An intelligent control system and method based on multi-modal dynamic boundary

Abstract

The application relates to the field of automatic control, in particular to an intelligent control system and method based on a multi-modal dynamic boundary, which comprises the following steps: taking a physical safety boundary as a limit constraint under a current working condition, obtaining a process constraint boundary, and then performing dynamic contraction; performing real-time operation on a proportional term, real-time operation on an integral term, and real-time operation on a differential term one by one, and synthesizing a desired control output; making a decision according to a comparison result, and obtaining a final control quantity; and based on the final control quantity, performing real-time driving on an executing mechanism. The application adopts an operation logic of hierarchical proportion, anti-integral saturation dynamic integral, and rate matching dynamic differential, compared with a fixed parameter PID, can effectively inhibit overshoot, oscillation and integral saturation, is more accurate in control response, synchronously optimizes system steady state and dynamic performance, and executes a hierarchical control strategy, maximizes system operation efficiency within a safety boundary, gets rid of scene and equipment limitations, and significantly improves cross-platform universality.

Description

Technical Field

[0001] This invention relates to the field of automatic control, and more specifically to an intelligent control system and method based on multimodal dynamic boundaries. Background Technology

[0002] Proportional-integral-derivative (PID) controllers are widely used in many fields such as industrial process control, motion control, and aerospace due to their simple structure, ease of implementation, and strong robustness.

[0003] However, traditional PID controllers have inherent limitations: they calculate and output based solely on the current error, resulting in short-sighted control behavior. In the event of a large disturbance or a sudden change in the setpoint, the PID controller's drastic response can easily cause the system state to exceed safety boundaries, leading to malfunctions or even danger. Existing boundary handling solutions, such as limiters and barrier functions, either crudely truncate the control chain at the end or rely on specific mathematical models, lacking versatility and failing to simultaneously ensure safety while maintaining dynamic performance and cross-platform adaptability.

[0004] Therefore, existing technologies lack a universal control method that can incorporate boundary constraints as an integral part of the control law. The industry urgently needs a general control framework that can understand the connotation of boundaries in different scenarios and make intelligent decisions in order to achieve the control objective of ensuring safety while achieving optimal dynamic performance within the boundaries. Summary of the Invention

[0005] This invention addresses the technical problems existing in the prior art by providing an intelligent control system and method based on multimodal dynamic boundaries.

[0006] The technical solution of the present invention to solve the above technical problems is as follows: an intelligent control system and method based on multimodal dynamic boundary, comprising the following steps: S1, obtaining dynamic trends through numerical fitting based on real-time acquired operating parameters, constructing a real-time mathematical model, constructing a physical safety boundary according to the operating parameters, using the physical safety boundary as the limit constraint under the current working condition, and then dynamically shrinking the process constraint boundary after obtaining it; S2. Based on PID, after calculating the control deviation of the real-time control setpoint, according to the current period control deviation value obtained from the deviation calculation, perform real-time calculation of the proportional term, real-time calculation of the integral term, and real-time calculation of the derivative term one by one, and synthesize the desired control output. S3. Based on the desired control output, the corresponding prediction period is matched according to the response speed of the controlled object, and the future single-cycle system state is calculated recursively cycle by cycle using a real-time mathematical model as the calculation framework. Then, the future multi-cycle state is continuously deduced using the single-cycle prediction value as the initial value. The predicted system state is obtained by comparing it with the physical safety boundary and process constraint boundary. Decisions are made according to the comparison results to obtain the final control quantity and obtain the multi-cycle continuous prediction state parameters. S4. The obtained multi-cycle continuous predicted state parameters are compared with the retrieved physical safety boundary and process constraint boundary. Decisions are made according to the comparison results to obtain the final control quantity. Real-time drive of the actuator is completed based on the final control quantity.

[0007] In a preferred embodiment, S1 collects operating parameters on a control cycle based on the sensors mounted on the controlled object. It should be noted that the operating parameters include speed, distance, temperature, vibration displacement, pressure, water level, rotational speed, trajectory, heading angle, and thermal stress. All operating parameters are original collected values ​​to obtain the original dataset. Based on the operating parameters in the original dataset, a dynamic mathematical model is constructed through real-time numerical fitting. Specifically, the changing trends of all operating parameters in the current cycle are extracted, and then the changing trends of the operating parameters are matched and fitted with the motion of the controlled object to generate a real-time mathematical model that is only suitable for the current cycle. Then, the operating parameters are extracted and safety boundaries are constructed according to categories, including vehicle speed safety boundary, temperature safety boundary, vibration safety boundary, and rotational speed safety boundary. Multiple real-time operating parameters are superimposed on the safety boundaries of each category for correction, and the physical safety boundaries are obtained.

[0008] In a preferred embodiment, step S1 extracts the real-time reaction acquisition value of the industrial equipment under the current operating conditions, arithmetically adds the real-time reactor temperature acquisition value with the temperature fluctuation value corresponding to the real-time reaction rate to obtain the current upper temperature limit constraint, which is used as the temperature constraint boundary. The product of the real-time reactor pressure acquisition value and the pressure coefficient corresponding to the real-time medium concentration is used to obtain the current upper pressure limit constraint, which is used as the pressure constraint boundary. The real-time boiler drum water level acquisition value is superimposed with the water level compensation value corresponding to the real-time steam flow rate to obtain the current water level limit constraint, which is used as the water level constraint boundary. The integrated water level constraint boundary, pressure constraint boundary, and temperature constraint boundary form the process constraint boundary.

[0009] In a preferred embodiment, S2 calculates the difference between the pre-input real-time control setpoint and the corresponding parameter value of the operating parameter obtained in S1 on a cycle-by-cycle basis to obtain the current cycle control deviation value, wherein a positive deviation value indicates that the actual value is lower than the setpoint, and a negative deviation value indicates that the actual value is higher than the setpoint. Next, the absolute value of the current cycle control deviation is classified into magnitude categories, and a real-time proportional coefficient is matched according to the magnitude categories. The current cycle control deviation value is multiplied by the matched real-time proportional coefficient to obtain the proportional calculation result. The control deviation value of the current cycle is continuously accumulated with the control deviation values ​​of all previous control cycles in chronological order to obtain the total accumulated control deviation value. The magnitude of the total accumulated control deviation value is then determined. The larger the total accumulated value, the smaller the matching real-time integral coefficient, to prevent integral saturation and complete the dynamic integral coefficient matching. The integral calculation result is obtained by multiplying the total accumulated control deviation value by the matched real-time integral coefficient. The change in control deviation is obtained by subtracting the control deviation value of the previous cycle from the current cycle control deviation value. The rate of change of the deviation is then determined to complete the dynamic differential coefficient matching. The differential operation result is obtained by multiplying the change in control deviation by the matched real-time differential coefficient. The results of proportional calculation, integral calculation, and differential calculation are added together to output the desired control output without boundary constraints.

[0010] In a preferred embodiment, S3 retrieves the real-time mathematical model, operating parameters, and desired control output. Based on the desired control output, the prediction period is determined according to the response speed of the controlled object. In this application, objects with fast response are set to 1 control period, and objects with slow response are set to 3 control periods, forming a set of prediction basic parameters. Using the current operating parameters as the initial state, the desired control output is substituted into the real-time mathematical model, and the predicted state parameters for the next control cycle are calculated recursively to obtain the single-cycle predicted state parameters. Using the single-cycle predicted state parameters as the new initial state, the desired control output is substituted into the real-time mathematical model, and the predicted state parameters for the second and third control cycles are recursively calculated to obtain multi-cycle continuous predicted state parameters.

[0011] In a preferred embodiment, step S4 synchronously retrieves the physical safety boundary and process constraint boundary, and performs a one-to-one numerical comparison between each parameter in the predicted system state—velocity, distance, temperature, vibration displacement, pressure, water level, rotational speed, trajectory, heading angle, and thermal stress—and the upper and lower limits of the corresponding physical safety boundary and process constraint boundary, including the following specific determinations: A zone is defined as a safe zone if all predicted parameters in the multi-period continuous prediction state parameters are greater than or equal to the lower limit of the physical safety boundary and less than or equal to the upper limit of the physical safety boundary. If all predicted parameters in the multi-cycle continuous prediction state parameters are greater than or equal to the physical safety boundary and less than or equal to the physical safety boundary, but at least one predicted parameter exceeds the upper limit and / or lower limit of the process constraint boundary, it is determined to be a warning zone. If any one of the predicted parameters is less than the lower limit of the physical safety boundary and / or the process constraint boundary, or greater than the upper limit of the physical safety boundary and / or the process constraint boundary, it is determined to be a hazardous area. The final output shows the region to which the system status belongs.

[0012] In a preferred embodiment, S4 retrieves the original value of the desired control output as the final control quantity based on the determination result of the safe zone, and synchronously transmits the final control quantity to the actuator. Obtain the upper limit, lower limit, and center reference value of the physical safety boundary for the current period. The center reference value is the arithmetic mean of the upper and lower limits. Subtract the predicted state parameter value from the center reference value of the physical safety boundary to obtain the single parameter deviation. A positive deviation indicates that the parameter deviates towards the upper limit of the corridor, and a negative deviation indicates that the parameter deviates towards the lower limit of the corridor. Sum the absolute values ​​of all parameter deviations that exceed the physical safety boundary to obtain the total system deviation. Substitute the total system deviation into the linear mapping formula (total deviation divided by deviation) to obtain the deviation ratio, which is used as the attenuation coefficient. Multiply the desired control output by the real-time attenuation coefficient to obtain the final control quantity after smooth attenuation. Note that there are positive and negative deviation attenuation. If the predicted parameter deviates towards the upper limit of the performance optimization corridor, multiply the desired control output by the real-time attenuation coefficient to reduce the positive amplitude of the control output and prevent the parameter from deviating further upward. If the predicted parameter deviates towards the lower limit of the performance optimization corridor, multiply the desired control output by the real-time attenuation coefficient to reduce the negative amplitude of the control output and prevent the parameter from deviating further downward.

[0013] In a preferred embodiment, based on the upper and lower limits of the current cycle process constraint boundary, a predicted parameter greater than the upper limit of the boundary is considered a positive boundary risk. When a positive boundary risk exists, the upper limit of the boundary is subtracted from the predicted state parameter value to obtain a positive boundary deviation value. When a negative boundary risk exists, the lower limit of the boundary is subtracted from the predicted state parameter value to obtain a negative boundary deviation value. Based on the positive boundary deviation value, the desired control output is a positive drive command, such as acceleration, pressure increase, position increase, and temperature increase. Based on the negative boundary deviation value, the desired control output is a negative drive command, such as deceleration, pressure decrease, position decrease, and temperature decrease. After the above determination is completed, the reverse direction of the determination is taken as the only direction of the avoidance control quantity without modification. The positive and / or negative boundary deviation value is divided by the maximum allowable boundary deviation value to obtain the real-time forced correction coefficient. The product of the positive and / or negative boundary deviation value and the real-time forced correction coefficient is used as the final control quantity.

[0014] In a preferred embodiment, the actuator receives the final control quantity and converts the digitized control quantity into mechanical actions, voltage and current signals and / or fluid regulation signals that can be recognized by the controlled object, thereby generating real-time operating actions of the controlled object.

[0015] The present invention also provides an intelligent control system based on multimodal dynamic boundaries, comprising: Real-time model and boundary construction module; used to extract the changing trends of operating parameters based on the original dataset, construct a mathematical model that is updated in real time on a cycle-by-cycle basis, and construct physical safety boundaries and process constraint boundaries based on the operating parameters; Dynamic PID calculation module; used to calculate the control deviation between the real-time control setpoint and the operating parameters, complete the real-time calculation of the proportional term, integral term, and derivative term, and synthesize the desired control output; The multi-cycle state prediction module is used to match the prediction cycle according to the response speed of the controlled object, and recursively calculate the future multi-cycle prediction system state using a real-time mathematical model as a framework. Boundary comparison and decision module: It is used to compare the predicted system state with the physical safety boundary and process constraint boundary, determine the system state region and generate the final control quantity, and receive the final control quantity and convert it into a drive signal that the actuator can recognize, so as to complete the real-time drive of the controlled object.

[0016] The beneficial effects of this invention are: by adopting a hierarchical proportional, anti-integral saturation dynamic integral, and rate-matching dynamic derivative operation logic, compared with fixed-parameter PID, it can effectively suppress overshoot, oscillation and integral saturation, resulting in more precise control response, simultaneous optimization of system steady-state and dynamic performance, and execution of hierarchical control strategy to maximize system operating efficiency within the safety boundary. At the same time, it combines real-time operating conditions to dynamically correct the safety boundary, breaking free from scene and equipment limitations, and significantly improving cross-platform universality. Attached Figure Description

[0017] Figure 1 This is a flowchart of the present invention; Figure 2 This is a visualization diagram of the final control quantity decision based on physical safety boundaries in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] As attached Figure 1-2As shown, this embodiment provides: an intelligent control system and method based on multimodal dynamic boundaries, including the following steps: S1. Based on the real-time acquired operating parameters, the dynamic trend is obtained through numerical fitting, and a real-time mathematical model is constructed. The physical safety boundary is constructed according to the operating parameters. The physical safety boundary is used as the limit constraint under the current working condition. After obtaining the process constraint boundary, dynamic contraction is performed. S1 collects operating parameters based on the sensors mounted on the controlled object in each control cycle. It should be noted that the operating parameters include speed, distance, temperature, vibration displacement, pressure, water level, rotational speed, trajectory, heading angle, and thermal stress. All operating parameters are raw collected values, thus obtaining the raw dataset. Based on the operating parameters in the original dataset, a dynamic mathematical model is constructed through real-time numerical fitting. Specifically, the changing trends of all operating parameters in the current cycle are extracted, and then the changing trends of the operating parameters are matched and fitted with the motion of the controlled object to generate a real-time mathematical model that is only suitable for the current cycle. In some other embodiments, this application provides a process for constructing a real-time mathematical model: The original real-time operating parameters of the current period and the previous three consecutive periods are arranged in chronological order to form a continuous parameter time series. The original value of each parameter in the corresponding period is retained without any filtering or correction. For adjacent period values ​​of the same parameter in the parameter time series, perform point-by-point difference calculation, subtract the parameter value of the previous period from the parameter value of the next period to obtain the single period change, divide the single period change by the period duration to obtain the parameter change rate, and summarize the change rates of all parameters to form a set of overall system change trends. The overall trend of the system is matched with the inherent motion / operation mechanism of the controlled object. Motion objects are matched with the kinematic laws of displacement, velocity and acceleration, and industrial process objects are matched with the thermodynamic / fluid dynamic laws of temperature, pressure and flow. The mechanism expression form that is completely adapted to the current trend is locked. Using the overall trend set of the system as the fitting input and the matched mechanism expression as the fitting framework, all parameter values ​​in the trend set are substituted into the mechanism framework one by one. The fitting error corresponding to each parameter is calculated. By continuously adjusting the coefficient values ​​in the mechanism framework, the sum of squared fitting errors of all parameters is minimized. At this point, the coefficient values ​​are combined with the mechanism framework to form a real-time mathematical model that is only suitable for the current cycle. After the model is built, each time a new control cycle begins, all the steps of parameter time series processing, trend extraction, operation mechanism matching, and least squares fitting are repeated. The newly generated model is used to cover the model of the previous cycle, so that the model is updated in real time with the collected data. Then, the operating parameters are extracted and safety boundaries are constructed according to categories, including vehicle speed safety boundary, temperature safety boundary, vibration safety boundary, and rotational speed safety boundary. Multiple real-time operating parameters are superimposed on the safety boundaries of each category for correction, and the physical safety boundaries are obtained.

[0020] It should be noted that the following steps are performed as follows: multiplying the real-time vehicle speed by the real-time safe headway, adding the real-time minimum stationary safe distance, and then adding the correction distance corresponding to the real-time road friction coefficient, finally yields the current safe following distance constraint, i.e., the vehicle speed safety boundary; dividing the real-time system temperature by the preset maximum allowable temperature ratio yields the real-time temperature coefficient, and multiplying the maximum allowable heating rate by the real-time temperature coefficient yields the current allowable heating rate constraint, i.e., the temperature safety boundary; multiplying the real-time vibration displacement acquisition value by the real-time equipment operating load coefficient yields the current maximum allowable vibration displacement constraint of the rotor, i.e., the vibration safety boundary; and subtracting the speed decay value corresponding to the real-time thermal stress from the real-time rated speed of the equipment yields the current maximum allowable speed constraint, i.e., the speed safety boundary.

[0021] S1 extracts the real-time reaction data collected by the industrial equipment under the current operating conditions, and arithmetically adds the real-time reactor temperature data collected with the temperature fluctuation value corresponding to the real-time reaction rate to obtain the current upper temperature limit constraint, which is used as the temperature constraint boundary. The real-time reactor pressure data collected with the pressure coefficient corresponding to the real-time medium concentration is used as the current upper pressure limit constraint, which is used as the pressure constraint boundary. The real-time boiler drum water level data collected is superimposed with the water level compensation value corresponding to the real-time steam flow to obtain the current water level limit constraint, which is used as the water level constraint boundary. The integrated water level constraint boundary, pressure constraint boundary, and temperature constraint boundary form the process constraint boundary.

[0022] S2. Based on PID, after calculating the control deviation of the real-time control setpoint, according to the current period control deviation value obtained from the deviation calculation, perform real-time calculation of the proportional term, real-time calculation of the integral term, and real-time calculation of the derivative term one by one, and synthesize the desired control output. S2 calculates the difference between the pre-input real-time control setpoint and the corresponding parameter value of the operating parameters obtained by S1 on a cycle-by-cycle basis to obtain the control deviation value of the current cycle. A positive deviation value indicates that the actual value is lower than the setpoint, and a negative deviation value indicates that the actual value is higher than the setpoint. Next, the absolute value of the current cycle control deviation is classified into magnitude categories, and a real-time proportional coefficient is matched according to the magnitude categories. The current cycle control deviation value is multiplied by the matched real-time proportional coefficient to obtain the proportional calculation result. The control deviation value of the current cycle is continuously accumulated with the control deviation values ​​of all previous control cycles in chronological order to obtain the total accumulated control deviation value. The magnitude of the total accumulated control deviation value is then determined. The larger the total accumulated value, the smaller the matching real-time integral coefficient, to prevent integral saturation and complete the dynamic integral coefficient matching. The integral calculation result is obtained by multiplying the total accumulated control deviation value by the matched real-time integral coefficient. The change in control deviation is obtained by subtracting the control deviation value of the previous cycle from the current cycle control deviation value. The rate of change of the deviation is then determined to complete the dynamic differential coefficient matching. The differential operation result is obtained by multiplying the change in control deviation by the matched real-time differential coefficient. In some other specific implementations, taking the real-time calculation of the proportional term as an example, the real-time proportional coefficient is obtained based on the following steps: First, take the absolute value of the current cycle control deviation value, divide the current operating conditions into three levels of thresholds: small deviation, medium deviation, and large deviation, compare the absolute value of the deviation with the threshold to determine the interval, and obtain the real-time proportional coefficient by online matching according to the interval. Small deviation is matched with a small coefficient, medium deviation with a medium coefficient, and large deviation with a large coefficient. Then, the current cycle control deviation value is multiplied by the matched real-time proportional coefficient to obtain the proportional calculation result. This application provides a specific example for illustration, taking the following distance control of an adaptive cruise control system as an example. The current operating condition is high-speed stable driving; the current cycle following distance setting is 30 meters, and the actual following distance collected by the sensor in real time is 25 meters, resulting in a calculated control deviation of +5 meters for the current cycle; taking the absolute value of the control deviation, we obtain an absolute deviation value of 5 meters; based on the high-speed stable driving condition, we divide the deviation level thresholds in real time: small deviation ≤ 2 meters, medium deviation 2 meters < deviation ≤ 5 meters, and large deviation > 5 meters. We compare the absolute deviation value of 5 meters with the threshold and determine that it belongs to the medium deviation range; we match the real-time proportional coefficient online according to the range: small deviation matching coefficient 0.3, medium deviation matching coefficient 0.8, and large deviation matching coefficient 1.2, and the current matching yields a real-time proportional coefficient of 0.8; we perform a multiplication operation: multiply the current cycle control deviation value of 5 meters by the real-time proportional coefficient of 0.8, and this value is the proportional calculation result. Of course, the above solution can also use AI or machine learning methods, utilizing historical experience formed by circular convolution, to quickly determine and obtain the proportional coefficient.

[0023] The results of proportional calculation, integral calculation, and differential calculation are added together to output the desired control output without boundary constraints.

[0024] S3. Based on the desired control output, the corresponding prediction period is matched according to the response speed of the controlled object, and the future single-cycle system state is calculated recursively cycle by cycle using a real-time mathematical model as the calculation framework. Then, the future multi-cycle state is continuously deduced using the single-cycle prediction value as the initial value. The predicted system state is obtained by comparing it with the physical safety boundary and process constraint boundary. Decisions are made according to the comparison results to obtain the final control quantity and obtain the multi-cycle continuous prediction state parameters. S3 retrieves the real-time mathematical model, operating parameters, and desired control output. Based on the desired control output, the prediction period is determined according to the response speed of the controlled object. In this application, objects with fast response are set to 1 control period, and objects with slow response are set to 3 control periods, forming a set of basic prediction parameters. Using the current operating parameters as the initial state, the desired control output is substituted into the real-time mathematical model, and the predicted state parameters for the next control cycle are calculated recursively to obtain the single-cycle predicted state parameters. Using the single-cycle predicted state parameters as the new initial state, the desired control output is substituted into the real-time mathematical model, and the predicted state parameters for the second and third control cycles are recursively calculated to obtain multi-cycle continuous predicted state parameters.

[0025] S4. The obtained multi-cycle continuous predicted state parameters are compared with the retrieved physical safety boundary and process constraint boundary. Decisions are made according to the comparison results to obtain the final control quantity. Real-time drive of the actuator is completed based on the final control quantity.

[0026] S4 synchronously retrieves the physical safety boundary and process constraint boundary, and performs a one-to-one numerical comparison between each parameter in the predicted system state—velocity, distance, temperature, vibration displacement, pressure, water level, rotational speed, trajectory, heading angle, and thermal stress—and the upper and lower limits of the corresponding physical safety boundary and process constraint boundary, including the following specific judgments: A zone is defined as a safe zone if all predicted parameters in the multi-period continuous prediction state parameters are greater than or equal to the lower limit of the physical safety boundary and less than or equal to the upper limit of the physical safety boundary. If all predicted parameters in the multi-cycle continuous prediction state parameters are greater than or equal to the physical safety boundary and less than or equal to the physical safety boundary, but at least one predicted parameter exceeds the upper limit and / or lower limit of the process constraint boundary, it is determined to be a warning zone. If any one of the predicted parameters is less than the lower limit of the physical safety boundary and / or the process constraint boundary, or greater than the upper limit of the physical safety boundary and / or the process constraint boundary, it is determined to be a hazardous area. The final output shows the region to which the system status belongs.

[0027] Based on the determination result of the safe zone, S4 retrieves the original value of the expected control output as the final control quantity and synchronously transmits the final control quantity to the actuator. Obtain the upper limit, lower limit, and center reference value of the physical safety boundary for the current period. The center reference value is the arithmetic mean of the upper and lower limits. Subtract the predicted state parameter value from the center reference value of the physical safety boundary to obtain the single parameter deviation. A positive deviation indicates that the parameter deviates towards the upper limit of the corridor, and a negative deviation indicates that the parameter deviates towards the lower limit of the corridor. Sum the absolute values ​​of all parameter deviations that exceed the physical safety boundary to obtain the total system deviation. Substitute the total system deviation into the linear mapping formula (total deviation divided by deviation) to obtain the deviation ratio, which is used as the attenuation coefficient. Multiply the desired control output by the real-time attenuation coefficient to obtain the final control quantity after smooth attenuation. Note that there are positive and negative deviation attenuation. If the predicted parameter deviates towards the upper limit of the performance optimization corridor, multiply the desired control output by the real-time attenuation coefficient to reduce the positive amplitude of the control output and prevent the parameter from deviating further upward. If the predicted parameter deviates towards the lower limit of the performance optimization corridor, multiply the desired control output by the real-time attenuation coefficient to reduce the negative amplitude of the control output and prevent the parameter from deviating further downward.

[0028] S4 also includes: Based on the upper and lower limits of the current cycle process constraint boundary, the predicted parameter being greater than the upper limit of the boundary is considered a positive boundary risk. When there is a positive boundary risk, the upper limit of the boundary is subtracted from the predicted state parameter value to obtain the positive boundary deviation value. When there is a negative boundary risk, the lower limit of the boundary is subtracted from the predicted state parameter value to obtain the negative boundary deviation value. Based on the positive boundary deviation value, the desired control output is a positive drive command, such as acceleration, pressure increase, position increase, and temperature increase. Based on the negative boundary deviation value, the desired control output is a negative drive command, such as deceleration, pressure decrease, position decrease, and temperature decrease. After the above determination is completed, the reverse direction of the determination is taken as the only direction of the avoidance control quantity without modification. The positive and / or negative boundary deviation value is divided by the maximum allowable boundary deviation value to obtain the real-time forced correction coefficient. The product of the positive and / or negative boundary deviation value and the real-time forced correction coefficient is used as the final control quantity.

[0029] The actuator receives the final control quantity and converts the digitized control quantity into mechanical motion, voltage and current signals and / or fluid regulation signals that can be recognized by the controlled object, thereby generating the real-time operation of the controlled object.

[0030] The present invention also provides an intelligent control system based on multimodal dynamic boundaries, comprising: Real-time model and boundary construction module; used to extract the changing trends of operating parameters based on the original dataset, construct a mathematical model that is updated in real time on a cycle-by-cycle basis, and construct physical safety boundaries and process constraint boundaries based on the operating parameters; Dynamic PID calculation module; used to calculate the control deviation between the real-time control setpoint and the operating parameters, complete the real-time calculation of the proportional term, integral term, and derivative term, and synthesize the desired control output; The multi-cycle state prediction module is used to match the prediction cycle according to the response speed of the controlled object, and recursively calculate the future multi-cycle prediction system state using a real-time mathematical model as a framework. Boundary comparison and decision module: It is used to compare the predicted system state with the physical safety boundary and process constraint boundary, determine the system state region and generate the final control quantity, and receive the final control quantity and convert it into a drive signal that the actuator can recognize, so as to complete the real-time drive of the controlled object.

Claims

1. A smart control method based on multimodal dynamic boundaries, characterized in that, Includes the following steps: S1. Based on the real-time acquired operating parameters, the dynamic trend is obtained through numerical fitting, and a real-time mathematical model is constructed. The physical safety boundary is constructed according to the operating parameters. The physical safety boundary is used as the limit constraint under the current working condition. After obtaining the process constraint boundary, dynamic contraction is performed. S2. Based on PID, after calculating the control deviation of the real-time control setpoint, according to the current period control deviation value obtained from the deviation calculation, perform real-time calculation of the proportional term, real-time calculation of the integral term, and real-time calculation of the derivative term one by one, and synthesize the desired control output. S3. Based on the desired control output, the corresponding prediction period is matched according to the response speed of the controlled object, and the future single-cycle system state is calculated recursively cycle by cycle using a real-time mathematical model as the calculation framework. Then, the future multi-cycle state is continuously deduced using the single-cycle prediction value as the initial value. The predicted system state is obtained by comparing it with the physical safety boundary and process constraint boundary. Decisions are made according to the comparison results to obtain the final control quantity and obtain the multi-cycle continuous prediction state parameters. S4. The obtained multi-cycle continuous predicted state parameters are compared with the retrieved physical safety boundary and process constraint boundary. Decisions are made according to the comparison results to obtain the final control quantity. Real-time drive of the actuator is completed based on the final control quantity.

2. The intelligent control method based on multimodal dynamic boundaries according to claim 1, characterized in that, S1 collects operating parameters in each control cycle based on the sensors mounted on the controlled object to obtain the raw dataset; Based on the operating parameters in the original dataset, a dynamic mathematical model is constructed through real-time numerical fitting. Specifically, the changing trends of all operating parameters in the current cycle are extracted, and then the changing trends of the operating parameters are matched and fitted with the motion of the controlled object to generate a real-time mathematical model that is only suitable for the current cycle. Then, the operating parameters are extracted and safety boundaries are constructed according to categories, including vehicle speed safety boundary, temperature safety boundary, vibration safety boundary, and rotational speed safety boundary. Multiple real-time operating parameters are superimposed on the safety boundaries of each category for correction to obtain the physical safety boundary.

3. The intelligent control method based on multimodal dynamic boundaries according to claim 1, characterized in that, S1 extracts the real-time reaction acquisition value of the industrial equipment under the current operating conditions, arithmetically adds the real-time reactor temperature acquisition value and the temperature fluctuation value corresponding to the real-time reaction rate to obtain the current upper temperature limit constraint, which is used as the temperature constraint boundary. The product of the real-time reactor pressure acquisition value and the pressure coefficient corresponding to the real-time medium concentration is used to obtain the current upper pressure limit constraint, which is used as the pressure constraint boundary. The real-time boiler drum water level acquisition value is superimposed with the water level compensation value corresponding to the real-time steam flow to obtain the current water level limit constraint, which is used as the water level constraint boundary. The integrated water level constraint boundary, pressure constraint boundary, and temperature constraint boundary form the process constraint boundary.

4. The intelligent control method based on multimodal dynamic boundaries according to claim 1, characterized in that, S2 calculates the difference between the pre-input real-time control setpoint and the corresponding parameter value of the operating parameter obtained in S1, and obtains the control deviation value of the current cycle. Next, the absolute value of the current cycle control deviation is classified into magnitude categories, and a real-time proportional coefficient is matched according to the magnitude categories. The current cycle control deviation value is multiplied by the matched real-time proportional coefficient to obtain the proportional calculation result. The control deviation value of the current cycle is continuously accumulated with the control deviation values ​​of all previous control cycles in chronological order to obtain the total accumulated control deviation value. The magnitude of the total accumulated control deviation value is then determined, and the result of the integral operation is obtained by multiplying the total accumulated control deviation value by the matched real-time integral coefficient. The change in control deviation is obtained by subtracting the control deviation value of the previous cycle from the current cycle control deviation value. The rate of change of the deviation is then determined to complete the dynamic differential coefficient matching. The differential operation result is obtained by multiplying the change in control deviation by the matched real-time differential coefficient. The results of proportional calculation, integral calculation, and differential calculation are added together to output the desired control output without boundary constraints.

5. The intelligent control method based on multimodal dynamic boundaries according to claim 4, characterized in that, The S3 retrieves the real-time mathematical model, operating parameters, and desired control output. Based on the desired control output, the prediction period is determined according to the response speed of the controlled object, forming a set of basic prediction parameters. Using the current operating parameters as the initial state, the desired control output is substituted into the real-time mathematical model, and the predicted state parameters for the next control cycle are calculated recursively to obtain the single-cycle predicted state parameters. By using the single-cycle predicted state parameters as the new initial state and substituting the desired control output into the real-time mathematical model, the predicted state parameters for future control cycles are recursively calculated, thus obtaining multi-cycle continuous predicted state parameters.

6. The intelligent control method based on multimodal dynamic boundaries according to claim 5, characterized in that, The S4 synchronously retrieves the physical safety boundary and process constraint boundary, compares each parameter in the predicted system state with the upper and lower limits of the corresponding dimension's physical safety boundary and process constraint boundary, and includes the following specific determinations: A zone is defined as a safe zone if all predicted parameters in the multi-period continuous prediction state parameters are greater than or equal to the lower limit of the physical safety boundary and less than or equal to the upper limit of the physical safety boundary. If all predicted parameters in the multi-cycle continuous prediction state parameters are greater than or equal to the physical safety boundary and less than or equal to the physical safety boundary, but at least one predicted parameter exceeds the upper limit and / or lower limit of the process constraint boundary, it is determined to be a warning zone. If any one of the predicted parameters is less than the lower limit of the physical safety boundary and / or the process constraint boundary, or greater than the upper limit of the physical safety boundary and / or the process constraint boundary, it is determined to be a hazardous area. The final output shows the region to which the system status belongs.

7. The intelligent control method based on multimodal dynamic boundaries according to claim 6, characterized in that, S4 retrieves the original value of the desired control output as the final control quantity based on the determination result of the safe zone, and synchronously transmits the final control quantity to the actuator. Obtain the upper limit, lower limit, and center reference value of the physical safety boundary for the current period. Subtract the predicted state parameter value from the center reference value of the physical safety boundary to obtain the deviation of a single parameter. Sum the absolute values ​​of all parameter deviations that exceed the physical safety boundary to obtain the total deviation of the system. Substitute the total deviation of the system into the linear mapping formula to obtain the deviation ratio, which is used as the attenuation coefficient. Multiply the desired control output by the real-time attenuation coefficient to obtain the final control quantity after smooth attenuation.

8. The intelligent control method based on multimodal dynamic boundaries according to claim 6, characterized in that, S4 further includes: Based on the upper and lower limits of the current cycle process constraint boundary, the predicted parameter being greater than the upper limit of the boundary is considered a positive boundary risk. When there is a positive boundary risk, the upper limit of the boundary is subtracted from the predicted state parameter value to obtain the positive boundary deviation value. When there is a negative boundary risk, the lower limit of the boundary is subtracted from the predicted state parameter value to obtain the negative boundary deviation value. Based on the positive boundary deviation value, the desired control output is converted into a positive drive command; based on the negative boundary deviation value, the desired control output is converted into a negative drive command. The positive and / or negative boundary deviation values ​​are divided by the maximum permissible boundary deviation value to obtain the real-time forced correction coefficient, and the product of the positive and / or negative boundary deviation values ​​and the real-time forced correction coefficient is used as the final control quantity.

9. The intelligent control method based on multimodal dynamic boundaries according to claim 1, characterized in that, The S4 includes: the actuator receiving the final control quantity, converting the digitized control quantity into mechanical actions, voltage and current signals and / or fluid regulation signals that can be recognized by the controlled object, and generating real-time operating actions of the controlled object.

10. An intelligent control system based on multimodal dynamic boundaries, applied to the intelligent control method based on multimodal dynamic boundaries as described in any one of claims 1-9, characterized in that, include: Real-time model and boundary construction module; used to extract the changing trends of operating parameters based on the original dataset, construct a mathematical model that is updated in real time on a cycle-by-cycle basis, and construct physical safety boundaries and process constraint boundaries based on the operating parameters; Dynamic PID calculation module; used to calculate the control deviation between the real-time control setpoint and the operating parameters, complete the real-time calculation of the proportional term, integral term, and derivative term, and synthesize the desired control output; The multi-cycle state prediction module is used to match the prediction cycle according to the response speed of the controlled object, and recursively calculate the future multi-cycle prediction system state using a real-time mathematical model as a framework. Boundary comparison and decision-making module; It is used to compare the predicted system state with the physical safety boundary and process constraint boundary, determine the system state region and generate the final control quantity, and receive the final control quantity and convert it into a drive signal that the actuator can recognize, so as to complete the real-time drive of the controlled object.

Images