An industrial collaborative system based on space-time intelligent agent

Abstract

The application belongs to the technical field of industrial collaborative system based on space-time agent, and particularly relates to an industrial collaborative system based on space-time agent, which is used to solve the problems of serious data island, high delay of centralized decision, low collaborative efficiency and lack of adaptive intelligence in industrial scenarios. The system comprises an edge device intelligent collaborative module, a space-time data fusion and cognitive modeling module, a collaborative decision and control execution module, a safety and operation monitoring module and a visualization and user interaction module. Through the collaborative operation of multi-level agents at the device end, edge and cloud, the system realizes multi-source heterogeneous data fusion, local autonomous response, global optimization decision driven by large models, end-to-end safety monitoring and human-computer collaborative interaction. The application significantly improves the intelligentization, autonomy and collaborative performance of the industrial system, reduces the communication delay and bandwidth pressure, enhances the system reliability and manageability, and realizes the self-evolution transition from single-point intelligence to group intelligence.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligence and multi-agent collaborative control technology, specifically relating to an industrial collaborative system based on spatiotemporal intelligent agents. Background Technology

[0002] Currently, similar technologies in the Industrial Internet field mainly focus on areas such as "industrial big data analysis platforms," ​​"digital twin systems," and "manufacturing execution systems (MES)." These systems typically rely on cloud computing, edge computing, and IoT devices to achieve remote monitoring of industrial equipment, production process optimization, and data visualization analysis. Existing technologies mostly collect field data through hardware devices such as sensor networks, PLC controllers, and industrial gateways, transmitting it to the cloud via an Industrial Internet platform, and utilizing AI algorithms for predictive maintenance, energy consumption optimization, and production scheduling. At the software level, the system generally includes a data acquisition module, a communication module, an analysis and decision-making module, and a visualization management module. Its workflow is as follows: the underlying hardware senses the operating status of industrial equipment, uploads it to the cloud platform via an edge gateway, the platform uses machine learning models to perform status analysis and decision-making, and then issues control commands to achieve a certain degree of "intelligent closed loop."

[0003] However, these systems suffer from significant technical flaws and application bottlenecks. First, existing industrial internet platforms primarily rely on static data analysis, lacking the ability to integrate and coordinate spatiotemporally dynamic information, thus hindering intelligent scheduling of complex industrial systems across regions and time scales. Second, most AI algorithms are limited to local optimization (such as predictive maintenance or single-point energy consumption optimization), lacking global intelligent agent decision-making mechanisms across devices and factories. Third, existing systems rely on manual configuration and centralized cloud training for model updates and intelligent strategy iterations, resulting in slow response times, poor adaptability, and difficulty in handling the rapid changes in equipment status in industrial scenarios. Furthermore, the problem of data silos remains prominent; inconsistent interface standards between devices and systems from different manufacturers lead to difficulties in data sharing and low collaborative efficiency. In terms of security, existing solutions mostly employ traditional access control or encryption mechanisms, failing to provide security constraints and anomaly detection for multi-agent behaviors across spatiotemporal dimensions. The interplay of these issues often leads to technical challenges in handling complex industrial collaborative tasks, including insufficient real-time decision-making, limited depth of intelligence, and difficulty in guaranteeing system reliability. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide an industrial collaborative system based on spatiotemporal intelligent agents to solve the problems of severe data silos, high latency in centralized decision-making, low collaborative efficiency and lack of adaptive intelligence in existing industrial Internet systems, and to realize intelligent collaboration, dynamic optimization and secure autonomy of industrial equipment across spatiotemporal dimensions.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide an industrial collaborative system based on spatiotemporal intelligent agents, deployed in a multi-level industrial control environment comprising an edge layer, an intelligent agent layer, a collaborative decision-making layer, and a visualization layer. The system includes an edge device intelligent collaboration module, a spatiotemporal data fusion and cognitive modeling module, a collaborative decision-making and control execution module, a safety and operation monitoring module, and a visualization and user interaction module, wherein:

[0006] The edge device intelligent collaboration module is deployed in the edge computing node or industrial control terminal with computing capabilities in the industrial field. It is directly connected to the field sensors, actuators, programmable logic controllers and industrial robots. It is configured to collect, preprocess and extract features from multi-source sensor data in real time, and execute a lightweight spatiotemporal reasoning and decision-making model through a local intelligent agent engine to achieve dynamic perception of equipment status, detection of local anomalies and autonomous adjustment of strategies.

[0007] The spatiotemporal data fusion and cognitive modeling module is deployed in a regional edge server or cloud computing cluster, maintaining a two-way data channel with each edge node. It is configured to access multi-source heterogeneous data through standardized interfaces and industrial communication protocols, construct a dynamic data index across time and space, and combine knowledge graphs and deep learning algorithms to perform semantic understanding and causal correlation mining on spatiotemporal data, thereby generating a multi-dimensional cognitive model of equipment health status, production efficiency, and energy consumption patterns.

[0008] The collaborative decision-making and control execution module is deployed in the cloud or regional main control server. It is configured to integrate a large-scale language model that has been fine-tuned with industrial scenario knowledge. Based on the industrial data context, it automatically generates control strategies, diagnostic reports or production scheduling schemes. Through multi-agent collaborative algorithms, it dynamically allocates resources and tasks, transforms high-level semantic strategies into executable control instructions that meet the real-time and security requirements of equipment, and finally sends them to the industrial control system, programmable logic controller and robot controller to form a closed-loop control link.

[0009] The security and operation monitoring module is deployed in a hybrid environment of cloud and edge nodes. It is configured to implement identity authentication and access control through hardware root of trust and software security policies, establish an end-to-end secure communication channel between the edge and the cloud, monitor the performance indicators, task execution status and network communication quality of each module of the system in real time, and automatically trigger policy rollback, local isolation or alarm notification mechanisms when an anomaly or fault is detected.

[0010] The visualization and user interaction module is deployed on a cloud server or management terminal. It is configured to visualize spatiotemporal data, device status, agent inference results, and global decision-making strategies in the form of charts, heat maps, or 3D digital twins. It provides an interactive control interface to support users in real-time adjustments to agent behavior, scheduling strategies, or security strategies. It also ensures that user input can be promptly transmitted to the collaborative decision-making and execution module and implemented in specific device operations through an operation command feedback channel.

[0011] The edge device intelligent collaboration module includes an edge computing unit, a communication management unit, and a local intelligent agent engine.

[0012] The edge computing unit is used to collect multi-source industrial data from sensors, actuators and industrial control terminals connected to it at a preset sampling period, and to perform time alignment, noise reduction and feature extraction on the multi-source industrial data to generate a device status feature vector containing timestamps and device identifiers.

[0013] The local intelligent agent engine is used to embed a lightweight spatiotemporal reasoning and decision-making model, calculate the device operating status statistics within a preset time window, and compare the statistics with a preset operating threshold or historical baseline to determine whether the device operating status is abnormal or whether the computing load has changed.

[0014] The communication management unit is used to send the corresponding device state feature vector and judgment result to the spatiotemporal data fusion and cognitive modeling module via industrial Ethernet or 5G network when the judgment result generated by the local intelligent agent engine meets the reporting conditions, so as to be used for cross-node state association modeling or collaborative decision processing.

[0015] The spatiotemporal data fusion and cognitive modeling module includes a data access layer, a spatiotemporal data storage layer, and a cognitive modeling engine.

[0016] The data access layer is used to receive device status feature vectors uploaded from multiple edge device intelligent collaboration modules, and based on device identifiers, data types and timestamp information, performs semantic alignment and format unification on data from different factories and different brands of devices to generate standardized status data.

[0017] The spatiotemporal data storage layer includes a distributed time-series database and a graph database. It writes the standardized state data received in the data access layer into the distributed time-series database and the graph database, and constructs a dynamic data index across time and space to realize state evolution, event tracing and trend prediction.

[0018] The cognitive modeling engine, based on the dynamic state index, performs statistical feature calculations on equipment state data within a preset time window, and combines the equipment topology relationships recorded in the graph database to perform correlation analysis on equipment operating status, production efficiency, and energy consumption characteristics, generating structured cognitive information that reflects the evolution of equipment state and their mutual influence.

[0019] The structured cognitive information includes equipment status assessment results, the influence relationships between related equipment, and status parameters used for collaborative decision-making.

[0020] The collaborative decision-making and control execution module includes a large model intelligent agent center, a collaborative scheduling unit, a strategy generation and optimization unit, and a control execution unit.

[0021] The large model intelligent agent center is configured to receive structured cognitive information output by the spatiotemporal data fusion and cognitive modeling module, and construct standardized context input for large model reasoning. Based on the industrial data context, it automatically generates control strategies, diagnostic reports or production scheduling schemes, and decomposes the global production target into a series of executable atomic tasks through the thinking chain reasoning technology.

[0022] The collaborative scheduling unit is configured to construct an integer linear programming (MILP) scheduling model based on the verified task graph and node load information, and generate task scheduling results with minimizing total delay and load peak as the objective function.

[0023] The strategy generation and optimization unit is configured to generate a control strategy representation and perform closed-loop constraint optimization based on the task scheduling results generated by the collaborative scheduling unit.

[0024] The control execution unit is configured to map the control strategy representation generated by the strategy generation and optimization unit into industrial control protocol instructions, and issue instructions through the encrypted communication channel established by the safety and operation monitoring module, collect execution feedback data to form execution receipts, and send the execution receipts back to the spatiotemporal data fusion and cognitive modeling module to update the state index and cognitive information.

[0025] The security and operation monitoring module further includes an identity authentication and access control subsystem, a data and communication encryption unit, an operation status monitoring engine, and an abnormal event response mechanism.

[0026] The identity authentication and access control subsystem is configured to assign a unique digital identity to each device and operator connected to the system, and to ensure the legitimacy of access by intelligent agents and operators through hardware root of trust and software security policies.

[0027] The data and communication encryption unit is configured to establish an encrypted communication channel between the edge device intelligent collaboration module, the spatiotemporal data fusion and cognitive modeling module, and the collaborative decision-making and control execution module, and to add hash verification and digital signature to the reported device status feature vector and structured cognitive information, so that the integrity verification is completed at the receiving end before entering the subsequent modeling or decision-making process.

[0028] The operation status monitoring engine is configured to collect operation indicators from each module and form a unified monitoring data stream;

[0029] The abnormal event response mechanism is configured to trigger a control strategy rollback and generate an abnormal event record when the operation status monitoring engine detects that the delay in issuing control commands exceeds a preset delay threshold or the number of execution receipt timeouts exceeds a preset number threshold.

[0030] The visualization and user interaction module further includes a data visualization engine, an interactive control interface, a running status display unit, and an operation instruction feedback channel.

[0031] The data visualization engine is configured to receive structured cognitive information from the spatiotemporal data fusion and cognitive modeling module and task scheduling results from the collaborative decision-making and control execution module, and to visualize and render the device operating status, node load status and task execution status based on device identifier, timestamp and spatial topology information.

[0032] The interactive control interface is configured to provide operators with a graphical interface-based interactive entry point, which is used to receive manual adjustment instructions within a preset permission range.

[0033] The operation status display unit is configured to centrally display key operation indicators during the system operation process;

[0034] The operation instruction feedback channel is configured to perform format verification and permission verification on the manual adjustment instruction after receiving it, and send the verified manual adjustment instruction to the collaborative decision-making and control execution module to participate in subsequent task scheduling or control strategy generation. At the same time, it records the manual adjustment instruction and its effective result to form an operation log.

[0035] The beneficial effects of the above technical solution of the present invention are as follows:

[0036] This invention constructs a multi-layered spatiotemporal intelligent agent collaborative architecture, which combines the cognitive and reasoning capabilities of large-scale intelligent agents to achieve distributed autonomous decision-making and adaptive optimization across the cloud, edge, and terminal. At the same time, it can integrate multi-source data in real time, break down data silos, improve the accuracy, collaboration, and response speed of decision-making, and significantly enhance the intelligence level and overall operating efficiency of industrial systems. Attached Figure Description

[0037] Figure 1 This is a system architecture diagram of the industrial collaborative system based on spatiotemporal intelligent agents in this invention;

[0038] Figure 2 This is a flowchart of the collaborative decision-making and control execution module in this invention. Detailed Implementation

[0039] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0040] This invention provides an industrial collaborative system based on spatiotemporal intelligent agents to address the problems of severe data silos, high latency in centralized decision-making, low collaborative efficiency, and lack of adaptive intelligence in existing industrial internet systems. Specifically, the data silos refer to the lack of unified data standards and poor interface compatibility among different devices, platforms, and production processes, making data interoperability and sharing difficult and hindering the full exploitation of information value. High latency in centralized decision-making means that traditional systems rely on a central node for unified scheduling and analysis, which can easily create data transmission and computing bottlenecks in complex and ever-changing industrial environments, preventing real-time decision-making. Multi-layered collaboration is inefficient; there is a lack of intelligent collaboration mechanisms between the cloud, edge, and terminals, with task allocation and status feedback relying on manual configuration, resulting in a lack of dynamic coordination capabilities for the system as a whole. Finally, existing solutions are mostly rule-driven or static models, lacking continuous learning, experience transfer, and self-optimization capabilities, making it difficult to automatically adjust strategies in response to environmental changes.

[0041] Therefore, the industrial collaborative system based on spatiotemporal intelligent agents provided by this invention, by deploying multi-layered intelligent agents with perception, reasoning, and learning capabilities at the cloud, edge, and device layers, achieves cross-spatial fusion of industrial data, distributed intelligent decision-making, and dynamic self-optimization. This effectively solves the technical bottlenecks in existing systems, such as difficulty in information sharing, non-real-time decision-making, and lack of system self-evolution, significantly improving the intelligence, autonomy, and collaborative performance of industrial production processes. Figure 1 As shown, the industrial collaborative system based on spatiotemporal intelligent agents includes an edge device intelligent collaboration module, a spatiotemporal data fusion and cognitive modeling module, a collaborative decision-making and control execution module, a safety and operation monitoring module, and a visualization and user interaction module.

[0042] The edge device intelligent collaboration module provides fundamental support and real-time response functions for the spatiotemporal intelligent agent system, and is a key component for realizing autonomous industrial site decision-making and low-latency decision-making. Deployed in edge computing nodes or industrial control terminals with computing capabilities in the industrial site, this module is directly connected to field sensors, actuators, programmable logic controllers, and industrial robots. It is configured to perform real-time acquisition, preprocessing, and feature extraction of multi-source sensor data, and execute lightweight spatiotemporal inference and decision-making models through a local intelligent agent engine to achieve dynamic perception of equipment status, detection of local anomalies, and autonomous strategy adjustment.

[0043] In this embodiment, the edge device intelligent collaboration module consists of three parts: an edge computing unit, a communication management unit, and a local intelligent agent engine. The edge computing unit collects multi-source industrial data from connected sensors, actuators, and industrial control terminals at a preset sampling period. It then performs time alignment, noise reduction, and feature extraction on the multi-source industrial data to generate a device status feature vector containing timestamps and device identifiers, ensuring data accuracy and timeliness. The local intelligent agent engine embeds a lightweight spatiotemporal reasoning and decision-making model, possessing the ability to dynamically perceive device status, detect local anomalies, and autonomously adjust strategies. Based on the device status feature vectors from sensors, actuators, and industrial control terminals, it calculates device operating status statistics within a preset time window and compares these statistics with a preset operating threshold or historical baseline to determine whether the device operating status is abnormal or whether the computational load has changed.

[0044] The communication management unit is used to conduct highly reliable communication with the upper-layer collaborative center through industrial Ethernet, 5G or TSN network. When the judgment result generated by the local intelligent agent engine meets the reporting conditions, it sends the corresponding device state feature vector and judgment result to the spatiotemporal data fusion and cognitive modeling module through industrial Ethernet or 5G network for cross-node state association modeling or collaborative decision processing, so as to realize high-speed data interaction across devices and nodes.

[0045] When the industrial environment fluctuates or external tasks change, this module can respond instantly and self-adjust without relying on the cloud, and share the processing results with the upper-level intelligent agent when necessary, thus forming a hybrid control mode that combines edge autonomy and cloud intelligence. This design not only significantly reduces network latency and bandwidth pressure, but also improves the safety, reliability, and task continuity of industrial field operations.

[0046] The spatiotemporal data fusion and cognitive modeling module is the core data hub of the system, responsible for unified modeling and intelligent analysis of multi-source heterogeneous data from edge devices, production line sensors, and the external environment. Deployed in a regional edge server or cloud computing cluster, it maintains bidirectional data channels with each edge node and is configured to access multi-source heterogeneous data through standardized interfaces and industrial communication protocols. It constructs a dynamic data index across time and space, and combines knowledge graphs and deep learning algorithms to perform semantic understanding and causal correlation mining on spatiotemporal data, thereby generating a multi-dimensional cognitive model of equipment health status, production efficiency, and energy consumption patterns.

[0047] In this embodiment, the spatiotemporal data fusion and cognitive modeling module includes a data access layer, a spatiotemporal data storage layer, and a cognitive modeling engine, wherein,

[0048] The data access layer is configured to perform semantic alignment on data from different factories and different brands of equipment, and receive device status feature vectors uploaded from multiple edge device intelligent collaboration modules. Based on device identifiers, data types, and timestamp information, it performs semantic alignment and format unification on data from different factories and different brands of equipment through standardized interfaces and industrial communication protocols (such as OPC UA, MQTT, Modbus TCP, etc.) to generate standardized status data, thereby realizing data collection and format unification for different factories and different equipment.

[0049] The spatiotemporal data storage layer includes a distributed time-series database and a graph database. Standardized state data generated by the data access layer is written into these two databases. The distributed time-series database uses TimescaleDB to store equipment operating status data in chronological order. The graph database uses Neo4j to store spatial and topological relationships between devices and between devices and processes, constructing a dynamic data index across time and space to achieve state evolution, event tracing, and trend prediction.

[0050] The cognitive modeling engine, based on the dynamic state index, performs statistical feature calculations on equipment state data within a preset time window. Combined with the equipment topology relationships recorded in the graph database, it conducts correlation analysis on equipment operating status, production efficiency, and energy consumption characteristics, generating structured cognitive information reflecting the evolution of equipment states and their interrelationships. This information is used to construct a multi-dimensional cognitive model of equipment health status, production efficiency, and energy consumption patterns. The structured cognitive information includes at least equipment state assessment results, the influence relationships between related equipment, and state parameters used for collaborative decision-making, and is provided to the collaborative decision-making and control execution module as decision input.

[0051] The spatiotemporal data fusion and cognitive modeling module provides structured knowledge support for upper-level decision-making and collaborative scheduling, enabling the system to understand the complex spatiotemporal relationships in industrial scenarios and thereby achieve predictive maintenance, process optimization, and intelligent resource scheduling. Through the dynamic linkage between edge real-time data and cloud-based cognitive models, this module effectively breaks down data silos and achieves information interconnection and knowledge sharing across the entire industrial domain.

[0052] The collaborative decision-making and control execution module is responsible for realizing intelligent collaboration, task allocation, and global control among multi-level intelligent agents, and is a key link in the system's "from perception to action" realization. Deployed in the cloud or regional master control server, it is configured to integrate a large-scale language model fine-tuned with industrial scenario knowledge. Based on the industrial data context, it automatically generates control strategies, diagnostic reports, or production scheduling schemes, and dynamically allocates resources and tasks through multi-agent collaborative algorithms. It transforms high-level semantic strategies into executable control instructions that meet the real-time and security requirements of equipment, and finally sends them to the industrial control system, programmable logic controller, and robot controller to form a closed-loop control link. This module is based on a large model-driven intelligent agent decision engine, combined with reinforcement learning, adaptive control, and knowledge reasoning mechanisms, to achieve real-time analysis, optimal decision-making, and dynamic execution control of complex industrial scenarios.

[0053] In this embodiment, the collaborative decision-making and control execution module includes a large model agent center, a collaborative scheduling unit, a policy generation and optimization unit, and a control execution unit. The specific data processing flow between each unit is as follows: Figure 2 As shown, where,

[0054] The large-scale model intelligent agent center, deployed in the cloud or a regional master server, is configured to receive structured cognitive information output by the spatiotemporal data fusion and cognitive modeling module. It integrates a large-scale language model (LLM) fine-tuned with industrial scenario knowledge, possessing cross-task and cross-device semantic understanding and instruction generation capabilities. It can automatically generate control strategies, diagnostic reports, or production scheduling schemes based on industrial data context and construct standardized contextual inputs for large-scale model inference. The large model can employ Deepseek or Qwen, and uses a LoRA-based domain incremental fine-tuning method to solidify industrial semantics and control instruction templates. The large-scale model intelligent agent center is configured to sequentially execute the following processing steps:

[0055] S1: Encode the structured cognitive information into a JSON structure input containing equipment identifiers, process segment identifiers, status evaluation parameters, topological adjacency relationships, and timestamps; S2: Generate task planning prompts based on preset system prompt words and control templates, driving the large language model to output a task graph. The task graph includes at least a set of atomic tasks, task dependencies, a set of target devices, and time window constraints for each task; S3: Call a consistency checker to perform syntax and constraint checks on the task graph. The consistency checker performs field integrity checks, topological reachability checks, and security threshold checks on the task graph, generating a task graph that passes the checks as scheduling input.

[0056] The collaborative scheduling unit is configured to construct an integer linear programming (MILP) scheduling model based on a verified task graph and node load information, and generate task scheduling results with the objective function of minimizing total delay and peak load. The MILP scheduling model includes at least: task start time variables, task allocation node variables, device mutual exclusion constraints, time window constraints, and node computing capacity constraints. The output task scheduling results include at least the task-node mapping, task execution order, and planned start time.

[0057] The strategy generation and optimization unit is configured to generate a control strategy based on the task scheduling result and perform closed-loop constraint optimization. The closed-loop constraint optimization adopts model predictive control (MPC). The MPC aims to minimize the tracking error and control increment within a fixed prediction step size, and applies upper and lower bound constraints and rate of change constraints on the control quantity, and outputs a control sequence that meets the real-time constraints.

[0058] The strategy generation and optimization unit is further configured to solidify the control sequence into a control strategy representation, the control strategy representation including at least a target device identifier, a control variable identifier, a target setpoint, and an execution cycle.

[0059] The control execution unit is configured to map the control strategy representation generated by the strategy generation and optimization unit into industrial control protocol instructions and issue them for execution. The mapping includes binding control variable identifiers to device control endpoint identifiers (such as OPC UA nodes or Modbus registers) and encoding target setpoints into corresponding protocol data frames.

[0060] The control execution unit issues instructions through the encrypted communication channel established by the security and operation monitoring module, and collects execution feedback data to form an execution receipt. The execution receipt is then sent back to the spatiotemporal data fusion and cognitive modeling module to update the state index and cognitive information, forming a closed-loop control link of decision-making-execution-feedback.

[0061] The security and operation monitoring module is responsible for security assurance, anomaly detection, and operational health management during industrial collaboration, ensuring the reliability of collaboration among intelligent agents and the overall stability of the system. Deployed in a hybrid environment of cloud and edge nodes, this module is tightly coupled with the sensing and acquisition, data fusion, and collaborative decision-making modules. It implements identity authentication and access control through hardware trust roots and software security policies, establishes an end-to-end secure communication channel between the edge and cloud, monitors the performance indicators, task execution status, and network communication quality of each system module in real time, and automatically triggers policy rollback, local isolation, or alarm notification mechanisms when anomalies or faults are detected. Working in conjunction with other modules, this module enables the industrial system to operate securely, reliably, and adaptively in complex and dynamic environments, providing a solid foundation for protection and assurance of intelligent collaboration.

[0062] In this embodiment, the security and operation monitoring module further includes an identity authentication and access control subsystem, a data and communication encryption unit, an operation status monitoring engine, and an abnormal event response mechanism.

[0063] The identity authentication and access control subsystem is configured to generate a device identity key based on TPM 2.0 for each edge computing node and server node, and to issue an X.509 device certificate for the device identity key for two-way authentication when the node accesses the network. The identity authentication and access control subsystem is further configured to assign access tokens with role fields to operators, and to verify the access tokens when the request reaches the collaborative decision-making and control execution module, so as to limit the scope of permissions for control policy issuance and manual intervention operations, and ensure the legitimacy of access by intelligent agents and operators.

[0064] The data and communication encryption unit is configured to establish an encrypted communication channel based on TLS 1.3 between the edge device intelligent collaboration module, the spatiotemporal data fusion and cognitive modeling module, and the collaborative decision-making and control execution module. It also adds an HMAC-SHA256 integrity check value to the reported device status feature vector and structured cognitive information, so that the integrity check is completed at the receiving end before entering the subsequent modeling or decision-making process, thus ensuring the confidentiality and integrity of the data during transmission and storage.

[0065] The operational status monitoring engine is configured to collect operational metrics on each module side and form a unified monitoring data stream. The operational metrics include at least edge node CPU / memory utilization, task queue length, control command issuance delay, execution receipt timeout count, and network round-trip latency. The operational metrics are then reported to the monitoring service for operational status assessment.

[0066] The abnormal event response mechanism is configured to trigger a control policy rollback and generate an abnormal event record when the operation status monitoring engine detects that the delay in issuing control commands continuously exceeds a preset delay threshold or the number of execution receipt timeouts continuously exceeds a preset number threshold. The control policy rollback includes stopping the issuance of new control commands and switching to preset safety control parameters, while simultaneously sending the abnormal event record to the visualization and user interaction module for alarm display.

[0067] The visualization and user interaction module provides an operation and monitoring interface for maintenance personnel and managers, undertaking real-time display, strategy management, and human-machine interaction functions for the industrial site and intelligent agent operating status. Deployed on a cloud server or management terminal, this module maintains real-time data connectivity with edge devices, the cognitive modeling module, and the collaborative decision-making module. It is configured to visualize spatiotemporal data, equipment status, intelligent agent inference results, and global decision-making strategies in the form of charts, heatmaps, or 3D digital twins. It provides an interactive control interface to support real-time adjustments to intelligent agent behavior, scheduling strategies, or security policies by users, and ensures that user input is promptly transmitted to the collaborative decision-making and execution module and implemented in specific equipment operations through an operation command feedback channel. Through this module, users can gain comprehensive control over the system's operating status and participate in intelligent agent strategy optimization and task scheduling, achieving a seamless integration of human-machine collaboration and intelligent control, further improving the manageability, transparency, and operational efficiency of the industrial system.

[0068] In this embodiment, the visualization and user interaction module further includes a data visualization engine, an interactive control interface, a running status display unit, and an operation command feedback channel, wherein,

[0069] The data visualization engine is configured to receive structured cognitive information from the spatiotemporal data fusion and cognitive modeling module and task scheduling results from the collaborative decision-making and control execution module, and to visualize and render the device operating status, node load status and task execution status based on device identifiers, timestamps and spatial topology information. The visualization includes at least a time series curve view and a topology relationship view to show the changes in device status over time and the relationships between devices, enabling real-time dynamic refreshing of large-scale data points and supporting synchronous display on multiple terminals.

[0070] The interactive control interface is configured to provide operators with a graphical interface-based interactive entry point. The interactive entry point is used to receive manual adjustment instructions within a preset permission range. The manual adjustment instructions include at least a request to modify the control target parameters or a request to adjust the task priority, allowing users to make real-time adjustments to the agent's behavior, scheduling strategy, or security strategy.

[0071] The operation status display unit is configured to centrally display key operation indicators during system operation. The key operation indicators include at least edge computing node load, control command issuance status, execution receipt status, and abnormal event identifiers, in order to assist operators in judging the system operation status.

[0072] The operation instruction feedback channel is configured to perform format verification and permission verification on the manual adjustment instruction after receiving it, and send the verified manual adjustment instruction to the collaborative decision-making and control execution module to participate in subsequent task scheduling or control strategy generation. At the same time, the manual adjustment instruction and its effective result are recorded to form an operation log, ensuring that the control instructions input by the user can be transmitted to the collaborative decision-making and execution module in a timely manner and implemented in the equipment operation.

[0073] In summary, the system of this invention, in actual operation, forms a loosely coupled interaction mechanism based on a publish-subscribe pattern among its modules. The edge device intelligent collaboration module, acting as a data publisher, publishes processed feature data to corresponding message topics; the spatiotemporal data fusion and cognitive modeling module subscribes to these topics, performing knowledge aggregation and modeling; the collaborative decision-making and control execution module subscribes to the cognitive results, generates control decisions, and publishes instruction topics; finally, each edge node subscribes to the instruction topics and completes the execution. This architecture ensures the system has extremely high scalability. When new production units or functional modules need to be added, they only need to be connected to the message bus and follow a unified semantic specification to achieve instant linkage, without requiring large-scale code refactoring of the existing system, greatly improving the flexibility and robustness of the industrial collaborative system.

[0074] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An industrial collaborative system based on spatiotemporal intelligent agents, deployed in a multi-level industrial control environment comprising an edge layer, an intelligent agent layer, a collaborative decision-making layer, and a visualization layer, characterized in that, The system includes an edge device intelligent collaboration module, a spatiotemporal data fusion and cognitive modeling module, a collaborative decision-making and control execution module, a security and operation monitoring module, and a visualization and user interaction module. The edge device intelligent collaboration module is connected to sensors, actuators and industrial control terminals in the industrial field. It is used to collect and extract features from multi-source sensor data in real time, and perform spatiotemporal reasoning through the local intelligent agent engine to realize equipment status perception and local strategy adjustment. The spatiotemporal data fusion and cognitive modeling module maintains a bidirectional data channel with the edge device intelligent collaboration module to access multi-source heterogeneous data, construct a dynamic data index across time and space, and generate a multi-dimensional cognitive model of device health status and production efficiency by combining knowledge graphs. The collaborative decision-making and control execution module is deployed on the server and configured to integrate a large language model for industrial scenarios. Based on the multi-dimensional cognitive model, it generates control strategies and allocates production resources and tasks through a multi-agent collaborative algorithm. The strategy is then converted into execution instructions and sent to the industrial control system. The security and operation monitoring module is used to implement identity authentication and access control through hardware root of trust, establish encrypted communication channels between modules, monitor system operation indicators in real time, and trigger a security response mechanism when an anomaly is detected. The visualization and user interaction module is used to display spatiotemporal data, device status, and intelligent agent inference results in a three-dimensional digital twin format, and provides an interactive interface to support real-time intervention by users in the behavior of the intelligent agent.

2. The industrial collaborative system based on spatiotemporal intelligent agents according to claim 1, characterized in that, The edge device intelligent collaboration module includes an edge computing unit, a communication management unit, and a local intelligent agent engine; The edge computing unit is used to collect multi-source industrial data from sensors, actuators and industrial control terminals connected to it at a preset sampling period, and to perform time alignment, noise reduction and feature extraction on the multi-source industrial data to generate a device status feature vector containing timestamps and device identifiers. The local intelligent agent engine is used to embed a lightweight spatiotemporal reasoning and decision-making model, calculate the device operating status statistics within a preset time window, and compare the statistics with a preset operating threshold or historical baseline to determine whether the device operating status is abnormal or whether the computing load has changed. The communication management unit is used to send the corresponding device state feature vector and judgment result to the spatiotemporal data fusion and cognitive modeling module via industrial Ethernet or 5G network when the judgment result generated by the local intelligent agent engine meets the reporting conditions, so as to be used for cross-node state association modeling or collaborative decision processing.

3. The industrial collaborative system based on spatiotemporal intelligent agents according to claim 1, characterized in that, The spatiotemporal data fusion and cognitive modeling module includes a data access layer, a spatiotemporal data storage layer, and a cognitive modeling engine. The data access layer is used to receive device status feature vectors uploaded from multiple edge device intelligent collaboration modules, and based on device identifiers, data types and timestamp information, performs semantic alignment and format unification on data from different factories and different brands of devices to generate standardized status data. The spatiotemporal data storage layer includes a distributed time-series database and a graph database. It writes the standardized state data received in the data access layer into the distributed time-series database and the graph database, and constructs a dynamic data index across time and space to realize state evolution, event tracing and trend prediction. The cognitive modeling engine, based on the dynamic state index, performs statistical feature calculations on equipment state data within a preset time window, and combines the equipment topology relationships recorded in the graph database to perform correlation analysis on equipment operating status, production efficiency, and energy consumption characteristics, generating structured cognitive information that reflects the evolution of equipment state and their mutual influence, and providing it to the collaborative decision-making and control execution module as decision input.

4. The industrial collaborative system based on spatiotemporal intelligent agents according to claim 3, characterized in that, The structured cognitive information includes equipment status assessment results, the influence relationships between related equipment, and status parameters used for collaborative decision-making.

5. The industrial collaborative system based on spatiotemporal intelligent agents according to claim 1, characterized in that, The collaborative decision-making and control execution module includes a large model agent center, a collaborative scheduling unit, a policy generation and optimization unit, and a control execution unit. The large model intelligent agent center is configured to receive structured cognitive information output by the spatiotemporal data fusion and cognitive modeling module, and construct standardized context input for large model reasoning. Based on the industrial data context, it automatically generates control strategies, diagnostic reports or production scheduling schemes, and decomposes the global production target into a series of executable atomic tasks through the thinking chain reasoning technology. The collaborative scheduling unit is configured to construct an integer linear programming (MILP) scheduling model based on the verified task graph and node load information, and generate task scheduling results with minimizing total delay and load peak as the objective function. The strategy generation and optimization unit is configured to generate a control strategy representation and perform closed-loop constraint optimization based on the task scheduling results generated by the collaborative scheduling unit. The control execution unit is configured to map the control strategy representation generated by the strategy generation and optimization unit into industrial control protocol instructions, and issue instructions through the encrypted communication channel established by the safety and operation monitoring module, collect execution feedback data to form execution receipts, and send the execution receipts back to the spatiotemporal data fusion and cognitive modeling module to update the state index and cognitive information.

6. The industrial collaborative system based on spatiotemporal intelligent agents according to claim 1, characterized in that, The security and operation monitoring module further includes an identity authentication and access management subsystem, a data and communication encryption unit, an operation status monitoring engine, and an abnormal event response mechanism. The identity authentication and access control subsystem is configured to assign a unique digital identity to each device and operator connected to the system, and to ensure the legitimacy of access by intelligent agents and operators through hardware root of trust and software security policies. The data and communication encryption unit is configured to establish an encrypted communication channel between the edge device intelligent collaboration module, the spatiotemporal data fusion and cognitive modeling module, and the collaborative decision-making and control execution module, and to add hash verification and digital signature to the reported device status feature vector and structured cognitive information, so that the integrity verification is completed at the receiving end before entering the subsequent modeling or decision-making process. The operation status monitoring engine is configured to collect operation indicators from each module and form a unified monitoring data stream; The abnormal event response mechanism is configured to trigger a control strategy rollback and generate an abnormal event record when the operation status monitoring engine detects that the delay in issuing control commands exceeds a preset delay threshold or the number of execution receipt timeouts exceeds a preset number threshold.

7. The industrial collaborative system based on spatiotemporal intelligent agents according to claim 1, characterized in that, The visualization and user interaction module further includes a data visualization engine, an interactive control interface, a running status display unit, and an operation command feedback channel. The data visualization engine is configured to receive structured cognitive information from the spatiotemporal data fusion and cognitive modeling module and task scheduling results from the collaborative decision-making and control execution module, and to visualize and render the device operating status, node load status and task execution status based on device identifier, timestamp and spatial topology information. The interactive control interface is configured to provide operators with a graphical interface-based interactive entry point, which is used to receive manual adjustment instructions within a preset permission range. The operation status display unit is configured to centrally display key operation indicators during the system operation process; The operation instruction feedback channel is configured to perform format verification and permission verification on the manual adjustment instruction after receiving it, and send the verified manual adjustment instruction to the collaborative decision-making and control execution module to participate in subsequent task scheduling or control strategy generation. At the same time, it records the manual adjustment instruction and its effective result to form an operation log.

Images