Industrial control network communication task scheduling method based on predictive time constraint and priority classification

Abstract

The invention discloses an industrial control network communication task scheduling method based on predictive time constraint and priority classification. The method comprises the following steps that S1, industrial communication data are collected and preprocessed; s2, an SARIMA model is modeled in an off-line mode, and residual analysis is carried out; s3, the MCU carries out online calculation on the current task through historical data, predictively and dynamically classifies the priority of the task, and carries out scheduling according to the priority of the task; and S4, constructing a scheduling result feedback link, forming a closed loop by overall prediction, scheduling and feedback, forming a closed loop updating module, and updating a closed loop parameter adjustment strategy in real time. According to the method, a dynamic priority classification mechanism based on time sequence prediction is realized by embedding time-load-urgency multi-dimensional prediction into threshold generation, so that the high-critical task assessment access priority is improved, dynamic balance between an emergency task and a conventional service is realized, and the comprehensive load capacity of an industrial control system is improved.

Description

Technical Field

[0001] This invention belongs to the field of information technology and relates to a scheduling method for network communication tasks in industrial control systems. Specifically, it relates to a communication scheduling method based on the SARIMA model (Seasonal Autoregressive Integrated Moving Average Model) and an improved EDF algorithm, which includes predictive time constraints and priority classification. Background Technology

[0002] With the development of industrial control networks and the increasing demand for intelligent manufacturing, communication scheduling mechanisms directly determine the timeliness of command response and the consistency of production rhythm in scenarios such as production line collaborative control, remote equipment monitoring, and distributed intelligent decision-making. Especially in critical links such as precision machining, robot collaboration, and process control, even microsecond-level delays can cause equipment malfunctions or quality fluctuations. Faced with periodic sensor data, sudden event alarms, and critical control commands generated by a large number of nodes in the industrial field, how to achieve efficient and orderly transmission over a shared communication medium has become a key bottleneck restricting the performance improvement of industrial control systems.

[0003] Current mainstream real-time scheduling algorithms, such as EDF (Earliest Deadline First), primarily make decisions based on static information of the current task queue, essentially operating on a passive response model. When unexpected traffic spikes or sudden changes in equipment status occur in industrial settings, traditional EDF algorithms struggle to adjust resource allocation strategies in advance, easily leading to increased deadline miss rates and network utilization imbalances. Furthermore, existing methods generally lack the ability to quantitatively predict communication load changes and cannot transform historical traffic characteristics into forward-looking scheduling parameters, making it difficult to maintain a dynamic balance between urgent tasks and routine operations. Therefore, a communication scheduling method with predictive time constraints and priority classification is needed to improve the task coordination and load balancing capabilities of industrial control communication systems. Summary of the Invention

[0004] To address the common problem that existing methods generally lack the ability to quantitatively predict communication load change trends, this invention provides a predictive time constraint and priority classification method for industrial control network communication task scheduling. This method achieves a dynamic priority classification mechanism based on time series prediction by embedding multi-dimensional prediction of "time-load-urgency" into threshold generation, thereby improving the evaluation and admission priority of highly critical tasks and realizing a dynamic balance between urgent tasks and routine business, which is conducive to improving the overall load capacity of industrial control systems.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A method for scheduling industrial control network communication tasks based on predictive time constraints and priority classification includes the following steps:

[0007] Step S1: Industrial communication data acquisition and preprocessing;

[0008] Step S2: Offline Modeling Input historical traffic data and output optimal results. Parameter vector and residual variance ,in These are, respectively, the non-seasonal autoregressive term, the non-seasonal difference frequency, the non-seasonal moving average term, the seasonal autoregressive term, the seasonal difference frequency, the seasonal moving average term, and the seasonal cycle;

[0009] Step S3: Perform residual analysis on the SARIMA model obtained in step S2. If the residual analysis results meet the requirements, proceed to step S4; otherwise, return to step S2 to re-determine the modeling parameters.

[0010] Step S4: Load the SARIMA model obtained in step S3 into the MCU;

[0011] Step S5: An online prediction algorithm is formed using the SARIMA model. The MCU calculates and predicts online, and the algorithm cycles every 100ms to obtain the real-time update results of the SARIMA model from the current time to 100ms ago.

[0012] Step S6: Introduce closed-loop correction coefficients to form a threshold calculation function and generate a dynamic threshold. and The conservative prediction upper limit of the task is compared with the threshold, and the task is divided into three categories: Cat-A, Cat-B, and Cat-C. Cat-A is urgent and hard preemption is used; Cat-B is real-time and uses an 8-level power function partitioning with improved EDF; Cat-C is non-real-time and uses background filling channels.

[0013] Step S7: Map the prediction model to the MCU scheduling parameter interface via DMA;

[0014] Step S8: The MCU decision-making end executes the improved EDF scheduling main loop to adjust the EDF priority queue;

[0015] Step S9: Construct a scheduling result feedback loop, forming a closed loop of overall prediction-scheduling-feedback, and constituting a closed-loop update module. and Driven by miss error and jitter error respectively, the gradient is updated every 1 second using error feedback, and then the process returns to step S6 to repeat. This allows the dynamic threshold to automatically loosen / tighten, always moving in the direction of "error reduction".

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] (1) In view of the time-varying characteristics of the communication load of industrial control network, this invention uses the SARIMA model to make accurate quantitative predictions and transforms historical traffic characteristics into forward-looking scheduling parameters to provide prerequisite information support for real-time scheduling.

[0018] (2) By introducing the SARIMA prediction results into the EDF algorithm, dynamic adjustments are made according to the future network situation, and a dynamic priority classification mechanism based on time series is established. While ensuring the time limit of emergency tasks, the load of regular business is adaptively balanced, thereby improving the system response determinism.

[0019] (3) Construct a prediction-scheduling-feedback closed loop, use the scheduling results to continuously feed back and optimize the decision model, so that the system has self-correction capability and continuously enhances the task coordination and load balancing level of the industrial control network under complex working conditions. Attached Figure Description

[0020] Figure 1 A flowchart of the overall process for scheduling industrial control network communication tasks based on predictive time constraints and priority classification.

[0021] Figure 2 A flowchart illustrating the specific process of scheduling industrial control network communication tasks based on predictive time constraints and priority classification. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0023] This invention provides a method for scheduling industrial control network communication tasks based on predictive time constraints and priority classification, such as... Figure 1 As shown, the method includes the following steps:

[0024] Step S1: Industrial communication data acquisition and preprocessing;

[0025] Step S2: Offline modeling of the SARIMA model and residual analysis;

[0026] Step S3: The MCU calculates the current task online using historical data and predictively and dynamically classifies the task priority, and executes scheduling accordingly;

[0027] Step S4: Construct a scheduling result feedback loop, forming a closed loop of overall prediction-scheduling-feedback, and construct a closed loop update module to update the closed loop parameter adjustment strategy in real time.

[0028] like Figure 2 As shown, the specific implementation steps are as follows:

[0029] Step S1: Industrial communication data acquisition and preprocessing. The specific steps are as follows:

[0030] Step S11: Collect 30 days of historical data using traffic probes. The data packet components include data packet size, arrival timestamp of each data stream, relative deadline, priority level, and network load. Process the data and form a triple (t, load, urgency), and write it into the prediction layer shared queue.

[0031] Step S12: Perform the ADF (Augmented Dickey-Fuller Test). If the test value meets the requirements, proceed to the next step; otherwise, perform a difference operation and select the difference order. This ensures that the test values ​​meet the requirements, allowing the stationarity hypothesis to be accepted and guaranteeing the smooth operation of SARIMA in the future.

[0032] Step S2: Offline Modeling Input historical traffic data and output optimal results. Parameter vector and residual variance The specific steps are as follows:

[0033] Step S21: Use the historical flow data after completing the ADF test. The model is used for modeling, where These are, respectively, the non-seasonal autoregressive term, the non-seasonal difference frequency, the non-seasonal moving average term, the seasonal autoregressive term, the seasonal difference frequency, the seasonal moving average term, and the seasonal cycle.

[0034] Step S22: Run the SARIMA model within the selected order range, and use the AIC (Akaike Information Criterion) to minimize the selected order. .

[0035] Step S23: Use MLE (Maximum Likelihood Estimation) to calculate the converged parameter vector using the "initial value → iteration → convergence" method. and residual variance , where the parameter vector Includes non-seasonal autoregressive coefficients Non-seasonal moving average coefficient Seasonal autoregression coefficient Seasonal moving average coefficient .

[0036] Step S3: Perform residual analysis on the SARIMA model obtained in Step S2. If the residual analysis results meet the requirements, proceed to Step S4; otherwise, return to Step S2 to re-determine the modeling parameters. The specific steps are as follows:

[0037] Step S31: Combine the predicted values ​​and differences after running the SARIMA model. The difference between stationary sequences after the order is used as the residual. Its variance The benchmark for establishing the confidence interval for online prediction is ensured by the Ljung-Box, Jarque-Bera, and ARCH-LM triple tests. It is approximately white noise;

[0038] Step S32: If the test fails, increase the order or transform and re-estimate until the residual passes the diagnosis; otherwise, mark the flow as high noise and increase the online confidence bandwidth.

[0039] Step S4: Load the SARIMA model obtained in step S3 into the MCU.

[0040] Step S5: An online prediction algorithm is generated using the SARIMA model. The MCU calculates and predicts online, looping every 100ms to obtain the real-time update result of the SARIMA model from the current moment to 100ms ago. The specific steps are as follows:

[0041] Step S51: Collect the accumulated values ​​in the acquisition window to obtain the original sample;

[0042] Step S52: Adjust the step size Perform recursion. To obtain the maximum number of recursion steps, we obtain the real-time residuals, and then further obtain the updated residual variance values. The formula is as follows:

[0043]

[0044] Where: forgetting factor , For the first Step defect.

[0045] Step S53: Calculate the 95% confidence band half-width The formula is as follows:

[0046]

[0047]

[0048] In the formula: is the variance propagation coefficient.

[0049] Step S54: Introduce the anomaly flag, if If so, then set the bit to anomaly. The value of the confidence band half-width under certain conditions is given by the formula above. Calculation. When anomaly=True: Immediately enlarge online. (The left arrow is the assignment symbol, the same below), the confidence band is widened and the threshold is more conservative; the next modeling cycle is shortened to 50ms to quickly track sudden situations.

[0050] Step S55: Following steps S51 to S54, calculate the load level of the current traffic to obtain the predicted load point value. The 95% confidence half-width of the load is ; Perform calculations on the urgency points to obtain the prediction. .

[0051] Step S56: Predict the relative deadline of the task to obtain the predicted value. 95% confidence half-width is Generate predictive parameter vectors The formula for calculating the relative deadline of the original sample message is:

[0052]

[0053] In the formula: This is the deadline for message processing. This represents the message arrival time.

[0054] Step S6: Introduce closed-loop correction coefficients to form a threshold calculation function and generate a dynamic threshold. and The conservative prediction upper limit of the task is compared with the threshold, and the task is divided into three categories: Cat-A, Cat-B, and Cat-C. Cat-A is urgent and uses hard preemption; Cat-B is real-time and uses an 8-level power function partitioning with an improved EDF; Cat-C is non-real-time and uses background-filled channels. The specific steps are as follows:

[0055] Step S61: Introduce closed-loop correction coefficient and The formula is as follows:

[0056]

[0057]

[0058] In the formula: The gradient step size is a fixed value, typically set to 1. `miss_ratio` is the percentage of deadline frames missed in the past second (target 0.01%), `jitter` is the average absolute deviation between the task completion time and the deadline, and `jitter_tgt` is its target deviation, set to 50. .

[0059] Step S62: Form a dynamic threshold calculation function and The calculation formula is as follows:

[0060]

[0061]

[0062] In the formula: and All are gain coefficients, and are taken as fixed values. and All are initial value constants, in units of .

[0063] Among the coefficients mentioned above, , It is adjustable in a closed loop. , , Fixed and unchanging.

[0064] Step S63: For each task in the waiting queue, use the conservative upper bound method to construct a conservative upper bound relative to its deadline, as shown in the following formula:

[0065]

[0066] In the formula: This is a conservative upper limit relative to the deadline. These are model predictions relative to the cutoff date. For safety factor, take , This represents the standard deviation of the prediction at the same step.

[0067] At this point, the SARIMA module is only used to predict the upper limit of the relative cutoff period, which is less than 100%. If the probability distribution is considered to be normally distributed, then the conservative upper limit is the 97.7% upper limit.

[0068] Step S64: Transfer the data for each task. Compared with dynamic thresholds, if If so, it is marked as a Cat-A task; if If so, it is marked as a Cat-B task; If so, it is marked as a Cat-C task.

[0069] Step S7: Map the prediction model to the MCU scheduling parameter interface via DMA.

[0070] Step S8: The MCU decision-making end executes the improved EDF scheduling main loop and adjusts the EDF priority queue.

[0071] In this step, the MCU executes scheduling decisions, and the following situations may occur:

[0072] Case (1) If there is at least one Cat-A task on the network load, then the Cat-A task has a higher priority than the Cat-B and Cat-C tasks, and among the Cat-A tasks, priority is determined by... A size comparison is performed, and then the strict EDF algorithm is executed; after the Cat-A task is completed, the scheduling strategy is handled according to case (2).

[0073] Case (2) If there are only a few Cat-B tasks and Cat-C tasks on the network load, the continuous deadlines are mapped to 8 discrete partitions using an power function partitioning method. Then, the Cat-B tasks are divided into 8 partitions. For different partitions, the smaller partition number takes priority. Within the same partition, the priority is still given to the smaller partition number. Ascending order (strict EDF); after the Cat-B task is completed, the scheduling strategy is handled according to situation (3).

[0074] Case (3) If only Cat-C tasks exist on the network load, the channel is relatively idle, and the Cat-C tasks are sent via FIFO.

[0075] Step S9: Construct a scheduling result feedback loop, forming a closed loop of overall prediction-scheduling-feedback, and constituting a closed-loop update module. and Driven by miss error and jitter error respectively, the gradient is updated every 1 second using error feedback, and then the process returns to step S6 to repeat. This allows the dynamic threshold to automatically loosen / tighten, always moving in the direction of "error reduction".

[0076] In this step, the MCU collects feedback from the past second: miss_ratio and jitter. The following situations may occur:

[0077] Case (1) If miss_ratio > target or jitter > jitter_tgt, an error occurs, and the closed-loop correction coefficient needs to be adjusted to raise the threshold and process the task as soon as possible. The formula is as follows:

[0078]

[0079]

[0080]

[0081] In the formula: , It is an initial constant.

[0082] In case (2), if the target is met for 5 consecutive seconds, the dynamic threshold gradually approaches the minimum allowable value while ensuring the deadline, thus reducing bandwidth pressure. The formula is as follows:

[0083]

[0084]

[0085] Through feedback adjustment, parameters are updated in real time and dynamic thresholds are determined. The process returns to step S6 and repeats, allowing the dynamic thresholds to automatically loosen / tighten, always moving in the direction of "error reduction", thus achieving self-optimization of the prediction-scheduling-feedback closed loop.

[0086] Example:

[0087] Step S1: Industrial communication data acquisition and preprocessing. 30 days of historical data are collected using a flow probe. The data packet components include packet size, arrival timestamps of each data stream, relative deadlines, priority levels, and network load, forming a triple (2025-01-05-194706, "payload", 5), which is written to the prediction layer's shared queue. Further, an ADF test is performed. It is found that after first-order differencing, the ADF test passes, allowing acceptance of the stationarity hypothesis and ensuring stable operation of subsequent SARIMA. Therefore, the differencing order... .

[0088] Step S2: Offline Modeling The model is obtained by selecting the minimum order of AIC. The converged parameter vector is calculated using the "initial value → iteration → convergence" method. and residual variance .

[0089] Step S3: Perform residual analysis on the obtained model. If the residual analysis results meet the requirements, proceed to step S4; otherwise, return to step S2 to re-determine the modeling parameters.

[0090] In this embodiment, for the case without autocorrelation, the significance probability is... For the normal case, the significance probability For the case without heteroscedasticity, the significance probability All three values ​​are greater than 0.05, as confirmed by residual analysis.

[0091] Step S4: Load the obtained SARIMA model into the MCU.

[0092] Step S5: The model forms an online prediction algorithm, the MCU calculates and predicts online, and cycles every 100ms to obtain the real-time update result of the model from the current time to 100ms ago.

[0093] In this embodiment, load prediction With urgency point prediction Normalization, i.e. Furthermore, the closer the value is to 1, the greater the network load and pressure. For example, the traffic segment captured in this example... , , , Predicted value vector , .

[0094] Step S6: Introduce closed-loop correction coefficients to form a threshold calculation function and generate a dynamic threshold. and The conservative prediction upper limit of the task is compared with the threshold to form a three-level classification: Cat-A / Cat-B / Cat-C, where Cat-A is urgent and hard preemption is used; Cat-B is real-time and uses an 8-level power function partitioning with improved EDF; and Cat-C is non-real-time and uses background filling channels.

[0095] In this embodiment, the coefficient value is selected. , , , anomaly=False; select an initial constant. , Calculated by the formula , .

[0096] Estimating the conservative upper bound vector .

[0097] The five tasks of Cat-B level and above in this traffic segment are labeled as tasks 1 to 5 from front to back. According to the calculation results, tasks 2 and 3 are Cat-A, tasks 1, 4 and 5 are Cat-B, and the others are Cat-C.

[0098] Step S7: Map the prediction model to the MCU scheduling parameter interface via DMA.

[0099] Step S8: The MCU decision-making end executes the improved EDF scheduling main loop and adjusts the EDF priority queue.

[0100] In this embodiment, Cat-A is executed first. Based on the comparison of relative deadlines, task 2 is executed first, followed by task 3. Tasks 1, 4, and 5 are Cat-B, where task 1 is in a smaller area code, and tasks 4 and 5 have the same area code. According to the algorithm, their priorities are 1, 4, and 5. Therefore, the scheduling order of this traffic segment is task 2, 3, 1, 4, and 5 in sequence.

[0101] Step S9: Construct a scheduling result feedback loop, forming a closed loop of overall prediction-scheduling-feedback, and constituting a closed-loop update module. and Driven by miss error and jitter error respectively, it performs error feedback gradient update every 1 second and returns to the S6 loop to execute, allowing the dynamic threshold to automatically loosen / tighten, always moving in the direction of "error reduction".

[0102] In this embodiment, miss_ratio=0% <target,jitter=53 >jitter_tgt, therefore, the closed-loop correction coefficient needs to be adjusted, and an initial constant value is taken. , .

[0103] Calculated by the formula , The coefficient is fed back to the dynamic threshold adjustment step to correct the loop and achieve prediction-scheduling-feedback closed-loop self-optimization.

Claims

1. A method for scheduling industrial control network communication tasks based on predictive time constraints and priority classification, characterized in that... The method includes the following steps: Step S1: Industrial communication data acquisition and preprocessing; Step S2: Offline Modeling Input historical traffic data and output optimal results. Parameter vector and residual variance ,in These are, respectively, the non-seasonal autoregressive term, the non-seasonal difference frequency, the non-seasonal moving average term, the seasonal autoregressive term, the seasonal difference frequency, the seasonal moving average term, and the seasonal cycle; Step S3: Perform residual analysis on the SARIMA model obtained in step S2. If the residual analysis results meet the requirements, proceed to step S4; otherwise, return to step S2 to re-determine the modeling parameters. Step S4: Load the SARIMA model obtained in step S3 into the MCU; Step S5: An online prediction algorithm is formed using the SARIMA model. The MCU calculates and predicts online, and the algorithm cycles every 100ms to obtain the real-time update results of the SARIMA model from the current time to 100ms ago. Step S6: Introduce closed-loop correction coefficients to form a threshold calculation function and generate a dynamic threshold. and The task's conservative prediction upper limit is compared with the threshold, and the task is divided into three categories: Cat-A, Cat-B, and Cat-C. Step S7: Map the prediction model to the MCU scheduling parameter interface via DMA; Step S8: The MCU decision-making end executes the improved EDF scheduling main loop to adjust the EDF priority queue; Step S9: Construct a scheduling result feedback loop, forming a closed loop of overall prediction-scheduling-feedback, and constituting a closed-loop update module. and Driven by miss error and jitter error respectively, an error feedback gradient update is performed every 1 second, and the process returns to step S6 to repeat, allowing the dynamic threshold to automatically loosen / tighten, always moving in the direction of "error reduction".

2. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 1, characterized in that... The specific steps of step S1 are as follows: Step S11: Collect 30 days of historical data using flow probes, process the data and form a triple (t, load, urgency), and write it into the prediction layer shared queue. Step S12: Perform an ADF test. If the test value meets the requirements, proceed to the next step; otherwise, perform a difference operation and select the difference order. This is to ensure that the test values ​​meet the requirements.

3. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 2, characterized in that... In step S11, the data packet components include data packet size, arrival timestamps of each data stream, relative deadlines, priority levels, and network load.

4. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 1, characterized in that... The specific steps of step S2 are as follows: Step S21: Use the historical flow data after completing the ADF test. Modeling is performed; Step S22: Run the SARIMA model within the selected order range, and use the minimum selected order in the AIC. ; Step S23: Use MLE to calculate the converged parameter vector using the "initial value → iteration → convergence" method. and residual variance , where the parameter vector Includes non-seasonal autoregressive coefficients Non-seasonal moving average coefficient Seasonal autoregression coefficient Seasonal moving average coefficient .

5. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 1, characterized in that... The specific steps of step S3 are as follows: Step S31: Combine the predicted values ​​and differences after running the SARIMA model. The difference between stationary sequences after the order is used as the residual. Its variance The benchmark for establishing the confidence interval for online prediction is ensured by the Ljung-Box, Jarque-Bera, and ARCH-LM triple tests. It is approximately white noise; Step S32: If the test fails, increase the order or transform and re-estimate until the residual passes the diagnosis; otherwise, mark the flow as high noise and increase the online confidence bandwidth.

6. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 1, characterized in that... The specific steps of step S5 are as follows: Step S51: Collect the accumulated values ​​in the acquisition window to obtain the original sample; Step S52: Adjust the step size Perform recursion. To obtain the maximum number of recursion steps, we obtain the real-time residuals, and then further obtain the updated residual variance values. The formula is as follows: In the formula: Forgetting factor, For the first Step residual; Step S53: Calculate the 95% confidence band half-width The formula is as follows: In the formula: The variance propagation coefficient; Step S54: Introduce the anomaly flag, if Then set anomaly. The value representing the confidence band half-width, when anomaly=True: immediately zoom in online. The next modeling cycle is shortened to 50ms. Step S55: Following steps S51 to S54, calculate the load level of the current traffic to obtain the predicted load point value. The 95% confidence half-width of the load is ; Perform calculations on the urgency points to obtain the prediction. ; Step S56: Predict the relative deadline of the task to obtain the predicted value. 95% confidence half-width is Generate predictive parameter vectors .

7. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 6, characterized in that... The formula for calculating the relative cutoff period is: In the formula: This is the deadline for message processing. This represents the message arrival time.

8. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 1, characterized in that... The specific steps of step S6 are as follows: Step S61: Introduce closed-loop correction coefficient and The formula is as follows: In the formula: The gradient step size is denoted as , miss_ratio is the percentage of deadline frames missed in the past 1 second, jitter is the average absolute deviation between the task completion time and the deadline, and jitter_tgt is the target deviation. Step S62: Form a dynamic threshold calculation function and The calculation formula is as follows: In the formula: and All are gain coefficients. and All are initial value constants; Step S63: For each task in the waiting queue, use the conservative upper bound method to construct a conservative upper bound relative to its deadline, as shown in the following formula: In the formula: This is a conservative upper limit relative to the deadline. These are model predictions relative to the cutoff date. For safety reasons, The standard deviation of the forecast at the same step; Step S64: Transfer the data for each task. Compared with dynamic thresholds, if If so, it is marked as a Cat-A task; if If so, mark it as a Cat-B task; If so, it is marked as a Cat-C task.

9. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 1, characterized in that... In step S8, the MCU executes a scheduling decision, and the following situations may occur: Case (1) If there is at least one Cat-A task on the network load, then the Cat-A task has a higher priority than the Cat-B and Cat-C tasks, and among the Cat-A tasks, priority is determined by... A size comparison is performed, and then the strict EDF algorithm is executed; after the Cat-A task is completed, the scheduling strategy is handled according to case (2); Case (2) If there are only a few Cat-B tasks and Cat-C tasks on the network load, the continuous deadlines are mapped to 8 discrete partitions using an power function partitioning method. Then, the Cat-B tasks are divided into 8 partitions. For different partitions, the smaller partition number takes priority. Within the same partition, the priority is still given to the smaller partition number. Ascending order; after the Cat-B task is completed, the scheduling strategy will be handled according to situation (3); Case (3) If only Cat-C tasks exist on the network load, the channel is relatively idle, and the Cat-C tasks are sent via FIFO.

10. The industrial control network communication task scheduling method based on predictive time constraints and priority classification according to claim 1, characterized in that... In step S9, the MCU collects feedback from the past second: miss_ratio and jitter, and the following situations exist: Case (1) If miss_ratio > target or jitter > jitter_tgt, an error occurs, and the closed-loop correction coefficient needs to be adjusted to raise the threshold and process the task as soon as possible. The formula is as follows: In the formula: , It is an initial constant; In case (2), if the target is met for 5 consecutive seconds, the dynamic threshold gradually approaches the minimum allowable value while ensuring the deadline, thus reducing bandwidth pressure. The formula is as follows: Through feedback adjustment, parameters are updated in real time and dynamic thresholds are determined. The process returns to step S6 and repeats, allowing the dynamic thresholds to automatically loosen / tighten, always moving in the direction of "error reduction", thus achieving self-optimization of the prediction-scheduling-feedback closed loop.

Images