Cloud-side integrated industrial agent distributed task execution method and system

Abstract

The invention discloses a cloud and edge integrated industrial agent distributed task execution method and system. The method comprises the following steps: constructing a global agent at a cloud end, and deploying a local agent at an edge device; the global agent receives the production task and decomposes the production task into sub-tasks; sub-tasks are dynamically allocated based on the real-time state and capability of each edge device; the local agent executes the distributed subtasks and keeps state synchronization with the global agent; and establishing a cloud and edge collaborative learning mechanism. The system comprises a task analysis module, a resource sensing module, a dynamic allocation module, a collaborative execution module and an incremental learning module. According to the method, the problems of insufficient collaboration of cloud computing and edge computing and rigid task allocation in a traditional industrial system are solved, and intelligent collaboration of cloud edge resources and efficient execution of tasks are realized.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and distributed computing technology, and in particular to a cloud-edge integrated method and system for distributed task execution of industrial intelligent agents. Background Technology

[0002] With the development of Industry 4.0 and intelligent manufacturing, industrial systems are exhibiting increasingly strong distributed characteristics. Traditional centralized control systems can no longer meet the real-time, flexibility, and intelligence requirements of modern manufacturing. The combination of cloud computing and edge computing provides a new technological path for industrial systems, but it still faces the following challenges in practical applications: Rigid task allocation: Most existing cloud-edge collaborative systems adopt fixed task allocation strategies, which cannot be dynamically adjusted according to device status and task characteristics, resulting in low resource utilization efficiency.

[0003] Limited intelligence: Edge devices are usually only responsible for data collection and simple processing, lacking true intelligent decision-making capabilities and struggling to cope with complex industrial scenarios.

[0004] Insufficient learning capacity: Existing cloud-edge systems lack effective collaborative learning mechanisms, making it impossible to continuously optimize system capabilities and accumulate knowledge.

[0005] Low collaboration efficiency: The collaboration mechanism between the cloud and edge devices is not perfect, resulting in response delays and coordination difficulties during task execution.

[0006] In existing technologies, although some research involves task allocation in cloud-edge collaboration, most of them focus on offloading computing tasks and lack specific optimizations for industrial scenarios, especially lacking support for deep collaboration and continuous learning mechanisms for industrial intelligent agents. Summary of the Invention

[0007] The purpose of this invention is to provide a cloud-edge integrated industrial intelligent agent distributed task execution method and system, which solves the problems of insufficient cloud-edge collaboration, rigid task allocation, and limited intelligence level in traditional industrial systems, and realizes intelligent allocation and efficient execution of industrial tasks. Technical solution

[0008] The present invention adopts the following technical solution: A cloud-edge integrated distributed task execution method for industrial intelligent agents includes the following steps: First, a layered intelligent agent architecture is constructed, deploying a global intelligent agent in the cloud and local intelligent agents on edge devices. The global intelligent agent is responsible for macro-level task planning, resource coordination, and overall optimization, possessing a global perspective and long-term decision-making capabilities. The local intelligent agents are responsible for specific task execution, real-time decision-making, and local optimization, possessing rapid response and context-aware capabilities. The two communicate efficiently and collaborate closely through standardized interfaces.

[0009] Secondly, the global agent receives production tasks and performs in-depth analysis and intelligent decomposition. The analysis process considers the task's technological flow, resource requirements, quality requirements, and time constraints, identifies parallel execution opportunities and sequential dependencies in the task, decomposes the task into appropriately granular sub-task units, and establishes data flow and control flow relationships between sub-tasks.

[0010] Then, sub-tasks are dynamically allocated based on multi-dimensional information. The system monitors the computing resources, storage resources, network status, and capability characteristics of each edge device in real time, evaluates the matching degree between device capabilities and sub-task requirements, and uses a multi-objective optimization algorithm to make task allocation decisions, supporting the dynamic migration and reallocation of tasks to adapt to environmental changes.

[0011] Next, the local agent executes the assigned sub-tasks and maintains close state synchronization with the global agent. The local agent periodically reports execution progress, resource consumption, and anomalies to the global agent. Based on the feedback, the global agent adjusts its task allocation strategy in a timely manner and establishes an instant reporting and handling mechanism for anomalies to ensure full traceability of the task execution process.

[0012] Finally, a collaborative learning mechanism between the cloud and the edge is established to achieve continuous optimization of system capabilities. Local agents perform incremental learning at the edge, accumulating local experience and contextual knowledge; global agents perform model aggregation and knowledge extraction in the cloud to form a globally optimized model; and model updates are achieved through federated learning technology while protecting data privacy, and the learning results are dynamically distributed to relevant edge devices.

[0013] Accordingly, the present invention also provides a cloud-edge integrated industrial intelligent agent distributed task execution system, comprising: The task parsing module is responsible for analyzing production tasks and breaking them down into subtasks, including a task understanding unit, a task decomposition unit, a dependency analysis unit, and a priority allocation unit.

[0014] The resource awareness module is responsible for monitoring the status and capabilities of edge devices, including a status monitoring unit, a capability assessment unit, a load prediction unit, and a health assessment unit.

[0015] The dynamic allocation module is responsible for optimizing the allocation of subtasks based on multiple objectives, including a matching degree calculation unit, an optimization decision unit, a task scheduling unit, and a dynamic adjustment unit.

[0016] The collaborative execution module is responsible for coordinating the task execution of cloud-edge intelligent agents, including a state synchronization unit, an exception handling unit, a performance monitoring unit, and a result collection unit.

[0017] The incremental learning module is responsible for enabling continuous learning through cloud-edge collaboration, and includes a local learning unit, a model aggregation unit, a knowledge distribution unit, and an effect evaluation unit. Beneficial effects

[0018] Compared with the prior art, the present invention has the following beneficial effects: Improved task execution efficiency: Through intelligent task decomposition and dynamic allocation, the use of cloud and edge resources was optimized, reducing task execution time and resource waste.

[0019] Enhanced system adaptability: The hierarchical intelligent agent architecture enables the system to simultaneously consider global optimization and local real-time decision-making, adapting to complex and ever-changing industrial environments.

[0020] The system's intelligence level has been improved: the collaborative learning mechanism enables the system to learn from experience and continuously improve, gradually increasing the efficiency and quality of task execution.

[0021] The system's reliability is guaranteed: the robust state synchronization and exception handling mechanisms ensure the reliability and traceability of task execution, reducing the risk of system failure. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the system architecture of the present invention. Figure 2 This is a flowchart illustrating the task breakdown and allocation process. Figure 3 This is a schematic diagram of the cloud-edge collaborative execution process. Figure 4 This is a schematic diagram of a collaborative learning mechanism. Detailed Implementation

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] refer to Figure 1 The architecture of this system includes five core modules: task parsing module, resource awareness module, dynamic allocation module, collaborative execution module, and incremental learning module.

[0025] The task parsing module is deployed in the cloud-based global intelligent agent and is responsible for receiving and analyzing production tasks. This module uses natural language processing and knowledge graph technology to understand task requirements, uses graph theory analysis methods to identify task structure and dependencies, and generates optimized task decomposition schemes.

[0026] The resource awareness module is distributed across the cloud and edge, collecting real-time resource status, capability characteristics, and workload of each edge device through a lightweight agent. The cloud portion is responsible for aggregating and analyzing global resource information, while the edge portion is responsible for local status monitoring and preprocessing.

[0027] The dynamic allocation module is located at the core of the global intelligent agent in the cloud, and makes task allocation decisions based on a multi-objective optimization algorithm. This module comprehensively considers task requirements, equipment capabilities, resource constraints, and optimization objectives to generate the optimal task allocation scheme and supports dynamic adjustments at runtime.

[0028] The collaborative execution module spans both the cloud and edge, establishing a collaborative working mechanism between global and local intelligent agents. This module ensures state synchronization, anomaly coordination, and result collection during task execution, maintaining system consistency and reliability.

[0029] The incremental learning module enables collaborative learning between the cloud and the edge. Local learning units at the edge are responsible for incremental learning and experience accumulation, while model aggregation units in the cloud are responsible for knowledge extraction and model optimization. Through a secure and efficient knowledge distribution mechanism, the system's capabilities are continuously improved.

[0030] refer to Figure 2 The specific process of task decomposition and allocation is as follows: When the system receives a new production task, the task parsing module first performs an in-depth analysis of the task, identifying key elements such as product specifications, process requirements, quality indicators, and delivery time. Based on the domain knowledge base and historical experience, the module decomposes the task into a series of interrelated sub-tasks.

[0031] During task decomposition, the module identifies various dependencies between subtasks, including data dependencies (the output of one task is the input of another), control dependencies (one task can only begin after another task has completed), and resource dependencies (multiple tasks require the same scarce resource). These dependencies constitute a directed acyclic graph representation of the tasks.

[0032] The resource awareness module simultaneously collects real-time status information from each edge device, including real-time metrics such as CPU utilization, memory usage, network bandwidth, and device temperature, as well as capability characteristics such as processing power, accuracy level, and dedicated functions. This information forms the system's resource status view.

[0033] The dynamic allocation module uses a multi-objective optimization algorithm based on a task graph and resource view to make task allocation decisions. Optimization objectives include minimizing task completion time, maximizing resource utilization, balancing device load, and ensuring task reliability. The algorithm outputs the optimal task allocation scheme, assigning each subtask to the most suitable edge device.

[0034] refer to Figure 3 The process of cloud-edge collaborative execution includes: After the global agent assigns subtasks to the corresponding local agents, the collaborative execution module establishes a lifecycle management mechanism for task execution. Each subtask has a clear status identifier: waiting, ready, executing, completed, failed, etc.

[0035] After a local agent begins executing a task, it periodically reports its execution status to the global agent, including progress percentage, resource consumption, intermediate results, and exceptions. The reporting frequency is dynamically adjusted based on task criticality and network conditions to balance communication overhead and state synchronization requirements.

[0036] The global agent monitors the execution status of all tasks and performs coordination and optimization from a global perspective. When it detects that a task's execution deviates from expectations, such as a delay in a task that may affect subsequent tasks, the global agent will promptly adjust resource allocation or task scheduling to ensure overall progress.

[0037] Exception handling is a crucial aspect of collaborative execution. When a local agent detects an execution exception, it adopts different handling strategies based on the exception level: minor exceptions are handled and recorded locally; moderate exceptions are handled locally while being reported to the global agent; severe exceptions are immediately reported and await global instructions. The global agent then decides whether to reassign tasks or adjust the execution strategy based on the exception situation.

[0038] After the task is completed, the collaborative execution module collects complete execution results and process data, conducts quality assessment and performance analysis, and provides data support for subsequent task optimization and learning.

[0039] refer to Figure 4 The specific implementation of the collaborative learning mechanism includes: The local agent performs incremental learning at the edge, extracting knowledge from local task execution experience. The learning content includes: equipment characteristic modeling, process parameter optimization, and fault prediction models. The learning process employs online learning algorithms to adapt to changes in equipment status and environment.

[0040] To protect data privacy and reduce communication overhead, local agents do not upload raw data. Instead, they periodically send locally learned model updates (such as gradient information and model parameter increments) to the global agent in the cloud.

[0041] The global agent receives model updates from multiple edge devices and aggregates them using a federated learning algorithm. The aggregation process considers the data quality, model performance, and business importance of each device to generate an improved global model. This global model incorporates experiential knowledge from multiple devices, resulting in better generalization ability and robustness.

[0042] After the updated global model is validated and optimized, it is dynamically pushed to relevant edge devices by the knowledge distribution unit. The distribution strategy considers the device's demand, network conditions, and the urgency of updates to ensure that the learning outcomes can be applied to production practices in a timely manner.

[0043] The learning effectiveness evaluation unit continuously monitors the effectiveness of collaborative learning. By comparing indicators such as task execution efficiency, resource utilization, and product quality before and after learning, it evaluates the value of the learning mechanism and optimizes learning strategies and parameter settings accordingly.

[0044] The above embodiments demonstrate the specific application of the present invention in industrial scenarios. Through the technical solution of the present invention, intelligent collaboration of cloud-edge resources and efficient task execution can be achieved, significantly improving the flexibility, intelligence, and reliability of industrial manufacturing systems.

Claims

1. A cloud-edge integrated industrial intelligent agent distributed task execution method and system, characterized in that, The method comprises the following steps: Building a global agent in the cloud and deploying local agents on edge devices to form a hierarchical agent architecture; The global agent receives production tasks and decomposes them into multiple interrelated subtasks; Based on the real-time state, capability characteristics and resource load of each edge device, the subtasks are dynamically allocated; The local agent executes the allocated subtasks and maintains state synchronization and result feedback with the global agent; A collaborative learning mechanism is established between the cloud and the edge to continuously optimize the capabilities of the agents.

2. The method of claim 1, wherein, The hierarchical agent architecture comprises: The global agent is responsible for task planning, resource coordination and overall optimization; The local agent is responsible for specific task execution, real-time decision-making and local optimization; The global agent and the local agent communicate and collaborate through standardized interfaces.

3. The method of claim 1, wherein, The task decomposition comprises: Analyzing the process flow and resource requirements of the production task; Identifying parallel execution opportunities and sequential dependency relationships in the task; Decomposing the task into subtask units with appropriate granularity; Establishing data flow and control flow relationships between subtasks.

4. The method of claim 1, wherein, The dynamic allocation of subtasks comprises: Real-time monitoring of the computing resources, storage resources and network status of each edge device; Evaluating the matching degree of the device's capability characteristics and the subtask requirements; Making task allocation decisions based on multi-objective optimization algorithms; Supporting dynamic migration and reallocation of tasks.

5. The method of claim 1, wherein, The state synchronization and result feedback comprises: The local agent regularly reports the execution status to the global agent; The global agent adjusts the task allocation strategy based on the feedback information; Establishing an immediate reporting and handling mechanism for abnormal situations; Achieving full traceability of the task execution process.

6. The method of claim 1, wherein, The collaborative learning mechanism comprises: The local agent performs incremental learning on the edge; The global agent performs model aggregation and knowledge refinement in the cloud; Through federated learning, the model is updated without exposing the original data; Dynamically distribute the learning results to related edge devices.

7. A cloud-edge integrated industrial intelligent agent distributed task execution system, characterized in that, Comprise: Task analysis module, for analyzing production tasks and decomposing them into subtasks; Resource perception module, for monitoring the state and capability of edge devices; Dynamic allocation module, for allocating subtasks based on multi-objective optimization; Collaborative execution module, for coordinating the task execution of cloud and edge agents; Incremental learning module, for realizing continuous learning of cloud and edge collaboration.

8. The system of claim 7, wherein, The task analysis module comprises: Task understanding unit, parsing the target and constraint conditions of the production task; Task decomposition unit, decomposing complex tasks into executable subtasks; Dependency analysis unit, identifying the dependency relationships between subtasks; Priority allocation unit, determining the execution priority of subtasks.

9. The system of claim 7, wherein, The resource perception module comprises: State monitoring unit, collecting real-time device resource state information; Capability evaluation unit, evaluating the processing capacity and characteristics of the device; Load prediction unit, predicting the future load of the device; Health assessment unit, assessing the reliability and stability of the device.

10. The system of claim 7, wherein, The dynamic allocation module comprises: Matching degree calculation unit, calculating the matching degree of device capability and task requirements; Optimization decision unit, making allocation decisions based on multi-objective optimization algorithms; Task scheduling unit, generating specific task scheduling plans; Dynamic adjustment unit, adjusting task allocation according to execution conditions.

Images