Distributed control platform and method for industrial internet

Abstract

The invention discloses a distributed control platform and method for an industrial internet, and relates to the technical field of distributed control, and the method comprises the steps: obtaining real-time state parameters of a network path, and predicting a future probabilistic network state through a prediction model; decomposing a control task; based on the predicted network state, the control task demand spectrum, node resources and algorithm library information, a collaborative optimization decision engine is used for executing predictive multi-objective optimization calculation, and the optimal deployment node, the optimal adaptation control algorithm / strategy and the refined operation parameter set of each module are jointly decided; and deploying an operation module according to a decision result, and executing low-disturbance state consistency migration as required. The method further comprises a step of carrying out digital twinborn pre-verification on the joint decision result. According to the method, control challenges caused by network uncertainty and dynamic working conditions in an industrial internet environment can be solved, and the performance certainty, the resource efficiency, the overall toughness and the decision security of a distributed control system can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of distributed control, and in particular to a distributed control platform and method for the Industrial Internet. Background Art

[0002] The development of the Industrial Internet has driven the evolution of control systems toward distributed and intelligent systems. Traditional distributed control systems often operate on deterministic or semi-deterministic networks, with relatively fixed control logic deployment. However, in the Industrial Internet environment, network topologies are complex and diverse, integrating wired, wireless, and 5G networks. Network conditions such as latency, jitter, bandwidth, and packet loss rate exhibit significant time-varying and uncertainty. Furthermore, modern industrial production increasingly demands flexibility and reconfigurability, and control tasks may also change dynamically.

[0003] Existing technologies often face the following challenges when addressing these challenges: Most solutions assume stable network conditions or adopt worst-case design scenarios, resulting in wasted resources. Performance can degrade dramatically or even fail when the network deteriorates, making it difficult to guarantee deterministic control performance in complex, heterogeneous networks. Furthermore, control functions are often deployed in static configurations or based on coarse-grained policies at the edge, fog, and cloud. These configurations fail to implement refined, dynamic, and collaborative optimization based on real-time, even predictive, network conditions and the performance requirements of specific control tasks, such as real-time performance and computational complexity. This can easily lead to resource and task mismatches, impacting control effectiveness and system resilience. Summary of the Invention

[0004] The present invention provides a distributed control platform and method for the Industrial Internet to solve the key technical problems in the existing technology that, in the Industrial Internet environment, the distributed control performance is difficult to ensure certainty due to the time-varying and uncertain nature of the network state, and the static deployment of control functions is difficult to adapt to dynamic working conditions and network changes, resulting in resource mismatch.

[0005] To achieve the above-mentioned objectives, the first aspect of the present invention provides a distributed control method for the industrial Internet. The core of the method is: by acquiring and predicting probabilistic network states, combining with the quantified control task demand spectrum, and using a collaborative optimization decision engine to perform predictive multi-objective optimization calculations, the optimal deployment location, adaptation algorithm and operating parameters of the atomic control function module are determined in an integrated manner, and a low-disturbance migration mechanism and a digital twin pre-verification step are included, thereby realizing adaptive adjustment of the control process.

[0006] Specifically, the methods include: Obtaining real-time status parameters of the distributed control system network path, and using a prediction model to infer the probabilistic expected network state within a future preset time window based on the parameters; Decompose the overall control task into multiple atomic control function modules based on functional cohesion and interaction relationships, and establish a quantitative control task requirement spectrum for each atomic control function module; Based on the expected network state, the control task demand spectrum of the atomic control function module, the real-time available computing and network resource status of each potential deployment node, and the control algorithm strategy library information containing quantifiable network adaptability characteristics, a collaborative optimization decision engine is used to perform predictive multi-objective collaborative optimization calculations. The calculations are designed to minimize the combined cost of performance deviation and resource consumption, thereby jointly deciding the following for each atomic control function module in the next time period: (i) Optimal deployment nodes; (ii) selection of the optimal adaptive control algorithm or strategy; (iii) a refined set of operating parameters for the selected algorithm that matches the expected network and task states; According to the joint decision result, the corresponding atomic control function module is deployed at the optimal deployment node, and the selected optimal adaptive control algorithm and the refined operating parameters are used to perform control operations; when the joint decision indicates that the atomic control function module needs to change the deployment node, the low-disturbance migration protocol that maintains state consistency is activated and executed, so as to ensure the continuity of the key states of the control loop and the smooth transition of the overall system performance during the migration process.

[0007] Preferably, a strategy pre-verification step based on runtime digital twin is added before executing the joint decision result.

[0008] In a second aspect, the present invention further provides a distributed control platform for the Industrial Internet, comprising: A network state prediction unit configured to calculate a probabilistic expected network state including uncertainty statistics based on real-time network parameters using a prediction model; a control task decomposition and requirement management unit configured to decompose the control task into atomic control function modules and manage a quantized control task requirement spectrum including end-to-end timing dependency constraints; a network adaptive control algorithm library storing a plurality of control algorithm strategies with quantifiable network adaptability characteristics, and including at least one control strategy using a state observer for compensation based on the expected network state; a collaborative optimization decision engine configured to utilize a model-based predictive control framework to perform integrated, predictive, multi-objective optimization calculations for joint decision-making based on the expected network state, the control task demand spectrum, the node resource state, and the algorithm library information, the calculations taking into account migration costs and system-level knock-on effects; a low-disturbance state migration controller configured to execute a low-disturbance migration protocol including incremental state synchronization and state consistency preservation of a transition strategy; A runtime digital twin simulation and decision verification unit is configured to perform strategy pre-verification before executing the joint decision result, wherein the pre-verification is based on rapid virtual execution and quantitative risk and benefit assessment in a high-fidelity digital twin environment.

[0009] It can be seen from the above technical solutions that this specification provides at least the following technical effects or advantages: Improving the performance determinism and stability of control systems in dynamic network environments: By introducing a probabilistic network state prediction mechanism and combining it with the selection of network adaptive control algorithms and refined online parameter self-tuning, this solution can proactively and proactively compensate for the effects of network delay, jitter, and packet loss, effectively maintaining the performance indicators of the control loop under complex network conditions.

[0010] Achieve dynamic optimization of control function deployment and efficient resource utilization: Based on the precise definition of the quantitative requirements of atomic control function modules and the real-time and predictive control of network and node resources, the collaborative optimization decision engine can intelligently and dynamically determine the optimal deployment location of each function module between the edge, fog, and cloud, and efficiently execute it through low-disturbance migration protocols to ensure the best match between resources and task requirements and improve resource utilization.

[0011] The digital twin pre-verification step provides prior risk assessment and effect verification for complex decisions generated by the collaborative optimization decision-making engine, such as algorithm switching, parameter adjustment, and function migration, effectively avoiding potential risks and ensuring the safe and reliable application of advanced adaptive control strategies in actual industrial environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is an architecture diagram of the distributed control platform for the Industrial Internet of the present invention; Figure 2 This is a flow chart of the distributed control method for the Industrial Internet. DETAILED DESCRIPTION

[0013] The present invention proposes a distributed control platform and method for the Industrial Internet, aiming to solve the impact of network uncertainty on control performance in the Industrial Internet environment and the problem of control function deployment under dynamic working conditions.

[0014] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention, rather than all, are shown in the drawings. Example 1

[0015] like Figure 1-Figure 2 As shown in Figure 1, the typical architecture of the distributed control platform and method for the Industrial Internet can be deployed on the edge layer, fog layer, and cloud layer infrastructure of the Industrial Internet. The platform can logically or physically integrate the following core units: The network state prediction unit collects real-time network parameters, such as end-to-end or point-to-point latency, jitter (latency variation), available bandwidth, and packet loss rate, through lightweight probes deployed on key network nodes (such as edge gateways, switches, and target compute nodes) or through network management protocols (such as SNMP and NetFlow / sFlow). The collected raw data forms time series. This unit uses machine learning models, preferably long short-term memory (LSTM) networks, gated recurrent units (GRUs), or their combined variants, to train and perform online predictions on these time series data, inferring probabilistic expected network states within one or more predefined time windows (e.g., 100 milliseconds to 5 seconds in the future). Crucially, the prediction model outputs not only the expected values of the network state parameters but, more importantly, their probability distributions, such as the variance, standard deviation, or specific quantiles (e.g., 95% confidence bounds) of the predicted values, to quantify the uncertainty of the predictions. These probabilistic predictions, including uncertainty measures, serve as inputs for subsequent optimization decisions and are used by the collaborative optimization decision engine for risk assessment and robust decision-making.

[0016] Control Task Decomposition and Requirements Management Unit: First, complex centralized or distributed control tasks are decomposed into a series of clearly defined atomic control function modules (ACFMs) with standardized interfaces, based on the functional cohesion of the control logic, data flow relationships, and real-time requirements. For example, a closed-loop control task can be decomposed into: a sensor data acquisition and filtering ACFM, a state estimation ACFM, a control law calculation ACFM (such as PID or MPC), and an actuator instruction generation ACFM. Subsequently, a quantitative control task requirement spectrum (CTRS) is established for each ACFM. This process involves an in-depth analysis of the ACFM's algorithmic complexity, data processing volume, control cycle, and sensitivity to network quality. The CTRS may include: maximum allowable computational latency (ms), CPU core count requirement, memory requirement (MB), maximum tolerable network transmission latency (ms), network delay jitter sensitivity (e.g., performance degradation percentage / ms jitter), minimum required bandwidth (Mbps), tolerable packet loss rate (%), and reliability level (such as IEC61508 SIL level). In particular, this unit is responsible for defining and managing timing dependency constraints across ACFMs. For example, for a control loop requiring fast response, it defines an upper bound on the total end-to-end delay from data acquisition ACFM input to actuator command generation ACFM output. These quantified CTRS and end-to-end constraints are the foundation for the collaborative optimization decision engine to perform precise matching and optimization, and are treated as high-priority objectives or hard constraints in the optimization calculations.

[0017] Network Adaptive Control Algorithm Library: This is a storage and management center that contains implementation code or configuration templates for various control algorithms or strategies. These algorithms have different network robustness or adaptability characteristics. For example, the library may include: Standard PID controller; PID variants with a Smith Predictor or similar mechanism to compensate for known network delays; Control algorithms based on packet loss compensation mechanisms (such as data interpolation, holdover, or model-based prediction); Adaptive control algorithm that can adjust sampling period or control gain online to adapt to changes; At least one advanced control strategy employing a state observer (e.g., a Kalman filter or Luenberger observer) must explicitly account for network transmission delay (using the expected delay d_predicted from the network state prediction unit) and dynamically adjust its internal parameters (e.g., process noise covariance Q, measurement noise covariance R, or delay compensation terms) based on network state (e.g., jitter variance). Each algorithm or strategy is associated with metadata describing its computational complexity, memory requirements, and the typical performance or trends of key performance indicators (e.g., steady-state error, overshoot, and settling time) under different network conditions (e.g., delay, jitter, and packet loss rate ranges). This metadata is used by the collaborative optimization decision engine for evaluation and selection.

[0018] Collaborative optimization decision engine: Preferably, a model predictive control (MPC)-based framework is used to perform integrated, predictive multi-objective optimization calculations for joint decision-making. Its workflow typically includes: Input: Receives the probabilistic expected network state (including uncertainty measurement) from the network state prediction unit; receives the ACFM list and its CTRS (including end-to-end constraints) from the task management unit; receives the real-time available computing resources (such as CPU load, memory margin) and network interface status from each potential deployment node; and accesses the algorithm library to obtain algorithm metadata.

[0019] System Modeling: A simplified system dynamic model is built internally. This model predicts the expected computational latency, network transmission latency (based on predictions), resource consumption, and possible control performance indicators (based on algorithm metadata and CTRS) of each ACFM within a short time horizon (prediction horizon) in the future, given the deployment scheme (which ACFM is on which node), algorithm selection, and parameter configuration.

[0020] Optimization Problem Construction: A multi-objective optimization problem is constructed. The objective function aims to minimize a weighted sum that combines, for example: (a) the deviation (degree of non-satisfaction) between the expected control performance of all ACFMs and their CTRS requirements; (b) the expected computational and network resource costs; and (c) if the decision involves migration, the quantified migration cost (considering migration time, resource overhead, potential control perturbation risk, and even assessing potential knock-on effects on associated ACFMs). Constraints include meeting all hard CTRS requirements (such as maximum latency), end-to-end timing constraints, and node resource capacity limits. When dealing with forecast uncertainty, the engine can employ risk-averse strategies, such as incorporating the worst-case performance metric within the forecast confidence interval into the optimization objective or directly penalizing the deployment of sensitive ACFMs on paths or nodes with high forecast uncertainty.

[0021] Solving and Decision-Making: Use appropriate optimization algorithms (such as sequential quadratic programming (SQP), genetic algorithms (GA), and reinforcement learning (RL) policy networks) to solve the MPC problem and obtain the optimal joint decision sequence for a series of future time steps (including optimal deployment locations, optimal algorithm selection, and refined parameter sets).

[0022] Rolling execution: Only the joint decision result of the first time step in the sequence is executed, and then this process is repeated in the next decision cycle (for example, at a fixed time interval such as 1 second, or based on event triggering), forming a rolling optimization.

[0023] Refined Operational Parameter Set and Execution: The "refined operational parameter set" output by the collaborative optimization decision engine is more than just a static value. For control algorithms that support online adjustment (such as the state observer or other adaptive algorithms mentioned above), this parameter set can include dynamic adjustment rules or functions. For example, the parameter set might specify "PID gain" or "Kalman filter measurement noise covariance." The platform's runtime environment on the deployment node is responsible for parsing these rules or functions and updating the control algorithm's internal parameters in real time based on the latest network state predictions, enabling proactive adaptation.

[0024] Low-disturbance state migration controller: When the decision engine decides to migrate an ACFM, this controller is responsible for executing the low-disturbance migration protocol that maintains state consistency. The specific steps are generally as follows: (a) Preparation phase: The migration controller coordinates the source and target nodes. The target node prepares the runtime environment (e.g., container, virtual machine) and receives the static configuration information from ACFM.

[0025] (b) State capture: The source node captures a complete execution state snapshot of the ACFM at an appropriate time (e.g., at the end of a control cycle), which may include all key runtime data such as internal variables, accumulators, filter states, and historical data buffers.

[0026] (c) State Transfer and Recovery: The state snapshot is transferred to the target node via a secure and reliable network connection. The target node loads the snapshot and restores the internal state of the ACFM.

[0027] (d) Incremental synchronization and transition: While the target node is recovering its state, the source node may still be running and generating new state changes. At this point, an incremental state synchronization mechanism is initiated. For example, the source node sends state changes (deltas) or key event information to the target node, and the target node continuously applies these updates. To ensure a smooth transition of the control output, a transition strategy is adopted. For example, within a short switching window, control instructions based on short-term predictions can be interpolated (mixing old source node instructions with new target node instructions), or the target node can temporarily maintain the last valid control instruction issued by the source node (temporary state retention) until the target node is fully synchronized and takes over control. This process may require a precise time synchronization mechanism (such as PTP) to coordinate the switching moment.

[0028] (e) Switchover and Cleanup: When predetermined conditions are met (e.g., when the state difference is less than a threshold or when the synchronization time point is reached), an atomic control switch is performed. The source node stops running the ACFM and cleans up related resources. The target node officially takes over control.

[0029] Runtime digital twin simulation and decision verification unit: Enhancements to ensure safe and reliable system operation: Pre-verification: After the collaborative optimization decision engine generates any joint decision plan that may change the system's operating status (such as new deployment, algorithm switching, major parameter adjustment, migration), the plan is not executed directly, but is first sent to this unit for pre-verification.

[0030] High-fidelity digital twin environment: This unit maintains a high-fidelity digital twin model that is tightly synchronized with the real-time state of the physical system (including the controlled plant, actuators, and sensors), compute nodes (CPU, memory, and load models), and the network (topology, dynamic behavior models, and models that predict the unit's state using the network state). This synchronization is typically achieved through real-time data streaming.

[0031] Fast virtual execution: Apply candidate decision scenarios to the digital twin environment and perform fast simulations to simulate system behavior over a period of time (e.g., seconds to minutes) after the decision is executed.

[0032] Quantitative Assessment and Decision Approval: During the simulation process, key performance indicators (KPIs) (e.g., control error, overshoot, settling time), resource utilization, CTRS or end-to-end constraint violations, and safety redline triggering are monitored. Based on the simulation results, quantitative risk scores (e.g., probability of instability, severity of constraint violations) and benefit scores (e.g., performance improvement, resource savings) are calculated. These scores are compared against pre-set, configurable safety and performance thresholds. Only when the risk score is below the threshold and the benefit score is above (or equal to) the threshold does the unit signal "approval for execution" to the rest of the platform; otherwise, the decision is rejected, potentially triggering a replanning of the decision engine.

[0033] Platform collaborative workflow example: Assume that a latency-sensitive ACFM_X deployed on edge node A begins to experience degradation in its control loop performance.

[0034] The network status prediction unit monitors that the predicted network path delay from node A to its cooperative node B will continue to increase and exceed the CTRS requirement (including uncertainty assessment) of ACFM_X.

[0035] The collaborative optimization decision engine receives this prediction information, the CTRS of ACFM_X (including high delay sensitivity), the real-time resource status of nodes A / B / C, and the algorithm library information.

[0036] The engine runs MPC optimization, evaluating several options: (a) switching to a more latency-robust algorithm at node A; (b) migrating ACFM_X to node C, which has better network conditions; and (c) adjusting the parameters of the existing ACFM_X algorithm. The engine finds that option (b), migrating to node C, is the best solution that meets performance and resource constraints.

[0037] This migration decision is sent to the digital twin validation unit.

[0038] The verification unit simulated the migration of ACFM_X to node C and its subsequent operation in a digital twin environment. The simulation showed a smooth migration process (low disturbance), post-migration control performance meeting the CTRS, sufficient resources at node C, and both risk and benefit scores meeting preset thresholds.

[0039] The verification unit sends an "approval" signal.

[0040] The low-disturbance state migration controller is started and ACFM_X is safely and smoothly migrated from node A to node C according to the protocol.

[0041] After the migration is complete, ACFM_X continues to run on node C with the optimized configuration, and control performance is restored.

[0042] Learning and Evolution: Running digital twin simulation and decision verification units is not only used for single decision verification, but also generates a large amount of data (specific decision plan -> simulation environment status -> simulation performance results). This data can be collected and used for: Feedback to the collaborative optimization decision engine: Especially when the engine adopts a learning-based method (such as reinforcement learning), these simulation data can be used as high-quality training samples (for example, constituting (State, Action, Reward, Next_State) tuples) for offline or online training and optimization of the decision policy network, enabling it to make more accurate and efficient decisions.

[0043] Improved models: Comparison of simulation results with actual system operation results can also be used to calibrate and improve the accuracy of the digital twin model itself and the network status prediction model.

[0044] The present invention realizes high performance, high reliability, high efficiency and high flexibility of distributed control systems in a complex and changeable industrial Internet environment through the collaborative work of the above-mentioned units, especially through the organic combination of the key links of predictive network perception, task requirement quantification, algorithm adaptation, dynamic collaborative optimization deployment / migration, and digital twin pre-verification. Based on the above description, those skilled in the art can understand the core idea of the present invention and implement it. It should be pointed out that the above embodiments are only preferred examples of the present invention and are not limitations of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A distributed control method for the industrial Internet, characterized in that: include: Obtaining real-time status parameters of the distributed control system network path, and using a prediction model to infer the probabilistic expected network state within a future preset time window based on the parameters; Decompose the overall control task into multiple atomic control function modules based on functional cohesion and interaction relationships, and establish a quantitative control task requirement spectrum for each atomic control function module; Based on the expected network state, the control task demand spectrum of the atomic control function module, the real-time available computing and network resource status of each potential deployment node, and the control algorithm strategy library information containing quantifiable network adaptability characteristics, a collaborative optimization decision engine is used to perform predictive multi-objective collaborative optimization calculations. The calculations are designed to minimize the combined cost of performance deviation and resource consumption, thereby jointly deciding the following for each atomic control function module in the next time period: Optimal deployment nodes; Selection of the optimal adaptive control algorithm or strategy; A refined set of operating parameters for the selected algorithm that matches the expected network and task states; According to the joint decision result, the corresponding atomic control function module is deployed at the optimal deployment node, and the selected optimal adaptive control algorithm and the refined operating parameters are used to perform control operations; when the joint decision indicates that the atomic control function module needs to change the deployment node, the low-disturbance migration protocol that maintains state consistency is activated and executed, so as to ensure the continuity of the key states of the control loop and the smooth transition of the overall system performance during the migration process.

2. The distributed control method for industrial Internet according to claim 1, characterized in that: When the parameters use the prediction model to infer the probabilistic expected network state within a preset time window in the future, time series analysis and machine learning model structure are applied to output a future state distribution prediction including expected values and statistics representing uncertainty; and when the collaborative optimization decision engine makes the joint decision, it uses the uncertainty statistics to evaluate the potential risks of different decision options, and gives priority to decision plans that show greater robustness to prediction uncertainty in the optimization calculation, wherein the probabilistic expected network state includes predictions of network delay, jitter and packet loss rate distribution.

3. The distributed control method for industrial Internet according to claim 2, characterized in that: The process of establishing the quantified control task requirement spectrum includes analyzing the characteristics of the atomic control function modules to quantify their resource and network quality requirements, and the control task requirement spectrum further defines the end-to-end timing dependency constraints across the atomic control function module chain; In the optimization calculation, the collaborative optimization decision engine treats satisfying the end-to-end timing dependency constraint as a high-priority optimization goal or hard constraint condition. The control task requirement spectrum defines the specific requirements of the atomic control function module for computing resources, maximum tolerable network delay, delay jitter sensitivity threshold, data throughput and reliability level.

4. The distributed control method for industrial Internet according to claim 3, characterized in that: The collaborative optimization decision engine uses a model-based predictive control framework to perform the integrated, predictive multi-objective optimization calculations. The calculations involve constructing an evaluation function that comprehensively evaluates the expected performance, resource consumption, and quantified migration costs and system-level chain effects under deployment selection, algorithm selection, and parameter configuration, and searches for the optimal joint decision result in the solution space through a rolling optimization mechanism.

5. The distributed control method for industrial Internet according to claim 4, characterized in that: The collaborative optimization decision engine processes the quantified migration cost and the system-level chain effect as one of the optimization objectives or constraints in the optimization calculation.

6. The distributed control method for industrial Internet according to claim 5, characterized in that: The control algorithm strategy library includes a control algorithm strategy library with quantifiable network adaptability characteristics, and the algorithm or strategy is internally associated with metadata describing its network adaptability. The control algorithm strategy library also includes at least one control strategy that uses a state observer for compensation based on the expected network state. The refined operation parameter set includes instructions or values for dynamically adjusting key model parameters of the state observer to improve control accuracy and stability under network changes.

7. The distributed control method for industrial Internet according to claim 6, characterized in that: The low-disturbance migration protocol with state consistency maintenance includes: before the migration starts, the source node captures a complete execution state snapshot of the atomic control function module; after the target node prepares the operating environment, the snapshot is transmitted and the atomic control function module is restored at the target node; during the migration transition, an incremental state synchronization mechanism is adopted between the source node and the target node, combined with a control instruction interpolation or temporary state maintenance strategy based on short-term prediction to minimize control output interruption or jitter caused by state handover and network switching.

8. The distributed control method for industrial Internet according to claim 7, characterized in that: The method further includes a strategy pre-verification step based on the runtime digital twin before executing the joint decision result. The pre-verification step includes: inputting the candidate joint decision plan generated by the collaborative optimization decision engine into a high-fidelity digital twin environment that is closely synchronized with the real-time state of the physical system and includes network behavior simulation and computing resource simulation for rapid virtual execution; based on the expected control performance trajectory, resource consumption simulation and potential failure mode analysis results generated by the virtual execution, the candidate joint decision plan is quantitatively scored and the benefit evaluated; only when the evaluation result meets the preset safety and performance thresholds, the candidate joint decision plan is approved for execution in the actual physical system.

9. A distributed control platform for the industrial Internet, configured to implement the distributed control method for the industrial Internet according to claim 8, characterized in that: The platform includes: A network state prediction unit configured to calculate a probabilistic expected network state including uncertainty statistics based on real-time network parameters using a prediction model; a control task decomposition and requirement management unit configured to decompose the control task into atomic control function modules and manage a quantized control task requirement spectrum including end-to-end timing dependency constraints; a network adaptive control algorithm library storing a plurality of control algorithm strategies with quantifiable network adaptability characteristics, and including at least one control strategy using a state observer for compensation based on the expected network state; a collaborative optimization decision engine configured to utilize a model-based predictive control framework to perform integrated, predictive, multi-objective optimization calculations for joint decision-making based on the expected network state, the control task demand spectrum, the node resource state, and the algorithm library information, the calculations taking into account migration costs and system-level knock-on effects; a low-disturbance state migration controller configured to execute a low-disturbance migration protocol including incremental state synchronization and state consistency preservation of a transition strategy; A runtime digital twin simulation and decision verification unit is configured to perform strategy pre-verification before executing the joint decision result, wherein the pre-verification is based on rapid virtual execution and quantitative risk and benefit assessment in a high-fidelity digital twin environment.

10. The distributed control platform for the Industrial Internet according to claim 9, characterized in that: The runtime digital twin simulation and decision verification unit is further configured to feed back the data generated during its simulation process to the collaborative optimization decision engine. The collaborative optimization decision engine uses the simulation data as training samples or experience playback to continuously optimize its own decision model and decision accuracy in an online or offline manner.

Images