Industrial intelligent control systems, methods, and storage media based on cloud-edge-device collaboration

Abstract

This application relates to an industrial intelligent control system, method, and storage medium based on cloud-edge-device collaboration, belonging to the field of industrial intelligent control technology. The system includes: a cloud module, which determines the production control target of an industrial unit according to scheduling instructions; and updates multiple computing power models based on multiple sets of historical data. An edge module, which performs state evaluation and prediction based on the production control target and first operating data, obtaining evaluation and prediction results. Based on the evaluation and prediction results, it determines the execution of optimized control tasks and determines the target control setpoint based on the first model. A terminal module, which generates a first control instruction based on the target control setpoint; controls multiple execution devices to execute corresponding instruction tasks according to the first control instruction, obtaining second operating data; performs operating status detection based on the second operating data, obtaining status detection results; and, based on the second model and the status detection results, obtains the target control instructions for multiple execution devices.

Description

Technical Field

[0001] This application relates to the field of industrial intelligent control technology, and in particular to industrial intelligent control systems, methods and storage media based on cloud-edge-device collaboration. Background Technology

[0002] With the expansion of industrial production scale and the improvement of automation levels, the number of devices in industrial scenarios has surged, resulting in diversified and complex demands for computing resources from control, monitoring, and data processing tasks. Traditional industrial control systems are limited by hardware, with data flowing within a local closed loop, failing to achieve integrated intelligent control across devices. Currently, existing general-purpose cloud-edge-device architectures, when applied to industrial control, cannot intelligently form a continuous learning and iterative closed-loop control system for control commands and model training. Furthermore, the resource scheduling of general-purpose cloud-edge-device architectures suffers from non-deterministic latency and kernel resource contention, failing to meet the hard real-time requirements of industrial control tasks, thus leading to low determinism and poor stability in industrial production control. Summary of the Invention

[0003] This invention provides an industrial intelligent control system, method, and storage medium based on cloud-edge-device collaboration to solve the real-time conflict and computing resource competition in task scheduling between traditional industrial control networks and general cloud-edge-device computing networks. This conflict causes delays in core control task response, leading to low determinism and poor stability in industrial production. The technical solution of this invention is as follows:

[0004] According to a first aspect of the present invention, an industrial intelligent control system based on cloud-edge-device collaboration is provided. The system is deployed in a cloud-edge-device computing power network architecture and includes: a cloud module configured to: determine the production control target of an industrial unit according to a scheduling instruction; and update multiple computing power models according to multiple sets of historical data to obtain multiple updated computing power models; the multiple updated computing power models include a first model and a second model; and an edge module configured to: perform state evaluation and prediction on the operating status of multiple execution devices according to the production control target and first operating data of multiple execution devices in the industrial unit to obtain evaluation and prediction results. Based on the evaluation and prediction results, the optimized control task is determined, and the target control setpoint is determined based on the first model. The evaluation and prediction results characterize whether the target indices corresponding to multiple execution devices in the current operating state meet the requirements of the production control target. The first operating data characterizes the operating data of multiple execution devices acquired at the current moment. The first model characterizes the correlation between the operating data, the target indices, and the target control setpoint. The terminal module is configured to: generate first control instructions for multiple execution devices based on the target control setpoint; control multiple execution devices to execute corresponding instruction tasks according to the first control instructions to obtain second operating data; perform operating state detection on the operating state of multiple execution devices based on the second operating data to obtain state detection results; obtain the target control instructions for multiple execution devices based on the first model and the state detection results; and the second model characterizes the mapping relationship between the operating state of multiple execution devices and the target control instructions.

[0005] As one implementation, the terminal module includes: a control unit and an operation monitoring unit; the control unit is configured to send real-time acquired first operation data or second operation data to the operation monitoring unit; and to issue a first control command or a target control command to multiple execution devices; the operation monitoring unit is configured to perform operation status detection on the operation status of multiple execution devices based on the first operation data or second operation data, and obtain a status detection result; the status detection result includes normal operation and abnormal operation; if the status detection result is detected as normal operation, the first control command is fine-tuned in real time based on the second model and the operation status to obtain a target control command; and the first operation data or second operation data is reported to the edge module to perform status assessment and prediction; if the status detection result is detected as abnormal operation, a safety interlock mechanism is triggered to generate an emergency control command; the emergency control command represents an instruction to immediately stop the execution of the current control task if one or more execution devices are identified or predicted to be in a dangerous operating condition.

[0006] In this implementation, by prioritizing preliminary operational status detection of the acquired operational data and monitoring the operational status of the executing equipment in real time, and determining that the operation is normal, real-time closed-loop fine-tuning is performed based on the equipment's operational status. This can instantly compensate for local high-frequency micro-disturbances not covered by the edge layer model, such as equipment temperature drift and subtle material changes. Furthermore, it can trigger emergency control commands in the event of operational anomalies, ensuring the operational safety of the industrial executing equipment. This enhances the real-time performance, determinism, and reliability of the industrial system.

[0007] As one implementation method, the edge module includes an edge computing unit; the edge computing unit includes a task execution unit and a task analysis unit; the task analysis unit is configured to: evaluate and predict the operating status of multiple execution devices based on first or second operating data to obtain a target index; obtain an evaluation prediction result based on the comparison between the target index and a preset threshold; the evaluation prediction result indicates whether the target index corresponding to the operating data obtained according to the currently executed control command meets the requirements corresponding to the production control target; if the target index is detected to be greater than or equal to the preset threshold, determine that the evaluation prediction result is that the current control command can meet the requirements corresponding to the production control target; if the target index is detected to be less than the preset threshold, determine that the evaluation prediction result is that the current control command cannot meet the requirements corresponding to the production control target; trigger the task execution unit to execute an optimized control task; the task execution unit is configured to: determine to execute the optimized control task, and based on the first model, determine the target control setpoint according to the deviation value between the target index and the preset threshold; and send the target control setpoint to the terminal module.

[0008] In this implementation, the determination of whether to trigger an optimization control task is based on the evaluation and prediction results. Only when the prediction results indicate that the current control path cannot meet the target (i.e., the target index is less than a preset threshold) is the task execution unit triggered to perform a computationally intensive optimization control task. This fundamentally reduces the probability and intensity of computing power competition, ensuring that critical control tasks can obtain the necessary computing resources when needed. Furthermore, based on the evaluation and prediction results, the system can proactively smooth the production process, avoiding large-scale fluctuations and frequent adjustments, significantly improving the stability and determinism of the production process. It upgrades the response to disturbances from passive remediation to proactive prevention.

[0009] As one implementation, the edge computing unit also includes an edge resource allocation unit. The edge resource allocation unit is configured to: when determining that the task execution unit is to execute an optimization control task, perform an edge resource pool workload assessment based on the edge resource data of the edge computing unit to obtain an edge workload assessment result; obtain an edge resource allocation result based on the edge workload assessment result; the edge workload assessment result includes idle state and high load state; the edge resource allocation result indicates whether to activate the resource preemption mode and prioritize the execution of the optimization control task; if the edge workload assessment result is detected as idle, determine that the edge resource allocation result is not to activate the resource preemption mode and directly allocate computing resources to the optimization control task; if the edge workload assessment result is detected as high load, determine that the resource preemption mode is activated, preempt the computing resources currently occupied by the task analysis unit, and prioritize the execution of the optimization control task.

[0010] In this implementation, the resource preemption mode is determined by assessing the busy / idle status of the edge modules. When the overall system load is low, all tasks are allowed to simultaneously occupy computing resources, maximizing resource utilization and improving system efficiency. When the system is busy and under high load, the resource preemption mode of the optimization control task is triggered, preempting the resources of the task analysis unit to ensure that the optimization control task always obtains the necessary computing resources and guarantees timely response.

[0011] As one implementation, the edge module also includes a computing power gateway; the computing power gateway is configured to receive first or second operating data reported by the terminal module, send the first or second operating data to the edge computing unit; and report the first operating data, the second operating data, the target control setpoint, and the edge resource data to the cloud module to update multiple computing power models; and receive the production control target and multiple updated computing power models issued by the cloud module; and distribute the multiple updated computing power models to the corresponding edge computing units and terminal modules.

[0012] In this implementation, the computing power gateway reports operational data and distributes computing power models in a targeted manner, realizing intelligent collaborative distribution of data, models, and instructions, as well as global state synchronization.

[0013] As one implementation method, the cloud module includes a global computing power scheduling unit, a global data analysis and visualization unit, and a global model training unit. The global model training unit is configured to execute model training tasks. The global computing power scheduling unit is configured to evaluate the cloud resource pool's workload based on the cloud resource data of the cloud module, and obtain the cloud module's workload evaluation result; based on the cloud module's workload evaluation result, obtain the cloud resource allocation result; the cloud resource allocation result indicates whether to activate the resource preemption mode and prioritize the execution of the model training tasks of the global model training unit; the global data analysis and visualization unit is configured to perform real-time data analysis on the first running data, the second running data, the target control setpoint, the edge resource data, and the cloud resource data, and obtain data analysis results; and visualize the data analysis results.

[0014] As one implementation method, the global model training unit is specifically configured to train multiple computing power models based on multiple sets of historical data to obtain multiple updated computing power models; within a preset time range, the multiple updated computing power models are distributed to the computing power gateway; historical data represents the data acquired within a preset time period before the current time period; multiple sets of historical data include historical operating data and historical target control settings.

[0015] In this implementation, the global model training unit continuously uses historical data from edge modules, terminal modules, and cloud modules to train the computing power models of different modules in real time, and then distributes the updated computing power models to each module to achieve intelligent updates and collaborative evolution of the system.

[0016] According to a second aspect of the present invention, an industrial intelligent control method based on cloud-edge-device collaboration is provided, applied to an industrial intelligent control system. The industrial intelligent control system includes: a cloud module, an edge module, and a terminal module. The evaluation and prediction results characterize whether the target indices corresponding to multiple execution devices in their current operating state meet the requirements of the production control target; first operating data characterizes the operating data of multiple execution devices acquired at the current moment; and a first model characterizes the mapping relationship between the operating state of the execution devices and the target control command. The method includes: determining the production control target of the industrial unit according to the scheduling command; and updating multiple computing power models according to multiple sets of historical data, and distributing the updated computing power models to the corresponding terminal module or edge module. The system comprises multiple computing power models, including a first model and a second model. Based on production control objectives and first operating data from multiple execution devices in the industrial unit, the operating status of these devices is assessed and predicted to obtain assessment and prediction results. Based on the assessment and prediction results, optimized control tasks are determined, and target control setpoints are determined based on the first model. First control instructions for the multiple execution devices are generated based on the target control setpoints. The system controls the multiple execution devices to execute corresponding instruction tasks according to the first control instructions, obtaining second operating data. Based on the second operating data, the operating status of the multiple execution devices is detected, obtaining status detection results. Based on the first model and the status detection results, target control instructions for the multiple execution devices are obtained.

[0017] As one implementation, the status detection result includes normal operation and abnormal operation; the emergency control command represents an instruction to immediately stop the execution of the current control task when the identified or predicted execution device is in a dangerous operating condition; the method also includes: if the status detection result is normal operation, based on the second model and according to the operating status, the first control command is fine-tuned in real time to obtain the target control command; and the first or second operating data is reported to the edge module to perform status assessment and prediction; if the status detection result is abnormal operation, a safety interlock mechanism is triggered to generate an emergency control command. The industrial intelligent control method also includes executing any operation of the system as described in the first aspect and any of its possible implementations.

[0018] According to a third aspect of the present invention, a computer-readable storage medium is provided, on which instructions are stored, such that when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform an industrial intelligent control method based on cloud-edge-device collaboration, as described in the second aspect and any possible implementation thereof.

[0019] According to a fourth aspect of the present invention, a computer program product is provided, the computer program product including computer instructions, which, when executed on an electronic device, cause the electronic device to perform the cloud-edge-device collaborative industrial intelligent control method of the second aspect and any possible implementation thereof.

[0020] The technical solutions provided by the embodiments of the present invention bring at least the following beneficial effects: By deploying the industrial intelligent control system in a cloud-edge-device computing network architecture, this application finely divides the computing power of various control tasks to cloud modules, edge modules, and terminal modules, realizing the separation and collaboration of determinism and intelligence in control tasks. Specifically, computationally complex, computationally demanding, but non-real-time production control target optimization tasks are processed by the cloud, and the generated production control targets are distributed to the edge modules. The edge modules then perform relatively lightweight and real-time-required state assessment and prediction, and optimization control tasks based on the production control targets, intelligently transforming the production control targets. The target control setpoint is set, and the closed-loop optimized target control setpoint is sent to the terminal module. In this way, the generation of closed-loop control no longer depends on round-trip network communication with the cloud, effectively reducing the resource competition of cloud computing power and providing deterministic low latency guarantee. The terminal module explicitly executes tasks with hard real-time requirements, collects operating data, and controls multiple execution devices to execute the generated control commands in real time, ensuring timely response to control tasks. Moreover, even if cloud services are interrupted, the edge module and the terminal module can still form a complete and working control system, ensuring that multiple execution devices continue to work, thereby ensuring the continuity and stability of the industrial production process.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0023] Figure 1 This is an exemplary embodiment illustrating an industrial intelligent control system framework based on cloud-edge-device collaboration. Figure 1 ;

[0024] Figure 2 This is a task hierarchy and scheduling priority diagram illustrated according to an exemplary embodiment;

[0025] Figure 3 This is a schematic diagram illustrating operation status detection according to an exemplary embodiment;

[0026] Figure 4This is an exemplary embodiment illustrating an industrial intelligent control system framework based on cloud-edge-device collaboration. Figure 2 ;

[0027] Figure 5 This is a flowchart illustrating the training and updating of a control closed-loop optimization model according to an exemplary embodiment;

[0028] Figure 6 This is a schematic diagram of a closed-loop optimization control process for the combustion process in a thermal power plant, according to an exemplary embodiment.

[0029] Figure 7 This is a flowchart illustrating an industrial intelligent control method based on cloud-edge-device collaboration according to an exemplary embodiment. Detailed Implementation

[0030] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0031] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0032] Before providing a detailed introduction to the cloud-edge-device collaborative industrial intelligent control system and method provided in the embodiments of this application, let's first briefly introduce the application scenarios and implementation environment involved in the embodiments of this application.

[0033] Research has revealed that with the expansion of industrial production scale and the improvement of automation levels, the number of devices in industrial settings has surged, resulting in diversified and complex demands for computing resources from control, monitoring, and data processing tasks. Traditional industrial control systems, such as distributed control systems (DCS) or programmable logic controllers (PLCs), while capable of real-time monitoring and control of equipment through localized deployment, ensuring real-time performance and reliability, are typically limited by hardware capabilities, making it difficult to support advanced artificial intelligence (AI) algorithms, big data modeling, and global optimization analysis. This leads to a situation where core control tasks excel in ensuring real-time performance but suffer from bottlenecks in integrated intelligence. To bridge this gap, cloud computing and edge computing technologies have been introduced into the industrial field to process massive amounts of data and support advanced analytics.

[0034] However, current general-purpose cloud-edge-device architectures often lack deep collaboration and integration with industrial control networks. On the one hand, they have not yet utilized the massive amounts of collected operational data for in-depth analysis, continuous learning, and model iteration, thus failing to achieve continuous adjustment and self-optimization of control commands and model parameters to form a self-evolving intelligent closed-loop system. On the other hand, these general-purpose architectures primarily focus on the balanced allocation of computing resources or maximizing overall throughput. Their general resource scheduling strategies lead to non-deterministic latency and kernel resource contention, failing to meet the stringent requirements of industrial control tasks for millisecond-level deterministic response (hard real-time) and maximum reliability. Therefore, when control tasks share edge computing resources with non-core tasks, general-purpose cloud-edge-device architectures and resource scheduling schemes can easily lead to response delays in core control tasks, affecting the determinism and stability of industrial production.

[0035] To address the aforementioned issues, this application proposes an industrial intelligent control system based on cloud-edge-device collaboration. By deploying the industrial intelligent control system within a cloud-edge-device computing network architecture, the computing power for various control tasks is finely distributed among cloud modules, edge modules, and terminal modules, achieving the separation and collaboration of determinism and intelligence in control tasks. This eliminates the dependence of closed-loop control generation on round-trip network communication with the cloud, effectively reducing resource competition for cloud computing power and providing deterministic low-latency guarantees.

[0036] For ease of understanding, the following section provides a detailed description of the cloud-edge-device collaborative industrial intelligent control system and method provided in this application, in conjunction with the accompanying drawings.

[0037] This application provides an embodiment of, as follows: Figure 1 The diagram shows the industrial intelligent control system framework based on cloud-edge-device collaboration. Figure 1 The system comprises a cloud module 1, an edge module 2, and a terminal module 3. These modules are connected via network communication. The cloud module 1 includes a global model training unit 11, a global computing power scheduling unit 12, and a global data analysis and visualization unit 13. The edge module 2 includes an edge computing unit 21 and a computing power gateway 22. The terminal module 3 includes a control unit 31 and an operation monitoring unit 32.

[0038] Cloud module 1 is configured to determine the production control objectives of industrial units based on scheduling instructions. It also updates multiple computing power models based on multiple sets of historical data, resulting in multiple updated computing power models.

[0039] Several updated computing power models include the first model and the second model.

[0040] Edge module 2 is configured to evaluate and predict the operating status of multiple execution devices based on the production control objectives and the first operating data of multiple execution devices in the industrial unit, and obtain the evaluation and prediction results; based on the evaluation and prediction results, determine the execution optimization control task, and determine the target control setpoint based on the first model.

[0041] Among them, the evaluation and prediction results characterize whether the target indices corresponding to multiple execution devices in the current operating state meet the requirements of production control objectives.

[0042] The first operational data represents the operational data of multiple execution devices acquired at the current moment. It is the operational data that is reported to the edge module 2 after real-time data preprocessing and operational status detection by the terminal module 3.

[0043] The first model represents the correlation between operating data and target indices and target control setpoints.

[0044] Terminal module 3 is configured to generate first control instructions for multiple execution devices according to the target control setpoint; control the multiple execution devices to execute corresponding instruction tasks according to the first control instructions to obtain second operating data; perform operating status detection on the multiple execution devices according to the second operating data to obtain status detection results; and obtain target control instructions for multiple execution devices based on the second model and the status detection results.

[0045] The second operational data represents the raw operational data collected in real time from multiple execution devices executed according to the first control command. Real-time operational status detection is performed in terminal module 3 to monitor the operational status of multiple execution devices in real time.

[0046] The second model represents the mapping relationship between the operating states of multiple execution devices and the target control commands.

[0047] In some implementations, this system is an industrial intelligent control system based on a cloud-edge-device computing network. To balance the real-time performance, reliability, and optimization requirements of the industrial control system's task execution, task hierarchy and scheduling priority are introduced. Based on the real-time requirements of the tasks for the industrial control system's operation, task execution levels are divided into four levels. The task types, descriptions, scheduling priorities, and mechanism diagrams corresponding to each level are shown below. Figure 2 As shown.

[0048] The task level is L1, the task type is a core control task, and the task description is real-time control of the device, requiring extremely low latency and the highest reliability. The scheduling priority and mechanism are dedicated to resources and do not perform dynamic allocation, ensuring the immediate execution of core control commands.

[0049] The task level is L2, the task type is AI-based control task, and the task description includes lightweight data preprocessing, equipment status monitoring, AI-based predictive production control model generation, and emergency control strategy generation. The scheduling priority and mechanism are dedicated to resources and do not involve dynamic allocation, ensuring normal equipment operation.

[0050] The task level is L3, the task type is control closed-loop optimization, and the task description is data preprocessing, lightweight data analysis, and AI-based system / process unit control closed-loop optimization. The scheduling priority and mechanism are absolute priority, employing a preemptive scheduling mechanism to interrupt lower-level tasks to ensure the timely execution of higher-order control strategies.

[0051] The task level is L4, the task type is data analysis and optimization, and the task description is global, non-real-time tasks such as edge data analysis, big data modeling, non-real-time optimization, historical data mining, model training, and unit / equipment / plant-wide control closed-loop optimization. The scheduling priority and mechanism is the lowest priority, utilizing massive storage and global computing resources for continuous analysis and model iteration.

[0052] The main task levels undertaken by each module are as follows.

[0053] The cloud module 1, also known as the cloud layer, is applied in cloud servers and resource scheduling platforms, providing massive storage and global computing power support. It provides computing resources for tasks by adopting resource sharing and priority resource preemption. The global model training unit 11, the global computing power scheduling unit 12, and the global data analysis and visualization unit 13 respectively carry out a variety of L4-level data analysis and optimization tasks.

[0054] Primarily responsible for global, non-real-time tasks such as edge data analysis, big data modeling, non-real-time optimization of production control targets, historical data mining, model training, and optimization of production control tasks for units / equipment / the entire plant.

[0055] Edge module 2, also known as the edge layer, provides computing resources for tasks using resource sharing and priority-based resource preemption. Edge computing unit 21 carries L3-level production control process closed-loop optimization tasks and L4-level data analysis and optimization tasks. The scheduling priority of L3-level tasks is higher than that of L4-level tasks.

[0056] It is mainly responsible for L3 level tasks such as data preprocessing, lightweight data analysis, and closed-loop optimization control of production control processes, as well as L4 level tasks such as analyzing system / process unit task response latency, failure rate, and resource utilization.

[0057] Terminal module 3, also known as the terminal layer, uses hard isolation for resource allocation. Control unit 31 carries out L1-level production control tasks, while operation monitoring unit 32 carries out L2-level control process monitoring tasks.

[0058] It is mainly responsible for real-time control tasks of industrial field equipment, L1 level production control tasks with extremely low latency requirements and the highest reliability requirements, and L2 level tasks. After receiving the secondary control command after closed-loop optimization, i.e. the target control setpoint, the industrial control system immediately converts it into hardware execution signals to drive the actuator for secondary control. It also uses its dedicated resource channels and safety interlocking mechanisms to ensure the immediate response of control commands and the highest reliability of L2 level tasks.

[0059] In this implementation, by classifying tasks, hard-isolating resources, allocating resources, and prioritizing scheduling, real-time conflicts and competition for computing resources during task scheduling are effectively avoided.

[0060] Through the separation and collaborative control of three modules, a closed-loop optimization process for production control is achieved. Terminal module 3 is responsible for real-time adjustment of control commands, edge module 2 is responsible for closed-loop optimization of system / process unit production control, and cloud module 1 is responsible for training the production control model and optimizing the production control of the unit / equipment / plant as a whole. It also outputs new production control model parameters and periodically feeds them back to and updates edge module 2 and terminal module 3, thereby driving the continuous intelligent optimization of the entire system.

[0061] like Figure 1 As shown, terminal module 3 includes control unit 31 and operation monitoring unit 32.

[0062] The control unit 31 is configured to send the first or second operating data acquired in real time to the operation monitoring unit 32; and to issue a first control command or a target control command to multiple execution devices.

[0063] The operation monitoring unit 32 is configured to perform operation status detection on multiple execution devices based on the first operation data or the second operation data, and obtain the status detection result.

[0064] The actuators include various actuators such as motors or valves, as well as various sensors such as vibration, temperature, pressure, or vision sensors.

[0065] Actuators are used to receive control commands and execute corresponding actions. Sensors are used to monitor various operating data and execution feedback data of industrial equipment in real time.

[0066] The status detection results include normal operation and abnormal operation. The target control instructions corresponding to the two status detection results are as follows.

[0067] First, if the status detection result indicates normal operation, based on the second model and the operating status, the first control command is fine-tuned in real time to obtain the target control command. Then, the first or second operating data is reported to edge module 2 to perform status assessment and prediction tasks.

[0068] Secondly, if the status detection result indicates an operational anomaly, the safety interlock mechanism is triggered, generating an emergency control command.

[0069] An emergency control command indicates that one or more actuators have been identified or predicted to be in a dangerous operating condition, and commands to immediately cease performing the current control task.

[0070] By prioritizing preliminary operational status detection of the acquired operational data and monitoring the operational status of the executing equipment in real time, and determining that operation is normal, real-time closed-loop fine-tuning is performed based on the equipment's operational status. This allows for instantaneous compensation for local high-frequency minute disturbances not covered by the edge layer model, such as equipment temperature drift and subtle material changes. Furthermore, it can trigger emergency control commands in cases of operational anomalies to ensure the operational safety of industrial executing equipment. This enhances the real-time performance, determinism, and reliability of the industrial system.

[0071] Terminal module 3 performs lightweight data preprocessing on the raw data, which involves data filtering and noise reduction followed by the extraction of amplitude, frequency, and trend features.

[0072] The collected raw state data is first processed and checked locally. After obtaining the state detection results, only lightweight data such as feature values ​​and state results are reported to edge module 2, which significantly reduces the amount of data reported. Edge module 2 transmits the aggregated evaluation results and targets to cloud module 1, which is even smaller and can effectively alleviate network congestion.

[0073] In one implementation of terminal module 3 resource occupancy, control unit 31 is deployed on intelligent dedicated controller, and resource allocation is performed using hard isolation.

[0074] Control unit 31, also known as the control area, prioritizes the allocation of high-performance physical resources to run L1 level tasks.

[0075] The operation monitoring unit 32, also known as the AI ​​usage area, allocates remaining resources and runs L2-level tasks through configured computing power units, including NPU (Neural Processing Unit) and GPU (Graphics Processing Unit).

[0076] The CPU (Central Processing Unit) core and memory are divided into a control unit 31 and a runtime monitoring unit 32. This ensures that L1-level and L2-level tasks each have their required computing resources, prohibiting cross-regional resource usage and avoiding task interference and latency that may occur in traditional shared architectures. The specific process of runtime status monitoring is as follows: Figure 3 As shown.

[0077] In one specific implementation, taking the Linux system as an example, the above-mentioned resource hard isolation scheme can be implemented in the following way.

[0078] Specifically, for the CPU, a specific core group (such as cores 0-4) can be bound to cpuset (central processing unit set). With the help of the cpuset subsystem of cgroup (control groups), the container process can be forced to be scheduled only within the limited core, thus preventing cross-core occupation.

[0079] Regarding memory, the cgroup memory subsystem intercepts excessive memory requests from containers and prohibits swap space usage, thus locking memory usage boundaries.

[0080] For GPUs, the configuration of assigning specific GPU devices from the host machine to containers and the device isolation features of containers are used to block containers from accessing GPU device files and driver commands, completely blocking GPU resource calls at the hardware mapping and command layer.

[0081] In this implementation, it is based on kernel-level control rather than application-layer restrictions, and has the characteristics of hard isolation and no escape. It can accurately limit the scope of resource usage of containers and avoid resource contention between containers. It is a lightweight and easy-to-implement resource isolation method.

[0082] like Figure 1 As shown, the edge module 2 includes an edge computing unit 21 and a computing gateway 22.

[0083] Edge computing unit 21 is configured to receive real-time environmental and equipment parameters, as well as control strategy data, from the terminal module 3. Edge computing unit 21 initially mines the value of the real-time data from the terminal module 3, performs a closed-loop optimization process, generates new control strategies, and deploys them to the terminal module 3. Most intelligent analysis and decision-making processes for system / process units (such as closed-loop optimization of combustion processes in thermal power units, closed-loop optimization of soot blowing processes, and optimal output optimization of multiple pump units) can be directly executed within edge computing unit 21, only reporting the results to cloud module 1, without requiring unified cloud-side calculation and the issuance of guidance instructions, thereby effectively improving operational efficiency.

[0084] The computing power gateway 22 is configured to receive first or second operating data reported by the terminal module 3, send the first or second operating data to the edge computing unit 21, and report the first operating data, the second operating data, the target control setting value and the edge resource data to the cloud module 1; and receive the production control target and multiple updated computing power models issued by the cloud module 1, and distribute the multiple updated computing power models to the corresponding edge computing unit 21 and terminal module 3.

[0085] like Figure 4 As shown, the edge computing unit 21 includes a task analysis unit 211, a task execution unit 212, and an edge resource allocation unit 213.

[0086] The task analysis unit 211 is configured to evaluate and predict the operating status of multiple execution devices based on the first operating data or the second operating data, and obtain a target index; and obtain an evaluation and prediction result based on the comparison between the target index and a preset threshold.

[0087] Among them, the evaluation and prediction results characterize whether the target index corresponding to the operating data obtained according to the currently executed control instructions meets the requirements of the production control objectives.

[0088] If the target index is detected to be greater than or equal to a preset threshold, it is determined that the evaluation and prediction results indicate that the current control command can meet the requirements corresponding to the production control target.

[0089] If the target index is detected to be less than the preset threshold, and the evaluation and prediction result is determined to be that the current control command cannot meet the requirements corresponding to the production control target, the task execution unit 212 is triggered to execute the optimization control task.

[0090] The task execution unit 212 is configured to determine the execution of the optimization control task and, based on the first model, determine the target control setpoint according to the deviation between the target index and the preset threshold.

[0091] Thus, by determining the evaluation and prediction results, it is decided whether to trigger an optimization control task. Only when the prediction results show that the current control path cannot meet the target, i.e., the target index is less than a preset threshold, is the task execution unit 212 triggered to execute a computationally intensive optimization control task. This fundamentally reduces the probability and intensity of computing power competition, ensuring that critical control tasks can obtain the necessary computing resources when needed. Furthermore, based on the evaluation and prediction results, the system can proactively smooth the production process, avoiding large-scale fluctuations and frequent adjustments, significantly improving the stability and determinism of the production process. It upgrades the response to disturbances from passive remediation to proactive prevention.

[0092] The edge resource allocation unit 213 is configured to, when determining that the trigger task execution unit 212 is executing the optimization control task, perform an edge resource pool busy / idle assessment based on the edge resource data of the edge computing unit 21, and obtain the edge busy / idle assessment result; and obtain the edge resource allocation result based on the edge busy / idle assessment result.

[0093] Edge workload assessment results include idle state and high load state.

[0094] The edge resource allocation result indicates whether the resource preemption mode is activated, and the optimization control task is executed first.

[0095] The edge workload assessment result was detected as idle, and the edge resource allocation result was determined to be no resource preemption mode, so the computing resources were directly allocated to the optimization control task.

[0096] If the edge workload assessment result is detected as a high-load state, the resource preemption mode is activated, preempting the computing resources currently used by the task analysis unit 211, and prioritizing the execution of optimization control tasks.

[0097] In one implementation of edge module 2 resource utilization, L4 and L3 level tasks are deployed as containers in the task analysis unit 211 and task execution unit 212 of the edge computing unit 21. Each task shares virtualized computing resources, and the edge computing unit 21 runs the edge resource allocation unit 213 to allocate virtual resources.

[0098] When resources are insufficient to support all task requirements, i.e., under high load, L3 tasks acquire resources from L4 tasks through resource preemption. Specifically, the resource preemption mode of the current computing node is measured by accurately calculating the edge busy / idleness, as shown in formulas (1) to (3) below.

[0099] (1).

[0100] (2).

[0101] (3).

[0102] in, This indicates the CPU resource utilization rate of the edge computing unit 21; This indicates the GPU resource utilization rate of the edge computing unit 21; This indicates the NPU resource utilization rate of edge computing unit 21; This indicates the memory resource utilization rate of the edge computing unit 21; This represents the maximum utilization rate of all resources, and also indicates the current resource consumption situation; This indicates the changes in resource utilization of edge computing unit 21; 'B' indicates the calculation period, which is generally greater than 100ms; 'B' indicates the busy / idleness index. This represents the threshold for resource utilization, which is adjustable, with an initial value of 80%.

[0103] when Furthermore, if the resource utilization rate continues to increase, then the host machine is considered to be under high load. At this time, the busy / idle index B is 1, and the resource preemption mode is activated when an L2 task arrives.

[0104] In this implementation, the resource preemption mode is determined by assessing the busy / idle status of edge module 2. When the overall system load is low, all tasks are allowed to simultaneously occupy computing resources, maximizing resource utilization and improving system operating efficiency. When the system is busy and under high load, the resource preemption mode of the optimization control task is triggered, preempting the resources of task analysis unit 211, ensuring that the optimization control task can always obtain the required computing resources and guaranteeing timely response.

[0105] like Figure 1 As shown, cloud module 1 includes a global model training unit 11, a global computing power scheduling unit 12, and a global data analysis and visualization unit 13.

[0106] The global model training unit 11 is configured to perform model training tasks. Based on multiple sets of historical data, it trains multiple computing power models to obtain multiple updated computing power models; within a preset time range, it distributes the multiple updated computing power models to the computing power gateway 22.

[0107] Historical data represents data acquired within a preset time period prior to the current time period.

[0108] Multiple sets of historical data also include historical operational data, historical target control settings, historical edge resource data, and historical cloud resource data.

[0109] Multiple computing power models also include edge computing power scheduling algorithms and cloud computing power scheduling algorithms.

[0110] The global model training unit 11 is configured to distribute the updated edge computing power scheduling algorithm and cloud computing power scheduling algorithm to the edge resource allocation unit 213 and the global computing power scheduling unit 12.

[0111] In this way, the global model training unit 11 continuously acquires historical operating data and historical resource data from the edge module 2, terminal module 3, and cloud module 1, trains the computing power models of different modules in real time, and distributes the updated computing power models to each module in a targeted manner, so as to realize the intelligent update and collaborative evolution of the system.

[0112] The global computing power scheduling unit 12 is configured to perform a cloud resource pool busy / idle assessment based on the cloud resource data of the cloud module 1, and obtain the cloud module busy / idle assessment result; and obtain the cloud resource allocation result based on the cloud module busy / idle assessment result.

[0113] Among them, the cloud resource allocation result indicates whether the resource preemption mode is activated, and the model training task of the global model training unit 11 is executed first. The busy / idleness of the cloud module is specifically represented by the above formulas (1) to (3).

[0114] The global data analysis and visualization unit 13 is configured to perform real-time data analysis on the first running data, the second running data, the target control setpoint, the edge resource data, and the cloud resource data to obtain data analysis results; and to visualize the data analysis results.

[0115] The global model training unit 11 is also configured to perform edge computing power scheduling algorithm optimization and cloud computing power scheduling algorithm optimization, and periodically update the edge module 2 and cloud module 1.

[0116] The flowchart for training and updating the control closed-loop optimization model is as follows: Figure 5 As shown.

[0117] S401 aggregates multi-edge data in the cloud.

[0118] S402 is a periodically retrained reinforcement learning model.

[0119] S403, Model and Parameter Evaluation.

[0120] S404 is sent to the edge module and the terminal module.

[0121] S405, Update the local scheduling engine.

[0122] S406, Model Execution and Monitoring.

[0123] S407 synchronizes terminal data and edge data to the cloud knowledge base.

[0124] Understandably, Cloud Module 1 is responsible for both the closed-loop control optimization of the unit / equipment / plant as a whole and the intelligent evolution of the end-point and edge production control models. Cloud Module 1 aggregates data from various field sites / production lines, enabling intelligent calculations within the plant to generate plant-level analysis and decisions. It performs macro-level optimization control on each production line, defining production control targets and distributing these targets to Terminal Module 3 for execution, such as the plant's electrical and thermal load distribution systems. The results are then sent back to Cloud Module 1. Furthermore, Cloud Module 1's cloud server continuously runs L4-level global analysis and optimization tasks. By aggregating historical production operation data, logs, and control execution results from all Terminal Module 3 and Edge Module 2, it trains a more accurate control optimization model. The training results are periodically transmitted back to Terminal Module 3 and Edge Module 2 via a secure communication network, for example, weekly or monthly. Upon receiving updated parameters, models, and strategies from Cloud Module 1, Terminal Module 3 and Edge Module 2 update their internal production control optimization models.

[0125] A specific implementation method, combined with a specific embodiment of closed-loop optimization of the combustion process in a thermal power plant and an integrated main and auxiliary power control scenario, is described in detail. Specifically, it includes the following three steps, and the specific closed-loop optimization control process of the thermal power plant combustion process is as follows: Figure 6 As shown.

[0126] Firstly, when the system starts up or the model is updated, cloud module 1 completes model training and deploys the model and parameters to the lower layer.

[0127] Among them, the first model is the edge layer unit-level control closed-loop optimization first model, and the second model is the end layer lightweight execution second model.

[0128] The lightweight execution model at the end layer is marked as an L2 level task, and the safety interlock parameters are sent and deployed to the terminal module 3.

[0129] The first model of the edge layer unit-level control closed-loop optimization is marked as L3 level, and the model and initial optimization parameters are sent to the edge computing unit 21 of the edge module 2.

[0130] Secondly, the system enters normalized operation, and the control process follows the sequence of cloud, edge, and endpoint, forming a multi-level closed loop. Specifically, this includes the following three steps:

[0131] First, cloud module 1 performs unit / equipment / plant-wide optimization and target setting.

[0132] Cloud module 1, based on plant-level optimized closed-loop control, executes unit / equipment / plant-wide optimization and target setting according to grid dispatch instructions, determining the optimal power generation load, heat load, auxiliary control system production indicators, and other macro-level production control targets for the current thermal power unit. These plant-level macro-level production control targets are then distributed to the corresponding edge computing unit 21 as constraints for unit-level optimization.

[0133] Secondly, edge module 2 performs closed-loop optimization and strategy generation for the system / process unit.

[0134] I. Data Acquisition and Combustion Status Monitoring. The system acquires the first operational data reported by the terminal module 3 after preprocessing in real time. This includes raw operational data collected by I / O and data processed by the terminal module, such as key operational parameters like boiler furnace temperature, furnace pressure, flue gas oxygen content, and damper opening of each burner layer.

[0135] Edge computing unit 2 preprocesses the collected data, including filtering and feature extraction, and calculates the combustion efficiency index in real time, as shown in the following formula (4).

[0136] (4).

[0137] in, Combustion efficiency index; This is the weighting coefficient for the oxygen content of flue gas; This is the weighting coefficient for flue gas temperature. Ideal oxygen content in flue gas; Real-time oxygen content in flue gas; The target combustion temperature; This represents the real-time flue gas temperature.

[0138] when Combustion efficiency index consistently below This means that a preset combustion efficiency threshold, such as 0.90, is determined to be an inefficient operating state, and an L3 level combustion process closed-loop optimization task is automatically triggered.

[0139] II. Closed-Loop Optimization Task Execution and Resource Preemption. After receiving the L3-level closed-loop optimization task, the edge resource allocation unit 213 of the edge computing unit 21 queries the current computing node resource status and the edge busy / idle status of the computing host machine.

[0140] If the edge workload assessment result is detected as idle (B=0), resources are directly allocated to run the L3 level task. If the edge workload assessment result is detected as high load (B=1), resource preemption mode is activated. The L3 level combustion process closed-loop optimization task immediately preempts the virtualized computing resources of the running L4 level task, such as historical data backup and data mining, to ensure the timely generation and execution of the optimization strategy of the L3 level combustion process closed-loop optimization task.

[0141] III. Optimization of Instruction Generation and Execution Strategy Generation. The L3-level combustion process closed-loop optimization task invokes the first edge-layer unit-level control closed-loop optimization model (e.g., a deep reinforcement learning model) within the edge computing unit 21. Based on the deviation between the current combustion efficiency index and the preset combustion efficiency threshold, it determines the optimal control strategy under the current operating conditions, i.e., the new target control setpoint, including: optimal... Concentration target value and optimal air volume distribution ratio for each floor.

[0142] IV. Result Reporting. Edge module 2 reports key data from the optimization process, such as target control settings before and after optimization, and control commands, to cloud module 1 via computing power gateway 22. Simultaneously, the optimized target control settings are sent to terminal module 3.

[0143] Finally, terminal module 3 performs fine-tuning and anomaly monitoring.

[0144] I. Data Acquisition and Preprocessing. Terminal module 3 acquires the optimized second operating data in real time, including boiler furnace temperature, furnace pressure, flue gas oxygen content, and damper opening of each burner layer, and performs data cleaning.

[0145] II. Status Detection. The operational status of the second set of preprocessed data is detected. The abnormal status monitoring model of terminal module 3 runs independently, monitoring in real time whether parameters exceed safety interlock parameters.

[0146] III. AI-based control execution and fine-tuning or emergency control strategies.

[0147] Based on the status detection results, it is determined that the execution equipment is operating normally. The terminal module 3 then uses the optimized control settings sent by the edge module 2, which are the optimal settings. Target concentration values ​​and optimal airflow allocation for each layer. The deployed end-layer lightweight execution second model transforms the target control setpoints into real-time, proactive target control commands for actuators (such as dampers and coal feeders), enabling precise fine-tuning at the millisecond or second level to ensure smooth and stable control.

[0148] Based on the status detection results, if the executing equipment is found to be operating abnormally, the safety interlock mechanism will be immediately triggered, and an emergency control command will be issued to the executing equipment, covering all control or edge commands, to ensure the safe operation of the boiler.

[0149] IV. Execution Feedback. Terminal module 3 feeds back the device's second operating data and execution feedback data after the operating status detection to edge module 2 and cloud module 1.

[0150] Third, the system periodically improves its overall performance through cloud module 1.

[0151] Cloud module 1 continuously aggregates historical data. It acquires historical operational data, control logs, and optimization results reported by terminal module 3 and edge module 2.

[0152] Cloud module 1 utilizes high-performance computing resources and runs global L4-level analysis and optimization tasks based on aggregated massive historical data to retrain and fine-tune the models used by terminal module 3 and edge module 2, thereby improving their accuracy and robustness under different coal qualities, loads and environments.

[0153] In this implementation, multi-level, bidirectional data flow and continuous intelligent evolution of the combustion process control in thermal power plants are realized, overcoming the limitations of traditional control systems and improving combustion efficiency and environmental performance.

[0154] This application provides an embodiment of, as follows: Figure 7 The present invention illustrates an industrial intelligent control method based on cloud-edge-device collaboration, which includes the following steps.

[0155] S61 determines the production control objectives of the industrial units based on scheduling instructions. It also updates multiple computing power models based on historical data, resulting in multiple updated computing power models.

[0156] Several updated computing power models include the first model and the second model.

[0157] S62, based on the production control objectives and the first operating data of multiple actuators in the industrial unit, perform state assessment and prediction on the operating status of multiple actuators to obtain assessment and prediction results; based on the assessment and prediction results, determine the execution optimization control task, and determine the target control setpoint based on the first model.

[0158] S63, based on the target control setpoint, generate first control instructions for multiple execution devices; control the multiple execution devices to execute corresponding instruction tasks according to the first control instructions, and obtain second operating data; based on the second operating data, perform operating status detection on the operating status of the multiple execution devices, and obtain status detection results; based on the second model, and based on the status detection results, obtain target control instructions for the multiple execution devices.

[0159] The status detection results include normal operation and abnormal operation.

[0160] If the status detection result indicates normal operation, the first control command is fine-tuned in real time based on the second model and the operating status to obtain the target control command; and the first or second operating data is reported to the edge module to perform status assessment and prediction.

[0161] If the status detection result indicates an operational anomaly, the safety interlock mechanism is triggered, and an emergency control command is generated.

[0162] To achieve the above functions, the cloud-edge-device collaborative industrial intelligent control device includes the corresponding hardware structure and / or software modules for performing each function. Those skilled in the art will readily recognize that, based on the algorithmic steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0163] This application also provides a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by the processor of an industrial intelligent control device or electronic device based on cloud-edge-device collaboration, the industrial intelligent control device or electronic device based on cloud-edge-device collaboration can perform the industrial intelligent control method based on cloud-edge-device collaboration as described in any of the above possible embodiments. And it can achieve the same technical effect; to avoid repetition, it will not be described again here.

[0164] This application also provides a computer program product, including a computer program or instructions, which are executed by a processor as described in any of the possible implementations above, representing a cloud-edge-device collaborative industrial intelligent control method. This method achieves the same technical effects and, to avoid repetition, will not be described further here.

[0165] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0166] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. An industrial intelligent control system based on cloud edge-end cooperation, characterized in that, The system is deployed in a cloud-edge-device computing network architecture, and the industrial intelligent control system includes: a cloud module, an edge module, and a terminal module; The system divides task execution levels into four task levels based on the real-time and reliability requirements of the task: core control task L1, AI-based control task L2, control closed-loop optimization task L3, and data analysis and optimization task L4. The cloud module is configured to: carry the data analysis and optimization task L4, determine the production control target of the industrial unit according to the scheduling instructions; and update multiple computing power models according to multiple sets of historical data to obtain multiple updated computing power models; the multiple updated computing power models include a first model and a second model; The edge module is configured to: carry the control closed-loop optimization task L3 and the data analysis and optimization task L4; the control closed-loop optimization task L3 includes an optimization control task; the scheduling priority of the control closed-loop optimization task L3 is higher than that of the data analysis and optimization task L4; based on the production control target and the first operating data of multiple execution devices in the industrial unit, the operating status of the multiple execution devices is evaluated and predicted to obtain an evaluation and prediction result; based on the evaluation and prediction result, the optimization control task is determined to be executed, and the target control setpoint is determined based on the first model; the evaluation and prediction result represents whether the target index corresponding to the multiple execution devices in the current operating state meets the requirements of the production control target; the first operating data represents the operating data of the multiple execution devices acquired at the current moment; the first model represents the correlation between the operating data and the target control setpoint; The terminal module is configured to: carry the core control task L1 and the AI-based control task L2; generate a first control instruction for the plurality of execution devices according to the target control setpoint; control the plurality of execution devices to execute corresponding instruction tasks according to the first control instruction to obtain second operating data; perform operating status detection on the operating status of the plurality of execution devices according to the second operating data to obtain a status detection result; and obtain the target control instruction for the plurality of execution devices based on the second model and the status detection result; the second model represents the mapping relationship between the operating status of the plurality of execution devices and the target control instruction. The terminal module includes a control unit and a running monitoring unit; wherein, the terminal module adopts kernel-level resource isolation technology to divide the task processing core and memory of the terminal module into the control unit and the running monitoring unit; The computing resources allocated to the control unit carrying the core control task L1 are physically isolated from the computing resources allocated to the operation monitoring unit carrying the AI-based control task L2. The control unit is configured to carry out the core control task L1, including the production control task; send the first operating data or the second operating data acquired in real time to the operation monitoring unit; and issue the first control command or the target control command to the plurality of execution devices. The operation monitoring unit is configured to carry the AI-based control task L2, which includes a control process monitoring task; to perform operation status detection on the operation status of the plurality of execution devices based on the first operation data or the second operation data, and to obtain the status detection result; the status detection result includes normal operation and abnormal operation. If the status detection result indicates normal operation, the first control command is fine-tuned in real time based on the second model and the operating status to obtain the target control command; and the first operating data or the second operating data is reported to the edge module to perform the status assessment and prediction. If the status detection result indicates an operational anomaly, a safety interlock mechanism is triggered, generating an emergency control command. The emergency control command indicates that one or more execution devices have been identified or predicted to be in a dangerous operating condition, and an instruction to immediately stop executing the current control task is issued. The edge module includes an edge computing unit; the edge computing unit includes a task execution unit, a task analysis unit, and an edge resource allocation unit. The task analysis unit is configured to: evaluate and predict the operating status of the plurality of execution devices based on the first operating data or the second operating data to obtain a target index; obtain an evaluation prediction result based on the comparison between the target index and a preset threshold; the evaluation prediction result indicates whether the target index corresponding to the operating data obtained according to the currently executed control instruction meets the requirements corresponding to the production control target; if the target index is detected to be greater than or equal to the preset threshold, determine that the evaluation prediction result indicates that the current control instruction can meet the requirements corresponding to the production control target; if the target index is detected to be less than the preset threshold, determine that the evaluation prediction result indicates that the current control instruction cannot meet the requirements corresponding to the production control target; and trigger the task execution unit to execute the optimized control task. The task execution unit is configured to determine to execute the optimization control task, obtain the target control setting value based on the first model and the deviation value between the target index and the preset threshold, and send the target control setting value to the terminal module. The edge resource allocation unit is configured to: when resource contention occurs between the control closed-loop optimization task L3 and the data analysis and optimization task L4, preempt the computing resources occupied by the data analysis and optimization task L4 and prioritize the execution of the control closed-loop optimization task L3; when it is determined that the task execution unit should be triggered to execute the optimization control task, perform an edge resource pool busy / idle assessment based on the edge resource data of the edge computing unit to obtain an edge busy / idle assessment result; and obtain an edge resource allocation result based on the edge busy / idle assessment result; the edge busy / idle assessment result includes idle state and high load state; and the edge resource allocation result indicates whether to activate the resource preemption mode and prioritize the execution of the optimization control task. If the edge workload assessment result is detected as an idle state, the edge resource allocation result is determined to not activate the resource preemption mode and directly allocate computing resources to the optimization control task; if the edge workload assessment result is detected as a high load state, the resource preemption mode is activated to preempt the computing resources currently occupied by the task analysis unit and prioritize the execution of the optimization control task.

2. The cloud edge-end collaborative based industrial intelligent control system according to claim 1, wherein, The edge module also includes a computing power gateway; The computing power gateway is configured to receive the first operating data or the second operating data reported by the terminal module, and send the first operating data or the second operating data to the edge computing unit. In addition, the first operating data, the second operating data, the target control setting value, and the edge resource data are reported to the cloud module to update the multiple computing power models; In addition, it receives the production control target and the multiple updated computing power models sent by the cloud module, and directs the multiple updated computing power models to the corresponding edge computing units and terminal modules. 3.The cloud edge-end collaboration based industrial intelligent control system according to claim 2, characterized in that, The cloud module includes a global computing power scheduling unit, a global data analysis and visualization unit, and a global model training unit; The global model training unit is configured to perform model training tasks; The global computing power scheduling unit is configured to: evaluate the busy / idle status of the cloud resource pool based on the cloud resource data of the cloud module, and obtain the busy / idle status evaluation result of the cloud module; obtain the cloud resource allocation result based on the busy / idle status evaluation result of the cloud module; the cloud resource allocation result indicates whether to activate the resource preemption mode and prioritize the execution of the model training task of the global model training unit. The global data analysis and visualization unit is configured to perform real-time data analysis on the first running data, the second running data, the target control setpoint, the edge resource data, and the cloud resource data to obtain data analysis results. The data analysis results are then visualized.

4. The industrial intelligent control system based on cloud-edge-device collaboration according to claim 3, characterized in that, The global model training unit is specifically configured to train the multiple computing power models based on the multiple sets of historical data to obtain the multiple updated computing power models; and to distribute the multiple updated computing power models to the computing power gateway within a preset time range; the historical data represents data acquired within a preset time period prior to the current time period. The multiple sets of historical data include historical operating data and historical target control settings.

5. An industrial intelligent control method based on cloud edge-end cooperation, characterized in that, The method is applied to an industrial intelligent control system, which includes a cloud module, an edge module, and a terminal module. The evaluation and prediction results characterize whether the target indices corresponding to multiple actuators in their current operating state meet the requirements of the production control objective. The first operating data characterizes the operating data of the multiple actuators acquired at the current moment. The first model characterizes the mapping relationship between the operating state of the actuators and the target control commands. The method includes: According to the scheduling instructions, the production control objectives of the industrial units are determined; and based on multiple sets of historical data, multiple computing power models are updated to obtain multiple updated computing power models; the multiple updated computing power models include a first model and a second model; Based on the production control target and the first operating data of multiple execution devices in the industrial unit, the operating status of the multiple execution devices is evaluated and predicted to obtain an evaluation and prediction result; based on the evaluation and prediction result, an optimized control task is determined, and a target control setpoint is determined based on the first model; specifically, this includes: evaluating and predicting the operating status of the multiple execution devices based on the first operating data to obtain a target index; obtaining an evaluation and prediction result based on the comparison between the target index and a preset threshold; the evaluation and prediction result indicates whether the target index corresponding to the operating data obtained according to the currently executed control command meets the requirements corresponding to the production control target; if the target index is detected to be greater than or equal to the preset threshold, it is determined that the evaluation and prediction result indicates that the current control command can meet the requirements corresponding to the production control target; if the target index is detected to be less than the preset threshold, it is determined that the evaluation and prediction result indicates that the current control command cannot meet the requirements corresponding to the production control target; triggering the task execution unit to execute the optimized control task; determining to execute the optimized control task, and obtaining the target control setpoint based on the deviation between the target index and the preset threshold, according to the first model; Specifically, when it is determined that the task execution unit will execute the optimization control task, the edge resource pool's workload is assessed based on the edge resource data of the edge computing unit to obtain an edge workload assessment result; based on the edge workload assessment result, an edge resource allocation result is obtained; the edge workload assessment result includes an idle state and a high-load state; the edge resource allocation result indicates whether to activate the resource preemption mode and prioritize the execution of the optimization control task; if the edge workload assessment result is detected as an idle state, it is determined that the edge resource allocation result is not to activate the resource preemption mode and directly allocate computing resources to the optimization control task; if the edge workload assessment result is detected as a high-load state, it is determined that the resource preemption mode is activated to preempt the computing resources currently occupied by the task analysis unit and prioritize the execution of the optimization control task; Based on the target control setting value, a first control instruction is generated for the plurality of execution devices; the plurality of execution devices are controlled to execute corresponding instruction tasks according to the first control instruction to obtain second operating data; based on the second operating data, the operating status of the plurality of execution devices is detected to obtain a status detection result; based on the second model, the target control instruction for the plurality of execution devices is obtained according to the status detection result.

6. The cloud edge-end collaboration based industrial intelligent control method according to claim 5, characterized in that, The status detection results include normal operation and abnormal operation; An emergency control command indicates that an execution device has been identified or predicted to be in a dangerous operating condition, and commands it to immediately stop executing the current control task. The method further includes: If the status detection result indicates that the operation is normal, the first control command is fine-tuned in real time based on the second model and the operating status to obtain the target control command; and the first operating data or the second operating data is reported to the edge module to perform the status assessment and prediction. If the status detection result indicates an operational abnormality, a safety interlock mechanism is triggered, generating an emergency control command. The industrial intelligent control method further includes performing any operation as performed by any system according to any one of claims 2 to 4.

7. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the cloud-edge-device collaborative industrial intelligent control method as described in any one of claims 5 and 6.

Images