A dual-inertia system position control method based on fusion model and data-driven

Abstract

The present invention discloses a dual-inertia system position control method that integrates a model and data drive. The method reduces the order of the system into a slow-changing subsystem and a fast-changing subsystem through a singular perturbation method, and obtains a quasi-steady-state model of a slow time scale and a boundary layer model of a fast time scale. In the design of the slow-changing subsystem controller, the present invention proposes a two-degree-of-freedom ADRC position control method based on DESO to achieve the decoupling of the system tracking performance and anti-disturbance performance. For the fast-changing subsystem, the present invention proposes a neural network PD control strategy that only relies on the input and output information of the system to update the neural network weights online, so as to achieve rapid suppression of system vibration when the stiffness is uncertain. Therefore, the advantages of the present invention are that the reference tracking performance and anti-disturbance performance of the controller are decoupled, the controller design does not rely on model parameters, the structure is simple, and there are few parameters to be adjusted.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor control, and in particular relates to a dual-inertia system position control method that integrates a model and data drive. Background Art

[0002] In industrial fields such as robots and machine tools, the Two Mass System (TMS) is a common model that describes the connection between the motor shaft and the load through a flexible transmission mechanism such as a gearbox, reducer, and belt. The flexible transmission mechanism makes the TMS a fourth-order under-actuated system and causes resonance points in the system, which seriously affects the positioning accuracy on the load side. In addition, unknown disturbances caused by modeling errors, external interference, and load changes will have a significant impact on the positioning accuracy of the under-actuated TMS load side. How to design a reasonable controller that can effectively suppress system vibration while timely compensating for unknown disturbances is the key to studying the TMS position control problem.

[0003] In recent years, singular perturbation method has been widely used in controller design of TMS. Its core idea is to separate the time scale of system dynamics, obtain a quasi-steady-state model for tracking control and a boundary layer model for vibration suppression. Active Disturbance Rejection Control (ADRC) has become a practical control method to replace classical PID control due to its strong anti-disturbance ability and independence from accurate models. It can be used in controller design for quasi-steady-state models. As the core part of ADRC, Extended State Observer (ESO) uniformly equates internal and external disturbances of the system to lumped disturbances for observation and compensation, thereby improving the disturbance suppression performance of the system; however, the traditional position loop ADRC controller belongs to a one-degree-of-freedom controller, and the tracking performance and anti-disturbance performance of the system are coupled.

[0004] The active damping method is a simple and effective method in the design of boundary layer model controller. This method improves the damping ratio of the boundary layer model based on the feedback compensation of the system transmission torque differential, thereby suppressing the vibration of the system. This method relies on the accurate acquisition of the transmission torque. In the scenario where the torque sensor is not installed and the dual encoder is used, it depends on the accurate acquisition of the stiffness of the flexible mechanism. However, in actual application scenarios, the stiffness calibration of the system is often a difficult task, and it is inevitable that the stiffness of the flexible mechanism will change due to system aging. Therefore, the difficulty in obtaining accurate stiffness parameters limits the application of the active damping method. Summary of the invention

[0005] In view of the above, the present invention provides a dual-inertia system position control method that integrates a model and data-driven fusion, which can compensate for unknown disturbances while completely decoupling the system tracking performance and anti-disturbance performance, achieve high-precision position control, and effectively suppress system vibration without knowing the system stiffness parameters.

[0006] A dual-inertia system position control method based on a fusion model and data drive comprises the following steps:

[0007] (1) The dual-inertia system is reduced to a slow-varying subsystem and a fast-varying subsystem by the singular perturbation method;

[0008] (2) Reconstructing the slow-changing subsystem and the fast-changing subsystem;

[0009] (3) Based on the reconstructed slow-varying subsystem, its controller is designed as a two-degree-of-freedom ADRC controller based on DESO (Decoupling Extended State Observer);

[0010] (4) Based on the reconstructed fast-changing subsystem, its controller is designed as a PD (proportional differential) controller based on data drive and neural network;

[0011] (5) The output torque of the slow-changing subsystem controller u s The output torque of the fast-changing subsystem controller u f Add to get the reference torque u , according to the reference torque u The load side position is controlled in a closed loop through a dual inertia system.

[0012] Furthermore, the model expression of the slow-varying subsystem is as follows:

[0013] ;

[0014] ;

[0015]

[0016] in: J m and J L are the rotational inertia of the motor side and the load side of the dual inertia system, B m and B L are the damping coefficients of the motor side and the load side of the dual inertia system, T L is the load torque, is the quasi-steady-state value of the transmission torque, and They are i L The second and first order differentials of i L is the position of the load side of the dual inertia system, d is an intermediate variable.

[0017] Furthermore, the model expression of the fast-changing subsystem is as follows:

[0018]

[0019] in: k 0= K s e 2 , K s is the spring stiffness coefficient, e is the singular perturbation parameter (0< e <1), or is the transmission torque error, , T r is the transmission torque T r = K s ( i m - i L ), i m is the position of the motor side of the dual inertia system, and They are or On fast time scales t ε The second and first order differentials under d t ε / d t =1 / e , t Indicates time.

[0020] Furthermore, the specific implementation method of step (2) is as follows: the flexible compensation gain is unified into the controller parameter design of the slow-changing subsystem and the fast-changing subsystem respectively, and the flexible compensation term is taken as a part of the fast-changing subsystem and combined with the damping compensation term to form a PD controller with the transmission torque as input. The controller outputs of the reconstructed slow-changing subsystem and the fast-changing subsystem are expressed as follows:

[0021]

[0022] in: i d is a given reference position instruction, k sp and k sd are the proportional coefficient and differential coefficient of the reconstructed slow-varying subsystem controller, k fp and k fd are the proportional coefficient and differential coefficient of the reconstructed fast-changing subsystem controller, , , They are i d , i m , i L The first-order differential of .

[0023] Furthermore, the two-degree-of-freedom ADRC controller in step (3) includes a tracking differentiator, a PD controller and a DESO, wherein the tracking differentiator is used to convert a step-change reference position instruction into a smoothly changing reference position signal, and provide the reference position signal and its differentiated reference speed signal to the PD controller; the PD controller uses the load side position and speed estimated by the DESO as feedback signals, and adjusts the motor output torque so that the load side position follows the reference position instruction; the DESO is used to estimate the load side position, speed and lumped disturbance, and suppress internal and external disturbances by compensating for the lumped disturbance. The discrete equation of the DESO is as follows:

[0024]

[0025] in: and Respectively k +1 control cycle and k The load side position estimate for the control cycle, and Respectively k +1 control cycle and k The load-side speed estimate for the control cycle, and Respectively k +1 control cycle and k The lumped disturbance estimate for the control period, , h 1. h 2. h 3 is the adjustment parameter of DESO, z express z Transformation operator, T sTo control the cycle length, i L ( k ) is the k The actual load-side position value of the control cycle, u s ( k ) is the k The output torque of the controller of the slowly changing subsystem with a control period, J n =J m +J L , k is a natural number.

[0026] Furthermore, the output of the two-degree-of-freedom ADRC controller is expressed as follows:

[0027]

[0028] in: u s ( k ) is the k The output torque of the controller of the slowly changing subsystem with a control period, i ref ( k )and oh ref ( k ) are respectively k Control cycle reference position signal and reference speed signal.

[0029] Furthermore, the PD controller in step (4) updates the neural network weights online according to the input and output information of the dual-inertia system to achieve rapid suppression of system vibration. The output of the PD controller is expressed as follows:

[0030]

[0031] in: u f ( k ) is the k The output torque of the subsystem controller with fast control period, X and a are the input layer and hidden layer of the neural network, respectively. a = VX =[ a 1( k ), a 2( k )] T , V is the hidden layer weight and , Wi ( k )for a i ( k ) corresponds to the output layer weight, X =[ e ( k ), e ( k- 1)] T , e ( k- 1)= i L ( k- 1)- i m ( k- 1), e ( k )= i L ( k )- i m ( k ), i L ( k- 1) and i L ( k ) are respectively k- 1 Control cycle and k The actual load-side position value of the control cycle, i m ( k- 1) and i m ( k ) are respectively k- 1 Control cycle and k The actual position value of the motor side during the control cycle, T Indicates transpose.

[0032] Furthermore, the loss function of the neural network is as follows:

[0033]

[0034] in: L represents the loss function.

[0035] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the above-mentioned fusion model and data-driven dual-inertia system position control method.

[0036] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements the above-mentioned fusion model and data-driven dual-inertia system position control method.

[0037] Based on the above technical solution, the present invention has the following beneficial technical effects:

[0038] 1. The decoupled two-degree-of-freedom auto-disturbance rejection position control method for the slow-varying subsystem of the present invention can independently optimize the system reference tracking performance and disturbance rejection performance by adjusting the controller parameters and the disturbance observer parameters.

[0039] 2. The neural network PD control strategy of the fast-changing subsystem of the present invention is a data-driven model-free control that can update controller parameters online to achieve effective suppression of system vibration under variable stiffness.

[0040] 3. The present invention makes full use of known model parameters and available data information. It is a control method that integrates model and data drive and has a simpler structure and fewer parameters to be adjusted than conventional intelligent algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of position-current dual closed-loop control of a dual-inertia system in an embodiment of the present invention.

[0042] Figure 2 It is a position control block diagram of the dual inertia system and the slow and fast changing subsystem controllers in an embodiment of the present invention.

[0043] Figure 3 Schematic diagram of experimental waveforms of traditional ADRC and decoupled two-degree-of-freedom ADRC of the present invention under position command step in the verification example of the present invention, wherein (a) corresponds to traditional ADRC, and (b) corresponds to decoupled two-degree-of-freedom ADRC of the present invention.

[0044] Figure 4 Schematic diagram of experimental waveforms of traditional ADRC and decoupled two-degree-of-freedom ADRC of the present invention under sudden change of load torque in the verification example of the present invention, wherein (a) corresponds to traditional ADRC, and (b) corresponds to decoupled two-degree-of-freedom ADRC of the present invention.

[0045] Figure 5 It is a schematic diagram of the experimental waveform of the neural network PD controller based on the present invention under the position command step in the verification example of the present invention. DETAILED DESCRIPTION

[0046] In order to describe the present invention more specifically, the technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0047] In this embodiment, a machine with an inertia of 1.2×10 -4kg·m 2 The motor under test is compared with a motor with an inertia of 0.8×10 - 4 kg·m 2 The load motor is connected by an elastic coupling with a stiffness of 5Nm / rad to simulate a dual-inertia system and perform closed-loop control on the load side position.

[0048] like Figure 1 As shown in the figure, the dual-inertia system adopts a dual-loop control framework based on the magnetic field oriented control strategy, which generally includes a position loop and a current loop control structure. The position loop is based on the reference position signal. i d , Motor side position feedback i m With load side position feedback i L , produces the desired reference torque u , through the torque coefficient K T Converted to the reference input of the current loop i ref ; The current loop samples the motor stator current i The output reference voltage is controlled, and then the inverter output voltage is adjusted through SVPWM (Space Vector Pulse Width Modulation) modulation technology to drive the motor. U DC is the DC voltage of the inverter. When the current loop bandwidth is high enough, the output torque T m Able to fully track the reference torque in a very short time u Therefore, the dynamics of the current loop is ignored when designing the position loop. u=T m .

[0049] In order to realize closed-loop control of the load side position of the dual-inertia system, this embodiment provides a dual-inertia system position control method that integrates a model and data drive, and the specific process is as follows:

[0050] (1) The dual-inertia system is reduced to a slow-varying subsystem and a fast-varying subsystem by the singular perturbation method.

[0051] 1.1 The dynamic model of the dual-inertia system is established as follows:

[0052]

[0053] in: i m and i L are the positions of the motor side and the load side respectively,J m and B m are the moment of inertia and damping coefficient of the motor side, J L and B L are the moment of inertia and damping coefficient on the load side, T m is the motor output torque, T L is the load torque, K s is the spring stiffness coefficient, and They are i L The second and first order differentials of and They are i m The second and first order differentials of .

[0054] 1.2 Definition of transmission torque T r = K s ( i m - i L ), thus converting the above dual-inertia system dynamics model into a singular perturbation form as follows:

[0055]

[0056] in: u To generate the desired reference torque, and They are T r The second and first order differentials of e is the singular perturbation parameter (0< e <1), k 0= K s e 2 .

[0057] make e = 0 makes the dual inertia system in a slow time scale and obtains the quasi-steady-state value of the transmission torque And the quasi-steady-state model of the slow-varying subsystem is as follows:

[0058] ;

[0059]

[0060] in: J n = J m +J L , B n = B m +B L , u s is the output of the slow-changing subsystem controller.

[0061] 1.3 Convert the dual-inertia system dynamics model in the singular perturbation form into i L The form of expression is as follows:

[0062]

[0063] in: , define a new scale variable t ε And let d t ε / d t =1 / e . In this fast-changing time scale, Treated as a constant, defining a variable And there is , the dual-inertia system model in singular perturbation form can be expressed as:

[0064]

[0065] make Right now or = 0, the slow-varying subsystem model under the slow-varying time scale is obtained as follows:

[0066]

[0067] Subtract the slow-changing subsystem model from the dual-inertia system model in the form of singular perturbation, and the boundary layer model of the fast-changing subsystem is obtained as follows:

[0068]

[0069] in: u f is the output of the fast-changing subsystem controller.

[0070] (2) Reconstruct the structure of the traditional singular perturbation method.

[0071] The traditional singular perturbation method slow-changing subsystem controller is a PD control law structure, and the fast-changing subsystem is designed according to the damping compensation method to obtain a D controller. This implementation method unifies the flexible compensation gain into the controller parameter design of the slow and fast-changing subsystems respectively, and at the same time incorporates the flexible compensation term as part of the fast-changing subsystem, and combines it with the damping compensation term to form a PD controller with the transmission torque as input, and obtains the output of the reconstructed slow and fast-changing subsystems u s , u f The expression is as follows:

[0072]

[0073] in: i d is the reference position instruction; k sp , k sd are the parameters of the reconstructed slow-varying subsystem PD controller; k fp , k fd are the parameters of the reconstructed fast-changing subsystem PD controller.

[0074] (3) Based on the reconstructed singular perturbation method structure, a completely decoupled two-degree-of-freedom active disturbance rejection controller (ADRC) is designed for the slow-varying subsystem.

[0075] The decoupled two-degree-of-freedom ADRC consists of three parts, including a tracking differentiator, a PD controller, and a decoupled extended state observer DESO. The tracking differentiator generates a transition process for the reference position command, making the motor torque change smoother, reducing the vibration and overshoot of the system, and providing the PD controller with a reference position signal and its differentiated reference speed signal; the PD controller uses the position signal and speed signal estimated by the DESO as feedback to adjust the motor output torque so that the load side position follows the reference position command; the DESO is used to estimate the position, speed and lumped disturbance of the system, and suppress internal and external disturbances by compensating for the lumped disturbance. Its discrete equation is:

[0076]

[0077] in: , , are the load side position, load side speed and lumped disturbance in the first k An estimate of the control period; T s To control the cycle; h 1. h 2. h3 is the adjustment parameter of DESO. In this embodiment, T s =5×10 -5 , design observer bandwidth based on bandwidth parameter tuning method oh o =600, h 1=3 oh o , h 2=3 oh o 2 , h 3= oh o 3 .

[0078] The expression of decoupled two-degree-of-freedom ADRC output is:

[0079]

[0080] in: k sp , k sd are the parameters of the PD controller, J n =J m +J L =2×10 -4 In this embodiment k sp =2, k sd =0.2.

[0081] The tracking performance of the decoupled two-degree-of-freedom ADRC is only related to the PD controller and has nothing to do with DESO. The anti-disturbance performance is only related to DESO and has nothing to do with the PD controller. The tracking performance and anti-disturbance performance are completely decoupled. By adjusting the PD controller parameters and DESO parameters, the tracking performance and anti-disturbance performance of the system position command can be optimized separately.

[0082] (4) Based on the reconstructed singular perturbation method structure, a data-driven neural network PD controller is designed for the fast-changing subsystem.

[0083] The data-driven neural network PD controller is improved on the basis of the original fast-changing subsystem PD controller. The neural network weights are updated online according to the input and output information of the system to achieve rapid suppression of system vibration without relying on accurate model parameters. The expression of the neural network PD controller is:

[0084]

[0085] in: X =[ e ( k ), e ( k- 1)] T is the input of the neural network, e ( k )= i L ( k )- i m ( k ) indicates the k The position difference between the load side and the motor side of the control cycle, is the weight of the hidden layer, a = VX is the output of the hidden layer, W is the output layer weight, and its initial value W (0)=[ k fp , k fd ]; In this embodiment k fp =10, k fd =0.2.

[0086] The data-driven neural network PD controller adopts an online training method. In each control cycle, the loss function of the neural network is calculated according to the error between the actual position and the reference position. The neural network is back-propagated through the gradient descent method, and the network output layer weight is updated until the loss function is minimized. The formula for online training of the neural network PD controller is as follows:

[0087] ;

[0088] ;

[0089] ;

[0090]

[0091] in: P ( k ) indicates the k The loss function of the iteration is, is the pseudo partial derivative used to fit the output and input relationship of the controlled system, Y ( k )= i m ( k )- i L ( k ), r is a given learning rate coefficient; in this embodiment r =0.01.

[0092] After completing the design of the slow and fast changing subsystem controllers, the fusion model and data driven dual inertia system position control structure are as follows: Figure 2 As shown in the figure, the tracking differentiator generates a transition process for the position command. The PD controller uses the output of the tracking differentiator as a reference. The position and velocity signals estimated by the observer are used as feedback to determine the tracking performance of the controller. The lumped disturbance estimated by the observer is used as disturbance compensation to determine the anti-disturbance performance of the controller.

[0093] Figure 3 The experimental waveforms of the traditional ADRC and the decoupled two-degree-of-freedom ADRC of the present invention are shown under the given position command with a transition time of 150ms and an amplitude of π rad. In actual scenarios, the dual-inertia system must have unknown disturbances such as friction and system parameter perturbation during the motion process. Figure 3 As can be seen from (a), although the ESO of the traditional ADRC can suppress the disturbance that occurs when the system tracks the reference position, the disturbance information will be transmitted to the estimated and This will affect the tracking performance of the PD controller. Figure 3 (b) shows that DESO effectively suppresses system disturbances while estimating and Does not contain unknown disturbance information, and the quality of disturbance suppression will not affect and The estimated value of will not affect the tracking performance of the PD controller; this shows that under the decoupled two-degree-of-freedom ADRC structure of the present invention, the disturbance suppression performance has no effect on the tracking performance.

[0094] Figure 4 The experimental waveforms of the conventional ADRC and the decoupled two-degree-of-freedom ADRC of the present invention are shown under a sudden load torque of 0.3Nm. Figure 4 (a) shows that due to the ESO estimate and Contains unknown disturbance information. To eliminate the disturbance, the PD controller participates in the disturbance suppression process. As a result, there is a certain coupling between the PD controller and the ESO in the disturbance suppression performance of the traditional ADRC. Figure 4 As can be seen from (b), the PD controller uses the DESO estimated and As feedback, it will not receive any disturbance information, and the output remains unchanged after the disturbance occurs, and it does not participate in the disturbance suppression process at all. This shows that the tracking performance under the decoupled two-degree-of-freedom ADRC will not affect the disturbance suppression performance. Through step reference and load mutation experiments, it is proved that the conventional ADRC reference tracking performance and the disturbance suppression performance are coupled, while the two performances of the decoupled two-degree-of-freedom ADRC of the present invention are decoupled.

[0095] Figure 5 The position tracking effect of the PD controller of the present invention and the convergence process of the neural network PD parameters under three consecutive step position commands with an amplitude of π rad are demonstrated. k fd The initial value is 0, which means that the PD controller does not perform damping compensation at all. It can be seen that the first position command tracking under the initial value cannot suppress the system vibration well. Then the neural network PD uses the vibration information of the system to k fd It was quickly adjusted to about 0.18, which quickly suppressed the vibration of the load side position. The vibration caused by the last two position changes was effectively suppressed. k fd Tends to converge.

[0096] The above description of the embodiments is to facilitate the understanding and application of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments. Improvements and modifications made by those skilled in the art to the present invention based on the disclosure of the present invention should be within the protection scope of the present invention.

Claims

1. A dual-inertia system position control method based on a fusion model and data drive, comprising the following steps: (1) The dual-inertia system is reduced to a slow-varying subsystem and a fast-varying subsystem by the singular perturbation method; The model expression of the slow-changing subsystem is as follows: in: J m and J L are the rotation inertia of the motor side and the load side of the dual inertia system, B m and B L are the damping coefficients of the motor side and the load side of the dual inertia system, T L is the load torque, is the quasi-steady-state value of the transmission torque, and They are θ L The second and first order differentials of θ L is the position of the load side of the dual inertia system, δ is the intermediate variable; The model expression of the fast-changing subsystem is as follows: Where: k0 = K s ε 2 , K s is the spring stiffness coefficient, ε is the singular perturbation parameter, and η is the transmission torque error. T r is the transmission torque, i.e. T r =K s (θ m -θ L ), θ m is the position of the motor side of the dual inertia system, η″ and η′ are respectively the ε The second and first order differentials under dt ε / dt=1 / ε, t represents time; (2) The slow-changing subsystem and the fast-changing subsystem are reconstructed. The specific implementation method is: the flexible compensation gain is unified into the controller parameter design of the slow-changing subsystem and the fast-changing subsystem respectively, and the flexible compensation term is taken as a part of the fast-changing subsystem and combined with the damping compensation term to form a PD controller with the transmission torque as input. The controller outputs of the reconstructed slow-changing subsystem and the fast-changing subsystem are expressed as follows: Where: θ d is the given reference position instruction, k sp and k sd are the proportional coefficient and differential coefficient of the reconstructed slow-varying subsystem controller, k fp and k fd are the proportional coefficient and differential coefficient of the reconstructed fast-changing subsystem controller, They are θ d ,θ m The first order differential of (3) Based on the reconstructed slow-varying subsystem, its controller is designed as a two-degree-of-freedom ADRC controller based on DESO; The two-degree-of-freedom ADRC controller includes a tracking differentiator, a PD controller and a DESO, wherein the tracking differentiator is used to convert a step-change reference position instruction into a smoothly changing reference position signal, and provide the reference position signal and its differentiated reference speed signal to the PD controller; the PD controller uses the load side position and speed estimated by the DESO as feedback signals to adjust the motor output torque so that the load side position follows the reference position instruction; the DESO is used to estimate the load side position, speed and lumped disturbance, and suppress internal and external disturbances by compensating for the lumped disturbance. The discrete equation of the DESO is as follows: in: and are the load side position estimation values ​​of the k+1th control cycle and the kth control cycle, respectively, and are the load side speed estimation values ​​of the k+1th control cycle and the kth control cycle, respectively, and are the lumped disturbance estimates of the k+1th control cycle and the kth control cycle, respectively, h1, h2, h3 are the adjustment parameters of DESO, z represents the z-transform operator, T s is the control cycle duration, θ L (k) is the actual load side position value of the kth control cycle, u s (k) is the output torque of the slow-changing subsystem controller in the kth control period, J n =J m +J L , k is a natural number; The output of the two-degree-of-freedom ADRC controller is expressed as follows: Where: u s (k) is the output torque of the slow-changing subsystem controller in the kth control period, θ ref (k) and ω ref (k) are the reference position signal and reference speed signal of the kth control cycle respectively; (4) Based on the reconstructed fast-changing subsystem, its controller is designed as a PD controller based on data drive and neural network; (5) The output torque u of the slow-changing subsystem controller s The output torque u of the fast-changing subsystem controller f The reference torque u is obtained by adding them together, and the load side position is closed-loop controlled through the dual inertia system based on the reference torque u.

2. The dual-inertia system position control method of a fusion model and data-driven according to claim 1, characterized in that: The PD controller in step (4) updates the neural network weights online according to the input and output information of the dual-inertia system to achieve rapid suppression of system vibration. The output of the PD controller is expressed as follows: Where: u f (k) is the output torque of the fast-changing subsystem controller in the kth control cycle, X and a are the input layer and hidden layer of the neural network respectively, a=VX=[a1(k),a2(k)] T , V is the hidden layer weight and W i (k) is a i (k) The corresponding output layer weight, X = [e(k), e(k-1)] T , e(k-1)=θ L (k-1)-θ m (k-1), e(k) = θ L (k)-θ m (k), θ L (k-1) and θ L (k) are the actual load position values ​​of the k-1th control cycle and the kth control cycle, respectively, θ m (k-1) and θ m (k) are the actual position values ​​of the motor side in the k-1th control cycle and the kth control cycle, respectively. T Indicates transpose.

3. The dual-inertia system position control method of a fusion model and data-driven according to claim 2, characterized in that: The loss function of the neural network is as follows: Where: L represents the loss function.

4. A computer device comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is used to execute the computer program to implement a dual-inertia system position control method that integrates a model and data drive as described in any one of claims 1 to 3.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for position control of a dual-inertia system driven by a fusion model and data is implemented as described in any one of claims 1 to 3.

Images