Variable load active-disturbance-rejection joint motor control method and device based on RBF optimization, medium and equipment

Abstract

The invention discloses a variable load active-disturbance-rejection joint motor control method and device based on RBF optimization, a medium and equipment, and the method comprises the steps: obtaining a target angle of a joint motor outputted by an upper controller, and carrying out the arithmetic subtraction operation based on the target angle and a rotor angle, and obtaining an angle error signal; parameters of an RBF neural network in the RBFADRC controller are initialized, and parameter updating is carried out on a third-order expansion state observer in the RBFADRC controller based on the ESO parameters; the target angle and the rotor angle are input into the updated RBFADRC controller, and a closed-loop control signal is obtained; taking the closed-loop control signal as the input of the MPC current inner loop module, and outputting a control parameter; and carrying out SVPWM (Space Vector Pulse Width Modulation) processing on the control parameters to obtain six paths of complementary and symmetrical PWM signals. According to the method, double-closed-loop design and intelligent optimization are achieved based on outer loop control of the RBFADRC controller and inner loop control of the MPC current inner loop module, and the control performance of the system under the variable load working condition is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of motor control, and particularly relates to a variable load active disturbance rejection joint motor control method, device, medium and equipment based on RBF optimization. BACKGROUND

[0002] With the development of modern mechanical equipment such as robots, numerical control machine tools and medical equipment, high-precision joint motor control has gradually become one of the key technologies in the field of intelligent manufacturing and advanced equipment. The joint motor usually takes a permanent magnet synchronous motor (PMSM) as a core driving component, which has advantages of high power density, high efficiency and superior dynamic performance, and is widely used in industrial robots, medical rehabilitation equipment and aerospace fields. However, under complex dynamic working conditions and variable load environment, how to realize accurate control, fast response and strong anti-interference ability of the joint motor is still a big problem in the technical field.

[0003] Traditional motor control methods generally adopt proportional-integral-derivative (PID) or vector control strategies to realize basic control functions through current loop and speed loop double closed loop. However, these control strategies have obvious shortcomings in dynamic performance and robustness, especially in the case of rapid load change or large external disturbance, which easily leads to system precision decline or even instability. In addition, due to the nonlinearity, strong coupling and load uncertainty of the PMSM system, the traditional control method is difficult to meet the accurate adjustment demand of complex working conditions.

[0004] As a new type of intelligent control technology, active disturbance rejection control (ADRC) has attracted widespread attention in recent years due to its low model dependence and strong anti-disturbance ability. ADRC can effectively compensate modeling errors and external disturbances through the extended state observer (ESO) to estimate the total disturbance of the system in real time. However, the traditional ESO parameter setting is difficult, and its adaptability is limited under complex dynamic working conditions. Therefore, introducing an optimization method based on artificial intelligence, such as RBF neural network optimization strategy, has become an important direction to improve the dynamic performance of ESO.

[0005] RBF (Radial Basis Function) neural network is widely used in the design of nonlinear control systems due to its superior nonlinear approximation ability, fast learning speed and global optimization characteristics. By combining RBF neural network with third-order ESO, an extended state observer based on RBF optimization (RBF_ESO) is constructed, which can realize high-precision online estimation of the dynamic state and total disturbance of the joint motor. The introduction of RBF_ESO can significantly improve the adaptive ability and dynamic response performance of the ADRC controller, thereby improving the robustness and stability of the system under complex load change scenarios.

[0006] On the other hand, traditional current inner loops typically employ PI controllers to regulate the dq-axis current of the PMSM (Motor-Modulated Joint Motor) to achieve decoupled control of electromagnetic torque (IQ) and flux linkage (ID). However, the performance of PI controllers is limited under large load disturbances or complex dynamic conditions. To address this issue, Model Predictive Control (MPC), due to its superior predictive and optimization capabilities, has been introduced into the current inner loop, replacing the traditional PI controller. The MPC current inner loop precisely regulates the dq-axis current of the joint motor through an optimized predictive control strategy, effectively enhancing the system's dynamic performance under varying load conditions.

[0007] Existing joint motor control technologies still have shortcomings in complex load variation scenarios. How to achieve variable load active disturbance rejection joint motor control based on RBF optimization is a technical problem that urgently needs to be solved. Summary of the Invention

[0008] In view of the above problems, the present invention is proposed to provide a variable load active disturbance rejection joint motor control method, apparatus, medium and equipment based on RBF optimization to overcome or at least partially solve the above problems.

[0009] Other features and advantages of the invention will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0010] According to a first aspect of the present invention, a variable load active disturbance rejection joint motor control method based on RBF optimization is provided, the variable load active disturbance rejection joint motor control method based on RBF optimization includes: The rotor angle of the joint motor is acquired by a magnetic encoder, and the three-phase current signal of the joint motor is acquired by a current detection module. Based on the three-phase current signal and the rotor angle, coordinate transformation is performed to obtain the direct-axis current and quadrature-axis current in the two-phase rotating coordinate system of the joint motor where the rotor rotates synchronously. Obtain the target angle of the joint motor output by the upper controller, and perform an arithmetic difference operation based on the target angle and the rotor angle to obtain the angle error signal; The parameters of the RBF neural network in the RBFADRC controller are initialized. The acquired angle error signal is input into the RBF neural network for online training. The RBF neural network dynamically updates the parameters through the backpropagation algorithm and outputs the optimal ESO parameters. The parameters of the third-order extended state observer in the RBFADRC controller are updated based on the ESO parameters. The target angle and the rotor angle are input into the updated RBFADRC controller to obtain a closed-loop control signal. The closed-loop control signal, direct-axis current, and quadrature-axis current are used as inputs to the MPC current inner loop module. The MPC current inner loop module executes a control strategy with a direct-axis current target value ID = 0, optimizes the selection of the optimal voltage vector through a cost function, and outputs control parameters. The control parameters are subjected to SVPWM modulation processing to obtain six complementary and symmetrical PWM signals, which are used to control the operating state of the joint motor.

[0011] In some embodiments of the present invention, the step of performing coordinate transformation based on the three-phase current signal and the rotor angle to obtain the direct-axis current and quadrature-axis current in the two-phase rotating coordinate system of the synchronous rotation of the joint motor includes: Based on the three-phase current signal, the current components in the two-phase static-stationary coordinate system are obtained through Clark transformation. Then, based on the rotor angle and the current components, the direct-axis current and quadrature-axis current in the two-phase rotating coordinate system that rotates synchronously with the rotor are obtained through Park transformation.

[0012] In some embodiments of the present invention, the parameters for initializing the RBF neural network within the RBFADRC controller include: The layers of the RBF neural network are determined, including an input layer with 3 input variables, a hidden layer with 6 nodes, and an output layer with 1 output. The input of the input layer is the angle error signal of the joint motor, and the output of the output layer is the ESO parameter. The 3 input variables are the position state error, the velocity state error, and the total perturbation estimated by the third-order extended state observer. The RBF neural network is initialized by initializing the parameters, which include the center vector of the hidden layer nodes, the variance of the basis functions, and the weights between the hidden layer and the output layer.

[0013] In some embodiments of the present invention, the step of inputting the target angle and the rotor angle into the updated RBFADRC controller to obtain a closed-loop control signal includes: The target angle of the joint motor is smoothed by the differential module in the RBFADRC controller to generate an ideal position trajectory and velocity trajectory. The rotor angle of the joint motor and the external disturbance are input into a third-order extended state observer to estimate the three input variables of the system, namely the position state error, the velocity state error and the total disturbance. The ideal position trajectory, velocity trajectory, and the three input variables are input into the nonlinear error feedback control module within the RFBADRC controller. Based on the ideal position trajectory, velocity trajectory, position state error, and velocity state error, error comparison processing is performed to generate a state error signal. Based on the total disturbance, the state error signal is subjected to feedforward compensation processing to obtain the closed-loop control signal.

[0014] According to a second aspect of the present invention, a variable load active disturbance rejection joint motor control device based on RBF optimization is provided, the variable load active disturbance rejection joint motor control device based on RBF optimization includes: A magnetic encoder is used to acquire the rotor angle of a joint motor. The current detection module is used to collect the three-phase current signal of the joint motor; The coordinate transformation module performs coordinate transformation based on the three-phase current signal and the rotor angle to obtain the direct-axis current and quadrature-axis current in the two-phase rotating coordinate system under synchronous rotor rotation of the joint motor. The RBFADRC controller includes an RBF module, a third-order extended state observer, a differential module, and a nonlinear error feedback control law module. The RBFADRC controller acquires the target angle of the joint motor output by the upper-level controller, and performs an arithmetic difference operation based on the target angle and the rotor angle to obtain an angle error signal. It initializes the parameters of the RBF neural network within the RBF module, inputs the acquired angle error signal into the RBF neural network for online training, dynamically updates the parameters through a backpropagation algorithm, and outputs optimal ESO parameters. Based on the ESO parameters, it updates the parameters of the third-order extended state observer. Finally, it obtains a closed-loop control signal based on the target angle and the rotor angle. The MPC current inner loop module takes the closed-loop control signal, direct-axis current, and quadrature-axis current as inputs, executes the control strategy of direct-axis current target value ID = 0, optimizes the selection of the optimal voltage vector through cost function, and outputs control parameters. The SVPWM modulation module is used to perform SVPWM modulation processing on the control parameters to obtain six complementary and symmetrical PWM signals, and output the PWM signals to the IGBT inverter. An IGBT inverter is used to control the operating state of the joint motor according to the received PWM signal.

[0015] In some embodiments of the present invention, the coordinate transformation module includes: The Clark transform module, based on the three-phase current signal, obtains the current components in a two-phase static-stationary coordinate system through Clark transform. The Park transformation module obtains the direct-axis current and quadrature-axis current in a two-phase rotating coordinate system that rotates synchronously with the rotor by performing Park transformation based on the rotor angle and the current components.

[0016] In some embodiments of the present invention, the parameters of the RBFADRC controller initializing the RBF neural network within the RBF module include: The layers of the RBF neural network are determined, including an input layer with 3 input variables, a hidden layer with 6 nodes, and an output layer with 1 output. The input of the input layer is the angle error signal of the joint motor, and the output of the output layer is the ESO parameter. The 3 input variables are the position state error, the velocity state error, and the total perturbation estimated by the third-order extended state observer. The RBF neural network is initialized by initializing the parameters, which include the center vector of the hidden layer nodes, the variance of the basis functions, and the weights between the hidden layer and the output layer.

[0017] In some embodiments of the present invention, the RFBADRC controller obtains the closed-loop control signal based on the target angle and the rotor angle, including: The target angle of the joint motor is smoothed by the differential module in the RBFADRC controller to generate an ideal position trajectory and velocity trajectory. The rotor angle of the joint motor and the external disturbance are input into a third-order extended state observer to estimate the three input variables of the system, namely the position state error, the velocity state error and the total disturbance. The ideal position trajectory, velocity trajectory, and the three input variables are input into the nonlinear error feedback control module within the RFBADRC controller. Based on the ideal position trajectory, velocity trajectory, position state error, and velocity state error, error comparison processing is performed to generate a state error signal. Based on the total disturbance, the state error signal is subjected to feedforward compensation processing to obtain the closed-loop control signal.

[0018] According to a third aspect of the present invention, a computer-readable storage medium is provided, wherein computer program instructions are stored therein, the computer program instructions being loaded and executed by a processor to perform the operations performed by the method described in any of the preceding claims.

[0019] According to a fourth aspect of the present invention, an electronic device is provided, including a processor and a memory, the memory storing computer program instructions executable by the processor, wherein when the processor executes the computer program instructions, it implements the instructions of any of the methods described above.

[0020] The technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: This invention provides a variable load active disturbance rejection joint motor control method, device, medium, and equipment based on RBF optimization. The method is based on an RBFADRC controller composed of a third-order extended state observer optimized by an RBF neural network and an MPC current inner loop module. The RBFADRC controller is used to precisely control the angle of the joint motor. By adaptively tuning the ESO parameters online through the RBF neural network, the system's observation accuracy of the actual state and total disturbance is significantly improved, thereby enhancing anti-interference capability and dynamic response performance. The MPC current inner loop module is used to precisely adjust the electromagnetic torque and flux linkage of the joint motor. By optimizing prediction and control strategies, high-precision dynamic performance and steady-state accuracy are ensured under variable load and disturbance conditions. Thus, based on the outer loop control of the RBFADRC controller and the inner loop control of the MPC current inner loop module, a dual closed-loop design and intelligent optimization are achieved, significantly improving the system's control performance under variable load conditions.

[0021] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a variable load active disturbance rejection joint motor control method based on RBF optimization provided in an embodiment of the present invention; Figure 2 A schematic diagram of the principle structure of a variable load active disturbance rejection joint motor control device based on RBF optimization provided in an embodiment of the present invention; Figure 3 A reference diagram showing the dynamic change curves of the ESO parameters of the third-order extended state observer for real-time tuning of the RBF neural network. Figure 4 A schematic diagram showing the position loop simulation results of the RBFADRC controller and ADRC controller tuned by the RBF neural network; Figure 5This is a reference schematic diagram showing the position loop simulation results of the RBFADRC controller and ADRC controller tuned by the RBF neural network under variable load. Detailed Implementation

[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings.

[0025] The accompanying drawings illustrate various structural schematics according to embodiments of the present disclosure. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0026] In the context of this disclosure, when a layer / component is referred to as being "above" another layer / component, that layer / component may be directly above the other layer / component, or there may be an intermediate layer / component between them. Additionally, if a layer / component is "above" another layer / component in one orientation, then when the orientation is reversed, that layer / component may be "below" the other layer / component. In the context of this disclosure, similar or identical components may be denoted by the same or similar reference numerals.

[0027] To better understand the above technical solutions, the following will describe the above technical solutions in detail with reference to specific implementation methods. It should be understood that the embodiments of this disclosure and the specific features in the embodiments are detailed descriptions of the technical solutions of the present invention, rather than limitations on the technical solutions of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0028] Figure 1 This is a flowchart illustrating a variable load active disturbance rejection joint motor control method based on RBF optimization provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the variable load active disturbance rejection joint motor control method based on RBF optimization includes the following steps: S1. The rotor angle of the joint motor 3 is acquired by the magnetic encoder 2, and the three-phase current signal of the joint motor 3 is acquired by the current detection module 4. Based on the three-phase current signal and the rotor angle, coordinate transformation is performed to obtain the direct-axis current and quadrature-axis current in the two-phase rotating coordinate system of the synchronous rotation of the rotor of the joint motor 3. Combination Figure 2As shown, in this embodiment of the invention, the rotor angle Θ_act of the joint motor 3 is acquired by the magnetic encoder 2, and the three-phase current signals IU, IV, and IW of the joint motor 3 are acquired by the current detection module 4. Then, the coordinate transformation module performs coordinate transformation based on the three-phase current signals and the rotor angle to obtain the feedback value of the rotor of the joint motor 3 in a two-phase rotating coordinate system for synchronous rotation. The feedback value includes the direct-axis current Id (used for feedback flux linkage) and the quadrature-axis current Iq (used for feedback torque).

[0029] The coordinate transformation module includes a Clark transformation module 51 and a Park transformation module 52. The Clark transformation module 51 is used to perform Clark transformation operations, which are used to transform variables (such as current and voltage) in a three-phase stationary coordinate system (a, b, c) to a two-phase stationary coordinate system (α, β), that is, to achieve a transformation from three-phase stationary to two-phase stationary. The Park transformation module 52 performs Park transformation operations, which are used to transform variables in a two-phase stationary coordinate system (α, β) to a two-phase rotating coordinate system (d, q), that is, to achieve a transformation from two-phase stationary to two-phase rotating.

[0030] Specifically, in this embodiment of the invention, coordinate transformation is performed based on the three-phase current signal and the rotor angle to obtain the direct-axis current and quadrature-axis current in the two-phase rotating coordinate system in which the rotor of the joint motor 3 rotates synchronously. This includes: obtaining the current components Iα and Iβ in the two-phase static-stationary coordinate system based on the three-phase current signal through Clark transformation; and then obtaining the direct-axis current Id and quadrature-axis current Iq in the two-phase rotating coordinate system in which the rotor rotates synchronously with the rotor through Park transformation based on the rotor angle and the current components.

[0031] S2. Obtain the target angle of the joint motor 3 output by the upper controller, and perform an arithmetic difference operation based on the target angle and the rotor angle to obtain the angle error signal; The target angle Θ_ref of the joint motor 3 is a command signal generated by the upper-level controller according to the application task planning, and serves as the external input of this control system. In this embodiment of the invention, the actual rotor angle Θ_act of the joint motor 3 is obtained in real time through the magnetic encoder 2, and the angle error signal is obtained by performing an arithmetic difference operation (i.e., Θ_ref - Θ_act) between the target angle Θ_ref and the actual rotor angle Θ_act.

[0032] S3. Initialize the parameters of the RBF neural network in the RBFADRC controller 1, input the acquired angle error signal into the RBF neural network for online training, the RBF neural network dynamically updates the parameters through the backpropagation algorithm and outputs the optimal ESO parameters, and update the parameters of the third-order extended state observer 12 in the RBFADRC controller 1 based on the ESO parameters. The RBFADRC controller 1 includes an RBF module 11, a third-order extended state observer 12, a differential module 13, and a nonlinear error feedback control law module 14. The RBF (Radial Basis Function) module is equipped with an RBF neural network. The third-order extended state observer 12 is used to estimate the system state (position, velocity, or position state error, velocity state error) and total disturbance (combination of internal and external disturbances) in real time, and dynamically adjusts parameters b01, b02, and b03 through the RBF neural network, enabling the observer to adapt to load changes and improve estimation accuracy. The differential module 13 is a tracking differentiator used to smooth the target angle Θ_ref generated by the upper-level controller, generating an ideal trajectory without overshoot and avoiding oscillations caused by sudden changes. The nonlinear error feedback control law module 14 is used to calculate the nonlinear error based on the outputs of the third-order extended state observer 12 and the differential module 13, and generate a closed-loop control signal IQ to achieve fast response and disturbance compensation.

[0033] The initialization parameters of the RBF neural network within the RBFADRC controller 1 in this embodiment of the invention include: determining the layers of the RBF neural network, wherein the layers include an input layer with 3 input variables, a hidden layer with 6 nodes, and an output layer with 1 output; the input of the input layer is the angle error signal of the joint motor 3; the output of the output layer is the ESO (third-order extended state observer 12) parameters (b01, b02, b03), wherein the ESO parameters are the parameters that the RBFADRC controller 1 needs to adjust, and b01, b02, and b03 in the ESO parameters respectively characterize the core observer gain, which respectively determines the estimation speed and accuracy of the system's position, velocity, and total disturbance, and are key parameters affecting the dynamic performance and stability of the observer; the 3 input layer The input variables are position state error, velocity state error, and total perturbation estimated by the third-order extended state observer 12. The RBF neural network is initialized by initial assignment, and the parameters of the initial assignment include the center vector ci of the hidden layer nodes, the variance bi of the basis functions, and the weight w between the hidden layer and the output layer.

[0034] In this embodiment of the invention, the acquired angle error signal is input into the RBF neural network for online training. The RBF neural network is integrated into the RBF module 11 and closely combined with the third-order extended state observer 12. The RBF neural network dynamically calculates the adjustment increment of the parameters in real time through the backpropagation algorithm, and then generates the optimal ESO parameters required for the current control cycle through the iterative formula and in combination with the learning rate and momentum factor. Based on the ESO parameters, the third-order extended state observer 12 in the RBFADRC controller 1 is updated with parameters.

[0035] In this embodiment of the invention, the RBF neural network dynamically calculates the parameter adjustment increment in real time using the backpropagation algorithm, specifically including: The hidden layer activation function is chosen to be the Gaussian function, with the following expression: , ; The outputs of each node in the hidden layer are: h(j)=exp(-norm(x-ci(:,j))^2 / (2*bi(j)*bi(j))); In the formula, j is an index used to distinguish these 6 different nodes, representing the j-th hidden layer node; The output of the RBF neural network is: ym=w'*h; In the formula, For the first Gaussian centers of neurons in the hidden layer For the first The base width of each hidden layer node is used to adjust the sensitivity of the neuron. Let be the Euclidean vector norm, representing the distance between the input neuron and the center point vector; w be the output layer weight vector; h be the output vector of all nodes in the hidden layer; and ym be the final output of the RBF network. w' represents the transpose of the output layer weight vector w. They have a direct mathematical relationship: w' is the transpose matrix / vector of w. w is a predefined weight column vector. In the formula for calculating the network output, its transpose w' is used to perform the correct vector dot product operation to obtain the final output result ym.

[0036] In this embodiment of the invention, the parameters initially assigned to the RBF neural network, such as the center vector ci of the hidden layer nodes, the variance bi of the basis functions, and the weights w between the hidden and output layers, are all updated and adaptively tuned online using a gradient descent algorithm with a momentum term. This online learning algorithm is widely used and mature in neural network training. Based on the error between the system output and the network's predicted output, and combined with the system Jacobian matrix calculated from the system input and output signals, the direction of parameter adjustment is determined, thereby continuously reducing the error. Its update rule is mainly adjusted by two key coefficients: Learning rate (η): Used to control the step size of each parameter update, with a value range of [0, 1]. Momentum factor (α): Used to adjust the influence of historical update amounts on the current update direction, in order to accelerate convergence and improve stability, with a value range of [0, 1]. Through this online learning algorithm, the RBF neural network of this embodiment of the invention can adjust its internal parameters in real time, thereby dynamically and adaptively adjusting the b01, b02, and b03 parameters of the third-order extended state observer 12 in the RBFADRC controller 1.

[0037] Combination Figure 3 The figure shows a reference diagram of the dynamic change curves of the parameters of the third-order extended state observer 12ESO in real time tuning by the RBF neural network. The curves in the figure show that the three key gain parameters b01, b02, and b03 are not fixed values, but change dynamically with time, which intuitively proves the adaptive characteristics of the embodiment of the present invention. Specifically: b01 (red curve): position observation gain, the value of which is adjusted in real time according to the dynamic requirements of the system to ensure accurate tracking of the motor position state; b02 (blue curve): speed observation gain, this parameter is also dynamically adjusted to optimize the estimation of motor speed and the response to dynamic changes in the system; b03 (green curve): total disturbance observation gain, its change amplitude is the largest, reflecting that the RBF neural network is dynamically adjusting the estimation and compensation intensity of total disturbances such as variable load and friction in the system according to the real-time operating conditions.

[0038] Figure 3 The results show that the RBF neural network in this embodiment of the invention can autonomously and continuously optimize ESO parameters, enabling the controller to actively adapt to complex operating conditions. This embodiment of the invention operates in real-time in the control of the articulated motor 3. Through the combination of RBF neural network optimization and the self-disturbance-resistant third-order extended state observer 12, it can significantly improve the control accuracy, dynamic response speed, and anti-interference capability of the articulated motor 3 under complex variable load conditions.

[0039] S4. Input the target angle and the rotor angle into the updated RBFADRC controller 1 to obtain a closed-loop control signal; In this embodiment of the invention, the target angle and the rotor angle are input into the updated RBFADRC controller 1 to obtain a closed-loop control signal. This includes: smoothing the target angle of the joint motor 3 through the differential module 13 in the RBFADRC controller 1 to generate an ideal position trajectory and velocity trajectory; inputting the rotor angle of the joint motor 3 and the external disturbance into a third-order extended state observer 12 to estimate the three input variables of the system, namely the position state error, velocity state error, and total disturbance; inputting the ideal position trajectory, velocity trajectory, and the three input variables into the nonlinear error feedback control law module 14 in the RBFADRC controller 1; performing error comparison processing based on the ideal position trajectory, velocity trajectory, position state error, and velocity state error to generate a state error signal; and performing feedforward compensation processing on the state error signal based on the total disturbance to obtain the closed-loop control signal.

[0040] In this embodiment of the invention, the differential module 13 is a second-order nonlinear differential tracker (TD), whose equation is expressed as follows: ; The equations of the third-order extended state observer 12 are expressed, for example: ; The equation for the nonlinear error feedback control law (NLSEF) module is expressed, for example: ; The above three equations define the three core modules of the RFBADRC controller 1, namely the differential module 13, the third-order extended state observer 12, and the nonlinear error feedback control law module 14. The three are closely coupled in function and sequentially connected in signal flow, together forming a complete closed-loop control system.

[0041] Specifically, the second-order nonlinear differential tracker (TD), acting as an instruction preprocessor, receives the external target angle Θ. * r (i.e., Θ_ref), and plan a smooth target position trajectory z for it. 11 and target velocity trajectory z 12 The third-order state extension observer (RBF_ESO) serves as the state and disturbance estimator, utilizing the actual angle Θ of the motor. * r (i.e., Θ_act) and the final control input i q *Real-time estimation of the system's actual position z 21 Actual speed z 22 and total internal and external disturbance z 23 Finally, the Nonlinear State Error Feedback Control Law (NLSEF) module, as the core controller, combines the results of the previous two to calculate the ideal trajectory (z) of the second-order nonlinear differential tracker TD output. 11 , z 12 ) and the actual state estimated by RBF_ESO (z 21 , z 22 The error between ) and the total perturbation z estimated using RBF_ESO. 23 Feedforward compensation is performed to ultimately generate control commands i that can actively suppress disturbances. q * (i.e., closed-loop control signal IQ), control command i q * On the one hand, it drives the joint motor 3, and on the other hand, it also serves as input feedback to RBF_ESO, forming a closed-loop control.

[0042] S5. The closed-loop control signal, direct-axis current, and quadrature-axis current are used as inputs to the MPC current inner loop module 6. The MPC current inner loop module 6 executes the control strategy of direct-axis current target value ID = 0, optimizes the selection of the optimal voltage vector through the cost function, and outputs control parameters. In this embodiment of the invention, the flux linkage loop and torque loop in the MPC current inner loop module 6 adopt model predictive control (MPC). Specifically, it performs predictive optimization calculations on the d-axis current zero-value control strategy (i.e., direct-axis current target value ID = 0) aimed at maximizing motor efficiency, the q-axis current target value (i.e., torque command IQ) output by the RBFADRC controller 1, and the feedback values ​​of the direct-axis current Id and quadrature-axis current Iq. The specific process of the predictive optimization calculation is as follows: The MPC current inner loop module 6 is based on the motor mathematical model to predict the future current response under different voltage vectors. It selects the optimal voltage vector that minimizes the current error through cost function optimization, and finally obtains the control output (i.e., control parameters): sector number (rank), the duration of action of the two adjacent base voltage vectors of the sector (ti, tj), and the duration of action of the zero voltage vector (tz). The MPC control is based on the prediction and optimization control strategy of system behavior. It can adjust the control input in real time and significantly reduce the impact of external disturbances on the regulation of electromagnetic torque (IQ) and flux linkage (ID), thereby ensuring the stability and precise control of the system under complex operating conditions.

[0043] S6. Perform SVPWM modulation processing on the control parameters to obtain six complementary and symmetrical PWM signals, which are used to control the operating state of the joint motor 3.

[0044] After obtaining the control parameters, ti, tj, tz, and rank are modulated by SVPWM to generate six complementary symmetrical PWM signals. These six complementary symmetrical PWM signals drive the six switching transistors of the three-phase bridge arm of the IGBT inverter 8. The complementary symmetry means that the drive signals of the upper and lower switching transistors of the same bridge arm are logically opposite and have a dead time to prevent power supply short circuit. These six PWM signals control the IGBT inverter 8 to drive the joint motor 3, realizing precise control of the electromagnetic torque (IQ) and flux linkage (ID) of the joint motor 3.

[0045] Combination Figure 4 The diagram shown is a reference schematic of the position loop simulation results for the RBFADRC controller 1 and ADRC controller tuned by the RBF neural network. Figure 4 In the simulation, a permanent magnet synchronous motor with a rated power of 360W was selected as the joint motor 3; the rated torque was Tn=8Nm; the stator resistance R=0.9585Ω; the stator inductance L=0.00525mH; the number of pole pairs p=24; and the moment of inertia J=5kg·cm². During the simulation, a load of 5 N·m was applied to the motor at startup, the initial target position was set to 1 radian, and adjusted to 2 radians at 0.2 seconds. Figure 4 As shown, the RFBADRC controller 1 (red line) of this embodiment approaches the desired value (green line) faster than the ADRC controller (purple line). In steady state, the RFBADRC controller 1 of this embodiment exhibits smaller error and higher control accuracy, while the ADRC controller shows significant error deviation in steady state. During transients, the RFBADRC controller 1 (red line) shows smaller overshoot and a faster stabilization process, while the ADRC controller (purple line) has larger overshoot and a slower return to steady state.

[0046] Combination Figure 5 The diagram shows a reference schematic of the position loop simulation results of the RBFADRC controller 1 and ADRC controller tuned by the RBF neural network under varying load. A permanent magnet synchronous motor with a rated power of 360W is selected as the joint motor 3; the rated torque is Tn=8Nm; the stator resistance R=0.9585Ω; the stator inductance L=0.00525mH; the number of pole pairs p=24; and the moment of inertia J=5kg·cm². In the simulation, a load of 5 N·m is applied to the motor at startup and the target position is set to 2 radians. At 0.2 seconds, a load of 10 N·m is applied to the motor. Figure 5As shown, when the load changes, the RFBADRC controller 1 (red line) of this embodiment exhibits better control performance than the ADRC controller (purple line). The RFBADRC controller 1 responds quickly to load changes, exhibits minimal overshoot, and smoothly transitions to steady state, ultimately achieving the target position accurately. Conversely, the ADRC controller experiences significant overshoot during load changes and has a slower recovery speed, resulting in a larger steady-state error. Particularly when the load changes from 5 N·m to 10 N·m, the RFBADRC controller 1 can quickly suppress overshoot and accurately track the target position.

[0047] Compared with existing technologies, the variable load active disturbance rejection joint motor control method based on RBF optimization described in this embodiment of the invention uses an RBFADRC controller 1 composed of a third-order extended state observer 12 (RBF_ESO) optimized by an RBF neural network and an MPC current inner loop module 6. The RBFADRC controller 1 is used to precisely control the angle of the joint motor 3. By adaptively tuning the ESO parameters (b01, b02, b03) online through the RBF neural network, the system's observation accuracy of the actual state and total disturbance is significantly improved, thereby enhancing anti-interference capability and dynamic response performance. The MPC current inner loop module 6 is used to precisely adjust the electromagnetic torque (IQ) and flux linkage of the joint motor 3. (ID) By optimizing prediction and control strategies, it ensures high-precision dynamic performance and steady-state accuracy under varying load and disturbance conditions. At the same time, the RBF neural network enables the ESO parameters to change adaptively to adapt to complex load conditions. The ESO estimates and compensates for disturbances, the MPC optimizes the current response, and the system remains stable during load changes. TD smooths the reference trajectory, and NLSEF nonlinear adjustment ensures fast response and low overshoot. SVPWM reduces switching losses, and MPC optimizes current control. Thus, based on the outer loop control of the RBFADRC controller 1 and the inner loop control of the MPC current inner loop module 6, a dual closed-loop design and intelligent optimization are achieved, which significantly improves the control performance of the system under varying load conditions.

[0048] Based on the above embodiments, as a supplement to the above... Figure 1 The present invention provides an embodiment of a variable load active disturbance rejection joint motor control device based on RBF optimization, which is similar to the method shown. Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices, see reference. Figure 2 As shown, the RBF-optimized variable load active disturbance rejection joint motor control device includes: Magnetic encoder 2 is used to acquire the rotor angle of articulated motor 3; The current detection module 4 is used to collect the three-phase current signal of the joint motor 3; The coordinate transformation module 5 performs coordinate transformation based on the three-phase current signal and the rotor angle to obtain the direct-axis current and quadrature-axis current in the two-phase rotating coordinate system of the synchronous rotation of the rotor of the joint motor 3. The RBFADRC controller 1 includes an RBF module 11, a third-order extended state observer 12, a differential module 13, and a nonlinear error feedback control law module 14. The RBFADRC controller 1 is used to acquire the target angle of the joint motor 3 output by the upper-level controller, and to obtain an angle error signal by performing an arithmetic difference operation based on the target angle and the rotor angle. It initializes the parameters of the RBF neural network within the RBF module 11, inputs the acquired angle error signal into the RBF neural network for online training, dynamically updates the parameters through a backpropagation algorithm, and outputs the optimal ESO parameters. Based on the ESO parameters, it updates the parameters of the third-order extended state observer 12. Finally, it obtains a closed-loop control signal according to the target angle and the rotor angle. The MPC current inner loop module 6 takes the closed-loop control signal, direct-axis current and quadrature-axis current as inputs, executes the control strategy of direct-axis current target value ID = 0, optimizes the selection of the optimal voltage vector through cost function, and outputs control parameters. The SVPWM modulation module 7 is used to perform SVPWM modulation processing on the control parameters to obtain six complementary and symmetrical PWM signals, and output the PWM signals to the IGBT inverter 8. IGBT inverter 8 is used to control the operating state of joint motor 3 according to the received PWM signal.

[0049] In this embodiment of the invention, the coordinate transformation module 5 includes: Clark transform module 51 obtains the current components in a two-phase static-stationary coordinate system based on the three-phase current signals through Clark transform. Park transformation module 52 obtains the direct-axis current and quadrature-axis current in a two-phase rotating coordinate system that rotates synchronously with the rotor by Park transformation based on the rotor angle and the current components.

[0050] In this embodiment of the invention, the parameters for initializing the RBF neural network within the RBF module 11 by the RBFADRC controller 1 include: The layers of the RBF neural network are determined, including an input layer with 3 input variables, a hidden layer with 6 nodes, and an output layer with 1 output. The input of the input layer is the angle error signal of the joint motor 3, and the output of the output layer is the ESO parameter. The 3 input variables are the position state error, the velocity state error, and the total disturbance estimated by the third-order extended state observer 12. The RBF neural network is initialized by initializing the parameters, which include the center vector of the hidden layer nodes, the variance of the basis functions, and the weights between the hidden layer and the output layer.

[0051] In this embodiment of the invention, the RBFADRC controller 1 obtains the closed-loop control signal based on the target angle and the rotor angle, including: The target angle of the joint motor 3 is smoothed by the differential module 13 in the RBFADRC controller 1 to generate an ideal position trajectory and velocity trajectory. The rotor angle of the joint motor 3 and the external disturbance are input into the third-order extended state observer 12 to estimate the three input variables of the system, namely the position state error, the velocity state error and the total disturbance. The ideal position trajectory, velocity trajectory, and the three input variables are input into the nonlinear error feedback control module 14 within the RFBADRC controller 1. Based on the ideal position trajectory, velocity trajectory, position state error, and velocity state error, error comparison processing is performed to generate a state error signal. Based on the total disturbance, the state error signal is subjected to feedforward compensation processing to obtain the closed-loop control signal.

[0052] The RBF-optimized variable load active disturbance rejection joint motor control device described in this embodiment can execute the RBF-optimized variable load active disturbance rejection joint motor control method provided in the above embodiments. The RBF-optimized variable load active disturbance rejection joint motor control device has the corresponding functional steps and beneficial effects of the RBF-optimized variable load active disturbance rejection joint motor control method described in the above embodiments. For details, please refer to the embodiments of the RBF-optimized variable load active disturbance rejection joint motor control method described above. The embodiments of this invention will not be repeated here.

[0053] This invention also provides an electronic device, which may include a processor and a memory, wherein the processor and memory can be connected via a bus or other means. The processor may be a Central Processing Unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations thereof. The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the RBF-optimized variable load active disturbance rejection joint motor control method in this invention embodiment. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the RBF-optimized variable load active disturbance rejection joint motor control method in the above method embodiment.

[0054] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. One or more modules are stored in the memory and, when executed by the processor, perform the RBF-optimized variable load active disturbance rejection joint motor control method as described in the above method embodiments. Specific details of the above electronic device can be understood by referring to the corresponding descriptions and effects in the above method embodiments, and will not be repeated here. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it may include the processes of the embodiments of the above methods. The storage medium may be a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above types of memory.

[0055] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0056] Similarly, it should be understood that, in order to simplify this disclosure and aid in understanding one or more of the various aspects of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the description of exemplary embodiments of the invention above. Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and it should be noted that the above embodiments are illustrative of the invention and not restrictive, and that alternative embodiments can be devised by those skilled in the art without departing from its scope.

Claims

1. A variable load active disturbance rejection joint motor control method based on RBF optimization, characterized in that, The variable load active disturbance rejection joint motor control method based on RBF optimization comprises: The rotor angle of the joint motor is collected through a magnetic encoder, the three-phase current signals of the joint motor are collected through a current detection module, the direct-axis current and the quadrature-axis current of the joint motor in a two-phase rotating coordinate system rotating synchronously with the rotor are obtained through coordinate transformation based on the three-phase current signals and the rotor angle; The target angle of the joint motor output by an upper controller is acquired, and an angle error signal is obtained through arithmetic difference operation based on the target angle and the rotor angle; The parameters of the RBF neural network in the RBFADRC controller are initialized, the acquired angle error signal is input into the RBF neural network for online training, the RBF neural network dynamically updates the parameters through a back propagation algorithm, and outputs optimal ESO parameters, and the three-order extended state observer in the RBFADRC controller is updated based on the ESO parameters; The target angle and the rotor angle are input into the updated RBFADRC controller to obtain a closed-loop control signal; The closed-loop control signal, the direct-axis current and the quadrature-axis current are taken as inputs of an MPC current inner loop module, the MPC current inner loop module executes a control strategy of a direct-axis current target value ID = 0, selects an optimal voltage vector through cost function optimization, and outputs a control parameter; The control parameter is subjected to SVPWM modulation processing to obtain six complementary and symmetrical PWM signals, and the PWM signals are used to control the operating state of the joint motor.

2. The RBF optimization-based variable load active disturbance rejection joint motor control method according to claim 1, characterized in that, The coordinate transformation based on the three-phase current signals and the rotor angle to obtain the direct-axis current and the quadrature-axis current in the two-phase rotating coordinate system rotating synchronously with the rotor comprises: The current components in a two-phase static-stationary coordinate system are obtained through Clark transformation based on the three-phase current signals, and the direct-axis current and the quadrature-axis current in the two-phase rotating coordinate system rotating synchronously with the rotor are obtained through Park transformation according to the rotor angle and the current components.

3. The RBF optimization-based variable load active disturbance rejection joint motor control method according to claim 1, characterized in that, The initialization of the parameters of the RBF neural network in the RBFADRC controller comprises: The level of the RBF neural network is determined, the level comprises one input layer containing three input variables, one hidden layer containing six nodes, and one output layer containing one output, the input of the input layer is the angle error signal of the joint motor, the output of the output layer is the ESO parameter, and the three input variables are respectively a position state error, a speed state error, and a total disturbance estimated by a three-order extended state observer; Initial assignment is performed on the RBF neural network to complete initialization, and the initial assignment parameters comprise a center vector of the hidden layer nodes, a variance of a basis function, and a weight value between the hidden layer and the output layer.

4. The RBF optimization-based variable load active disturbance rejection joint motor control method according to claim 3, characterized in that, The input of the target angle and the rotor angle into the updated RBFADRC controller to obtain a closed-loop control signal comprises: The target angle of the joint motor is subjected to smoothing processing through a differential module in the RBFADRC controller to generate an ideal position trajectory and a speed trajectory; The rotor angle of the joint motor and external disturbance input is input into a third-order extended state observer, and three input variables of the system, namely the position state error, the speed state error and the total disturbance, are estimated; The ideal position trajectory and speed trajectory and the three input variables are input into a nonlinear error feedback control rate module in the RBFADRC controller, error comparison processing is performed based on the ideal position trajectory and speed trajectory and the position state error and the speed state error, a state error signal is generated, the state error signal is fed forwardly compensated based on the total disturbance, and the closed-loop control signal is obtained.

5. A variable load active disturbance rejection joint motor control device based on RBF optimization, characterized in that, The RBF-optimized variable-load active disturbance rejection joint motor control device comprises: A magnetic encoder is configured to collect the rotor angle of the joint motor. A current detection module is configured to collect three-phase current signals of the joint motor. A coordinate conversion module is configured to perform coordinate transformation based on the three-phase current signals and the rotor angle to obtain direct-axis current and quadrature-axis current in a two-phase rotating coordinate system that rotates synchronously with the rotor of the joint motor. An RBFADRC controller comprises an RBF module, a third-order extended state observer, a differential module and a nonlinear error feedback control rate module. The RBFADRC controller is configured to obtain a target angle of the joint motor output by an upper controller, perform arithmetic difference operation based on the target angle and the rotor angle to obtain an angle error signal, initialize parameters of an RBF neural network in the RBF module, input the obtained angle error signal into the RBF neural network for online training, dynamically update the parameters of the RBF neural network through a back propagation algorithm, output optimal ESO parameters, and update the parameters of the third-order extended state observer based on the ESO parameters. An MPC current inner loop module is configured to take the closed-loop control signal, the direct-axis current and the quadrature-axis current as inputs, execute a control strategy of a direct-axis current target value ID = 0, select an optimal voltage vector through cost function optimization, and output control parameters. An SVPWM modulation module is configured to perform SVPWM modulation processing on the control parameters to obtain six complementary and symmetrical PWM signals, and output the PWM signals to an IGBT inverter.

6. The RBF optimization-based variable load active disturbance rejection joint motor control device according to claim 5, characterized in that, The IGBT inverter is configured to control the operating state of the joint motor according to the received PWM signals. The coordinate conversion module comprises: A Clark transformation module is configured to obtain current components in a two-phase static-stationary coordinate system through Clark transformation based on the three-phase current signals, 7. The RBF optimization-based variable load active disturbance rejection joint motor control device according to claim 5, characterized in that, A Park transformation module is configured to obtain direct-axis current and quadrature-axis current in a two-phase rotating coordinate system that rotates synchronously with the rotor through Park transformation based on the rotor angle and the current components. The RBFADRC controller initializes the parameters of the RBF neural network in the RBF module. determining a hierarchy of the RBF neural network, the hierarchy comprising an input layer comprising 3 input variables, an implicit layer comprising 6 nodes, and an output layer comprising 1 output, the input of the input layer being the angle error signal of the joint motor, the output of the output layer being the ESO parameter, the 3 input variables being the position state error, the speed state error, and the total disturbance estimated by the third-order extended state observer; initially assigning parameters to the RBF neural network to complete initialization, the parameters of the initial assignment including the center vector of the implicit layer nodes, the variance of the basis function, and the weight between the implicit layer and the output layer.

8. The RBF optimization-based variable load active disturbance rejection joint motor control device according to claim 7, characterized in that, the RBF ADRC controller obtaining a closed-loop control signal according to the target angle and the rotor angle comprises: smoothing the target angle of the joint motor through a differential module in the RBF ADRC controller to generate ideal position and speed trajectories; inputting the rotor angle of the joint motor and the external disturbance into a third-order extended state observer to estimate 3 input variables of the system, i.e., the position state error, the speed state error, and the total disturbance; inputting the ideal position and speed trajectories and the three input variables into a nonlinear error feedback control module in the RBF ADRC controller, comparing errors based on the ideal position and speed trajectories and the position state error and the speed state error to generate a state error signal, and performing feedforward compensation on the state error signal based on the total disturbance to obtain the closed-loop control signal.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer program instructions, which are loaded and executed by the processor to implement the operations performed by the method of any one of claims 1-4.

10. An electronic device comprising a processor and a memory, characterized in that The memory stores computer program instructions executable by the processor, and the processor executes the computer program instructions to implement the instructions of the method of any one of claims 1-4.

Images