Model predictive control system and method thereof

Abstract

A model predictive control system and method thereof are disclosed. The model predictive control system includes a model simulation unit, a monitoring unit and a proxy model predictive control unit. The model simulation unit is configured to simulate dynamic behavior of a target object according to a behavior model of the target object and a plurality of control decision parameters, and to generate a plurality of actual state parameters accordingly. The monitoring unit is configured to monitor the plurality of control decision parameters and the plurality of actual state parameters, and to generate a plurality of estimated state parameters. The proxy model predictive control unit is configured to perform an approximate model predictive control operation according to the plurality of estimated state parameters and a plurality of target parameters, to replace a model predictive control operation with higher computational complexity.

Description

Technical Field

[0001] This disclosure relates to a control system and a control method, and more particularly to a model predictive control system and method for predicting and controlling the behavior of a target object. Background Technology

[0002] Model predictive control mechanisms establish a dynamic model of the target object based on its dynamic behavior, predict the dynamic changes of the target object based on the dynamic model, and then use optimization algorithms to calculate the optimal solution of the control decision input value so that the behavior of the target object can meet the predetermined goal.

[0003] However, the optimization algorithm of the model predictive control mechanism involves a large-dimensional optimization solution process. When the dynamic changes of the target object are rapid or the dynamic model of the target object is complex, the optimization algorithm will consume a lot of computing time, resulting in the inability to complete the optimization solution within a limited decision-making cycle. This will cause delays in the control decision of the target object, affecting the performance of the target object and even causing accidents.

[0004] Therefore, those skilled in the art in related industries are dedicated to improving model predictive control mechanisms, which can quickly calculate control decision input values ​​suitable for control objectives in various application scenarios (such as financial transactions, self-driving cars, drones, factory production lines, etc.) and avoid control decision delays; furthermore, the improved model predictive control mechanism can also significantly reduce the amount of computation to lower hardware specification requirements. Summary of the Invention

[0005] This disclosure provides a model predictive control system, including a model simulation unit, a monitoring unit, and a proxy model predictive control unit. The model simulation unit simulates the dynamic behavior of a target object based on its behavioral model and multiple control decision parameters from previous states, generating multiple actual state parameters accordingly. The monitoring unit monitors the control decision parameters from previous states and the multiple actual state parameters, and generates multiple estimated state parameters based on these parameters. The proxy model predictive control unit performs approximate model predictive control calculations based on the estimated state parameters and target parameters to obtain multiple control decision parameters for the current state, and then feeds these current state control decision parameters back to the model simulation unit. The proxy model predictive control unit performs these approximate model predictive control calculations instead of computationally complex model predictive control calculations.

[0006] This disclosure also provides a model predictive control method, comprising the following steps: Establishing a behavioral model of a target object; Simulating the dynamic behavior of the target object based on this behavioral model and multiple control decision parameters from previous states; Generating multiple actual state parameters accordingly; Monitoring the multiple control decision parameters from previous states and the multiple actual state parameters; Generating multiple estimated state parameters based on the multiple control decision parameters from previous states and the multiple actual state parameters; Setting up a surrogate model predictive control unit; Performing approximate model predictive control calculations using the surrogate model predictive control unit based on the multiple estimated state parameters and the multiple target parameters, replacing computationally complex model predictive control calculations; and Feeding back the multiple control decision parameters from the current states to this model simulation unit. Attached Figure Description

[0007] Other aspects and advantages of this disclosure will become apparent from the following figures, detailed description and scope of the patent application.

[0008] Figure 1 This is a block diagram of a model predictive control system according to an embodiment of the present disclosure.

[0009] Figure 2 This is a schematic diagram of an embodiment of the model predictive control system of this disclosure applied to an autonomous vehicle.

[0010] Figure 3 This is a schematic diagram illustrating the model predictive control operation performed by the proxy model predictive control unit of this disclosure.

[0011] Figure 4 This is a schematic diagram of an embodiment of the model predictive control system of this disclosure applied to a factory production line.

[0012] Figure 5 This is a flowchart of a model predictive control method according to an embodiment of the present disclosure.

[0013] [Explanation of Labels in the Attached Image]

[0014] 1000: Model Predictive Control System

[0015] 100: Model Simulation Unit

[0016] 200: Monitoring Unit

[0017] 300: Optimized Model Predictive Control Unit

[0018] 400: Proxy Model Predictive Control Unit

[0019] 500: Update Unit

[0020] 700: Target

[0021] 2000: Factory Production Line

[0022] u1, u2, u3: Control decision parameters

[0023] y1, y2: Actual state parameters

[0024] x1, x2: Estimated state parameters

[0025] S: Set of target parameters

[0026] s1~sN: Target parameters

[0027] track1: Target movement trajectory

[0028] track2: Actual movement trajectory

[0029] u1_mean, u2_mean: average values

[0030] u1_var, u2_var: Mutated values

[0031] 2100: Feed rate

[0032] 2200: Steam volume

[0033] 2300: Reboiling temperature and pressure

[0034] S110~S180: Steps Detailed Implementation

[0035] The technical terms used in this specification refer to those commonly used in the field. Where this specification provides explanations or definitions for certain terms, the interpretation of those terms shall be based on the explanations or definitions provided in this specification. Each embodiment of this disclosure has one or more technical features. Where feasible, those skilled in the art may selectively implement some or all of the technical features in any embodiment, or selectively combine some or all of the technical features in these embodiments.

[0036] Figure 1 This is a block diagram of a model prediction control (MPC) system 1000 according to an embodiment of this disclosure. Please refer to... Figure 1 The model prediction control system 1000 includes a model simulation unit 100, a monitoring unit 200, an optimal model prediction control unit 300, a proxy model prediction control unit 400, and an update unit 500.

[0037] The model simulation unit 100 can establish a behavioral model of the target object 700 to simulate its dynamic behavior. In one example, the target object 700 (or "target object" or "operation target," generally referring to the control target for predictive control by the model predictive control system 1000) is, for example, a self-driving car. The model simulation unit 100 can establish a behavioral model of the self-driving car to simulate its dynamic behavior such as "vehicle speed" and "direction of travel." The control object associated with the self-driving car's behavior "vehicle speed" is the "accelerator," while the control object associated with "direction of travel" is the "steering wheel." In operation, the model simulation unit 100 can receive multiple control decision parameters, one of which, u1, is, for example, "accelerator pedal position," and its associated control object is the "accelerator." The model simulation unit 100 can simulate controlling the self-driving car's "accelerator" according to the control decision parameter u1 to adjust the self-driving car's behavior "vehicle speed." Furthermore, the model simulation unit 100 can simulate and calculate the actual measurement value of "vehicle speed", that is, the actual state parameter y1 of "vehicle speed"; in other words, the actual state parameter y1 represents the "actual vehicle speed" of the self-driving car.

[0038] On the other hand, another control decision parameter u2 of the model simulation unit 100 is, for example, "steering wheel rotation amount," and its associated control object is the "steering wheel." The model simulation unit 100 can simulate and control the "steering wheel" of the self-driving car according to the control decision parameter u2 to adjust the behavior "direction of travel" of the self-driving car. Furthermore, the model simulation unit 100 can simulate and calculate the actual measured value of the "direction of travel," that is, the actual state parameter y2; in other words, the actual state parameter y2 represents the "actual direction of travel" of the self-driving car.

[0039] In summary, the model simulation unit 100 can simulate controlling the "throttle" of the self-driving car based on the control decision parameter u1 "throttle size" to adjust the self-driving car's behavior "vehicle speed," and the model simulation unit 100 will output the corresponding actual state parameter y1 "actual vehicle speed." On the other hand, the model simulation unit 100 can simulate controlling the "steering wheel" of the self-driving car based on the control decision parameter u2 "steering wheel rotation amount" to adjust the self-driving car's behavior "direction of travel," and the model simulation unit 100 will output the corresponding actual state parameter y2 "actual direction of travel."

[0040] The monitoring unit 200 is connected to the model simulation unit 100. The monitoring unit 200 receives actual state parameters y1 and y2 from the model simulation unit 100 and generates estimated state parameters x1 and x2 based on the control decision parameters u1 and u2. Here, the estimated state parameter x1 represents the "estimated vehicle speed," and the estimated state parameter x2 represents the "estimated direction of travel" of the autonomous vehicle. More specifically, the monitoring unit 200 receives the actual state parameters y1(k) and y2(k) of the current state (or the current iteration, i.e., the k-th iteration) and the control decision parameters u1(k-1) and u2(k-1) of the previous state (or the previous iteration, i.e., the (k-1)-th iteration). The monitoring unit 200 generates the estimated state parameters x1(k) and x2(k) of the current iteration based on the actual state parameters y1(k) and y2(k) of the current iteration and the control decision parameters u1(k-1) and u2(k-1) of the previous iteration.

[0041] Furthermore, the optimization model prediction control unit 300 is connected to the monitoring unit 200, and can receive estimated state parameters x1(k) and x2(k) from the monitoring unit 200. In addition, the optimization model prediction control unit 300 can receive a target parameter set S, which may include multiple target parameters s1, s2, ..., sN and can be represented as S = {s1, s2, ..., sN}. Please refer to [reference needed]. Figure 2 The diagram illustrates an embodiment of the model predictive control system 1000 disclosed herein applied to an autonomous vehicle; the target parameter set S can represent the target behavior that the user expects the target object 700 (autonomous vehicle) to achieve. Figure 2The target movement trajectory track1 that the self-driving car should achieve is the target parameter set S. Furthermore, the coordinates of each point on the target movement trajectory track1 are the target parameters s1, s2, ..., sN of the target parameter set S. Based on the target parameter set S and the estimated state parameters x1(k) and x2(k), the optimized model prediction control unit 300 can execute model predictive control to calculate the corresponding control decision parameters u1(k) and u2(k). In other words, to ensure that the behavior of the self-driving car satisfies the target parameter set S (i.e., to make the actual movement trajectory track2 of the self-driving car satisfy the target movement trajectory track|), the optimized model prediction control unit 300 calculates the corresponding control decision parameters u1(k) and u2(k) (i.e., the self-driving car's "estimated speed" and "estimated direction of travel") based on the estimated state parameters x1(k) and x2(k) (i.e., the self-driving car's "estimated speed" and "estimated direction of travel") to control the self-driving car's "throttle" and "steering wheel" respectively, thereby adjusting the self-driving car's behavior "speed" and "direction of travel". The adjusted "vehicle speed" and "direction of travel" can be reflected in the actual state parameters y1 and y2 (i.e., the "actual vehicle speed" and "actual direction of travel" of the self-driving car) through the simulation of the model simulation unit 100.

[0042] In this embodiment, the optimization model predictive control unit 300 performs model predictive control according to the optimization algorithm, thereby calculating the optimal solution for the control decision parameters u1, u2 that satisfy the target parameter set S. However, if other control objects of the autonomous vehicle (e.g., vehicle registration, engine temperature, etc.) are considered, the optimization model predictive control unit 300 must perform optimization calculations for a large number of control decision parameters u1, u2, u3, ..., uM. This will consume a lot of computation time and will not be able to calculate the optimal solution for the control decision parameters u1, u2, u3, ..., uM within a short decision cycle (e.g., the decision cycle for the autonomous vehicle's driving can be as short as about 0.1 seconds). Therefore, real-time control cannot be achieved, which may cause performance degradation or safety accidents of the autonomous vehicle.

[0043] To address this, a pre-established surrogate model predictive control unit 400 can be used instead of the optimization model predictive control unit 300. The surrogate model predictive control unit 400 performs model predictive control calculations with lower computational complexity (and therefore shorter decision cycles) instead of the higher computational complexity (and therefore longer decision cycles) model predictive control calculations performed by the optimization model predictive control unit 300. Furthermore, although the surrogate model predictive control unit 400 does not perform optimization calculations, it can still obtain results approximate to the optimal solution, thus effectively adjusting the behavior of the controlled object. In this embodiment, the model of the surrogate model predictive control unit 400 can be pre-established in an offline manner (i.e., offline modeling), and the model establishment method can include machine learning methods such as imitation learning. See also... Figure 3 The diagram illustrates the surrogate model predictive control unit 400 performing model predictive control calculations. The surrogate model predictive control unit 400 can perform model predictive control based on a random program, such as a Gaussian process (GP). Taking the two control decision parameters u1 and u2 of a self-driving car (i.e., the "throttle input" and "steering wheel rotation") as an example, the surrogate model predictive control unit 400 performs Gaussian process calculations based on the estimated state parameters x1 and x2 and the target parameter set S to calculate the average values ​​u1_mean and u2_mean, and the variability values ​​u1_var and u2_var of the control decision parameters u1 and u2, respectively. In one example, the average values ​​u1_mean and u2_mean are used as the final output control decision parameters u1 and u2 (i.e., control decision parameter u1 = average value u1_mean, control decision parameter u2 = average value u2_mean). On the other hand, the variation values ​​u1_var and u2_var of the control decision parameters u1 and u2 are used as the model reliability index rindex to determine whether the surrogate model prediction control unit 400 needs to be updated.

[0044] Continuing from the above, the update unit 500 can set a threshold t1 for the model reliability pointer rindex and compare the model reliability pointer rindex with the threshold t1. When the model reliability pointer rindex is greater than the threshold t1 (i.e., the variation values ​​u1_var and u2_var of the control decision parameters u1 and u2 are greater than the threshold t1), it indicates that the reliability of the currently used surrogate model prediction control unit 400 is low and an adaptive update is required. In one example, various operational data (such as operational data related to the "vehicle speed" and "driving direction" of the autonomous vehicle) can be collected online, and the surrogate model prediction control unit 400 can be adaptively updated using multiple operational data.

[0045] Therefore, the model of the surrogate model predictive control unit 400 can be pre-built offline (i.e., offline modeling), and the surrogate model predictive control unit 400 can be adaptively updated online (i.e., online updating). Furthermore, when the surrogate model predictive control unit 400 is undergoing adaptive updates, it cannot calculate the control decision parameters u1 and u2 in real time. In this situation, it reverts to using the optimized model predictive control unit 300 to perform optimized model predictive control (i.e., model predictive control with higher computational complexity (and therefore longer decision cycle)) to calculate the control decision parameters u1 and u2.

[0046] Figure 4 A schematic diagram illustrating an embodiment of the model predictive control system 1000 disclosed herein applied to a factory production line 2000 is shown, as follows. Figure 4 As shown, the target object to be controlled is the "distillation column" in factory production line 2000. The controlled objects of the model predictive control system 1000 include the control variables of each PID controller setpoint in the entire distillation column system, such as "feed" 2100, "steam" 2200, and "reboiler temperature and pressure" 2300, etc. The control decision parameters u1, u2, and u3 of the model predictive control system 1000 correspond to the PID controller setpoints of the above-mentioned controlled objects, respectively. For example, control decision parameter u1 is the PID controller setpoint for "feed" 2100. Control decision parameter u2 is the PID controller setpoint for "steam" 2200. And control decision parameter u3 is the PID controller setpoint for "reboiler temperature or pressure" 2300.

[0047] Furthermore, the target parameter s1 of the target object "distillation column" is, for example, "output per unit time of distillate": In order to achieve the user's expected output per unit time, approximate model predictive control can be performed by the proxy model predictive control unit 400 of the model predictive control system 1000 to calculate control decision parameters u1, u2, u3 that approximate the optimal solution, and then the control decision parameters u1, u2, u3 are imported into the corresponding PID controller setpoint of the entire distillation column system in the actual factory production line 2000. The decision cycle of the aforementioned factory production line 2000 (taking a chemical plant with a distillation column system as an example) is approximately 15 to 30 seconds. The proxy model predictive control unit 400 can perform faster approximate model predictive control, and can calculate the control decision parameters u1, u2, u3 that approximate the optimal solution within a decision cycle of 15 to 30 seconds.

[0048] Furthermore, in the factory production line 2000, the proxy model predictive control system 1000 of this embodiment can also selectively cooperate with the control loop performance evaluation (CLPA) system, the plant-wide disturbance source tracing (PDT) system, the controller parameter tuning (PIDtunner) system, or the system identification (modeling) system to control and adjust various variables in the factory production line 2000.

[0049] Figure 5 This is a flowchart of a model predictive control method according to an embodiment of this disclosure. Please refer to section [number missing]. Figure 5 Model predictive control methods can be combined with Figure 1 Model predictive control system 1000 and Figure 3 This is implemented using a proxy model prediction control unit 400. First, in step S110, the model simulation unit 100 receives multiple control decision parameters, such as two control decision parameters u1 "throttle position" and u2 "steering wheel rotation amount" for the autonomous vehicle. Furthermore, the model simulation unit 100 simulates and generates actual state parameters y1 "actual vehicle speed" and y2 "actual direction of travel" based on the control decision parameters u1 and u2.

[0050] Then, in step S120, the model simulation unit 100 transmits the actual state parameters y1(k) and y2(k) of the current state to the monitoring unit 200, and the monitoring unit 200 also receives the control decision parameters u1(k-1) and u2(k-1) of the previous state. Accordingly, the monitoring unit 200 generates the estimated state parameter x1(k) "estimated vehicle speed" and the estimated state parameter x2(k) "estimated direction of travel" based on the actual state parameters y1(k) and y2(k) and the control decision parameters u1(k-1) and u2(k-1).

[0051] Then, in step S130, it is determined whether the decision period of the model predictive control system 1000 is within the decision cycle; taking a self-driving car as an example, the decision cycle is approximately 0.1 seconds. If it is still within the decision cycle, step S135 is executed, using the proxy model predictive control unit 400 to perform approximate model predictive control; wherein, the proxy model predictive control unit 400 receives the target parameter set S and the estimated state parameters x1(k) and x2(k) and generates the average values ​​u1_mean and u2_mean, and the variability values ​​u1_var and u2_var of the control decision parameters u1 and u2 respectively, and uses the variability values ​​u1_var and u2_var as the model reliability pointer rindex. The generated control decision parameters u1 and u2 can then be provided to the model simulation unit 100 in step S110 to perform simulation.

[0052] Then, in step S140, the threshold t1 of the model reliability pointer rindex is set; and it is determined whether the model reliability pointer rindex is greater than the threshold t1. If the model reliability pointer rindex is less than the threshold t1 (i.e., the variability values ​​u1_var and u2_var are less than the threshold t1), then step S150 is executed, and the average values ​​u1_mean and u2_mean are used as the final output control decision parameters u1 and u2 (i.e., control decision parameter u1 = average value u1_mean, control decision parameter u2 = average value u2_mean).

[0053] If the model reliability pointer rindex is greater than the threshold t1 (i.e., the variation values ​​u1_var and u2_var are greater than the threshold t1), it indicates that the surrogate model prediction control unit 400 must be updated. Then, step S160 is executed to prepare for updating the surrogate model prediction control unit 400. Next, step S170 is executed to calculate the control decision parameters u1 and u2 using the optimized model prediction control unit 300. Then, step S180 is executed to update the surrogate model prediction control unit 400; for example, an online adaptive update method can be used to collect various operational data related to the target object's behavioral model online, and update the surrogate model prediction control unit 400 using multiple data points. After the update, step S150 can be executed to calculate the control decision parameters u1 and u2 using the updated surrogate model prediction control unit 400.

[0054] In summary, the model predictive control system 1000 of this disclosure, in conjunction with the implementation of the model predictive control method, pre-establishes a surrogate model predictive control unit 400 through offline modeling, imitation learning, or machine learning. The surrogate model predictive control unit 400 performs approximate model predictive control calculations, replacing the computationally complex (and therefore longer decision-making cycle) optimization model predictive control calculations performed by the optimization model predictive control unit 300. Therefore, the surrogate model predictive control unit 400 can significantly reduce computation time and calculate the control decision parameters of the approximate optimal solution within a shorter decision cycle (especially for autonomous vehicles with a decision cycle of approximately 0.1 seconds). Furthermore, the surrogate model predictive control unit 400 performs model predictive control calculations based on a Gaussian program model to obtain the average and variability values ​​of the control decision parameters, and uses the variability values ​​of the control decision parameters as a model reliability indicator. If the model reliability indicator exceeds a threshold, the surrogate model predictive control unit 400 is updated online adaptively.

[0055] While the present invention has been disclosed above in detail with reference to preferred embodiments and examples, it is understood that the various embodiments are intended to illustrate rather than limit. It is anticipated that those skilled in the art will conceive of numerous modifications and combinations, all of which fall within the spirit of the invention and the scope of the appended claims.

Claims

1. A model predictive control system characterized by, The model simulation unit simulates the dynamic behavior of the target object according to the behavior model of the target object and the control decision parameters of the plurality of previous states, and correspondingly generates a plurality of actual state parameters; The monitoring unit monitors the control decision parameters of the previous states and the actual state parameters, and generates a plurality of estimated state parameters according to the control decision parameters of the previous states and the actual state parameters; and The proxy model predictive control unit performs approximate model predictive control operations according to the estimated state parameters and a plurality of target parameters to obtain a plurality of current state control decision parameters, and returns the current state control decision parameters to the model simulation unit; Wherein, the proxy model predictive control unit performs the approximate model predictive control operations to replace the model predictive control operations with higher computational complexity. The target object has a plurality of control objects, and the control objects are controlled according to the control decision parameters, so that the actual state parameters of the target object satisfy the target parameters.

2. The model predictive control system of claim 1, wherein, The proxy model predictive control unit performs a stochastic program operation to perform the approximate model predictive control operation.

3. The model predictive control system of claim 1, wherein, The stochastic program operation performed by the proxy model predictive control unit is a Gaussian program operation, and the proxy model predictive control unit performs the Gaussian program operation to obtain the average value and the variation value of each of the control decision parameters.

4. The model predictive control system of claim 3, wherein, The average values are the final output control decision parameters, and the variation values are model reliability indicators.

5. The model predictive control system of claim 4, wherein, Further comprising:

6. The model predictive control system of claim 5, wherein, An update unit for determining whether the model reliability indicator is greater than a threshold value, and if the model reliability indicator is greater than the threshold value, the proxy model predictive control unit is updated. Further comprising:

7. The model predictive control system of claim 6, wherein, An optimized model predictive control unit for performing the model predictive control operation with higher computational complexity to obtain the control decision parameters satisfying the optimal solution when the proxy model predictive control unit is updated. The proxy model predictive control unit is updated in an online adaptive manner.

8. The model predictive control system of claim 6, wherein, The proxy model predictive control unit pre-establishes the behavior model of the target object in an offline manner.

9. The model predictive control system of claim 1, wherein, The proxy model predictive control unit pre-establishes the behavior model in an imitation learning manner or a machine learning manner.

10. The model predictive control system of claim 9, wherein, The method comprises:

11. A model predictive control method, characterized by, Establishing a behavior model of a target object; Simulating the dynamic behavior of the target object according to the behavior model of the target object and the control decision parameters of a plurality of previous states; Correspondingly generating a plurality of actual state parameters; Monitoring the control decision parameters of the previous states and the actual state parameters; Generating a plurality of estimated state parameters according to the control decision parameters of the previous states and the actual state parameters; Setting a proxy model predictive control unit; Performing approximate model predictive control operations by the proxy model predictive control unit according to the estimated state parameters and a plurality of target parameters to replace the model predictive control operations with higher computational complexity; and Returning a plurality of current state control decision parameters to a model simulation unit. The target object has a plurality of control objects, and the model predictive control method comprises:

12. The model predictive control method of claim 11, wherein, ​ The control decision parameters are simulated to control the control objects so that the actual state parameters of the target object satisfy the target parameters.

13. The model predictive control method of claim 11, wherein, The stochastic program operation is performed by the proxy model predictive control unit to perform the approximate model predictive control operation.

14. The model predictive control method of claim 13, wherein, The stochastic program operation performed by the proxy model predictive control unit is a Gaussian program operation, and the model predictive control method comprises: The Gaussian program operation is performed by the proxy model predictive control unit to obtain respective mean values and variation values of the control decision parameters.

15. The model predictive control method of claim 14, wherein, Further comprising: The mean values are used as the final output control decision parameters. And The variation values are used as model reliability indicators.

16. The model predictive control method of claim 15, wherein, Further comprising: Judging whether the model reliability indicator is greater than a threshold value; and If the model reliability indicator is greater than the threshold value, the proxy model predictive control unit is updated.

17. The model predictive control method of claim 16, wherein, Further comprising: Setting an optimized model predictive control unit; and When the proxy model predictive control unit is updated, the model predictive control operation with higher operation complexity is performed by the optimized model predictive control unit to obtain the control decision parameters satisfying the optimal solution. The proxy model predictive control unit is updated in an online adaptive manner.

18. The model predictive control method of claim 16, wherein, The behavior model of the target object is pre-established by the proxy model predictive control unit in an offline manner.

19. The model predictive control method of claim 11, wherein, The behavior model is pre-established by the proxy model predictive control unit in an imitation learning manner or a machine learning manner.

20. The model predictive control method of claim 19, wherein, ​

Images