Integrated Intelligent Learning Control Method and System Based on Non-Strictly Repeating Motion System

Abstract

The present invention discloses a comprehensive intelligent learning control method based on a non-strictly repetitive motion system, which relates to the technical field of artificial intelligence and includes: transforming the initial error with variable iteration for the non-strictly repetitive motion system; intelligently identifying the uncertain part of the dynamic system model of the non-strictly repetitive motion system; constructing a comprehensive intelligent learning control law for the dynamic system to achieve the switching of control modes; and simulating and verifying the intelligent control method. The present invention also provides a comprehensive intelligent learning control system based on the non-strictly repetitive motion system. The comprehensive intelligent learning controller designed for the non-strictly repetitive motion system proposed by the present invention combines intelligent tools such as neural networks and fuzzy systems with iterative learning control technology, uses intelligent tools such as neural networks to obtain more accurate information about the system model, can handle the situation where the length of the repetitive motion time period of the system is not constant, and can be applied to more complex dynamic systems.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically to a comprehensive intelligent learning control method based on a non-strictly repetitive motion system. Background Art

[0002] Industrial robots belong to the core equipment of intelligent manufacturing and are a highly integrated manifestation of cutting-edge automation technology in modern industry. Industrial robots mostly perform repetitive tasks such as handling, installation, cutting, and painting during the intelligent manufacturing process. In addition, modern logistics and rail transit mostly have repetitive workflow changes every day; batch production of pharmaceutical products and chemical products also has the characteristic of repetition. These repetitive tasks require advanced intelligent control technology to ensure their accurate and effective operation.

[0003] Iterative learning control is applicable to dynamic systems that repeatedly execute the same control task over a fixed time period, and its goal is to achieve tracking control over a finite time period. When the mathematical model of a dynamic system with repetitive motion cannot be accurately known, the iterative learning control method can learn from the control experience of previous repeated operations, thereby making up for the lack of the system in terms of model knowledge. Industrial robots, with their complex model dynamics and repetitive motion process characteristics, have become a typical application field of iterative learning control.

[0004] However, the iterative learning control method commonly applied to industrial robots has strict restrictions on the repeatability of system motion in terms of the initial state, desired trajectory, and desired trajectory time length, etc., that is, it is required that the initial state, desired trajectory, and desired trajectory time length, etc. of the controlled system for each iteration (or repetitive motion) do not change with the number of iterations. In the actual industrial robot control applications, the above strict requirements for motion repeatability of the conventional iterative learning control method are very idealistic and are often not met, which will inevitably affect the accuracy of industrial robot trajectory tracking control. Especially under non-strictly repetitive motion conditions, when the desired trajectory of the iterative learning control system fluctuates with the number of iterations, it is more difficult to ensure the consistency of the system iteration initial state and the desired initial state. Summary of the Invention

[0005] In view of the above deficiencies of the prior art, the present invention provides a comprehensive intelligent learning control method and system based on a non-strictly repetitive motion system.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] A comprehensive intelligent learning control method based on a non-strictly repetitive motion system includes the following steps:

[0008] Perform transformation of variable iteration initial error on the non-strictly repetitive motion system;

[0009] Intelligently identify the uncertain part of the dynamic system model of the non-strictly repetitive motion system;

[0010] Construct a comprehensive intelligent learning control law for the dynamic system to achieve the switching of control modes;

[0011] Simulate and verify the intelligent control method.

[0012] As a further technical solution of the present invention, the transformation of the variable iteration initial error for the non-strictly repetitive motion system specifically includes:

[0013] Intercept an initial time period from the repetitive time period, construct a trajectory on the initial time period to naturally connect the initial state of each iteration and the desired trajectory point at the end of the initial time period, form a new desired trajectory, and transform the bounded variable iteration initial error into the desired trajectory under the condition of zero iteration initial error.

[0014] As a further technical solution of the present invention, the intelligent identification of the uncertain part of the dynamic system model of the non-strictly repetitive motion system specifically includes:

[0015] Use neural networks, fuzzy systems, and adaptive identification techniques to model and estimate the relevant unknown parts or signals in the non-strictly repetitive motion process of the system, and use the input-output measurement data of the system in each repetitive time period to train and adjust the neural network in real time; obtain system model knowledge and integrate the system model knowledge into the iterative learning controller.

[0016] As a further technical solution of the present invention, the construction of a comprehensive intelligent learning control law for the dynamic system to achieve the switching of control modes specifically includes:

[0017] The construction of the control law is divided into two methods in terms of time period. When the time period of the next repetitive motion is less than or equal to the time period of the previous repetitive motion, iterative learning control is implemented throughout the time period of the next repetitive motion; when the time period of the next repetitive motion is greater than the time period of the previous repetitive motion, the control throughout the time period of the next repetitive motion is divided into two parts. The same time period part in the front adopts the iterative learning control method, and in the remaining time period part in the back, according to the system model information real-time identified in this repetitive motion, a general tracking control method is adopted.

[0018] As a further technical solution of the present invention, the simulation verification of the comprehensive learning controller for the non-strictly repetitive motion system specifically includes:

[0019] The comprehensive intelligent learning controller for the non-strictly repetitive motion system is simulated through the MATLAB simulation experiment platform. Numerical simulation experiments are carried out according to the comprehensive intelligent learning control algorithm to verify the effectiveness of the comprehensive intelligent learning control method.

[0020] The present invention also provides a comprehensive intelligent learning control system based on a non-strictly repetitive motion system, including:

[0021] A variable iteration initial error conversion unit for converting the variable iteration initial error of the non-strictly repetitive motion system;

[0022] An intelligent identification technology unit for intelligently identifying the uncertain part of the dynamic system model of the non-strictly repetitive motion system;

[0023] A control switching unit for constructing a comprehensive intelligent learning control law of the dynamic system to realize the switching of the control mode;

[0024] A simulation verification unit for simulating and verifying the intelligent control method.

[0025] The beneficial effects of the present invention are:

[0026] 1. The comprehensive intelligent learning controller designed for the non-strictly repetitive motion system proposed by the present invention combines intelligent tools such as neural networks and fuzzy systems with iterative learning control technology, so that the tracking error of iterative learning control converges to a sufficiently small bound determined by the approximation error of the intelligent tool outside a very small initial time period.

[0027] 2. The present invention is oriented to industrial robot applications, designs a comprehensive intelligent learning controller for non-strictly repetitive motion conditions for more general dynamic systems, generally only requires boundedness for the variation laws of variable iteration initial errors and variable desired trajectories, has no strict restrictions on the data model of the controlled system, and the number of inputs and outputs of the system can be set as needed.

[0028] 3. The comprehensive intelligent learning control method designed by the present invention uses intelligent tools such as neural networks to obtain more accurate information about the system model, can cope with the situation where the length of the repetitive motion time period of the system is not constant, and can be applied to more complex dynamic systems. Description of the Drawings

[0029] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention.

[0030] Figure 1 It is a flow chart of the comprehensive intelligent learning control method based on the non-strictly repetitive motion system proposed by the present invention;

[0031] Figure 2Structural diagram of the integrated intelligent learning control system based on the non-strictly repetitive motion system proposed by the present invention. Detailed implementation manners

[0032] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0033] Under the condition of non-strictly repetitive motion, the present invention is directed to an industrial robot system, and organically combines artificial neural networks, fuzzy systems, adaptive identification technologies, etc. with the iterative learning control method, so as to design an integrated intelligent learning controller for a dynamic system under non-strictly repetitive motion conditions, and achieve good tracking of the desired trajectory by the system within a finite time period.

[0034] See Figure 1 , the integrated intelligent learning control method based on the non-strictly repetitive motion system includes:

[0035] Step 101, transform the variable iterative initial error of the non-strictly repetitive motion system;

[0036] Step 102, intelligently identify the uncertain part of the dynamic system model of the non-strictly repetitive motion system;

[0037] Step 103, construct an integrated intelligent learning control law for the dynamic system to realize the switching of the control mode;

[0038] Step 104, perform simulation verification on the intelligent control method.

[0039] In the embodiment of the present invention, when the iterative initial error and the desired trajectory of the iterative learning control system vary arbitrarily within a certain range, the conventional robust iterative learning control design often can only ensure that the tracking error of the iterative learning control is bounded, and the size of this "bound" cannot be controlled and adjusted. The integrated intelligent learning controller designed for the non-strictly repetitive motion system proposed by the present invention combines intelligent tools such as neural networks and fuzzy systems with the iterative learning control technology, so that the tracking error of the iterative learning control converges to a sufficiently small bound determined by the approximation error of the intelligent tool outside a very small initial time period.

[0040] The present invention will be directed to industrial robot applications, and design an integrated intelligent learning controller for more general dynamic systems under non-strictly repetitive motion conditions. Generally, only boundedness requirements are made for the variation laws of the variable iterative initial error and the variable desired trajectory, and there are no strict restrictions on the data model of the controlled system. The number of inputs and outputs of the system can be set according to needs.

[0041] The integrated intelligent learning control method designed by the present invention uses intelligent tools such as neural networks to obtain more accurate information of the system model, can cope with the situation where the length of the repeated motion time period of the system is not constant, and can be applied to more complex dynamic systems.

[0042] In the embodiment of the present invention, the transformation of the variable iteration initial error for the non-strictly repeating motion system is specifically as follows:

[0043] Intercept an initial time period from the repeated time period, construct a trajectory on the initial time period, so that it naturally connects the initial state of each iteration and the desired trajectory point at the end of the initial time period to form a new desired trajectory, and transform the bounded variable iteration initial error into the desired trajectory under the condition of zero iteration initial error.

[0044] In the embodiment of the present invention, intelligent identification is performed on the uncertain part of the dynamic system model of the non-strictly repeating motion system; specifically, it includes: using neural networks, fuzzy systems, and adaptive identification technologies to model and estimate the relevant unknown parts or signals in the non-strictly repeating motion process of the system, and using the input and output measurement data of the system in each repeated time period to perform real-time training and adjustment on the neural network; obtaining system model knowledge and integrating the system model knowledge into the iterative learning controller.

[0045] Obtaining the estimation information of the non-strictly repeating motion system model, and integrating the estimation information of the controlled system model into the iterative learning controller to compensate for the iterative tracking error caused by the non-strictly repeating motion. Therefore, neural networks, fuzzy systems, and adaptive identification technologies are used to obtain the estimation information of the controlled system or signal.

[0046] For a dynamic system whose model is not accurately known, the present invention selects neural networks, fuzzy systems, and adaptive identification technologies, etc. to model and estimate the relevant unknown parts or signals in the system motion process, and uses the input and output measurement data of the system in each repeated time period to perform real-time training and adjustment on the neural network, etc. The design of the integrated intelligent controller includes both partial estimation information of the system model generated by neural networks and fuzzy systems, etc., and an iterative learning control mechanism.

[0047] In the embodiment of the present invention, constructing a comprehensive intelligent learning control law for the dynamic system to realize the switching of the control mode specifically includes: the construction of the control law is divided into two methods in the time period. When the time period of the next repeated motion is less than or equal to the time period of the previous repeated motion, iterative learning control is implemented in the entire time period of the next repeated motion; when the time period of the next repeated motion is greater than the time period of the previous repeated motion, the control in the entire time period of the next repeated motion is divided into two parts. The same time period part in the front adopts the iterative learning control method, and in the remaining time period part in the back, according to the system model information real-time identified in this repeated motion, a general tracking control method is adopted.

[0048] Under non-strictly repetitive motion conditions, a comprehensive intelligent learning control law for dynamic systems, specifically: for appropriate models of continuous and discrete dynamic systems, appropriate comprehensive intelligent learning control laws are constructed respectively. This comprehensive intelligent learning control law contains both partial estimation (identification) information of the system model generated by neural networks, fuzzy systems, adaptive identification techniques, etc., and an iterative learning control mechanism.

[0049] When the non-strictly repetitive motion conditions satisfy a certain variation law, the discussion of the comprehensive intelligent learning control problem of the dynamic system is taken as a special case. When the non-strictly repetitive motion conditions satisfy a certain variation law (such as the state connection conditions often existing in actual systems;

[0050] x k (T) = x k+1 (0), that is, the terminal state of the previous iteration run is equal to the initial state of the current iteration run), for parameterized or non-parameterized dynamic systems, we expect to derive a comprehensive intelligent learning control law with a more concise form and better control effect.

[0051] In the embodiments of the present invention, the iterative learning control adopted obtains a control input that can generate an expected output trajectory by repeatedly applying the information obtained from previous experiments to improve the control quality. Different from traditional control methods, iterative learning control can handle dynamic systems with a relatively high degree of uncertainty in a very simple way, and only requires less prior knowledge and computational effort. At the same time, it has strong adaptability and is easy to implement; more importantly, it does not depend on the exact mathematical model of the dynamic system and is an algorithm that iteratively generates an optimal input signal to make the system output as close as possible to the ideal value. Its research is of great significance for those problems with nonlinearity, complexity, difficult modeling, and high-precision trajectory control.

[0052] In the embodiments of the present invention, a simulation verification is carried out on the comprehensive learning controller of the non-strictly repetitive motion system; specifically including: simulating the comprehensive intelligent learning controller of the non-strictly repetitive motion system through the MATLAB simulation experiment platform, and conducting numerical simulation experiments according to the comprehensive intelligent learning control algorithm to verify the effectiveness of the comprehensive intelligent learning control method.

[0053] In the embodiments of the present invention, when the iterative initial error of the dynamic system changes boundedly, it is impossible for the system to perfectly track the expected trajectory from the initial point. Through the comprehensive intelligent learning controller, it can be ensured that the controlled system achieves ideal tracking of the expected trajectory outside a very small initial time period.

[0054] When the desired trajectory of a dynamic system changes boundedly with the number of iterations, on the premise that the desired trajectory can be achieved, the comprehensive intelligent learning controller can ensure the ideal tracking of the controlled system to the desired trajectory. The tracking accuracy depends on the identification error of the system by the intelligent identification technology.

[0055] When the length of the repeated motion time period of the dynamic system is not constant, the comprehensive intelligent learning controller with a switching function adopts different control strategies for the controlled system in different time periods to realize the organic switching combination of iterative learning control and conventional tracking control.

[0056] When the periods of the system's repeated motion are different, the control inputs of the system in different time periods cannot be learned and adjusted either. Since iterative learning control is often applied to the situation where the model of the controlled system is not fully known, when the strict repeated motion conditions cannot be met, more model knowledge is naturally needed for compensation. The present invention uses neural networks, fuzzy systems, adaptive identification technologies, etc. to model and estimate the relevant unknown parts or signals in the non-strict repeated motion process of the system, so as to obtain more system model knowledge and integrate it into the design of the iterative learning controller.

[0057] For the construction of the comprehensive intelligent learning controller, for different dynamic systems, an appropriate iterative learning control law is constructed, and the intelligent estimation information of model uncertainty is added, so that the iterative tracking error outside the initial time period converges to a very small range. For the case where the length of the repeated motion time period is not constant, the switching problem between iterative learning control and conventional tracking control also needs to be considered. This process is a repeated process and needs to be continuously corrected according to the proof requirements of tracking error convergence and experimental simulation conditions.

[0058] Iterative learning control in the sense of strict repeated motion can continuously learn functions from the control experience of the previous repeated operation. Therefore, its greatest advantage is that it does not require precise knowledge of the system model. When the repeated motion conditions of the system are no longer strict, in order to achieve good tracking of the desired trajectory of the system, it is necessary to rely on the model knowledge of the system and integrate the estimated information of the system model into the design of the iterative learning controller to cope with the non-strict repeated learning conditions; combine model estimation means such as artificial neural networks, fuzzy systems and adaptive technologies with iterative learning control to form a comprehensive intelligent learning control design.

[0059] The specific implementation process of the present invention is as follows:

[0060] (1) The present invention sets aside a small initial time period [0, h), constructs a small segment of trajectory on the initial time period [0, h) so that it naturally connects the initial state of each iteration and the (variable) desired trajectory point at time h, and the whole forms a new (variable) desired trajectory. Thus, the original problem is transformed into an iterative learning control problem of the (variable) desired trajectory under the condition of zero iterative initial error. Zero iterative initial error is of crucial significance for the convergence of iterative learning control. We aim to track the (variable) desired output trajectory in the time period [h, T].

[0061] (2) Only by accurately mastering the mathematical model of the system can we cope with the changes in the iterative initial state, desired trajectory, and tracking time period length of the non-strictly repetitive motion system. For a dynamic system with an inaccurately known model, the present invention selects neural networks, fuzzy systems, adaptive identification techniques, etc. to model and estimate the relevant unknown parts or signals during the system's motion process, and uses the input-output measurement data of the system in each repetitive time period to train and adjust the neural network, etc. in real time. The design of the comprehensive intelligent controller includes both partial estimation information of the system model generated by neural networks and fuzzy systems, etc., and an iterative learning control mechanism.

[0062] (3) The construction of the control law is divided into two methods in terms of time period. When the time period of the next repetitive motion is less than or equal to the time period of the previous repetitive motion, iterative learning control is implemented throughout the entire time period of the next repetitive motion; when the time period of the next repetitive motion is greater than the time period of the previous repetitive motion, the control throughout the entire time period of the next repetitive motion is divided into two parts. For the same time period part in the front, the iterative learning control method is adopted, and for the remaining time period part in the back, according to the system model information identified in real time during this repetitive motion, a general tracking control method is adopted.

[0063] After the above three steps (1)-(3), a comprehensive intelligent learning controller for the non-strictly repetitive motion system is theoretically designed. Then, with the help of the MATLAB simulation experiment platform, a suitable industrial robot model for non-strictly repetitive motion is selected, program codes are written according to the comprehensive intelligent learning control algorithm, and numerical simulation experiments are carried out to verify the effectiveness of the designed comprehensive intelligent learning control method.

[0064] As the number of repetitive iterations increases, the approximation degree of artificial neural networks and fuzzy systems, etc. to the real system continuously improves, and accordingly, the accuracy of the comprehensive intelligent learning control generated will also continuously improve. We expect to ultimately control the iterative learning tracking error of the system within a very small bound determined by the identification accuracy of intelligent tools such as neural networks.

[0065] See Figure 2 , the present invention also provides a comprehensive intelligent learning control system based on a non-strictly repetitive motion system, including:

[0066] The variable iteration initial error conversion unit 201 is used to convert the variable iteration initial error of the non-strictly repetitive motion system;

[0067] The intelligent identification technology unit 202 is used to intelligently identify the uncertain part of the dynamic system model of the non-strictly repetitive motion system;

[0068] The control switching unit 203 is used to construct a comprehensive intelligent learning control law for the dynamic system to realize the switching of the control mode;

[0069] The simulation verification unit 204 is used to simulate and verify the intelligent control method.

[0070] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Comprehensive intelligent learning control method based on a non-strictly repetitive motion system, Characterized in that, It includes the following steps: Transform the variable iteration initial error of the non-strictly repetitive motion system; Intelligently identify the uncertain part of the dynamic system model of the non-strictly repetitive motion system; Construct a comprehensive intelligent learning control law for the dynamic system to achieve the switching of control modes; Conduct simulation verification on the intelligent control method; The transformation of the variable iteration initial error of the non-strictly repetitive motion system specifically includes: Intercept an initial time period from the repetitive time period, construct a trajectory on the initial time period, so that it naturally connects the initial state of each iteration and the desired trajectory point at the end of the initial time period, form a new desired trajectory, and transform the bounded variable iteration initial error into the desired trajectory under the condition of zero iteration initial error.

2. The method according to claim 1, Characterized in that, The intelligent identification of the uncertain part of the dynamic system model of the non-strictly repetitive motion system specifically includes: Using neural networks, fuzzy systems and adaptive identification techniques, model and estimate the relevant unknown parts or signals in the non-strictly repetitive motion process of the system, and use the input and output measurement data of the system in each repetitive time period to train and adjust the neural network in real time; obtain system model knowledge and integrate the system model knowledge into the iterative learning controller.

3. The method according to claim 1, Characterized in that, The construction of the comprehensive intelligent learning control law for the dynamic system to achieve the switching of control modes specifically includes: The construction of the control law is divided into two methods in terms of time period. When the time period of the next repetitive motion is less than or equal to the time period of the previous repetitive motion, iterative learning control is implemented throughout the time period of the next repetitive motion; when the time period of the next repetitive motion is greater than the time period of the previous repetitive motion, the control throughout the time period of the next repetitive motion is divided into two parts. The same time period part in the front adopts the iterative learning control method, and in the remaining time period part in the back, according to the system model information real-time identified in this repetitive motion, a general tracking control method is adopted.

4. The method according to claim 1, Characterized in that, The simulation verification of the comprehensive learning controller for the non-strictly repetitive motion system specifically includes: Simulate the comprehensive intelligent learning controller of the non-strictly repetitive motion system through the MATLAB simulation experiment platform, conduct numerical simulation experiments according to the comprehensive intelligent learning control algorithm, and verify the effectiveness of the comprehensive intelligent learning control method.

5. Comprehensive intelligent learning control system based on a non-strictly repetitive motion system, Characterized in that, Adopt the method described in any one of claims 1-4, including: A variable iteration initial error transformation unit for transforming the variable iteration initial error of the non-strictly repetitive motion system; An intelligent identification technology unit for intelligently identifying the uncertain part of the dynamic system model of the non-strictly repetitive motion system; A control switching unit for constructing a comprehensive intelligent learning control law for the dynamic system to achieve the switching of control modes; A simulation verification unit for conducting simulation verification on the intelligent control method.

Images