An automation control system and an automation control method

Abstract

The application discloses an automatic control system and method, and relates to the technical field of industrial automatic control. The method comprises the following steps: collecting real-time data of a plurality of control points at a fixed period; calculating a multi-dimensional state feature vector containing mean value, standard deviation and slope for each control point; comparing the state feature vector with a preset stable operation interval to screen a point to be adjusted; obtaining a basic control action and an intensity coefficient by looking up a table; determining associated points according to a topological relationship, predicting state changes of the associated points by using a pre-trained influence relationship model, and iteratively reducing the intensity coefficient when the predicted state of the associated points exceeds the stable interval of the associated points, to generate a final control instruction; evaluating control convergence and system fluctuation after executing the instruction; and adjusting the weight of a strategy table according to the evaluation result. The application realizes rapid suppression of local deviation under coupling constraints, and helps to maintain stable operation of the whole system.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and more specifically, to an automated control system and an automated control method. Background Technology

[0002] In complex industrial automated production lines, multiple control units are interconnected through material and information flows to collaboratively complete production tasks. When a unit experiences performance deviations due to equipment wear and tear or external disturbances, conventional control methods typically employ independent adjustments based on local models or rely on computationally complex global optimization algorithms. The former fails to adequately consider the coupling relationships between units, and adjustments may cause unexpected disturbances to related units, leading to cascading system fluctuations. The latter, while seeking a global solution, suffers from heavy computational burdens, response delays, and insufficient support for the millisecond-level real-time requirements of high-cycle production scenarios. Furthermore, its effectiveness depends on a high-precision system model, limiting its adaptability in time-varying production line environments. Existing technologies have explored game theory and reinforcement learning-based methods for handling system interactions, but online solution and network training incur significant computational overhead. Other scheduling optimization methods based on big data analysis have relatively long control cycles, resulting in insufficient intervention timeliness for rapidly suppressing real-time dynamic disturbances between production line equipment. Therefore, existing technologies face a challenge in balancing real-time performance with collaborative control effectiveness when dealing with the real-time control of tightly coupled multi-unit systems. Therefore, an automated control system and an automated control method are proposed to address the above problems. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide an automated control system and an automated control method to solve the problem that the prior art is difficult to balance real-time response capability and global collaborative control effect when facing tightly coupled multi-unit control scenarios, thereby suppressing local deviations while avoiding disturbances to related units that exceed the allowable range.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an automated control method, comprising the following steps: S1. Obtain the data defined in the production process at a fixed sampling period. Real-time data of each control point The value is a positive integer greater than 1. This sampling period ensures that the state changes of each control point can be captured in a timely manner, providing a data basis for subsequent real-time control. S2. For each control point, based on its latest... A state feature vector is calculated from real-time data of several consecutive sampling periods. The state feature vector is a positive integer greater than 2, and is used to quantify the operating state of the control point from multiple dimensions in order to identify abnormal trends. S3. Compare the state feature vector of each control point with the preset stable operating range of that control point, and select control points whose at least one feature component in the state feature vector exceeds its corresponding stable operating range as adjustment points, so as to calibrate them in the early stage of local deviation. S4. For each point to be adjusted, based on the category and magnitude of its out-of-limit characteristic components, a basic control action and a basic intensity coefficient are matched from the preset first strategy table. The response is graded according to the type and degree of deviation to obtain a preliminary adjustment strategy. S5. Based on the preset production process topology, identify the points directly related to the current point to be adjusted. One related point, It is an integer greater than or equal to 0; Based on this The current state feature vector of each associated point is used to predict the predicted change in the state feature vector of each associated point after the basic control action is executed under the basic intensity coefficient using a pre-trained influence relationship model, and the predicted state feature vector of each associated point is calculated based on the predicted change. If the predicted state feature vector of any associated point exceeds its own stable operating range, the basic intensity coefficient is reduced by a preset step size, and prediction and judgment are performed again until the predicted state feature vectors of all associated points do not exceed their own stable operating range, or the intensity coefficient reaches the preset lower limit value. The adjusted strength coefficient is used as the final strength coefficient, which together with the basic control action constitutes the final control command. Before the adjustment action is executed, this step quantitatively verifies the impact on the associated points and generates a safe and feasible control command under coupling constraints by iteratively adjusting the strength coefficient with an upper bound. S6. Execute the final control command; S7. Continuous execution of instructions Within each sampling period, The value is a positive integer greater than 1. Real-time data of the point to be adjusted and its associated points are collected, and the trajectory of the change of its state feature vector is calculated. Based on the trajectory of change, the convergence index and volatility index of this control are calculated to quantify the degree of deviation elimination of the point to be adjusted and the magnitude of the transient disturbance of the associated points, respectively. S8. Establish an empirical data set, which is at least associated with the state feature vector of the point to be adjusted in S3, the final control command generated in S5, and the convergence index and volatility index calculated in S7. Based on the convergence index and volatility index, adjust the weight parameters of the relevant entries in the first strategy table so that the selection of the control strategy can be optimized online based on the historical control effect.

[0005] Furthermore, the calculation of the state feature vector in step S2 includes the following sub-steps: For each control point, calculate its nearest neighbor. The arithmetic mean of real-time data from each sampling period is used as the first characteristic component, which reflects the average operating level of the control point within the observation window. Calculate the The standard deviation of each data point relative to the arithmetic mean is used as a second characteristic component, which reflects the degree of fluctuation in the operating state of the control point. Based on sampling point number As the independent variable, with the above The data is used as the dependent variable, and the slope is calculated by linear regression as the third characteristic component, which reflects the changing trend of the control point's operating status. The first feature component, the second feature component, and the third feature component constitute the state feature vector, which comprehensively quantifies the state of the control point through three dimensions: mean, fluctuation, and trend.

[0006] Furthermore, the preset stable operating range in step S3 includes: a first numerical range set for the first feature component, a second numerical range set for the second feature component, and a third numerical range set for the third feature component, thereby realizing multi-dimensional monitoring and anomaly identification of the control point status.

[0007] Furthermore, in step S4, the first strategy table is a two-dimensional lookup table, with its row index being the feature component category and its column index being the exceedance level; the exceedance level is divided according to the percentage range of the feature component exceeding its stable operating range boundary value, thereby establishing a hierarchical correspondence between the degree of deviation and the response strength.

[0008] Furthermore, the influence relationship model described in step S5 is a multiple linear regression model or a shallow neural network model; For each state feature component of each associated point, an independent sub-model is established. This decoupling reduces the complexity of a single model and makes the prediction more targeted. The input features of the sub-model include: the current state feature vector of the point to be adjusted, the vectorized encoding of the control action identifier of the point to be adjusted, the intensity coefficient of the point to be adjusted, and the current state feature vector of the associated point, so that the input information covers the state of the disturbance source, the adjustment action attribute and the current state of the disturbed object. The output of the sub-model is the predicted change of the state feature components corresponding to the associated points; The influence relationship model is pre-trained using historical operating data, which includes records of control actions and corresponding state changes of the production line under normal operating conditions and controlled disturbances. Wherein, the predicted state feature vector of the associated point is equal to the sum of the current state feature vector of the associated point and the predicted change.

[0009] Furthermore, the preset step size in step S5 is a fixed positive value, and the preset lower limit value is a fixed value greater than 0. This setting ensures that the magnitude of each adjustment is determined and that there is a clear upper limit to the number of iterations, thereby guaranteeing the time predictability of the decision-making process.

[0010] Furthermore, the calculation of the convergence index and volatility index in step S7 includes the following sub-steps: For the point to be adjusted, compare it with the one described in the above. In the state feature vector at the end of each cycle, the component value corresponding to the over-limit feature component identified in S3 is the difference between the component value and the median value of the stable operating range corresponding to the over-limit feature component; if the difference is less than the preset tolerance, the convergence index is assigned a convergence state value, otherwise it is assigned a non-convergence state value, thereby determining whether the point to be adjusted has been effectively pulled back to the expected range. For each associated point, its statistics are presented in the... The maximum offset of each component of the state feature vector within a period relative to its initial value is taken as the volatility index, and the maximum value among all components of all related points is used to reflect the maximum transient impact of this adjustment on the system as a whole.

[0011] Furthermore, the specific rules for adjusting the weight parameters in step S8 are as follows: If the convergence index indicates convergence and the volatility index is below the set threshold, then the weight parameter of the corresponding entry in the first strategy table matched this time will be increased by a preset positive adjustment step size to strengthen the selection tendency of the strategy under similar working conditions in the future. If the convergence index indicates non-convergence or the volatility index is higher than the set threshold, the weight parameter of the corresponding entry in the first strategy table matched this time will be reduced by a preset positive adjustment step size, and the weight parameter will not be lower than the preset minimum value, thereby weakening the selection tendency of the strategy, so as to achieve continuous optimization of the control strategy through online experience accumulation.

[0012] Furthermore, it also includes step S9: If, during the execution of step S6 or the monitoring process of step S7, the real-time data of any control point exceeds its preset safety boundary, the current control process will be terminated immediately, and the preset second strategy table will be invoked to perform a safety protection action. The safety boundary is a threshold set based on the physical limits of the equipment or the red line of process safety. Its range is larger than the stable operating range, thus forming a two-level response mechanism in which the stable operating range is used for optimization and adjustment, and the safety boundary is used for mandatory protection.

[0013] An automated control system, comprising: The data acquisition unit is used to acquire data from the production process at a fixed sampling period. Real-time data from each control point provides real-time status input for the system. The feature calculation unit, connected to the data acquisition unit, is used to calculate the feature value based on the latest data at each control point. Real-time data from consecutive sampling periods are used to calculate the state feature vector, and the operating state is quantified using multi-dimensional features. The state judgment unit, connected to the feature calculation unit, is used to compare the state feature vector with a preset stable operating range, filter out the points to be adjusted, and realize the early location of abnormal states. The strategy matching unit, connected to the state judgment unit, is used to match the basic control action and the basic intensity coefficient from the first strategy table according to the over-limit characteristics of the point to be adjusted, and to complete the generation of the preliminary control strategy. The coupling coordination unit, connected to the strategy matching unit, is used to determine the associated points based on the production process topology, call the influence relationship model for prediction, and generate the final control command by iteratively adjusting the basic strength coefficient, and solve the feasible adjustment range under the premise of considering coupling constraints. An instruction execution unit, connected to the coupling coordination unit, is used to execute final control instructions; The effect evaluation unit, connected to the instruction execution unit, is used to monitor the change trajectory of the state feature vector after the instruction is executed, and to calculate the convergence index and volatility index to make a quantitative evaluation of the control effect. The learning and updating unit, connected to the effect evaluation unit and the strategy matching unit, is used to adjust the weight parameters in the first strategy table based on empirical data, forming a closed loop of online self-optimization of the control strategy.

[0014] The technical effects and advantages of this invention are as follows: This method constructs a multi-dimensional state feature vector containing mean, standard deviation, and slope to quantify the state of control points from three dimensions: average operating level, degree of fluctuation, and trend of change. Based on this, the system makes decisions using an iterative prediction-downsizing mechanism based on an influence relationship model. For the initial adjustment intensity obtained from a lookup table, the system uses pre-defined topological relationships to locate associated units and calls a pre-trained lightweight model to predict the impact of the adjustment action on each state component of the associated units. If the prediction results show that any associated unit will deviate from its stable operating range, the system determines the step size for downsizing intensity coefficients and re-verifies until the predicted states of all associated units return to their respective allowable ranges. This process replaces complex online optimization with bounded iterative calculations, providing a foundation for predictable response times within millisecond-level control cycles.

[0015] By iteratively adjusting the intensity coefficient, the system can generate a final control command that all directly related units can safely withstand. This mechanism allows for adaptive correction of adjustments made to local adjustment points beforehand, under the constraints of system coupling relationships, helping to avoid negative chain reactions that may be caused by traditional independent adjustment actions. Correspondingly, in the effect evaluation phase, this method not only focuses on the convergence degree of deviation of the adjustment point itself, but also incorporates the maximum offset of all related points in the transient process into the volatility index for quantitative assessment. This evaluation method, which takes the overall stability of the system as its observation perspective, integrates single-point correction and global stability control within the control logic.

[0016] This method constructs a closed-loop learning mechanism based on control effect feedback. After each control command is executed, the system records the convergence status of the point to be adjusted and the fluctuation level of related points, and associates this empirical data with the strategy entry invoked this time. Based on whether convergence is achieved and whether the fluctuation is below a set threshold, the selection weight of the corresponding strategy entry is dynamically adjusted. If the effect meets the conditions, the weight is increased to strengthen the selection tendency of the strategy; otherwise, the weight is decreased. Through continuous accumulation and online updates of operational data, the configuration of the strategy table can adapt to changes in the actual operating conditions of the production line and the slow deviation of equipment performance, enabling continuous improvement of control quality through experience accumulation.

[0017] This method establishes an independent safety monitoring unit based on an absolute safety boundary outside the intelligent control loop. This safety boundary is set based on the physical limits of the equipment or the process safety red line, and its range is wider than the stable operating range used to trigger optimization adjustments, forming a hierarchical protection mechanism. The safety monitoring unit operates through independent hardware or as a highest-priority task, logically or physically isolated from the main control flow. When the instantaneous value of real-time data at any control point touches the safety boundary, the system immediately interrupts the main control flow and directly executes the preset mandatory safety protection action. This hierarchical safety architecture ensures that the time and path of the protection response are not affected by the complex decision-making process of the main control logic, providing a safety net for system security under abnormal operating conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0019] Figure 2 This is a system composition framework diagram of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1 As attached Figures 1 to 2 The automated control system and method shown can be implemented by deploying corresponding control software on the main industrial computer of the production line. This computer is equipped with a data acquisition interface, a control signal output interface, a memory, and a processor. The core execution logic of the software program completely corresponds to steps S1 to S8 and optional S9 as described in claim 1.

[0022] For ease of explanation, it is assumed that the production line includes There are several key control points, each connected to the main control system via sensors and actuators. The following details each step.

[0023] Step S1: Real-time data acquisition With a fixed sampling period Obtain the definition in the production process Real-time data for each control point.

[0024] The value is set based on the production line's maximum cycle time requirement and the controller's processing capacity to ensure timely capture of state changes. It is usually in the millisecond range, for example, a value between 50ms and 200ms.

[0025] The data packets acquired in each sampling period are timestamped and stored in a circular buffer, providing a historical data sequence for subsequent analysis. The circular buffer design allows the system to efficiently access the most recent historical data for each control point while avoiding unlimited memory growth.

[0026] Step S2: Calculate the state feature vector For each control point ( ), based on its latest Calculate the state feature vector from real-time data of consecutive sampling periods. ,in It is a positive integer greater than 2.

[0027] The value of needs to balance real-time performance and noise immunity: If the size is too small, it is easily affected by transient noise. If the value is too large, the response to changes in state will be sluggish; a value between 5 and 20 is usually chosen.

[0028] Extract control points from the circular buffer The latest Data points constitute a sequence. The increase in the index represents a shift in time forward. for The oldest data from one period ago, This is the latest sampled value.

[0029] The specific calculation steps are as follows: First characteristic component (Arithmetic mean), reflecting the average operating level of the control points in the recent period: This component can smooth out transient noise and obtain a stable operating level estimate.

[0030] Second characteristic component (Standard deviation) quantifies the volatility or dispersion of the operating state of the control point: Those skilled in the art can choose the population standard deviation or sample standard deviation formula based on the characteristics of the data, both of which can effectively reflect the degree of dispersion. The larger the value, the more unstable the process, which may indicate equipment malfunction or inconsistent materials.

[0031] Third characteristic component (Slope of the rate of change) characterizes the recent trend of the control point's operating status. It is represented by the sampling point number. As the independent variable, with data sequence Using the least squares method as the dependent variable, a linear fit was performed to obtain the slope: in , .

[0032] This slope reflects the linear trend of the control point within the observation window; positive values ​​indicate an increase, negative values ​​indicate a decrease, and larger absolute values ​​indicate more drastic changes. This component can detect slow parameter drift in advance, enabling early warning.

[0033] Based on the above calculations, each control point The state is quantized into a three-dimensional vector This provides a foundation for subsequent anomaly identification and decision-making.

[0034] Step S3: Filter the points to be adjusted Each control point The preset stable operating range contains three sub-ranges, which correspond to... The three components: The corresponding first numerical range , The corresponding second numerical range , The corresponding third numerical interval .

[0035] These intervals are determined based on process requirements, equipment capabilities, and historical statistics, representing the permissible range for normal operation of control points. The setting of interval boundaries must balance production efficiency and quality requirements: too wide an interval may lead to quality problems, while too narrow an interval will trigger frequent adjustments.

[0036] The system will use the state feature vector of each control point Compare with the corresponding stable operating range. If , , If any component exceeds its preset range, the control point is identified as a "point to be adjusted".

[0037] Record two key pieces of information simultaneously: Deviation type: The category of the component that exceeds the limit, for example, A type indicates that the mean exceeds the limit, S type indicates that the standard deviation exceeds the limit, and T type indicates that the slope exceeds the limit; Exceedance range: The absolute value of the difference between the actual value and the nearest interval boundary value.

[0038] If there are multiple points to be adjusted in the same cycle, the system processes them in a preset order, completing the closed-loop control of one point before processing the next, in order to avoid instruction conflicts and resource contention.

[0039] Step S4: Match the basic control strategy The system has a pre-set "first strategy table", which is a two-dimensional lookup table. The row index is the feature component category, such as A, S, and T, and the column index is the exceedance level.

[0040] The levels are classified based on the percentage of the exceedance relative to the width of the component's stable operating range. For example, Level 1 is the exceedance level. Interval width Level 2 is to Level 3 is This hierarchical classification method correlates policy matching with the severity of deviation, enabling tiered responses.

[0041] Each cell stores three data items: Basic control action identifier: a predefined string or enumeration code that points to a specific control logic, such as "depressurize", "increase speed", "reduce feed", etc. Foundation strength coefficient : Dimensionless scaling factor, representing the default intensity of the basic control action, with an initial value of 1.0; different initial factors can be configured for different levels, such as 0.8 for level 1 (mild adjustment) and 1.5 for level 3 (stronger adjustment). Select weights: Integer parameter, initialized to... The update mechanism for strategy optimization is described in step S8.

[0042] Based on the deviation type and exceedance level obtained in step S3, the corresponding basic control action and basic strength coefficient are obtained by looking up the table. This serves as an initial control strategy.

[0043] Step S5: Adjustment of Synergy Strength Sub-step 1: Determine the associated points Based on the pre-defined production process topology, determine the current point to be adjusted. There is a direct coupling relationship One related point.

[0044] Topology relationships, defined in the form of adjacency lists or matrices, define the dependencies of material flow, information flow, or energy flow between control points, such as feeding relationships, downstream connection relationships, or signal interaction relationships. This topology is configured offline and reflects the physical layout and process connections of the production line.

[0045] Sub-step 2: Iteration intensity adjustment Predict the impact of adjustment actions on associated points using a pre-trained influence relationship model.

[0046] The model can be a multiple linear regression or a shallow neural network. Its training data comes from historical production line operation records, including data under normal operating conditions and data from controlled disturbances. After training, the model parameters are stored and can be accessed online. The system can periodically update the model with new empirical data to maintain prediction accuracy.

[0047] For each associated point ( Each state component of ), i.e. , , Each of these is a separate sub-model.

[0048] The input features of the sub-model include: Points to be adjusted Current state ; Vectorized encoding of the control action identifiers of the points to be adjusted, such as one-hot encoding, converts discrete action identifiers into numerical vectors, which facilitates model processing. Strength coefficient ; Related points Current state .

[0049] The output of the sub-model is the predicted change in the state component corresponding to the associated point, denoted as . , , .

[0050] Let the current strength coefficient be... The initial value is obtained in step S4. Preset down adjustment step size , which is a fixed positive value, such as 0.1, and the lower limit of strength. , which is a fixed positive value, such as 0.2. The choice needs to balance search efficiency and accuracy. The adjustment must be physically meaningful; an excessively small strength coefficient may render the adjustment ineffective.

[0051] The iterative adjustment process is as follows: 1. For each associated point Call the corresponding sub-model and input. The predicted changes of each component are obtained. , , .

[0052] 2. Calculate the associated points Predicted state: 3. Check each associated point Are all predicted states within their preset stable operating range? 4. If all associated points satisfy the above conditions, the iteration terminates, and the final strength coefficient is obtained. .

[0053] 5. If any predicted component of any associated point exceeds the limit, then let .like Then let And terminate; otherwise return to step 1.

[0054] This iterative cycle is essentially a process of finding the maximum feasible adjustment strength under the system's coupling constraints. It ensures that the final control command is executed within the tolerance range of all related points, achieving a balance between single-point adjustment and global stability.

[0055] The final control command is derived from the basic control action and final intensity coefficient obtained in step S4. Together they constitute.

[0056] Step S6: Execute the final control command The instruction execution unit maps basic control action identifiers to specific execution functions and scales the control quantity according to the final intensity coefficient.

[0057] Specifically, if the reference adjustment amount of the basic control action is That is, the adjustment range when the strength coefficient is 1.0, then the actual adjustment amount is... Commands are sent to the actuators through the control system output interface, such as analog output modules or fieldbuses.

[0058] After execution, wait for a fixed instruction stabilization time. For example, 50ms to 200ms can be set according to the actuator response characteristics to ensure that the actuator's action has been initially completed and fed back to the sensor readings, providing effective data for subsequent monitoring.

[0059] This waiting time needs to be calibrated based on the physical response time of the actuator. If it is too short, the monitoring data may be inaccurate; if it is too long, it will delay subsequent decisions.

[0060] Step S7: Effectiveness Evaluation Set the number of observation periods , which is a positive integer greater than 1, such as 5 to 10. The value of needs to be sufficient to observe the convergence trend of the control effect, but not so long as to delay subsequent decisions due to excessive monitoring time.

[0061] After the instruction is executed and a stabilization period has elapsed, continuous monitoring is performed. One sampling period, collecting the points to be adjusted. and all of them Real-time data of each associated point. In each cycle, the state feature vector of each point is recalculated according to step S2 to form a change trajectory.

[0062] Two indicators are calculated based on the change trajectory: Convergence index: In After each cycle, obtain the point to be adjusted. State feature vector .

[0063] Find the out-of-limit component in step S3 that caused the point to be marked, and let its corresponding stable operating interval be [a, b]. Then the midpoint of the interval is... Calculate the value of this component and... absolute difference .

[0064] Preset tolerance This value can be set based on the process tolerance or the standard deviation of historical normal fluctuations, and is usually taken as the width of the stable range. to .like The convergence index Assigned value , indicates convergence, otherwise assign a value , indicating that the convergence has not occurred.

[0065] This indicator directly reflects whether the control action pulls the controlled variable back to the vicinity of the desired range center.

[0066] Volatility index: for each correlation point Calculate its in Each component within a cycle is relative to its initial value before instruction execution. Maximum offset: in These are the status values ​​of the associated points recorded before the negotiation begins in step S5.

[0067] The maximum value among the three is taken as the volatility of the correlation point. .

[0068] The maximum volatility of all related points is taken as the volatility index for this control: The larger the value, the more severe the overall system fluctuation caused by this adjustment. The design of this indicator reflects the core objective of collaborative control: to minimize interference with other parts of the system when solving problems.

[0069] Step S8: Experience Learning and Strategy Optimization The system generates an experience data record and stores the following information in association: The point to be adjusted in step S3 Original state feature vector ; The final control command generated in step S5 includes the basic control action identifier and the final intensity coefficient. ; The convergence index calculated in step S7 With volatility indicators .

[0070] This record is stored in a cyclical experience database with a fixed capacity. When the database is full, the oldest record is overwritten, ensuring that learning is based on recent operational experience and adapts to the time-varying characteristics of the production line.

[0071] Adjust the weight parameters of the corresponding entries in the first strategy table based on the control effect.

[0072] System preset fluctuation threshold For example, taking the width of the stable operating range of the associated points. This threshold serves as a benchmark for assessing whether the disturbance is excessive. It can be configured based on the process's sensitivity to disturbances.

[0073] The adjustment rules are as follows: like and This indicates that the strategy is effective and the disturbance is controllable. Therefore, in step S4, the selection weight of the corresponding entry in the first strategy table matched this time is increased by a preset positive adjustment step size. ,For example ; like or This indicates that the strategy is ineffective or the disturbance is too large, so the selection weight of this item is reduced by a preset positive number and the adjustment step size is increased. ,For example And the weight is not lower than the preset minimum value, for example .

[0074] The weight parameter is used for optimal selection during subsequent strategy matching: when there are multiple candidate strategies in the same cell, the system prioritizes the strategy with the highest weight value; if the cell only stores a single strategy, the weight parameter can be used for long-term evaluation. The system can periodically analyze the weight distribution, solidify strategies with consistently high weights, mark strategies with consistently low weights as inefficient strategies, and replace them with new candidate strategies after review by operators during the maintenance cycle.

[0075] Through this mechanism, the system achieves self-optimization based on historical control results, and the control quality continues to improve as operational experience accumulates.

[0076] Step S9: Safety Protection Each control point Also, a safety boundary is preset. Its range is wider than the stable operating range and is set based on the physical limits of the equipment or the red line of process safety.

[0077] The safety boundary and the stable operating range form a two-level protection mechanism: the stable range is used to trigger intelligent optimization and adjustment, and the safety boundary is used to trigger mandatory protection. The two do not interfere with each other.

[0078] The security monitoring unit operates as independent hardware or as a highest-priority task, continuously monitoring raw real-time data. This unit is physically or logically isolated from the main control flow to ensure its reliability.

[0079] If the instantaneous value of the real-time data at any control point exceeds the safety boundary, immediately perform the following actions: 1. Send the highest priority interrupt signal to the main control flow to suspend all operations from steps S1 to S8 and clear the pending instruction queue; 2. Call the preset second strategy table to send mandatory safety protection commands to the actuator, such as emergency stop, power source cut-off, and audible and visual alarm activation. These commands are executed with the shortest possible delay without going through complex decision-making logic. 3. Record security event logs, including trigger time, trigger point, and out-of-limit data, to facilitate post-event analysis and maintenance.

[0080] After the safety status is cleared, a manual reset is required to restart the main control process.

[0081] This design ensures that the safety response is unaffected by the state of the main control process, executes protective actions with the shortest possible delay, and provides reliable fallback protection for the system.

[0082] System initialization During system initialization, the following configuration file must be loaded from non-volatile memory: Control point definition, including 1. Mapping of sensors and actuators at each control point; The production process topology is stored in the form of an adjacency list or matrix; Stable operating range boundaries and safety boundaries for each control point; Initial content of the first strategy table; Influence relationship model parameters; Preset parameters, including , , , , , , , , wait.

[0083] After each unit completes its startup and self-test, the system enters a periodic operation state.

[0084] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automated control method, characterized in that, Includes the following steps: S1. Obtain the data defined in the production process at a fixed sampling period. Real-time data of each control point It is a positive integer greater than 1; S2. For each control point, based on its latest... A state feature vector is calculated from real-time data of several consecutive sampling periods. It is a positive integer greater than 2; S3. Compare the state feature vector of each control point with the preset stable operating range of that control point, and select control points whose state feature vector has at least one feature component that exceeds its corresponding stable operating range as adjustment points. S4. For each point to be adjusted, based on the category and magnitude of its out-of-limit characteristic components, a basic control action and a basic intensity coefficient are matched from the preset first strategy table. S5. Based on the preset production process topology, determine the points directly related to the current point to be adjusted. One related point, It is an integer greater than or equal to 0; Based on this The current state feature vector of each associated point is used to predict the predicted change in the state feature vector of each associated point after the basic control action is executed under the basic intensity coefficient using a pre-trained influence relationship model, and the predicted state feature vector of each associated point is calculated based on the predicted change. If the predicted state feature vector of any associated point exceeds its own stable operating range, the basic intensity coefficient is reduced by a preset step size, and prediction and judgment are performed again until the predicted state feature vectors of all associated points do not exceed their own stable operating range, or the intensity coefficient reaches the preset lower limit value. The adjusted intensity coefficient is used as the final intensity coefficient, which together with the basic control action constitutes the final control command; S6. Execute the final control command; S7. Continuous execution of instructions Within each sampling period, For positive integers greater than 1, collect real-time data of the point to be adjusted and its associated points, and calculate the change trajectory of its state feature vector; based on the change trajectory, calculate the convergence index and volatility index of this control. S8. Establish an empirical data set, which is at least associated with the state feature vector of the point to be adjusted in S3, the final control command generated in S5, and the convergence index and volatility index calculated in S7; based on the convergence index and volatility index, adjust the weight parameters of the relevant entries in the first strategy table.

2. The automated control method according to claim 1, characterized in that, The calculation of the state feature vector in step S2 includes the following sub-steps: For each control point, calculate its nearest neighbor. The arithmetic mean of the real-time data from each sampling period is used as the first feature component. Calculate the The standard deviation of each data point relative to the arithmetic mean is used as the second feature component; Based on sampling point number As the independent variable, with the above The data points are used as the dependent variable, and the slope is calculated using linear regression as the third feature component. The first feature component, the second feature component, and the third feature component constitute the state feature vector.

3. The automated control method according to claim 2, characterized in that, The preset stable operating range mentioned in step S3 includes: a first numerical range set for the first feature component, a second numerical range set for the second feature component, and a third numerical range set for the third feature component.

4. The automated control method according to claim 1, characterized in that, In step S4, the first strategy table is a two-dimensional lookup table, with its row index being the feature component category and its column index being the exceedance magnitude level. The exceedance level is classified based on the percentage range of the characteristic component exceeding its stable operating range boundary value.

5. The automated control method according to claim 1, characterized in that, The influence relationship model mentioned in step S5 is a multiple linear regression model or a shallow neural network model; For each state feature component of each associated point, an independent sub-model is established. The input features of the sub-model include: the current state feature vector of the point to be adjusted, the vectorized encoding of the control action identifier of the point to be adjusted, the intensity coefficient of the point to be adjusted, and the current state feature vector of the associated point; The output of the sub-model is the predicted change of the state feature components corresponding to the associated points; The influence relationship model is pre-trained using historical operating data, which includes records of control actions and corresponding state changes of the production line under normal operating conditions and controlled disturbances. Wherein, the predicted state feature vector of the associated point is equal to the sum of the current state feature vector of the associated point and the predicted change.

6. The automated control method according to claim 1, characterized in that, The preset step size mentioned in step S5 is a fixed positive value, and the preset lower limit value is a fixed value greater than 0.

7. The automated control method according to claim 1, characterized in that, Step S7, which involves calculating the convergence and volatility indices, includes the following sub-steps: For the point to be adjusted, compare it with the one described in the above. At the end of each cycle, the difference between the component value corresponding to the out-of-limit feature component identified in S3 and the median value of the stable operating interval corresponding to the out-of-limit feature component in the state feature vector; if the difference is less than the preset tolerance, the convergence index is assigned a converged state value, otherwise it is assigned a non-converged state value. For each associated point, its statistics are presented in the... The maximum offset of each component of the state feature vector within a period relative to its initial value is taken as the volatility index, and the maximum value among all components of all associated points is used.

8. The automated control method according to claim 1, characterized in that, The specific rules for adjusting the weight parameters in step S8 are as follows: If the convergence index indicates convergence and the volatility index is below the set threshold, then the weight parameter of the corresponding entry in the first strategy table matched this time will be increased by a preset positive adjustment step. If the convergence index indicates non-convergence or the volatility index is higher than the set threshold, the weight parameter of the corresponding entry in the first strategy table matched this time will be reduced by a preset positive adjustment step size, and the weight parameter will not be lower than the preset minimum value.

9. The automated control method according to claim 1, characterized in that, It also includes step S9: If, during the execution of step S6 or the monitoring process of step S7, the real-time data of any control point exceeds its preset safety boundary, the current control process will be terminated immediately, and the preset second strategy table will be invoked to perform a safety protection action. The safety boundary is a threshold set based on the physical limits of the equipment or the red line of process safety, and its range is greater than the stable operating range.

10. An automated control system, characterized in that, include: The data acquisition unit is used to acquire data from the production process at a fixed sampling period. Real-time data of each control point; The feature calculation unit, connected to the data acquisition unit, is used to calculate the feature value based on the latest data at each control point. Calculate the state feature vector from real-time data of consecutive sampling periods; A state judgment unit, connected to the feature calculation unit, is used to compare the state feature vector with a preset stable operating range and filter out the points to be adjusted. The strategy matching unit, connected to the state judgment unit, is used to match the basic control action and the basic intensity coefficient from the first strategy table based on the over-limit characteristics of the point to be adjusted. The coupling coordination unit, connected to the strategy matching unit, is used to determine the associated points based on the production process topology, call the influence relationship model for prediction, and generate the final control command by iteratively adjusting the basic strength coefficient. An instruction execution unit, connected to the coupling coordination unit, is used to execute final control instructions; The effect evaluation unit, connected to the instruction execution unit, is used to monitor the change trajectory of the state feature vector after instruction execution and to calculate the convergence index and volatility index. The learning update unit, connected to the effect evaluation unit and the strategy matching unit, is used to adjust the weight parameters in the first strategy table based on empirical data.

Images