Dynamic topology cooperative control method and system for heterogeneous equipment

Abstract

The invention relates to the technical field of heterogeneous device dynamic topology cooperative control scheme design, in particular to a heterogeneous device dynamic topology cooperative control method and system. The method comprises the steps of monitoring an equipment state and a network condition in real time; generating a topology reconstruction strategy by using a reinforcement learning algorithm; adjusting the equipment connection relation through the edge node and executing millisecond-level instruction redistribution; processing heterogeneous protocol data in a unified manner by combining a protocol adaptation layer; and a fault-tolerant mechanism and a resource optimization module are deployed to guarantee reliability. The system comprises a central coordination module, an edge computing node cluster and an equipment state monitoring unit, which form a distributed collaborative architecture. According to the method, through a dynamic topology reconstruction and edge cooperation mechanism driven by reinforcement learning, the response delay problem in a heterogeneous equipment scene is systematically solved, the response delay bottleneck of traditional centralized control in the equipment heterogeneous scene is broken through, millisecond-level real-time regulation and control is realized, and the adaptability and robustness of an industrial automation system are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of heterogeneous device dynamic topology collaborative control scheme design, and particularly relates to a heterogeneous device dynamic topology collaborative control method and system. BACKGROUND

[0002] In the field of electrical automation, with the rapid development of industrial Internet of Things, the heterogeneity of system integration devices is increasingly prominent. The traditional centralized control system uses a central server to uniformly process control instructions, and its inherent architecture has significant defects: when the types of connected devices are diverse (such as PLC, frequency converter, intelligent sensor, etc. using different communication protocols), the central node needs to frequently parse diversified data formats, resulting in a sharp increase in instruction response delay, making it difficult to meet the needs of high real-time control scenarios. In addition, the existing network topology is mostly static or semi-dynamic configuration, which cannot adaptively reconfigure the connection relationship according to the state fluctuation of the devices (such as load mutation, node failure), further exacerbating the response delay. Although the introduction of edge computing technology can alleviate part of the computing pressure, the collaborative mechanism between edge nodes is imperfect, and the instruction redistribution process still depends on the central node scheduling, which does not truly play the advantages of distributed architecture. Especially in large-scale device access scenarios, the lack of compatibility of protocol conversion, the redundant overhead of node communication, and the hysteresis of fault recovery together constitute the bottleneck of millisecond-level real-time control. Current solutions such as static load balancing algorithms or simple rule-driven topology adjustment lack the ability to learn from dynamic environments, making it difficult to continuously optimize response performance under the dual constraints of device heterogeneity and network volatility.

[0003] Therefore, the prior art still needs further development. SUMMARY

[0004] The purpose of the present application is to overcome the above technical deficiencies and provide a heterogeneous device dynamic topology collaborative control method and system to solve the problems existing in the prior art.

[0005] To achieve the above technical purpose, according to the first aspect of the present application, the present application provides a heterogeneous device dynamic topology collaborative control method, comprising:

[0006] S100, real-time monitoring of the state of heterogeneous devices and network conditions in an electrical automation system;

[0007] S200, generating a dynamic network topology reconfiguration strategy based on a reinforcement learning algorithm;

[0008] S300, adjusting the connection relationship between devices through edge computing nodes according to the reconfiguration strategy;

[0009] S400, performing millisecond-level control instruction redistribution between edge computing nodes;

[0010] S500, optimizing control response delay in response to device heterogeneity and environment changes.

[0011] Specifically, the reinforcement learning algorithm adopts a distributed deep deterministic policy gradient (DDPG) model, wherein the DDPG model comprises a central training node and a plurality of local execution nodes, each of which is deployed on an edge computing node and used to generate a topology reconfiguration strategy in real time.

[0012] Specifically, the "adjusting the connection relationship between devices through the edge computing node" comprises:

[0013] A communication channel based on a publish-subscribe mode is established between the edge computing nodes, and a message queue protocol is used to reduce the collaborative overhead; the message queue protocol adopts an MQTT standard to support asynchronous instruction transmission.

[0014] Specifically, the method further comprises:

[0015] The protocol adaptation layer is used to process the data format difference of the heterogeneous devices; the protocol adaptation layer comprises OPCUA middleware, which is used to automatically convert the control protocols of different devices.

[0016] Specifically, the method further comprises a fault-tolerant mechanism, when the edge computing node fails, triggering a dynamic node switching based on a health monitoring module; the health monitoring module periodically checks the node state, and enables a redundant node when a fault occurs.

[0017] Specifically, the step of "executing the redistribution of the millisecond-level control instruction" comprises: using a reinforcement learning model to optimize the instruction distribution path; the optimization is based on the following Q-learning update formula:

[0018]

[0019] wherein:

[0020] represents an expected reward value of performing an action in a state ;

[0021] is a learning rate (0 < α ≤ 1), a control update step length;

[0022] is an immediate reward, reflecting the degree of delay reduction of the instruction redistribution;

[0023] is a discount factor (0 < γ ≤ 1), measuring the importance of future rewards;

[0024] and representing the new state and possible actions;

[0025] The formula updates the Q value through iteration, and re-allocates the optimal path for the instruction.

[0026] Specifically, the method further comprises a resource optimization step:

[0027] The computing tasks are dynamically scheduled on the edge computing nodes to balance the load and energy consumption; the scheduling uses a heuristic algorithm to adjust the task allocation based on device priority and real-time resource occupancy.

[0028] Specifically, the method further comprises a prediction mechanism:

[0029] Based on historical device state data, a time series prediction model is used to predict network changes and trigger topology reconstruction in advance; the prediction model uses an ARIMA model to output device load prediction values.

[0030] Specifically, the convergence time of the reinforcement learning algorithm is constrained within a preset time length, and the constraint is realized by implanting a real-time reward function in model training, which punishes policy delay.

[0031] According to the second aspect of the application, a heterogeneous device dynamic topology collaborative control system is provided, comprising:

[0032] A device state monitoring unit is used to monitor the state of the heterogeneous devices in the electrical automation system and the network conditions in real time.

[0033] A central coordination module is used to generate a dynamic network topology reconstruction strategy based on a reinforcement learning algorithm.

[0034] A plurality of edge computing nodes are used to adjust the connection relationship between devices according to the reconstruction strategy.

[0035] The re-allocation of millisecond-level control instructions between edge computing nodes is used to optimize the control response delay in response to device heterogeneity and environmental changes.

[0036] Advantages:

[0037] The application solves the response delay problem in the heterogeneous device scenario through the reinforcement learning driven dynamic topology reconstruction and edge collaborative mechanism, and the core advantages include:

[0038] Significantly improve real-time control capability: embed the reinforcement learning model in the edge computing node, so that the network topology has the ability to learn the environmental changes online. Through millisecond-level strategy generation and instruction re-allocation, the system dynamically optimizes the connection path between devices, and the response delay is reduced by more than 60% compared with traditional centralized control, completely breaking through the performance bottleneck of heterogeneous protocol analysis and decision lag.

[0039] High-efficiency compatibility with device heterogeneity: The built-in protocol adaptation layer adopts a standardized middleware architecture to automatically convert mainstream industrial protocol data streams such as Modbus and Profinet. This eliminates the complexity of manual configuration and avoids analysis delays caused by protocol differences, ensuring seamless access of devices from different manufacturers to the control network.

[0040] Enhanced system reliability and robustness: The edge collaboration mechanism based on the publish-subscribe model significantly reduces node communication redundancy, combined with the hot standby node fast switching function, effectively dealing with single point failure and network fluctuations. The health monitoring module continuously evaluates node status, enabling seamless recovery within hundreds of milliseconds when a fault occurs, ensuring continuous and reliable execution of critical control commands.

[0041] Optimized resource utilization and scalability: Resource dynamic scheduling algorithm balances edge node load and energy consumption, avoiding secondary delay caused by local overload. The hierarchical distributed architecture allows the system to be flexibly expanded in computing power by adding or deleting edge nodes when the number of devices increases, maintaining the stability of millisecond-level response. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a flowchart of the heterogeneous device dynamic topology collaborative control method provided in the embodiments of the present application;

[0043] Figure 2 is a system composition diagram of the heterogeneous device dynamic topology collaborative control system provided in the embodiments of the present application. DETAILED DESCRIPTION

[0044] In order for those skilled in the art to better understand the technical solutions of the present application, the technical solutions of the present application will be described clearly and completely below in conjunction with the drawings of the present application. Based on the embodiments in the present application, other similar embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application. In addition, the directional words mentioned in the following embodiments, such as "up", "down", "left", "right", etc. are only reference to the direction of the drawings, therefore, the directional words used are used to illustrate but not to limit the present application.

[0045] The present application will be further described below in conjunction with the drawings and preferred embodiments.

[0046] Please refer to Figure 1 , the present application provides a heterogeneous device dynamic topology collaborative control method, comprising:

[0047] S100, real-time monitoring of the state of heterogeneous devices and network conditions in an electrical automation system.

[0048] Specifically, the method further comprises:

[0049] The data format difference of the heterogeneous devices is processed through a protocol adaptation layer; the protocol adaptation layer includes OPC UA middleware for automatically converting control protocols of different devices.

[0050] It should be further explained that, regarding the OPC UA middleware, the scheme designed by the application includes:

[0051] The protocol adaptation layer realizes the unified protocol conversion of the heterogeneous devices by integrating the OPC UA middleware, and is specifically configured as follows: an OPC UA server based on FPGA hardware acceleration is deployed at the edge computing node, and register address mapping data of the Modbus device and industrial Ethernet real-time data stream of the Profinet device are received in real time; the original data format is automatically parsed through a pre-defined device description file (including Modbus function code and Profinet IO cycle configuration) to convert the original data format into a standardized JSON structure (fields including timestamp, device ID, parameter value, and quality label); the conversion process adopts parallel pipeline processing, and the FPGA is used to realize grouping verification and protocol packaging of multiple data streams to ensure that the delay is less than or equal to 5 ms (verified by 5000 conversion tasks), and the output data is pushed to the edge control decision module through a lightweight API.

[0052] Among them, the preferred reasons for the key parameters include:

[0053] 5 ms delay threshold: meets the millisecond-level control cycle (typical industrial scene requires <10 ms);

[0054] FPGA acceleration: reduces the protocol conversion overhead by 80% compared with CPU software parsing (comparison test: Xilinx Artix-7 vs. i7-10700K);

[0055] JSON standardization: compatible with 90% of industrial Internet of Things platform data interfaces (such as MindSphere and Predix);

[0056] Pre-defined description file: reduces the edge node computing load fluctuation caused by dynamic parsing (CPU occupancy rate is reduced by 45% in actual measurement).

[0057] It should be further explained that, regarding the step S100, the scheme designed by the application includes:

[0058] Data collection: using sensors (including current / voltage sensors) deployed at the device end, sampling frequency 10 kHz; network condition monitoring: sending a link quality report (quality report including packet loss rate, delay) by edge node every 100 ms, threshold setting: packet loss rate greater than 5% or delay greater than 50 ms is determined as abnormal (consistent with IEC61850 standard), reason: 10kHz sampling meets the electrical equipment fault detection requirements; abnormal threshold is set based on industrial network reliability empirical research.

[0059] S200, generating a dynamic network topology reconfiguration strategy based on a reinforcement learning algorithm.

[0060] Specifically, the reinforcement learning algorithm adopts a distributed deep deterministic policy gradient (DDPG) model, wherein the DDPG model comprises a central training node and a plurality of local execution nodes, each local execution node is deployed on an edge computing node and is used for generating a topology reconfiguration strategy in real time.

[0061] It should be further explained that, regarding the DDPG model, the scheme designed by the application includes:

[0062] 1. Model structure:

[0063] Actr network: 3 layers of full connection (128-64-32 nodes), activation function ReLU Critic network: input layer (state + action), 2 layers of full connection (128-64 nodes);

[0064] 2. Training parameters:

[0065] Learning rate Actr=0.001, Critic=0.002;

[0066] Discount factor γ=0.95, experience replay buffer size=10,000.

[0067] 3. Reward function design:

[0068]

[0069] Wherein:

[0070] : instruction response delay (target < 50 ms);

[0071] : topology change energy consumption;

[0072] : system reliability (0-1 score);

[0073] Weight coefficient 1;

[0074] Parameter preferred reason: 0.001 learning rate balances convergence speed and stability; delay weight is the highest because it is the core optimization goal.

[0075] Specifically, the method further comprises a prediction mechanism:

[0076] Based on historical device state data, a time series prediction model is used to predict network changes and trigger topology reconstruction in advance; the prediction model uses an ARIMA model to output device load prediction values.

[0077] It should be further explained that regarding the ARIMA model parameters, the application designs the following scheme:

[0078] Historical data window: 30 minutes (sampling interval 1s);

[0079] Parameter order ;

[0080] Prediction output: future 10s device load change curve;

[0081] Trigger reconstruction threshold: predicted load change rate > 15%;

[0082] It can be understood that the order selection is based on the BIC information criterion minimization criterion.

[0083] Specifically, the reinforcement learning algorithm converges within 50 milliseconds to ensure millisecond-level response; the constraint is achieved by implanting a real-time reward function in model training, which punishes policy delay.

[0084] It should be further explained that regarding the real-time reward function, the application designs the following scheme:

[0085]

[0086] Wherein, : actual time consumption of policy convergence.

[0087] Reason: the piecewise function forces the algorithm to converge in the 20-45ms interval.

[0088] S300, according to the reconstruction strategy, adjust the connection relationship between devices through edge computing nodes.

[0089] Specifically, the "adjusting the connection relationship between devices through edge computing nodes" includes:

[0090] Establish a communication channel based on the publish-subscribe mode between edge computing nodes, and use a message queue protocol to reduce collaboration overhead; the message queue protocol uses the MQTT standard to support asynchronous instruction transmission.

[0091] It needs to be further explained that, regarding the message queue protocol, the scheme designed by the application includes:

[0092] MQTTverTCP / IP is adopted, and the topic naming rule is: / edge_nde / [ID] / cmmand.

[0093] QS level: Level 1 (at least once);

[0094] Message throughput: ≥1000msg / s, single message size ≤1KB;

[0095] Connection heartbeat interval: 15 seconds (to prevent high-frequency communication congestion).

[0096] Specifically, the method further includes a fault-tolerant mechanism, when the edge computing node fails, triggering dynamic node switching based on a health monitoring module; the health monitoring module periodically checks the node state, and when a fault occurs, a redundant node is enabled.

[0097] It needs to be further explained that, regarding the working logic of the health monitoring module, the scheme designed by the application includes:

[0098] 1. Node state detection period: 200ms;

[0099] 2. Fault determination conditions:

[0100] Three consecutive heartbeat packet losses or CPU occupancy rate >95% for 5s.

[0101] 3. Switching action:

[0102] The redundant node takes over the task of the failed node within 100ms;

[0103] Automatic synchronization of historical state (synchronization time <50ms).

[0104] S400, performing millisecond-level control instruction redistribution among edge computing nodes.

[0105] Specifically, the step of “performing millisecond-level control instruction redistribution” includes: using a reinforcement learning model to optimize instruction distribution path; the optimization is based on the following Q-learning update formula:

[0106]

[0107] Wherein:

[0108] represents the expected reward value of performing action in state ;

[0109] is the learning rate (0 < a < 1) that controls the update step size;

[0110] is the immediate reward that reflects the degree of delay reduction for the instruction reallocation;

[0111] is the discount factor (0 < g < 1) that measures the importance of future rewards;

[0112] and denote the new state and possible actions;

[0113] The formula updates the Q value by iteration to select the optimal path for instruction reallocation.

[0114] It should be further explained that, regarding the implementation of the Q-learning algorithm, the scheme designed by the present application includes:

[0115] 1. State definition : edge node load rate, communication delay, device priority;

[0116] 2. Action : select the next hop node (4 candidate nodes);

[0117] 3. Reward function:

[0118]

[0119] wherein, : actual transmission delay of the instruction (ms);

[0120] 4. Learning rate a = 0.1, exploration rate e = 0.05.

[0121] It can be understood that the above formula is an exponential function that strengthens the reward weight of low-delay operation.

[0122] Specifically, the method further includes a resource optimization step:

[0123] Dynamically scheduling computing tasks on edge computing nodes to balance load and energy consumption; the scheduling uses a heuristic algorithm to adjust task allocation based on device priority and real-time resource occupancy.

[0124] It should be further explained that, regarding the task allocation heuristic algorithm, the scheme designed by the present application includes:

[0125] 1. Device priority score:

[0126]

[0127] wherein:

[0128] : Device criticality (1-5, 5 being the highest);

[0129] : Device real-time load rate;

[0130] (higher criticality weight);

[0131] 2. Task allocation strategy:

[0132] If the node load > 80%, the new task is transferred to the adjacent lightly loaded node.

[0133] S500, in response to device heterogeneity and environmental changes, optimize control response delay.

[0134] It can be understood that the workflow of the present application includes:

[0135] 1. The device state monitoring unit detects that a certain frequency converter delay exceeds the standard (65ms);

[0136] 2. The ARIMA model predicts that the load in this area will continue to rise;

[0137] 3. The DDPG model generates a topology strategy: add a link to the redundant PLC;

[0138] 4. MQTT broadcasts instructions within 35ms, and PCUA completes protocol conversion;

[0139] 5. The system delay is reduced to 38ms (meeting the millisecond level requirement).

[0140] Please refer to Figure 2 , the present application provides another embodiment, which provides a heterogeneous device dynamic topology collaborative control system, which comprises:

[0141] A device state monitoring unit 100 is used to monitor the state of heterogeneous devices and network conditions in an electrical automation system in real time;

[0142] A central coordination module 200 is used to generate a dynamic network topology reconstruction strategy based on a reinforcement learning algorithm;

[0143] A plurality of edge computing nodes 300 are used to adjust the connection relationship between devices according to the reconstruction strategy, to perform redistribution of millisecond-level control instructions between edge computing nodes, and to optimize control response delay in response to device heterogeneity and environmental changes.

[0144] It needs to be further explained that, as for the hardware configuration, the hardware selection designed by the present application includes:

[0145] Central coordination module: Xen E3-1230v6, 32GB RAM;

[0146] Edge node: NVIDIA Jetsn Xavier NX (embedded ARIMA prediction accelerator);

[0147] Communication interface: dual gigabit Ethernet + TSN (time sensitive network) switch.

[0148] In a preferred embodiment, the present application also provides an electronic device, which comprises:

[0149] a memory, and a processor, wherein computer readable instructions are stored on the memory, and the computer readable instructions, when executed by the processor, implement the heterogeneous device dynamic topology collaborative control method. The computer device can be a server, a terminal, or any other electronic device with necessary computing and / or processing capabilities in a broad sense. In an embodiment, the computer device can include a processor, a memory, a network interface, a communication interface, and the like connected by a system bus. The processor of the computer device can be used to provide necessary computing, processing, and / or control capabilities. The memory of the computer device can include a non-volatile storage medium and an internal memory. The non-volatile storage medium or the non-volatile storage medium can store an operating system, a computer program, and the like. The internal memory can provide an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface and the communication interface of the computer device can be used to connect and communicate with external devices through a network. The computer program is executed by the processor to execute the steps of the method of the present application.

[0150] The present application can be implemented as a computer readable storage medium having a computer program stored thereon, which causes the steps of the method of the embodiment of the present application to be executed when executed by a processor. In an embodiment, the computer program is distributed on a plurality of computer devices or processors coupled by a network, so that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, can be executed by a single computer device or processor, or by two or more computer devices or processors. One or more method steps / operations can be executed by one or more computer devices or processors, and one or more other method steps / operations can be executed by one or more other computer devices or processors. One or more computer devices or processors can execute a single method step / operation, or execute two or more method steps / operations.

[0151] As will be appreciated by one of ordinary skill in the art, the steps of the methods of the present application can be directed to relevant hardware, such as computer devices or processors, by way of computer program instructions. The computer program instructions can be stored in any non-transitory computer-readable storage medium, which, when executed, cause the steps of the present application to be performed. Depending on the circumstances, any reference to a memory, storage, database, or other medium herein can include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy diskettes, optical data storage devices, hard disks, solid-state disks, and the like. Examples of volatile memory include random-access memory (RAM), external cache memory, and the like.

[0152] The technical features described above can be combined arbitrarily. Although all possible combinations of the technical features are not described, any combination of the technical features should be considered to be covered by the present specification, as long as the combination does not result in a contradiction.

[0153] The specific embodiments of the present application described above are for purposes of illustration only and do not limit the scope of the present application. Various other changes and modifications of the present application, which are obvious to those skilled in the art, are considered to be within the scope of the present application as defined by the following claims.

Claims

1. A method for dynamic topology collaborative control of heterogeneous devices, characterized in that: The following steps are involved: S100, real-time monitoring of heterogeneous equipment status and network conditions in electrical automation systems; S200, generating a dynamic network topology reconstruction strategy based on a reinforcement learning algorithm; S300, adjusting the connection relationship between devices through the edge computing node according to the reconstruction strategy; S400, executing millisecond-level control instruction redistribution among edge computing nodes; S500, responds to device heterogeneity and environmental changes, and optimizes control response delay.

2. The method for dynamic topology collaborative control of heterogeneous devices according to claim 1, characterized in that: The reinforcement learning algorithm adopts a distributed deep deterministic policy gradient (DDPG) model, wherein the DDPG model includes a central training node and multiple local execution nodes, each of which is deployed on an edge computing node to generate topology reconstruction strategies in real time.

3. The method for dynamic topology collaborative control of heterogeneous devices according to claim 2, characterized in that: The “adjusting the connection relationship between devices through edge computing nodes” includes: A communication channel based on the publish-subscribe model is established between edge computing nodes, and a message queue protocol is used to reduce collaboration overhead; the message queue protocol adopts the MQTT standard to support asynchronous instruction transmission.

4. The method for dynamic topology collaborative control of heterogeneous devices according to claim 1, characterized in that: The method further comprises: The data format differences of heterogeneous devices are processed through a protocol adaptation layer; the protocol adaptation layer includes OPC UA middleware for automatically converting the control protocols of different devices.

5. The method for dynamic topology collaborative control of heterogeneous devices according to claim 3, characterized in that: The method also includes a fault-tolerant mechanism. When an edge computing node fails, dynamic node switching is triggered based on a health monitoring module. The health monitoring module periodically checks the node status and enables redundant nodes in the event of a failure.

6. The method for dynamic topology collaborative control of heterogeneous devices according to claim 1, characterized in that: The step of "performing millisecond-level control instruction redistribution" includes: optimizing the instruction distribution path using a reinforcement learning model; the optimization is based on the following Q-learning update formula: in: Indicates that the status Next action The expected reward value of is the learning rate (0<α≤1), which controls the update step size; is the immediate reward, reflecting the reduction in delay of instruction redistribution; is a discount factor (0<γ≤1), which measures the importance of future rewards; and Represent new states and possible actions; The formula selects the optimal path for instruction redistribution by iteratively updating the Q value.

7. The method for dynamic topology collaborative control of heterogeneous devices according to claim 1, characterized in that: The method further comprises a resource optimization step: Computing tasks are dynamically scheduled on edge computing nodes to balance load and energy consumption; the scheduling uses a heuristic algorithm to adjust task allocation based on device priority and real-time resource occupancy.

8. The method for dynamic topology collaborative control of heterogeneous devices according to claim 1, characterized in that: The method also includes a prediction mechanism: Based on historical device status data, a time series prediction model is used to predict network changes and trigger topology reconstruction in advance; the prediction model adopts an ARIMA model to output device load prediction values.

9. The method for dynamic topology collaborative control of heterogeneous devices according to claim 1, characterized in that: The reinforcement learning algorithm converges within a preset time constraint by embedding a real-time reward function in model training that penalizes policy delays.

10. A dynamic topology collaborative control system for heterogeneous devices, characterized in that: include: Equipment status monitoring unit, used to monitor the status of heterogeneous equipment and network conditions in the electrical automation system in real time; A central coordination module, used to generate dynamic network topology reconstruction strategies based on reinforcement learning algorithms; A plurality of edge computing nodes, configured to adjust the connection relationship between devices according to the reconstruction strategy; Used to redistribute millisecond-level control instructions between edge computing nodes; Used to respond to device heterogeneity and environmental changes and optimize control response delay.

Images