-inducing nodes that each candidate robot can cover are determined. The specific content of these nodes is described in detail, such as the object of the operation and the requirements of the operation. For example, if a candidate robot can cover operational nodes for the maintenance of a specific type of railway signaling equipment, the model of the equipment and the specific content of the maintenance are detailed in the supplementary coverage node description.

[0113] For the description of the timing sequence for filling in gaps, consider the work plan of the original task-executing robot and the real-time operating status of the candidate robot. Analyze when the candidate robot can start participating in the gap task and how it connects with the work nodes of the original task-executing robot. For example, if the original task-executing robot completes part of the work within a certain time period, the candidate robot can start executing the gap task immediately after that time period ends. This time sequence and connection relationship should be clearly defined in the description of the timing sequence for filling in gaps.

[0114] The descriptions of the replacement coverage nodes and the replacement timing connections for each candidate robot are compiled to form the replacement adaptability analysis results. These results can be represented using data structures such as lists or dictionaries, where each element corresponds to a candidate robot and includes information such as its replacement coverage node description and replacement timing connection description.

[0115] Step S350: Based on the results of the fill-in adaptability analysis, select candidate robots that can cover all potential blind spot operation nodes and position them as candidate maintenance robots with precise fill-in capabilities.

[0116] For example, the results of the fill-in-the-blank adaptability analysis can be studied in depth. For the fill-in-the-blank coverage node description of each candidate robot, it can be checked whether it can cover all potential blind spot operation nodes. The list of potential blind spot operation nodes can be compared one by one with the fill-in-the-blank coverage node description of each candidate robot. If the fill-in-the-blank coverage node description of a candidate robot contains the content of all potential blind spot operation nodes, it means that the robot can cover all these nodes.

[0117] During the comparison process, it is essential to ensure that the content of each task node is completely matched, including the task object, task requirements, and task type. If a task node with a potential capability blind spot is found to be absent or inconsistent in the candidate robot's description of the replacement coverage node, then that candidate robot cannot meet the requirement of covering all nodes.

[0118] Step S360: Encapsulate the multi-dimensional task adaptation gap prediction description and associated risk warning reference information in a format, generate a collaborative fill request data packet with a unique identity code, and control the original task execution railway maintenance robot to send the collaborative fill request data packet to the candidate maintenance robot with accurate fill capability.

[0119] For example, first, the multi-dimensional task adaptation gap prediction description and related risk warning reference information are formatted and encapsulated. A preset data format, such as JSON or XML, can be used to organize this information according to a specific structure. For instance, in JSON format, an object can be created, storing the multi-dimensional task adaptation gap prediction description and related risk warning reference information as attributes of that object.

[0120] During the encapsulation process, a unique identification code for the original task-executing railway maintenance robot is added to the data packet. This unique identification code can be stored as a field in the data packet so that candidate maintenance robots can identify the source of the request. After generating the collaborative replacement request data packet with the unique identification code, the AI ​​agent of the original task-executing robot controls the original task-executing railway maintenance robot to send the data packet to the candidate maintenance robot with accurate replacement capabilities through a two-way capability-aware communication network.

[0121] Step S400: The AI ​​agent of the candidate maintenance robot that receives the collaborative replacement request combines its own robot's real-time capability description and complete prediction information, and initiates a two-way capability complementarity negotiation with the AI ​​agent of the original task execution robot. The negotiation covers the task split boundary, the operation sequence connection node, and the scope of cross-authorization of permissions, and outputs a standardized replacement collaborative consensus result.

[0122] In one implementation, step S400 may specifically include the following steps S410 to S460:

[0123] Step S410: The AI ​​agent of the candidate maintenance robot receives the collaborative replacement request data packet with a unique identity code sent by the original task execution robot, decodes and verifies the collaborative replacement request data packet, extracts the multi-dimensional task adaptation gap prediction description and associated risk warning reference information in the packet, and obtains the replacement information to be negotiated.

[0124] The AI ​​agent of the candidate maintenance robot is responsible for managing the operation and decision-making of the candidate maintenance robot. For example, the AI ​​agent of the candidate maintenance robot receives a collaborative fill-in request data packet with a unique identification code sent by the original task execution robot through interaction with the communication interface. After receiving the data packet, it first decodes it. Depending on the data format of the data packet, such as JSON or XML, the corresponding decoding algorithm is used to parse the data packet into a readable data structure.

[0125] The decoded data is validated. The integrity of the data packet is checked, ensuring all necessary fields are present. Simultaneously, the accuracy of the data is verified, such as checking if the data format meets requirements and if the data values ​​are within reasonable ranges. Validation can be implemented using preset validation rules and algorithms. For example, checking if the list of potential capability blind spots in the multi-dimensional task adaptation gap prediction description contains valid task node information. The multi-dimensional task adaptation gap prediction description and associated risk warning reference information are extracted from the decoded and validated data packet. This information can be obtained by accessing the corresponding fields in the data structure. For example, in the parsed JSON data, the multi-dimensional task adaptation gap prediction description and associated risk warning reference information are extracted based on field names. The extracted information is then integrated to obtain the information to be negotiated and supplemented.

[0126] Step S420: The AI ​​agent of the candidate maintenance robot retrieves the real-time capability description from the local storage unit of the railway maintenance robot it controls, compares the information to be negotiated for replacement with the real-time capability description, identifies the replacement operation nodes that it can undertake, and obtains the set of fields for its own replacement operation range.

[0127] The AI ​​agent of the candidate maintenance robot is responsible for managing its operation and decision-making. Local storage units are devices used by the candidate maintenance robot to store its own data, such as hard drives and memory. The real-time capability description is a detailed description of the candidate maintenance robot's current operational capabilities, including operational coverage, operational efficiency, and technical capabilities.

[0128] For example, the AI ​​agent of a candidate maintenance robot retrieves the real-time capability description of the railway maintenance robot it controls through a data interface with the local storage unit. The real-time capability description can be stored in a specific file or database table on the local storage unit, and the AI ​​agent obtains the description by executing the corresponding query operation.

[0129] The information to be negotiated for replacement is compared with the real-time capability description. For the list of potential capability blind spots and descriptions of operational risk factors in the information to be negotiated for replacement, the real-time capability description is checked to see if there is a matching operational capability. If the real-time capability description includes the ability to complete a potential capability blind spot operational node or to deal with a operational risk factor node, it means that the candidate maintenance robot can undertake the replacement operational node.

[0130] Text matching algorithms and data association analysis methods can be used for comparison. For example, a string matching algorithm can be used to compare the similarity between the content of the task node and the task capability description in the real-time capability description. If the similarity exceeds a preset threshold, the candidate maintenance robot is considered capable of undertaking the replacement task node. The information of all capable replacement task nodes is organized into a set of fields to obtain the set of fields for its own replacement task scope.

[0131] Step S430: The AI ​​agent of the candidate maintenance robot generates a negotiation request data packet based on its own supplementary work range field set, adds its own unique identity code for controlling the railway maintenance robot, and sends it to the AI ​​agent of the original task execution robot.

[0132] For example, the AI ​​agent of the candidate maintenance robot generates a negotiation request data packet based on its own supplementary work scope field set. A preset data format, such as JSON or XML, can be used to organize and encapsulate the information in its own supplementary work scope field set. A unique identification code for the railway maintenance robot it controls is added to the negotiation request data packet. This unique identification code can be added as a field in the data packet so that the AI ​​agent of the original task-executing robot can accurately identify the source of the request.

[0133] Step S440: The AI ​​agent of the original task execution robot receives the negotiation request data packet, and generates an initial negotiation proposal containing task split boundary description, operation sequence connection node description, and permission cross-authorization scope description based on the multi-dimensional task adaptation gap prediction description and its own set of operation capability fields for controlling the railway maintenance robot, and sends it to the AI ​​agent of the candidate maintenance robot.

[0134] In one implementation, step S440 may specifically include the following steps S441 to S446:

[0135] Step S441: The AI ​​agent of the original task execution robot extracts the list of potential capability blind spots and the description of the nodes causing operational risks from the multi-dimensional task adaptation gap prediction description. Combined with the set of operational capability fields of its own railway maintenance robot, it identifies the operational nodes that it can complete independently and the operational nodes that need to be filled, and obtains the set of basic fields for task splitting.

[0136] For example, the AI ​​agent of the original task execution robot first extracts a list of potential capability blind spots and descriptions of task risk inducing nodes from the multi-dimensional task adaptation gap prediction description. These two parts can be accurately extracted from the text information of the multi-dimensional task adaptation gap prediction description using text parsing algorithms. For example, using regular expression matching, relevant fields for the potential capability blind spot task node list and task risk inducing node description can be found in the description.

[0137] Next, the extracted list of potential capability blind spots and descriptions of operational risk-causing nodes are compared with the set of operational capability fields of the self-controlled railway maintenance robot. For each operational node in the list of potential capability blind spots, the self-controlled operational capability field set is checked to see if there is a capability to complete the operational node. If so, it means that the original task-executing robot can complete the operational node independently; if not, the operational node is a node that needs to be filled. Similarly, a similar comparison and judgment is performed on the nodes in the descriptions of operational risk-causing nodes.

[0138] Text matching algorithms and data association analysis methods can be used for comparison. For example, a string similarity algorithm can be used to calculate the similarity between the content of a task node and its own task capability description. If the similarity exceeds a preset threshold, the robot is considered capable of independently completing the task node. The information of all task nodes that can be completed independently and those that need to be filled in is organized into a set of fields to obtain the basic field set for task splitting.

[0139] Step S442: The AI ​​agent of the original task execution robot defines the work node boundaries between the original task execution robot and the candidate maintenance robot based on the task splitting basic field set, generates the task splitting boundary description field, and clarifies the scope of the work nodes undertaken by each party.

[0140] For example, the AI ​​agent of the original task execution robot analyzes the set of basic fields for task splitting. Based on the work nodes that the original task execution robot can complete independently and the work nodes that need to be filled in, and combined with the filling willingness and capability range of the candidate maintenance robot (obtained from the negotiation request data packet), the work node boundary between the original task execution robot and the candidate maintenance robot is defined.

[0141] When defining the boundaries of work nodes, factors such as the type of work, the area of ​​work, and the difficulty of work are considered. For example, if the work involves the maintenance of equipment along a railway line, the maintenance task is divided into different parts according to the distribution area and type of the equipment, and assigned to the original task-executing robot and candidate maintenance robots respectively. For work nodes that the original task-executing robot has sufficient capacity to complete, it continues to be undertaken by that robot; for work nodes that require supplementary work, they are reasonably assigned to candidate maintenance robots based on their capabilities.

[0142] The defined task node boundaries are described in detail to generate a task splitting boundary description field. This description field can be in text format and lists in detail the specific content, object, and requirements of the task nodes undertaken by the original task execution robot and the candidate maintenance robot.

[0143] Step S443: The AI ​​agent of the original task execution robot extracts the execution time data of similar task nodes from the historical task completion records of the railway maintenance robot it controls, arranges the execution time period for the task nodes of the original task execution robot and the candidate maintenance robot, defines the connection time point between the nodes, and generates the task sequence connection node description field.

[0144] In one implementation, step S443 may specifically include the following steps S4431 to S4436:

[0145] Step S4431: The AI ​​agent of the original task execution robot extracts the historical node execution time data that is the same as the operation node of the maintenance task to be performed from the historical operation completion record of the railway maintenance robot controlled by itself, removes abnormal content in the time data, and obtains the historical node execution time benchmark data set.

[0146] For example, the AI ​​agent of the original task-executing robot retrieves the historical task completion records of the railway maintenance robot it controls through a data interface with the local storage unit. From these historical records, it filters out historical nodes that match the task node to be performed. Precise matching can be achieved using features such as the task node name, the object of the task, and the task type. For instance, if one of the tasks in the pending maintenance is to debug railway signaling equipment X, the system searches the historical task completion records for all task nodes named "Debugging Railway Signaling Equipment X".

[0147] Extract the execution time data for these historical nodes. Organize this execution time data into a dataset. Remove outliers from this dataset. Statistical analysis methods can be used, such as calculating the mean and standard deviation of the data. Data that deviates from the mean by more than a certain multiple of the standard deviation is considered outlier. For example, if the execution time of a historical node is significantly greater or less than the execution time of most other nodes, and exceeds the range of the mean plus or minus three times the standard deviation, then this data is removed as outlier.

[0148] Step S4432: The AI ​​agent of the original task execution robot allocates an execution duration to each job node undertaken by the original task execution robot based on the historical node execution duration benchmark data set, and generates a set of original robot job node execution time period fields.

[0149] For example, the AI ​​agent of the original task execution robot determines each job node undertaken by the original task execution robot based on the task split boundary description field. For each job node, it searches for the corresponding historical node execution time in the historical node execution time benchmark data set. If a corresponding historical node execution time exists, that time is used as the execution time of the job node. If no completely corresponding historical node execution time exists, it can be estimated based on the execution times of similar job nodes.

[0150] For example, if the current work node is to repair railway signal equipment Y, and there is no completely identical work node in the historical node execution time benchmark data set, but there are historical node execution times for repairing similar type of signal equipment, the execution time of the work node can be estimated by making appropriate adjustments based on factors such as the complexity of the equipment and the difficulty of the repair.

[0151] Step S4433: The AI ​​agent of the original task execution robot allocates an execution time to each work node undertaken by the candidate maintenance robot based on the historical node execution time benchmark data set, ensuring that the start time of the candidate robot's work node is completely matched with the end time of the corresponding connecting node of the original robot, and generating a set of candidate robot work node execution time period fields.

[0152] For example, the AI ​​agent of the original task execution robot determines each work node to be undertaken by the candidate maintenance robot based on the task splitting boundary description field. For each candidate maintenance robot work node, the corresponding historical node execution time is searched in the historical node execution time benchmark data set. If a corresponding historical node execution time exists, that time is used as the execution time of the work node. If no completely corresponding historical node execution time exists, it can be estimated based on the execution times of similar work nodes, in a method similar to allocating execution times to the original task execution robot work nodes.

[0153] When assigning execution time slots to candidate maintenance robot work nodes, it is essential to ensure that the start time of each candidate robot work node perfectly matches the end time of its corresponding connecting node. Based on the sequence and connection relationship of the work nodes as determined in the work sequence connecting node description field, the start time of each candidate maintenance robot work node is scheduled. The execution time slot information for each candidate maintenance robot work node is then compiled into a set of fields, generating a candidate robot work node execution time slot field set.

[0154] Step S4434: The AI ​​agent of the original task execution robot merges the original robot task node execution time period field set with the candidate robot task node execution time period field set, sorts them according to the time sequence, and generates a time-series task sorting set.

[0155] For example, the AI ​​agent of the original task-executing robot merges the original robot's set of task node execution time period fields and the candidate robot's set of task node execution time period fields. Data structures such as lists or arrays can be used to merge elements from the two sets into a new set. For example, the execution time period information of each task node in both the original and candidate robot task node execution time period fields can be added as an element to a new list.

[0156] Sort the merged set according to chronological order. Sorting algorithms such as quicksort and mergesort can be used. Sort the sets according to the start time of each job node's execution period, ensuring that the job nodes in the set are arranged in chronological order. Through merging and sorting operations, a time-series task sorted set is generated.

[0157] Step S4435: The AI ​​agent of the original task execution robot performs conflict detection on the time-series task sorting set, identifies task nodes with overlapping execution periods, adjusts the execution duration and start time of the corresponding nodes, and eliminates the overlap of time periods.

[0158] For example, the AI ​​agent of the original task execution robot traverses the time-series task sorting set. For each task node in the set, it checks whether its execution time overlaps with the execution time of other task nodes. This can be determined by comparing the start and end times of the task nodes. Once overlapping task nodes are identified, the cause is analyzed and adjustments are made. If the overlap is due to inaccurate execution time estimation, the execution time can be re-estimated based on historical node execution time benchmark data or the actual situation. For example, if the actual execution time of task node A may be shorter than expected, its execution time can be appropriately shortened. If adjusting the execution time cannot eliminate the overlap, the start time of the task nodes needs to be adjusted. The order of task nodes is reasonably arranged according to the priority and importance of the tasks. For example, if task node A has a higher priority, the start time of task node B is postponed until task node A finishes.

[0159] After adjustments, check again for any remaining overlapping execution periods. Repeat this process until all overlapping periods are eliminated. Through conflict detection and adjustment, ensure that the work nodes of the original task-executing robot and the candidate maintenance robot do not conflict with each other in terms of time, guaranteeing the smooth progress of the operation.

[0160] Step S4436: The AI ​​agent of the original task execution robot converts the corrected time-series task sorting set into a job time-series connection node description field.

[0161] For example, the AI ​​agent of the original task execution robot processes the revised time-series task order set. It organizes and describes the execution time period information of each task node in the set. For each task node, it records its start time, end time, and its connection relationship with other task nodes. This can be done using text descriptions, converting the execution time periods and connection relationships of task nodes into clear and easy-to-understand textual information. For example, the original task execution robot's work node A: 9:00 AM - 11:00 AM; the candidate maintenance robot's work node B: 11:00 AM - 1:00 PM; work node A and work node B are closely connected with no interval time.

[0162] Step S444: Based on the resource access requirements of the work node, the AI ​​agent of the original task execution robot defines the scope of storage resources and control resources that both railway maintenance robots can access during the operation, and generates a description field of the cross-authorization scope.

[0163] The resource access requirements of a work node refer to the usage requirements of storage resources (such as databases and file systems) and control resources (such as equipment controllers and sensors) by the original task execution robot and the candidate maintenance robot when executing the work node. The scope of storage and control resources defines the range of resources that both railway maintenance robots can access and use during the operation. The cross-authorization scope description field provides a detailed description of the resource access permissions for both railway maintenance robots during the operation. For example, the AI ​​agent of the original task execution robot first analyzes the resource access requirements of each work node. Based on the task splitting boundary description field and the work sequence connection node description field, it determines the storage and control resources required by each work node. For example, work node A may need to access equipment fault records in the database, while work node B may need to control the operating status of a certain piece of equipment.

[0164] Based on the analysis results, the scope of storage and control resources that both railway maintenance robots can access during operation is defined. Considering resource security and sharing, resource access permissions are allocated reasonably. For example, for certain sensitive storage resources, only pre-defined robots are allowed access; for some control resources that can be shared, access permissions and usage rules for both parties are set.

[0165] The defined storage and control resource scopes are described in detail to generate a cross-authorization scope description field. This description field can be in the form of a list or table, listing the resource names that each robot can access, the access methods (e.g., read-only, read-write, etc.), and the specific conditions for access. By generating the cross-authorization scope description field, the resource access permissions of the original task-executing robot and the candidate maintenance robot during the operation are clearly defined, avoiding the impact on the operation due to resource access conflicts.

[0166] Step S445: The AI ​​agent of the original task execution robot will structure the task split boundary description field, the operation timing connection node description field, and the permission cross-authorization scope description field, add its own exclusive identity code for controlling the railway maintenance robot, and generate the initial negotiation proposal.

[0167] For example, the AI ​​agent of the original task execution robot can structure the task decomposition boundary description field, the operation sequence connection node description field, and the permission cross-authorization scope description field. A preset data format, such as JSON or XML, can be used to organize these three description fields according to a certain structure. During the structured arrangement process, a unique identification code for the railway maintenance robot it controls is added to the proposal. This unique identification code can be added as a field in the proposal so that the AI ​​agent of the candidate maintenance robot can identify the source of the proposal.

[0168] Step S446: The AI ​​agent of the original task execution robot encrypts the initial negotiation proposal, adds a generation timestamp of the initial negotiation proposal, and sends it to the AI ​​agent of the candidate maintenance robot, recording the sending status and timestamp of the initial negotiation proposal.

[0169] For example, the AI ​​agent of the original task-performing robot uses an encryption algorithm to encrypt the initial negotiation proposal. Symmetric or asymmetric encryption algorithms can be used, such as AES symmetric encryption or RSA asymmetric encryption. During encryption, a pre-negotiated encryption key is used to encrypt the proposal. For example, using the AES algorithm, the plaintext data of the initial negotiation proposal is converted into ciphertext data.

[0170] Add a timestamp to the initial negotiation proposal. The timestamp can be the current system's precise time, such as a value accurate to milliseconds. Add this timestamp to the encrypted proposal as an attribute. Send the encrypted initial negotiation proposal to the AI ​​agent of the candidate maintenance robot via the communication interface of the bidirectional capability-aware communication network. During transmission, use network communication protocols such as TCP / IP to ensure reliable data transmission. Record the transmission status and timestamp of the initial negotiation proposal. Log recording can be used to record the transmission status (e.g., success, failure) and transmission time. If transmission fails, retry the operation until successful transmission.

[0171] Step S450: The AI ​​agent of the candidate maintenance robot receives the initial negotiation proposal, evaluates each description field in the initial negotiation proposal, generates a negotiation feedback proposal, and sends it to the AI ​​agent of the original task execution robot.

[0172] For example, the AI ​​agent of the candidate maintenance robot receives the initial negotiation proposal sent by the AI ​​agent of the original task execution robot through the communication interface of the two-way capability-aware communication network. The initial negotiation proposal is decrypted, and the encrypted proposal is converted into plaintext data using a pre-negotiated decryption key.

[0173] Evaluate each descriptive field in the initial negotiation proposal. For the task splitting boundary description field, check whether it matches the operational capabilities and willingness of the candidate maintenance robots. For example, check whether the work nodes assigned to the candidate maintenance robots are within their capabilities, and whether there are situations where the tasks are too heavy or too light. For the work sequence connection node description field, evaluate whether the time arrangement is reasonable and whether it can guarantee the continuity and efficiency of the work. For example, check whether the start time of the work nodes of the candidate maintenance robots matches their own preparation time and the connection time of the original task execution robot. For the permission cross-authorization scope description field, evaluate whether the resource access permissions meet the operational needs of the candidate maintenance robots, and whether there are issues of permissions being too large or too small.

[0174] If any description field is found to be unreasonable or infeasible, explain the problem in detail and propose modifications. For example, if the task split boundary description field assigns more job nodes to a candidate maintenance robot than its capacity allows, it is recommended to reduce the number of job nodes or adjust the job content. The proposed adjustments can be organized using the same data format as the initial negotiation proposal, such as JSON or XML, according to a defined structure.

[0175] Step S460: The AI ​​agent of the original task execution robot and the AI ​​agent of the candidate maintenance robot interact in multiple rounds based on the initial negotiation proposal and the negotiation feedback proposal until the content of the proposals of both parties is completely consistent. The consistent content is structured and organized to generate a standardized supplementary collaborative consensus result that includes task splitting boundaries, operation timing connection nodes, and cross-authorization scope of permissions.

[0176] For example, the AI ​​agent of the original task-performing robot receives a negotiation feedback proposal from the AI ​​agent of the candidate maintenance robot. It analyzes the proposal to understand the adjustments suggested by the candidate maintenance robot. Based on these adjustments, it modifies the initial proposal, generating a new one. This new proposal is then sent to the candidate maintenance robot's AI agent. The candidate maintenance robot's AI agent receives the new proposal and evaluates it again. If there are still unsatisfactory aspects, it generates a new negotiation feedback proposal and sends it to the original task-performing robot's AI agent. This process is repeated continuously, involving multiple rounds of interaction.

[0177] In each round of interaction, both AI agents must communicate fully to understand each other's needs and opinions. Through negotiation and compromise, they gradually narrow the differences in their proposals. When the proposals are completely consistent, it indicates that a consensus has been reached.

[0178] The consistent content is structured and organized. Using a preset data format, such as JSON or XML, information such as task splitting boundaries, job sequence connections, and cross-authorization scopes are organized according to a specific structure. Unique identification codes for both robots and consensus generation timestamps are added to the structured data for subsequent identification and querying.

[0179] A standardized consensus result for complementary collaboration is generated, including task splitting boundaries, job timing connection nodes, and cross-authorization scope. This result provides clear guidance for the collaborative operation of the original task execution robot and the candidate maintenance robot, ensuring that both parties can cooperate according to a consistent plan during task execution.

[0180] Step S500: The AI ​​agents of the original task execution robot and the candidate maintenance robot jointly generate a collaborative scheduling instruction based on the standardized supplementary collaborative consensus result, which includes a dynamic operation sequence table, cross-authorization rules, and dynamic risk avoidance strategies. The maintenance operation is then completed collaboratively according to the collaborative scheduling instruction.

[0181] In one implementation, step S500 may specifically include the following steps S510-S560:

[0182] Step S510: The AI ​​agents of the original task execution robot and the candidate maintenance robot jointly retrieve the standardized supplementary collaborative consensus results from the local storage unit and extract the fields of task split boundary, operation timing connection node, and permission cross-authorization scope.

[0183] For example, the AI ​​agents of the original task-performing robot and the candidate maintenance robot jointly retrieve standardized supplementary collaboration consensus results through a data interface with the local storage unit. These standardized supplementary collaboration consensus results can be stored in specific files or database tables within the local storage unit, and the AI ​​agents obtain these results by performing corresponding query operations.

[0184] Extract the task splitting boundary, job timing connection node, and permission cross-authorization range fields from the retrieved standardized supplementary collaborative consensus results. This can be done by parsing the data structure of the standardized supplementary collaborative consensus results, such as JSON or XML format, and extracting the corresponding information based on the field names. For example, in JSON data, the values ​​of the task splitting boundary, job timing connection node, and permission cross-authorization range fields can be obtained using key-value pairs.

[0185] Step S520: Based on the task splitting boundary field, the AI ​​agents of the original task execution robot and the candidate maintenance robot allocate the operation nodes of the maintenance task to be executed to the original task execution robot and the candidate maintenance robot, and generate the original robot operation task list and the candidate robot supplementary operation task list.

[0186] For example, the AI ​​agents of the original task execution robot and the candidate maintenance robot allocate the work nodes of the maintenance task to be performed based on the task split boundary field. According to the work scope of both parties as specified in the task split boundary, the work nodes are classified under the original task execution robot and the candidate maintenance robot respectively.

[0187] For example, if the task splitting boundary field specifies that the original task execution robot is responsible for flaw detection of sections A and B of the railway track, and the candidate maintenance robot is responsible for flaw detection of section C of the railway track, then the task nodes for flaw detection of sections A and B will be added to the original robot task list, and the task nodes for flaw detection of section C will be added to the candidate robot supplementary task list.

[0188] Step S530: Based on the job sequence connection node field, the AI ​​agents of the original task execution robot and the candidate maintenance robot bind each task in the original robot job task list and the candidate robot replacement job task list with the corresponding execution time period to generate a dynamic job sequence table.

[0189] For example, the AI ​​agents of the original task execution robot and the candidate maintenance robot extract the start and end times of each task node from the task sequence connection node field. These times are then converted into a standardized timestamp format for easier comparison and sorting later.

[0190] Bind each task in the original robot task list to the corresponding timestamped task node time period field. Data structures such as dictionaries or lists can be used to associate task names with execution time periods. Similarly, bind each task in the candidate robot replacement task list to the corresponding timestamped task node time period field.

[0191] Sort the original robot time-series task field set and the candidate robot time-series task field set according to their timestamp order. Sorting algorithms such as quicksort and mergesort can be used to ensure the tasks are arranged in chronological order. The resulting sorted time-series task set is then generated.

[0192] Perform conflict detection on the sequenced task set. Check for overlaps in the execution times of the tasks. If overlaps are found, analyze the reasons and make adjustments. The execution times of the tasks can be adjusted or the tasks can be reassigned based on factors such as task priority and actual conditions.

[0193] Add a job description and a robot-specific identification code to each task node in the time-series task sorting set. The job description details the specific content of the job task, while the robot-specific identification code distinguishes whether the job task belongs to the original task-executing robot or a candidate maintenance robot. Convert the time-series task sorting set, with the added job description and robot-specific identification code, into a robot-recognizable instruction format.

[0194] Step S540: Based on the cross-authorization scope field, the AI ​​agents of the original task execution robot and the candidate maintenance robot clarify the resource types and resource paths that the original task execution robot and the candidate maintenance robot can access during the operation, and generate cross-authorization rules.

[0195] For example, the AI ​​agents of the original task execution robot and the candidate maintenance robot analyze the cross-authorization scope field. Based on the resource access permission scope specified in the field, the types of resources and resource paths that the original task execution robot and the candidate maintenance robot can access during the operation are determined.

[0196] For example, if the cross-authorization scope field specifies that the original task execution robot can access the device information table in database A, and the candidate maintenance robot can access the job record table in database B, then the resource type is clearly defined as database, and the resource paths are the corresponding tables in database A and database B, respectively.

[0197] Based on clearly defined resource types and paths, cross-authorization rules are generated. These rules specify the access permissions for each resource for both the original task-executing robot and the candidate maintenance robot, such as read-only, read-write, and execute permissions. For example, the original task-executing robot may be granted read-only permissions to the equipment information table in database A, while the candidate maintenance robot may be granted read-write permissions to the job record table in database B.

[0198] Step S550: The AI ​​agents of the original task execution robot and the candidate maintenance robot formulate countermeasures for the risk-inducing nodes based on the multi-dimensional task adaptation gap prediction description and related risk warning reference information, and generate dynamic risk avoidance strategies.

[0199] For example, the AI ​​agents of the original task-performing robot and the candidate maintenance robot retrieve multi-dimensional task adaptation gap prediction descriptions and associated risk warning reference information from the local storage unit. They then analyze the operational risk trigger nodes to understand the specific circumstances of each risk trigger node and the potential risks it may cause.

[0200] For each operational risk trigger node, specific countermeasures are formulated based on the set of response direction fields in the associated risk warning reference information. For example, if the operational risk trigger node is due to equipment failure caused by severe weather, countermeasures may include protecting the equipment before the arrival of severe weather, such as adding protective covers and checking the waterproof performance of the equipment; strengthening equipment monitoring during severe weather to obtain real-time equipment operating status; and promptly activating emergency maintenance plans if equipment failure occurs.

[0201] The corresponding countermeasures developed for each operational risk trigger will be integrated to generate a dynamic risk avoidance strategy. This strategy may include risk monitoring mechanisms, risk warning thresholds, and emergency response procedures. For example, warning thresholds may be set for parameters such as equipment temperature and humidity, and a warning may be issued promptly when these parameters exceed the thresholds; emergency response procedures may be developed, clearly defining the handling steps and division of responsibilities under different risk conditions.

[0202] Step S560: The AI ​​agents of the original task execution robot and the candidate maintenance robot encapsulate the dynamic operation sequence table, cross-authorization rules, and dynamic risk avoidance strategies, generate collaborative scheduling instructions with unique identity codes, and send them to the railway maintenance robots they control. The control robots then execute maintenance operations according to the collaborative scheduling instructions.

[0203] For example, the AI ​​agents of the original task-executing robot and the candidate maintenance robot encapsulate dynamic work sequence tables, cross-authorization rules, and dynamic risk avoidance strategies. These three parts of information can be organized according to a pre-defined data format, such as JSON or XML.

[0204] During the encapsulation process, unique identification codes for both the original task-executing robot and the candidate maintenance robot are added to the collaborative scheduling instructions. These unique identification codes can be added as a field to the instructions, allowing the robots to identify the source and applicable objects of the instructions. After generating the collaborative scheduling instructions with unique identification codes, they are sent to the respective controlled railway maintenance robots via the communication interface of the bidirectional capability-aware communication network. During transmission, network communication protocols, such as TCP / IP, are used to ensure reliable data transmission.

[0205] After receiving the collaborative scheduling instruction, the original task-executing robot and the candidate maintenance robot parse the instruction content. Based on the time arrangement of the dynamic job sequence table, they begin executing the corresponding job tasks. During the operation, they adhere to the cross-authorization rules and access resources appropriately. Simultaneously, dynamic risk avoidance strategies are applied in real time to address potential risks.

[0206] This invention also provides a collaborative scheduling system, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements the steps in the collaborative scheduling method for railway maintenance robots based on AI agents provided in this invention.

[0207] Please see details. Figure 2 This is a schematic diagram of a collaborative scheduling system provided in an embodiment of the present invention. Figure 2 As shown, the aforementioned collaborative scheduling system 1000 may include: a processor 1001, a network interface 1004, and a memory 1005. Furthermore, the collaborative scheduling system 1000 may also include: a user interface 1003, and at least one communication bus 1002. The communication bus 1002 is used to implement communication between these components. The user interface 1003 may include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1005 may also be at least one storage device located remotely from the aforementioned processor 1001. Figure 2 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a device control application.

Claims

1. A collaborative scheduling method for railway maintenance robots based on AI agents, characterized in that, The method includes: Each railway maintenance robot detects heterogeneous railway maintenance robots in the neighborhood, establishes a two-way capability perception and communication network among multiple robots, and each AI agent synchronously controls its own robot to upload real-time operating status data and historical work completion records, generating a neighborhood real-time capability sharing map that associates the capabilities of all networked robots. Based on the real-time capability sharing map of the neighborhood, combined with the demand benchmark of the current preset maintenance task to be executed, the AI ​​agent of the original task execution robot simulates the task deduction and predicts the potential capability blind spots and operational risk factors of its own task execution, and outputs multi-dimensional task adaptation gap prediction description and related risk warning reference information. The AI ​​agent of the original task-executing robot performs cross-robot capability complementarity matching calculations based on the real-time capability sharing graph of the neighborhood, the multi-dimensional task adaptation gap prediction description, and the associated risk warning reference information. It then locates candidate maintenance robots with precise replacement capabilities and schedules the original task-executing railway maintenance robot to send a collaborative replacement request data packet carrying complete prediction information to the candidate maintenance robot. Specifically, the AI ​​agent of the original task-executing robot extracts the set of operational capability fields of all other railway maintenance robots in the network from the real-time capability sharing graph of the neighborhood, extracts a list of potential capability blind spots and descriptions of operational risk inducing nodes from the multi-dimensional task adaptation gap prediction description, and extracts directional guidance from the associated risk warning reference information to obtain a basic data set for replacement matching. The AI ​​agent then compares the list of potential capability blind spots and descriptions of operational risk inducing nodes in the basic data set with the operational capability field sets of other railway maintenance robots to identify railway maintenance robots that can cover the corresponding node content. The robot obtains a preliminary set of robot-specific identity codes; extracts the real-time operating status data fields corresponding to the preliminary set of robot-specific identity codes from the neighborhood real-time capability sharing map, filters out robot-specific identity codes whose current real-time operating status is not idle, and obtains a set of candidate robot-specific identity codes for supplementary positions; analyzes the set of operational capability fields corresponding to the set of candidate robot-specific identity codes for supplementary positions, and generates a supplementary adaptation analysis result containing a supplementary coverage node description and a supplementary timing connection description for each candidate robot; based on the supplementary adaptation analysis result, selects candidate robots that can cover all potential capability blind spot operational nodes and positions them as candidate maintenance robots with precise supplementary capabilities; encapsulates the multi-dimensional task adaptation gap prediction description and the associated risk warning reference information in a formatted manner to generate a collaborative supplementary request data packet with a unique identity code, and controls the original task executing railway maintenance robot to send the collaborative supplementary request data packet to the candidate maintenance robot with precise supplementary capabilities; The AI ​​agent of the candidate maintenance robot that receives the collaborative replacement request combines its own robot's real-time capability description with the complete prediction information, and initiates a two-way capability complementarity negotiation with the AI ​​agent of the original task execution robot. The negotiation covers the task split boundary, the operation sequence connection node, and the scope of cross-authorization of permissions, and outputs a standardized replacement collaborative consensus result. The AI ​​agents of the original task execution robot and the candidate maintenance robot jointly generate a collaborative scheduling instruction based on the standardized supplementary collaborative consensus result. This instruction includes a dynamic operation sequence table, cross-authorization rules, and dynamic risk avoidance strategies. The maintenance operation is then completed collaboratively according to the collaborative scheduling instruction.

2. The method as described in claim 1, characterized in that, Each railway maintenance robot detects heterogeneous railway maintenance robots in its neighborhood, establishes a two-way capability-aware communication network among multiple robots, and each AI agent synchronously controls its own robot to upload real-time operating status data and historical task completion records, generating a neighborhood real-time capability sharing map that associates the capabilities of all networked robots, including: Each railway maintenance robot sends a detection signal with a unique identification code to the surrounding space, receives response signals with unique identification codes returned by heterogeneous railway maintenance robots in the surrounding space, decodes and verifies the signal codes, filters invalid coded signals, and obtains a set of unique identification codes for valid heterogeneous railway maintenance robots in the surrounding space. Each AI agent initiates a two-way communication link handshake request with the corresponding coded robot based on the set of unique identity codes for the effective heterogeneous railway maintenance robots in the neighborhood, performs link encryption verification and connectivity status confirmation, and obtains a two-way capability perception communication network among multiple robots. Each AI agent retrieves real-time operating status data and historical work completion records from the local storage unit of the railway maintenance robot it controls, performs unified data format conversion, and generates standardized operation information. Each AI agent uploads the standardized operation information to the shared storage node of the two-way capability perception communication network, and simultaneously obtains the standardized operation information uploaded by other railway maintenance robots in the network, thus obtaining a set of standardized operation information for the entire network. Based on the standardized operation information set of the entire network, the operation coverage, historical operation completion records and real-time operation status data of each railway maintenance robot are mapped by field association to generate a neighborhood real-time capability sharing map that associates the capabilities of all networked robots. The neighborhood real-time capability sharing map is synchronized to the local storage unit of all AI agents in the bidirectional capability-aware communication network, and the synchronization timestamp of the neighborhood real-time capability sharing map is updated.

3. The method as described in claim 2, characterized in that, Based on the standardized operational information set of the entire network, the system performs field association mapping on the operational coverage, historical operation completion records, and real-time operational status data of each railway maintenance robot to generate a neighborhood real-time capability sharing map that associates the capabilities of all networked robots, including: Extract the operation coverage field, historical operation completion record field, and real-time operation status data field of each railway maintenance robot from the standardized operation operation information set of the whole network, remove redundant content in the fields, and obtain the basic field set of single robot operation capability. Each field in the basic field set of single robot operation capabilities is bound to the unique identification code of the corresponding railway maintenance robot to generate a set of operation capability fields with unique identification codes; Cross-compare the set of operational capability fields with unique identification codes to identify overlapping content in the operational coverage field of different railway maintenance robots, similar content in the historical operation completion record field, and complementary content in the real-time operation status data field, to obtain a set of capability association fields within the network. Among them, similar content refers to content with a similarity exceeding a preset threshold. Based on the set of capability association fields within the network, the operational capability fields of each railway maintenance robot are mapped topologically according to the associated content to generate underlying topology data with unique identification codes, operational coverage areas, and capability association links. The real-time operating status data fields of each railway maintenance robot are embedded into the topology node area corresponding to its unique identity code, and similar content of historical operation completion record fields are embedded into the middle area of ​​the corresponding capability association link to generate a neighborhood real-time capability sharing map. The integrity of the neighborhood real-time capability sharing graph is verified to confirm that all fields and associated content have been embedded, and a verification pass mark is generated for the neighborhood real-time capability sharing graph.

4. The method as described in claim 1, characterized in that, The AI ​​agent of the original task execution robot, based on the real-time capability sharing map of the neighborhood and combined with the demand benchmark of the current preset maintenance task to be executed, simulates task deduction and predicts the potential capability blind spots and operational risk factors of its own task execution, and outputs multi-dimensional task adaptation gap prediction descriptions and related risk warning reference information, including: The AI ​​agent of the original task execution robot retrieves the set of requirement benchmark fields for the current preset maintenance task to be executed from the local storage unit, and extracts the set of operation capability fields for controlling the railway maintenance robot from the real-time capability sharing map of the neighborhood. The set of requirement benchmark fields for the current preset maintenance task to be executed is compared with the set of operation capability fields of the railway maintenance robot controlled by itself, and the fields in the requirement benchmark that are not covered by its own capabilities are identified to obtain a preliminary set of requirement difference fields. Based on the aforementioned set of preliminary requirement difference fields, the data of each job node of the task to be executed is mapped, a simulated job underlying data flow scenario is generated, and the node-by-node flow of the simulated job nodes is initiated. The process of simulated work nodes is monitored in real time, and situations such as node stagnation and node jump failure during the process are identified. The corresponding work node data is recorded to obtain a set of abnormal node data in the simulation. The simulated abnormal node data set is classified and organized. The work nodes corresponding to node stagnation are classified into the potential capability blind spot field, and the work nodes corresponding to node jump failure are classified into the work risk cause field, thus obtaining the classified abnormal node field set. Based on the classified set of abnormal node fields, a multi-dimensional task adaptation gap prediction description is generated, which includes a list of potential capability blind spots in operation nodes and descriptions of operation risk-causing nodes. A set of response direction fields is generated for each abnormal node, and related risk warning reference information is obtained.

5. The method as described in claim 4, characterized in that, The process of mapping data for each job node of the task to be executed based on the initial set of demand difference fields, generating the underlying data flow scenario of the simulated job, and initiating the node-by-node flow of the simulated job nodes includes: Based on each job node field in the preliminary demand difference field set, corresponding node flow rule data is generated, defining the node's input data requirements, output data format, and flow triggering conditions; The set of operational capability fields of the self-controlled railway maintenance robot is input into the underlying data flow scenario of the simulated operation. The data drives the flow of each operation node according to the flow rules. The input data, output data, and flow duration of each node are recorded to obtain the set of simulated operation node flow data. The simulated operation node flow data set is analyzed in real time, and the matching of the output data of each node with the preset output format is compared. Nodes with mismatched output data are identified and marked as flow abnormal nodes. For the abnormal nodes in the flow, trace the matching between their input data and the flow rule data, identify nodes whose input data does not meet the rule requirements as nodes that are stagnant, identify nodes that are not covered by the flow rules as nodes that have failed to jump, and obtain a set of abnormal node type tags. The abnormal node type tag set is bound to the corresponding job node data to generate an abnormal node data set with type tags; The abnormal node data set with type tags is synchronized to the temporary storage unit of the original task execution robot AI agent, and the record timestamp of the abnormal node data set with type tags is updated.

6. The method as described in claim 1, characterized in that, The process involves comparing the potential capability blind spot operation node list and operation risk inducement node description in the basic data set of the supplementary matching with the operation capability field set of other railway maintenance robots, one by one, to identify railway maintenance robots that can cover the corresponding node content, and obtaining a preliminary matching robot-specific identity code set, including: For each node in the potential capability blind spot operation node list and operation risk cause node description in the supplementary matching basic data set, set a unique requirement identifier field; Extract the operation coverage node field of each robot from the set of operation capability fields of other railway maintenance robots in the network, and set a unique capability identifier field for each operation coverage node; The requirement identifier field and the capability identifier field are compared to identify identifier pairs with the same job node content, and a set of requirement-capability identifier association fields is generated. Traverse the capability identifier field of each other railway maintenance robot in the network, count the number of identifiers that appear in the demand-capability identifier association field set, and obtain the matching identifier count for each robot; Bind the matching identifier count of each robot to its corresponding unique identity code to generate a set of matching count fields with unique identity codes; Select robot-specific identification codes that can cover all the aforementioned requirement identification fields to obtain a preliminary set of matching robot-specific identification codes.

7. The method as described in claim 6, characterized in that, The comparison of the demand identifier field and the capability identifier field identifies identifier pairs with the same job node content, generating a set of demand-capability identifier association fields, including: The content of the job node corresponding to the demand identifier field is numerically converted to generate the demand node value; the content of the job coverage node corresponding to the capability identifier field is numerically converted to generate the capability node value. The numerical values ​​of the demand nodes are compared with the numerical values ​​of the capability nodes to identify pairs of identifiers with exactly the same value, and a set of numerical matching identifier pairs is generated. The content of the job node corresponding to the numerical matching identifier pair set is verified a second time to confirm that the text description and job requirements of the node content are completely consistent, and a set of identifier pairs that have passed the verification is generated. The set of verified identifier pairs is stored in the shared storage node of the bidirectional capability-aware communication network, and the generation timestamp of the set of verified identifier pairs is updated. Retrieve the historical demand-capability identifier association field set from the shared storage node, compare it with the current set of verified identifier pairs, update duplicate associations, add new associations, and obtain the updated demand-capability identifier association field set; The updated set of requirement-capability identifier association fields is synchronized to the local storage unit of all AI agents within the bidirectional capability-aware communication network.

8. The method as described in claim 1, characterized in that, The AI ​​agent of the candidate maintenance robot receiving the collaborative replacement request combines its own robot's real-time capability description with the complete prediction information, and initiates a two-way capability complementarity negotiation with the AI ​​agent of the original task execution robot. The negotiation covers task splitting boundaries, work sequence connection nodes, and cross-authorization scope, and outputs a standardized replacement collaboration consensus result, including: The AI ​​agent of the candidate maintenance robot receives a collaborative replacement request data packet with a unique identity code sent by the original task execution robot. It decodes and verifies the collaborative replacement request data packet, extracts the multi-dimensional task adaptation gap prediction description and associated risk warning reference information in the packet, and obtains the replacement information to be negotiated. The AI ​​agent of the candidate maintenance robot retrieves the real-time capability description from the local storage unit of the railway maintenance robot it controls, compares the information to be negotiated for replacement with the real-time capability description, identifies the replacement operation nodes that it can undertake, and obtains the set of fields for its own replacement operation range. The AI ​​agent of the candidate maintenance robot generates a negotiation request data packet based on the set of fields for its own supplementary work scope, adds its own unique identity code for controlling the railway maintenance robot, and sends it to the AI ​​agent of the original task execution robot. The AI ​​agent of the original task execution robot receives the negotiation request data packet, and based on the multi-dimensional task adaptation gap prediction description and the set of operational capability fields of its own control of the railway maintenance robot, generates an initial negotiation proposal containing a task split boundary description, an operational timing connection node description, and a permission cross-authorization scope description, and sends it to the AI ​​agent of the candidate maintenance robot. The AI ​​agent of the candidate maintenance robot receives the initial negotiation proposal, evaluates each description field in the initial negotiation proposal, generates a negotiation feedback proposal, and sends it to the AI ​​agent of the original task execution robot. The AI ​​agent of the original task execution robot and the AI ​​agent of the candidate maintenance robot interact in multiple rounds based on the initial negotiation proposal and the negotiation feedback proposal until the content of the proposals of both parties is completely consistent. The consistent content is structured and organized to generate a standardized supplementary collaborative consensus result that includes task splitting boundaries, operation timing connection nodes, and cross-authorization scope of permissions.

9. A collaborative scheduling system, characterized in that, include: processor; And a memory, wherein the memory stores computer-readable code that, when executed by the processor, causes the processor to perform the method as described in any one of claims 1 to 8.

Images