A self-learning anti-disturbance control method

Abstract

The invention discloses a self-learning anti-disturbance control method. The method introduces a self-learning mechanism into an extended state observer and a state error feedback device in a self-disturbance anti-control system respectively to construct a self-learning anti-disturbance control system structure, including: establishing a self-learning extended state observer algorithm, introducing an extended state equation into a classical linear and nonlinear extended state observer, wherein the gain of the extended state equation changes dynamically with an estimated error, thereby reducing or eliminating an initial peak phenomenon; establishing a self-learning state error feedback algorithm, adding a self-learning term in series or in parallel on the basis of a proportional differential control of a classical state error feedback device, wherein the self-learning term includes a self-learning control error function and a first-order low-pass filter, which is used to reduce a following error and improve control stability; the invention can alleviate and eliminate the peak phenomenon while realizing a rapid and accurate estimation of a total disturbance, and reduce a following error while realizing a rapid following of a set value.

Description

Technical Field

[0001] The invention relates to the technical field of industrial self-disturbance rejection control, and in particular to a self-learning anti-disturbance control method. Background Art

[0002] Disturbance rejection control understands control systems and control methods from the perspective of disturbance rejection, and constructs specific control algorithms based on this. Active Disturbance Rejection Control (ADCC) is a typical disturbance rejection control method and has been widely used in industrial control. Figure 1 As shown in the figure, the traditional ADRC consists of a tracking differentiator (TD), an extended state observer (ESO) and a state error feedback (SEF), where ESO realizes the estimation and disturbance rejection of the unmodeled dynamics of the controlled system and the external disturbance of the system, that is, the total disturbance of the system, TD realizes the softening of the set value, and SEF realizes the disturbance rejection and following of the system output to the set value.

[0003] The selection of bandwidth of existing linear and nonlinear ESO requires a compromise between total disturbance estimation and initial peak phenomenon, which increases the workload of parameter tuning in practical applications; classical linear and nonlinear SEF have the form of PD, which can improve the convergence speed of the control system, but has no contribution to the type of control system. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art. A self-learning anti-disturbance control method is proposed. The estimation accuracy and peak phenomenon of the total disturbance are solved by modifying the bandwidth change of ESO at different errors. The SEF is improved by increasing the quasi-integral, thereby improving the anti-disturbance control performance of the entire control system. The state estimation error of SL-ESO and the output error information of SL-SEF are used to construct a special nonlinear integral function to efficiently learn from historical error information to improve the estimation and control performance. While realizing fast and accurate estimation of the total disturbance, the peak phenomenon is alleviated and eliminated; while realizing fast following of the set value, the following error is reduced.

[0005] The purpose of the present invention is achieved through the following technical solution: a self-learning anti-disturbance control method, which introduces a self-learning mechanism into an extended state observer ESO and a state error feedback SEF in an automatic anti-disturbance control system ADRC, respectively, to construct a self-learning anti-disturbance control system structure, including:

[0006] A self-learning extended state observer algorithm SL-ESO is established, and an extended state equation is introduced into the classical linear and nonlinear extended state observer ESO, where the gain of the extended state equation changes dynamically with the estimation error, thereby reducing or eliminating the initial peak phenomenon;

[0007] A self-learning state error feedback algorithm SL-SEF is established. Self-learning items are added in series or parallel on the basis of the proportional differential control PD of the classic state error feedback SEF. The self-learning items include a self-learning control error function and a first-order low-pass filter to reduce following error and improve control stability.

[0008] Further, the establishment of a self-learning extended state observer algorithm SL-ESO includes:

[0009] The extended state equation is introduced for the classical linear extended state observer and the nonlinear extended state observer ESO. , the expanded state equation is shown in the following formula (1);

[0010] (1);

[0011] is the exponential of the error. Formula (1) is weighted according to the exponential function of the error. , is the self-learning rate of SL-ESO, , the first term of formula (1) is The dynamic attenuation term improves the stability of the extended state observer ESO. At the same time, this term and the subsequent integral term form a dynamic weighting coefficient with a sum of 1. The absolute value of is used to dynamically adjust the weighting coefficient, realizing partial quasi-integral effect, and obtaining the total disturbance from the historical error through the self-learning mechanism according to the estimated error.

[0012] Further, the establishment of the self-learning state error feedback algorithm SL-SEF includes:

[0013] Based on the classic nonlinear SEF and classic linear SEF, a self-learning term is added to the PD control of the state error feedback SEF. , respectively realizing the SL-SEF structure in series and realizing the SL-SEF structure in parallel;

[0014] is the self-learning control error function, It is a non-increasing function of the absolute value of the error not greater than 1, as shown in the following formula (2):

[0015] (2);

[0016] in is the self-learning coefficient; when hour, , fully learn the historical errors; when The bigger the The smaller it is, the smaller the control amount learned from the tracking error is, which avoids over-learning of errors caused by external disturbances and improves stability.

[0017] is a first-order low-pass filter, s is the differential operator in the Lagrange transform, and is expressed in the form of sampling control as shown in the following formula (3):

[0018] (3);

[0019] in is the inertia time constant, is the sampling period; Reduced reliance on historical errors and increased learning from current errors; The error is self-learning The accumulation of effective errors after the relevant structure processing;

[0020] when When , in the SL-SEF structure in series arrive The transfer function is shown in the following formula (4):

[0021] (4);

[0022] In the parallel SL-SEF structure, arrive The transfer function is shown in the following formula (5):

[0023] (5);

[0024] when When , the series SL-SEF structure becomes a series PID, and the parallel SL-SEF structure becomes a parallel PID; when , SL-SEF automatically quits the pure integral action, that is, the self-learning mechanism with automatic exit from the pure integral action is used to overcome the side effects of integration. The integral effect is enhanced nearby to improve the tracking accuracy of the control system;

[0025] Then the control quantity output by the series SL-SEF structure is as shown in the following formula (6):

[0026] (6);

[0027] The control quantity output by the parallel SL-SEF structure is shown in the following formula (7):

[0028] (7);

[0029] When setting the parameters in formula (6) and formula (7), the The effect of bandwidth is set according to the bandwidth method. , , …, Parameters, then add , setting , and ,choose , and , is the sampling period.

[0030] A non-transitory computer-readable medium storing instructions, when the instructions are executed by a processor, performs the steps according to the above self-learning anti-disturbance control method.

[0031] A computing device comprises a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, the above-mentioned self-learning anti-disturbance control method is implemented.

[0032] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0033] 1. The present invention adds a self-learning mechanism to the linear and nonlinear extended state observers ESO, realizes automatic adjustment of dynamic gain, and reduces the initial peak phenomenon of the extended state observer ESO;

[0034] 2. The present invention adds a self-learning item to the state error feedback device SEF to obtain the self-learning state error feedback SL-SEF, thereby improving the type of the control system, and improving the steady-state accuracy while ensuring the tracking speed;

[0035] 3. The present invention adopts the self-learning state error feedback algorithm SL-SEF and the self-learning extended state observer algorithm SL-ESO in the linear self-learning anti-disturbance control system structure ADRC to obtain the performance of nonlinear control, but at the same time has the convenience of parameter setting of linear ADRC.

[0036] 4. The present invention improves the adaptability of linear ESO. Even if the first n equations of SL-ESO are kept in the form of linear equations, and the n+1th equation of SL-ESO adopts the self-learning algorithm with nonlinear characteristics of the present invention, satisfactory performance can be obtained. The commonly used bandwidth method can still be used to adjust the parameters of ESO, which is convenient for understanding and practical engineering application. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is the schematic diagram of the existing ADRC structure.

[0038] Figure 2 It is a structural principle diagram of the present invention.

[0039] Figure 3 This is the structural principle diagram of SL-ESO.

[0040] Figure 4 This is the structural principle diagram of the series SL-SEF.

[0041] Figure 5 This is the structural principle diagram of the parallel SL-SEF. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with specific embodiments.

[0043] Example 1

[0044] The self-learning anti-disturbance control method provided in this embodiment is to introduce a self-learning mechanism into an extended state observer ESO and a state error feedback SEF in an automatic anti-disturbance control system ADRC, respectively, to construct a self-learning anti-disturbance control system structure, and the self-learning anti-disturbance control system is applied to control a controlled object, and the controlled object is applicable to general motion control objects and process control objects, such as motors, magnetic suspension systems, liquid level systems, etc. The method specifically includes:

[0045] A self-learning extended state observer algorithm SL-ESO is established, and an extended state equation is introduced into the linear and nonlinear extended state observers ESO, where the gain of the extended state equation changes dynamically with the estimation error, thereby reducing or eliminating the initial peak phenomenon;

[0046] A self-learning state error feedback algorithm SL-SEF is established. Self-learning items are added in series or parallel on the basis of the proportional differential control PD of the state error feedback device SEF. The self-learning items include a self-learning control error function and a first-order low-pass filter to reduce following errors and improve control stability.

[0047] See also Figure 2 As shown, the structure of the self-learning anti-disturbance control system and the system description of the classic self-disturbance rejection control are as follows:

[0048] (1);

[0049] in, is the system order, It's time. is the control gain, is the system status, is the first-order derivative of the system state, , is the system output, and are the internal disturbance and external disturbance of the system, is the control input of the system. According to the classical ADRC theory, a variable representing the sum of internal and external disturbances is introduced as the expanded state, that is, the total disturbance of the system is represented by the expanded state. Then formula (1) is written as the equation with the expanded state as follows:

[0050] (2);

[0051] in is the expanded state representing the total disturbance of the system, is a rough estimate of the given control gain;

[0052] Tracking Differentiator TD is used to arrange the transition process of the input and obtain the input Approximate differentials of various orders ;

[0053] The classic nonlinear extended state observer ESO is shown in the following formula (3):

[0054] (3);

[0055] in, is a symbolic function; , , …, and In order to make the ESO a stable positive number, the bandwidth method is used for tuning; , , …, and are the estimated values ​​of the state and expansion state of ESO, respectively; , ; and is the parameter of the fal function, satisfying ,when When , the classical nonlinear extended state observer ESO of formula (3) becomes the classical linear extended state observer ESO of formula (4), that is:

[0056] (4);

[0057] When using the classical nonlinear SEF, the control quantity of ADRC is shown in the following formula (5):

[0058] (5);

[0059] in , , …, To make the closed-loop system stable, and is the parameter of the fal function, satisfying ; is the reference input obtained by tracking the input of the differentiator TD Approximate differentials of various orders.

[0060] When it is a classic linear SEF, it is as shown in the following formula (6):

[0061] (6)

[0062] in , , …, The bandwidth method can be used to adjust the , , …, , , for Approximate differential of , j =2, …,n.

[0063] In order to achieve accurate estimation of the total disturbance by ESO, the gain of the classical ESO ( ) are often large, with a more serious initial peak phenomenon, and it is difficult to adapt to the total disturbance of a large range of changes. In order to solve these problems encountered in the practical application of ESO, see Figure 3 As shown, M21 is the linear ESO The equation part can also be replaced by nonlinear ESO; the extended state equation M22 is introduced to establish the self-learning extended state observer algorithm SL-ESO, including:

[0064] The extended state equation M22 is introduced into formula (3) and formula (4) to calculate , the expanded state equation is shown in the following formula (7);

[0065] (7);

[0066] in is the exponential of the error. Formula (7) is weighted according to the exponential function of the error. , is the self-learning rate of SL-ESO, , formula (7) adds The dynamic attenuation term, i.e. the first term of formula (7), improves the stability of the extended state observer ESO. At the same time, this term and the subsequent integral term form a dynamic weighting coefficient whose sum is 1. The absolute value of the weighting coefficient is used to dynamically adjust the weighting coefficient, realizing partial quasi-integral effect, and obtaining the total disturbance from the historical error through the self-learning mechanism according to the estimated error; The nonlinear function is used as the weighting coefficient. Even if the linear gain coefficient is used in M21, the effect of nonlinear ESO can be achieved, which is convenient for the parameter setting of ESO.

[0067] In order to eliminate the side effects of the integral term in PID, the SEF of the classic ADRC uses linear or nonlinear proportional-derivative (PD). The classic nonlinear SEF is shown in formula (5), and the classic linear SEF is shown in formula (6).

[0068] In order to improve the SEF's setpoint steady-state tracking capability, a self-learning item is added based on the SEF's PD. Figure 4 As shown, it is the SL-SEF structure realized in series; see Figure 5 As shown, it is the SL-SEF structure realized in parallel. Figure 4 and Figure 5 Classic-SEF represents the classic SEF algorithm.

[0069] Adding self-learning items based on PD control of state error feedback device SEF , respectively realizing the SL-SEF structure in series and realizing the SL-SEF structure in parallel;

[0070] is the self-learning control error function, is a non-increasing function of the absolute value of the error not greater than 1, as shown in the following formula (8):

[0071] (8);

[0072] in is the self-learning coefficient; when hour, , fully learn the historical errors; when The bigger the The smaller it is, the smaller the control amount learned from the tracking error is, which avoids over-learning of errors caused by external disturbances and improves stability.

[0073] is a first-order low-pass filter, s is the differential operator in the Lagrange transform, and is expressed in the form of sampling control as shown in the following formula (9):

[0074] (9);

[0075] in is the inertia time constant, is the sampling period; Reduced reliance on historical errors and increased learning from current errors; The error is self-learning The accumulation of effective errors after the relevant structure processing;

[0076] when When , in the SL-SEF structure in series arrive The transfer function is shown in the following formula (10):

[0077] (10);

[0078] In the parallel SL-SEF structure, arrive The transfer function is shown in the following formula (11):

[0079] (11);

[0080] when When , the series SL-SEF structure becomes a series PID, and the parallel SL-SEF structure becomes a parallel PID; when , SL-SEF automatically quits the pure integral action, that is, the self-learning mechanism with automatic exit from the pure integral action is used to overcome the side effects of integration. The integral effect is enhanced nearby to improve the tracking accuracy of the control system;

[0081] The control quantity output by the series SL-SEF structure is as shown in the following formula (12):

[0082] (12);

[0083] The control quantity output by the parallel SL-SEF structure is shown in the following formula (13):

[0084] (13);

[0085] When setting the parameters in formula (12) and formula (13), the The effect of bandwidth is set according to the bandwidth method. , , …, Parameters, then add , setting , and ,choose , and , is the sampling period.

[0086] Example 2

[0087] This embodiment discloses a non-transitory computer-readable medium storing instructions. When the instructions are executed by a processor, the steps of the self-learning anti-disturbance control method according to Embodiment 1 are executed.

[0088] The non-temporary computer-readable medium in this embodiment may be a disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), a USB flash drive, a mobile hard disk, or other media.

[0089] Example 3

[0090] This embodiment discloses a computing device, including a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, the self-learning anti-disturbance control method described in Embodiment 1 is implemented.

[0091] The computing device described in this embodiment may be a desktop computer, a laptop computer, a smart phone, a PDA handheld terminal, a tablet computer, a programmable logic controller (PLC), or other terminal devices with a processor function.

[0092] The embodiments described above are only preferred embodiments of the present invention and are not intended to limit the scope of implementation of the present invention. Therefore, all changes made according to the shape and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A self-learning anti-disturbance control method, characterized in that: The method introduces self-learning mechanism into the extended state observer ESO and the state error feedback SEF in the ADRC to construct the self-learning anti-disturbance control system structure, including: A self-learning extended state observer algorithm SL-ESO is established. The extended state equation is introduced into the classical linear and nonlinear extended state observer ESO. The gain of the extended state equation changes dynamically with the estimation error, thereby reducing or eliminating the initial peak phenomenon, including: The extended state equation is introduced for the classical linear extended state observer and the nonlinear extended state observer ESO. , the expanded state equation is shown in the following formula (1); (1)； In order to make the ESO a stable positive number, the bandwidth method is used for adjustment. is the system order, is the exponential of the error. Formula (1) is weighted according to the exponential function of the error. , is the self-learning rate of SL-ESO, , the first term of formula (1) is The dynamic attenuation term improves the stability of the extended state observer ESO. At the same time, this term and the subsequent integral term form a dynamic weighting coefficient with a sum of 1. The absolute value of is used to dynamically adjust the weighting coefficient, realizing partial quasi-integral effect, and obtaining the total disturbance from the historical error according to the estimated error through the self-learning mechanism; A self-learning state error feedback algorithm SL-SEF is established. On the basis of the proportional differential control PD of the classic state error feedback SEF, a self-learning term is added in series or parallel. The self-learning term includes a self-learning control error function and a first-order low-pass filter to reduce the following error and improve the control stability, including: Based on the classic nonlinear SEF and classic linear SEF, a self-learning term is added to the PD control of the state error feedback SEF. , respectively realizing the SL-SEF structure in series and realizing the SL-SEF structure in parallel; is the self-learning control error function, It is a non-increasing function of the absolute value of the error not greater than 1, as shown in the following formula (2): (2)； in is the self-learning coefficient; when hour, , fully learn the historical errors; when The bigger it is, The smaller it is, the smaller the control amount learned from the tracking error is, which avoids over-learning of errors caused by external disturbances and improves stability. is a first-order low-pass filter, s is the differential operator in the Lagrange transform, and is expressed in the form of sampling control as shown in the following formula (3): (3)； in is the inertia time constant, is the sampling period; Reduced reliance on historical errors and increased learning from current errors; The error is self-learning The accumulation of effective errors after the relevant structure processing; when When , in the SL-SEF structure in series arrive The transfer function is shown in the following formula (4): (4)； In the parallel SL-SEF structure, arrive The transfer function is shown in the following formula (5): (5)； when When , the series SL-SEF structure becomes a series PID, and the parallel SL-SEF structure becomes a parallel PID; when , SL-SEF automatically quits the pure integral action, that is, the self-learning mechanism with automatic exit from the pure integral action is used to overcome the side effects of integration. The integral effect is enhanced nearby to improve the tracking accuracy of the control system; Then the control quantity output by the series SL-SEF structure is as shown in the following formula (6): (6)； The control quantity output by the parallel SL-SEF structure is shown in the following formula (7): (7)； When setting the parameters in formula (6) and formula (7), the The effect of bandwidth is set according to the bandwidth method. , , …, Parameters, then add , setting , and ,choose , and , is the sampling period, is a rough estimate of the given control gain.

2. A non-transitory computer-readable medium storing instructions, characterized in that: When the instruction is executed by a processor, the steps of the self-learning anti-disturbance control method according to claim 1 are performed.

3. A computing device comprising a processor and a memory for storing a program executable by the processor, characterized in that: When the processor executes the program stored in the memory, the self-learning anti-disturbance control method described in claim 1 is implemented.

Images