Servo driver control system based on multi-modal fusion

Abstract

The invention relates to the technical field of control systems, in particular to a servo driver control system based on multi-modal fusion, and adopts the technical scheme that motor current, rotating speed and load torque signals are acquired in real time through a dynamic parameter identification module, so that the dynamic response speed of the system is increased; an extended state observer (ESO) is adopted to estimate unmodeled dynamics and disturbance through the nonlinear observer module, the observation precision and the real-time performance and adaptability of the system are improved, a hybrid algorithm of fuzzy PID and model predictive control (MPC) is integrated through the adaptive control module, the control precision of the system is improved, and the control precision of the system is improved. The anti-interference compensation module generates a dynamic compensation amount according to the output of the observer, the dynamic compensation amount is injected into a control loop, the anti-interference capability of the system is enhanced, the energy consumption optimization unit adjusts a PWM modulation strategy through an online efficiency optimization algorithm, and the energy efficiency of the system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of control systems, and particularly to a servo drive control system based on multi-modal fusion. Background Art

[0002] Servo drive is a control system that realizes precise control of a load by controlling the speed, position, and torque of a motor. The servo drive system uses a feedback mechanism to monitor and adjust the output to achieve precise control. At present, the servo drive control system has problems such as low dynamic response speed, low observation accuracy and control accuracy, low real-time performance and adaptability, and low energy efficiency.

[0003] In view of this, we propose a servo drive control system based on multi-modal fusion to solve the existing problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a servo drive control system based on multi-modal fusion to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A servo drive control system based on multi-modal fusion, including a dynamic parameter identification module, a non-linear observer module, an adaptive control module, an anti-interference compensation module, and an energy consumption optimization unit; wherein, the dynamic parameter identification module collects motor current, speed, and load torque signals in real time, the non-linear observer module uses an extended state observer (ESO) to estimate the unmodeled dynamics and disturbances, the adaptive control module integrates a hybrid algorithm of fuzzy PID and model predictive control (MPC), the anti-interference compensation module generates a dynamic compensation amount according to the observer output and injects it into the control loop, and the energy consumption optimization unit adjusts the PWM modulation strategy through an online efficiency optimization algorithm.

[0006] Furthermore, the dynamic parameter identification module uses the recursive least squares method to identify the electrical parameters of the motor online, constructs the relationship between torque and acceleration using a load inertia identification algorithm, and integrates a parameter anomaly detection mechanism to trigger system protection when the identification value fluctuates by more than ±15%.

[0007] Furthermore, the non-linear observer module includes a gain adaptive mechanism, a multi-disturbance frequency division estimation unit, and a sliding mode compensation module; wherein, the gain adaptive mechanism dynamically adjusts the ESO bandwidth parameter according to the disturbance estimation error, the multi-disturbance frequency division estimation unit processes low-frequency load disturbances and high-frequency mechanical resonances respectively through a parallel ESO structure, the frequency of the low-frequency load disturbance is less than or equal to 100 Hz, the frequency of the high-frequency mechanical resonance is greater than or equal to 1 kHz, and the sliding mode compensation module injects a sliding mode control term at the observer output to suppress the observation error chattering.

[0008] Furthermore, the adaptive control module includes a fuzzy logic layer, a predictive control layer, and a dynamic coupling mechanism. Among them, the fuzzy logic layer adjusts the PID parameters online according to the error and the rate of change of the error. The predictive control layer generates a feedforward control quantity based on rolling optimization in a finite time domain. The dynamic coupling mechanism realizes the output fusion of fuzzy PID and model predictive control (MPC) through a weight factor.

[0009] Furthermore, the anti-interference compensation module includes a high-frequency vibration suppression sub-module, a low-speed pulsation compensation sub-module, and a temperature drift compensation sub-module. Among them, the high-frequency vibration suppression sub-module uses a notch filter bank to eliminate mechanical resonance. The low-speed pulsation compensation sub-module eliminates cogging effect based on the harmonic injection method. The temperature drift compensation sub-module corrects the gain parameters in real time through a thermistor network.

[0010] Furthermore, the energy consumption optimization unit includes an efficiency optimization engine, a sub-region modulation strategy, and a flux compensation unit. Among them, the efficiency optimization engine synchronously optimizes the Pareto front solutions of the switching loss model and the conduction loss model. The sub-region modulation strategy switches the CPWM / DPWM / THI hybrid mode according to the real-time bus voltage and the phase current amplitude. The flux compensation unit injects a third harmonic component into the modulation wave to reduce the current harmonic distortion rate.

[0011] Furthermore, the gain adaptive mechanism establishes a non-linear mapping relationship between the bandwidth parameter and the system state variables, introduces a gradient descent algorithm to optimize the mapping coefficient online, and sets a parameter adjustment dead zone.

[0012] Furthermore, the multi-disturbance frequency division estimation unit includes a low-frequency ESO channel, a high-frequency ESO channel, and a synthesizer. Among them, the low-frequency ESO channel adopts a third-order ESO structure, and the observation bandwidth is set to 3-5 times the fundamental frequency of the motor. The high-frequency ESO channel adopts band-pass filtering preprocessing, and the adjustable range of the center frequency is 500Hz - 5kHz. The synthesizer fuses the observation results of the two channels through a wavelet packet reconstruction algorithm.

[0013] Furthermore, the sliding mode compensation module adopts a sliding mode surface function and a quasi-sliding mode control law, and the output limit of the compensation amount does not exceed 20% of the total control amount.

[0014] Furthermore, the weight factor is dynamically adjusted according to the system working state. The PID dominant mode is adopted in the steady state stage, and it switches to the MPC dominant mode in the transient state stage. The smooth switching is realized through an S-shaped function in the transition region. Among them, the stage with an error less than 1% is set as the steady state stage, the stage with an error greater than or equal to 5% is set as the transient state stage, and the stage with an error greater than or equal to 1% and less than 5% is set as the transition region.

[0015] Compared with the prior art, the beneficial effects of the present invention are:

[0016] The present invention improves the dynamic response speed of the system by a dynamic parameter identification module that real-time collects motor current, speed, and load torque signals. The nonlinear observer module uses an Extended State Observer (ESO) to estimate unmodeled dynamics and disturbances, improving the observation accuracy, real-time performance, and adaptability of the system. The adaptive control module integrates a hybrid algorithm of fuzzy PID and Model Predictive Control (MPC) to improve the control accuracy of the system. The anti-interference compensation module generates a dynamic compensation amount based on the observer output and injects it into the control loop to enhance the anti-interference ability of the system. The energy consumption optimization unit adjusts the PWM modulation strategy through an online efficiency optimization algorithm to improve the energy efficiency of the system. Description of the Drawings

[0017] Figure 1 is a schematic flowchart of a servo drive control system based on multi-modal fusion of the present invention;

[0018] Figure 2 is a schematic flowchart of the dynamic parameter identification module of the present invention;

[0019] Figure 3 is a schematic flowchart of the nonlinear observer module of the present invention;

[0020] Figure 4 is a schematic flowchart of the adaptive control module of the present invention;

[0021] Figure 5 is a schematic flowchart of the anti-interference compensation module of the present invention;

[0022] Figure 6 is a schematic flowchart of the energy consumption optimization unit of the present invention. Detailed Embodiments

[0023] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0024] Embodiment 1

[0025] As Figure 1 shown, a servo drive control system based on multi-modal fusion includes a dynamic parameter identification module, a nonlinear observer module, an adaptive control module, an anti-interference compensation module, and an energy consumption optimization unit. Among them, the dynamic parameter identification module real-time collects motor current, speed, and load torque signals, the nonlinear observer module uses an Extended State Observer (ESO) to estimate unmodeled dynamics and disturbances, the adaptive control module integrates a hybrid algorithm of fuzzy PID and Model Predictive Control (MPC), the anti-interference compensation module generates a dynamic compensation amount based on the observer output and injects it into the control loop, and the energy consumption optimization unit adjusts the PWM modulation strategy through an online efficiency optimization algorithm.

[0026] AsFigure 2 As shown, the dynamic parameter identification module uses the recursive least squares method to identify the electrical parameters of the motor online, constructs the relationship between torque and acceleration using the load inertia identification algorithm, and integrates a parameter anomaly detection mechanism. When the identified value fluctuates by more than ±15%, the system protection is triggered.

[0027] As Figure 3 shown, the non-linear observer module includes a gain adaptive mechanism, a multi-disturbance frequency division estimation unit, and a sliding mode compensation module. Among them, the gain adaptive mechanism dynamically adjusts the ESO bandwidth parameter according to the disturbance estimation error. The multi-disturbance frequency division estimation unit processes low-frequency load disturbances and high-frequency mechanical resonances respectively through a parallel ESO structure. The frequency of the low-frequency load disturbance is less than or equal to 100Hz, and the frequency of the high-frequency mechanical resonance is greater than or equal to 1kHz. The sliding mode compensation module injects a sliding mode control term at the observer output to suppress the observation error chattering.

[0028] As Figure 4 shown, the adaptive control module includes a fuzzy logic layer, a predictive control layer, and a dynamic coupling mechanism. Among them, the fuzzy logic layer online adjusts the PID parameters according to the error and the error change rate. The predictive control layer generates a feedforward control quantity based on the finite-time domain rolling optimization. The dynamic coupling mechanism realizes the output fusion of fuzzy PID and model predictive control (MPC) through a weight factor.

[0029] As Figure 5 shown, the anti-interference compensation module includes a high-frequency vibration suppression sub-module, a low-speed pulsation compensation sub-module, and a temperature drift compensation sub-module. Among them, the high-frequency vibration suppression sub-module uses a notch filter bank to eliminate mechanical resonance. The low-speed pulsation compensation sub-module eliminates the cogging effect based on the harmonic injection method. The temperature drift compensation sub-module corrects the gain parameter in real time through a thermistor network.

[0030] As Figure 6 shown, the energy consumption optimization unit includes an efficiency optimization engine, a sub-region modulation strategy, and a flux compensation unit. Among them, the efficiency optimization engine synchronously optimizes the Pareto front solutions of the switching loss model and the conduction loss model. The sub-region modulation strategy switches the CPWM / DPWM / THI hybrid mode according to the real-time bus voltage and the phase current amplitude. The flux compensation unit injects a third harmonic component into the modulation wave to reduce the current harmonic distortion rate.

[0031] The working principle of a servo drive control system based on multi-modal fusion according to Embodiment 1 is:

[0032] The Recursive Least Squares (RLS) method is a parameter estimation method that uses minimizing the sum of the squares of measurement errors as the optimization criterion and can be used in fields such as parameter identification and adaptive control. The core idea of RLS is to recursively calculate new parameter estimates using previous estimation results and new observation data, thus avoiding the need to reprocess the entire dataset. The main feature of the RLS algorithm is its ability to quickly adapt to system dynamic changes and also provide relatively accurate parameter estimates in the presence of noise. Compared with the batch least squares method, RLS does not require storing a large amount of historical data, reducing the computational and storage burden, and is particularly suitable for real-time or near-real-time processing. In a motor control system, appropriate input signals (such as voltage, current) and output signals (such as rotational speed, electromagnetic torque) are selected, and the relationships between them and the parameters to be identified are determined. Based on the mathematical model of the motor, the parameters to be identified are extracted and rewritten in a form suitable for the RLS algorithm. According to the basic principle of the RLS algorithm, the recursive formula for parameter estimation is derived.

[0033] The load inertia identification algorithm is a method for estimating and identifying the load inertia in a mechanical system. Load inertia is a physical quantity that describes the inertia of an object during rotational motion and has an important impact on the control and dynamic performance of the motor. The load inertia identification algorithm estimates the rotational inertia of the load based on the dynamic response of the motor and known control inputs. Once the rotational inertia of the load is obtained, it can be substituted into the relationship between torque and angular acceleration to establish the specific relationship between torque and acceleration.

[0034] The gain adaptive mechanism establishes a non-linear mapping relationship between the bandwidth parameter and the system state variable, introduces the gradient descent algorithm to optimize the mapping coefficient online, and sets a parameter adjustment dead zone. Let the ESO bandwidth parameters be β1, β2, β3, and the system state variable be e. The non-linear mapping relationship between the bandwidth parameter and the system state variable is established as β1 = k1·tanh(|e| / ε) + min(β1, β2, β3), where k1 and ε are the mapping coefficients. For setting the parameter adjustment dead zone, the adjustment is frozen when |e| ≤ 0.5%.

[0035] The gradient descent algorithm is a commonly used optimization algorithm, mainly used to solve the minimum value of the loss function in machine learning and deep learning. It calculates the gradient of the loss function with respect to the model parameters and then updates the parameters in the negative direction of the gradient to find the parameter values that can minimize the loss function. In the process of introducing the gradient descent algorithm to optimize the mapping coefficients online, first select an appropriate loss function according to the specific problem and data characteristics; then initialize the model parameters, which can be randomly set before starting training or set according to some prior knowledge; then whenever new data flows in, calculate the gradient of the loss function with respect to the mapping coefficients under the current parameters; then use the gradient descent algorithm to update the mapping coefficients, and adjust the parameter values according to the calculated gradient and a predetermined learning rate; finally, as new data continues to flow in, repeat the gradient calculation and parameter update until a certain stopping condition is reached (the loss function converges below a certain threshold or a predetermined number of iterations is reached).

[0036] The multi-disturbance frequency division estimation unit includes a low-frequency ESO channel, a high-frequency ESO channel, and a synthesizer; among them, the low-frequency ESO channel adopts a third-order ESO structure, and the observation bandwidth is set to 3-5 times the fundamental frequency of the motor; the high-frequency ESO channel adopts band-pass filtering preprocessing, and the adjustable range of the center frequency is 500Hz-5kHz; the synthesizer fuses the observation results of the two channels through the wavelet packet reconstruction algorithm.

[0037] The third-order ESO (Extended State Observer) structure is one of the core components based on the Active Disturbance Rejection Control (ADRC) theory. The Extended State Observer (ESO) is a model-free observer that can estimate the disturbances brought by factors such as model inaccuracy, internal coupling, and external intervention in the system. In the third-order ESO, through the processing of the system input and output signals, it can estimate three state variables of the system and an extended state (usually the total disturbance).

[0038] The third-order ESO is designed for a system of the following form:

[0039]

[0040] where y is the system output, u is the system input, f() represents the internal dynamics, ω represents the external disturbance, and b is the system parameter.

[0041] Regarding the total disturbance as an extended state, the system can be transformed into state-space form, and a third-order ESO can be constructed to estimate these states.

[0042] The observer equation of the third-order ESO is usually expressed as:

[0043]

[0044] Among them, are the estimates of the system state y, respectively, and l1, l2, and l3 are the observer gains.

[0045] The observer gains l1, l2, and l3 are determined by pole placement or optimization algorithms to ensure that the ESO can quickly and accurately estimate the system state.

[0046] Band-pass filtering preprocessing is a signal processing technique that uses a band-pass filter to preprocess the signal. The specific process of band-pass filtering preprocessing is as follows:

[0047] S1. Select a band-pass filter: According to the characteristics of the signal and the processing requirements, select a suitable band-pass filter. This includes determining parameters such as the center frequency and bandwidth of the filter to ensure that the desired frequency components can be accurately extracted or enhanced.

[0048] S2. Apply the filter: Input the signal to be processed into the band-pass filter. The filter will allow signals within a specific frequency range to pass through according to its frequency response characteristics and attenuate signals of other frequencies.

[0049] S3. Output the processed signal: The signal processed by the band-pass filter will mainly contain the desired frequency components, while other frequency components are effectively suppressed. This processed signal can be used for subsequent signal processing, analysis, or transmission.

[0050] The wavelet packet reconstruction algorithm is a signal restoration method. Based on the wavelet packet transform theory, it can recombine the signal decomposed by wavelet packets into the original signal. The wavelet packet reconstruction algorithm is a process of performing inverse transformation using wavelet packet coefficients. In wavelet packet decomposition, the signal is decomposed into multiple wavelet packets with different frequencies and time resolutions, and each wavelet packet contains certain signal information. By recording the coefficients of these wavelet packets and using these coefficients for inverse transformation during the reconstruction process, the original signal can be restored. The implementation steps of the wavelet packet reconstruction algorithm are as follows:

[0051] A1. Obtain wavelet packet coefficients: The wavelet packet coefficients of the signal need to be obtained through the wavelet packet decomposition algorithm. These coefficients contain information about the signal at different frequencies and time resolutions.

[0052] A2. Inverse transformation process: Perform inverse transformation using the obtained wavelet packet coefficients. The inverse transformation process is the inverse of wavelet packet decomposition. By gradually combining wavelet packets with different frequencies and time resolutions, the original signal is finally restored.

[0053] A3. Reconstruct the signal: After the inverse transformation, the reconstructed signal can be obtained. This signal should be consistent with the original signal within the allowable error range.

[0054] The sliding mode compensation module adopts the sliding mode surface function and the quasi-sliding mode control law, and the output limit of the compensation amount does not exceed 20% of the total control amount.

[0055] The sliding mode surface function s(x) is a function of the system state variable x, which describes the characteristics of the system running along the sliding mode surface. In sliding mode control, the choice of the s(x) function directly determines whether the system can quickly enter the sliding mode surface and stably operate on the sliding mode surface. This function usually needs to be selected in combination with the dynamic characteristics of the system, the control objectives, and the stability requirements of the system. The sliding mode surface function can divide the state space into two parts. When the system state motion trajectory is above the sliding mode surface, the sliding mode control law will pull it down; conversely, when the state motion trajectory is below the sliding mode surface, the sliding mode control law will pull it up, so that the system state trajectory always moves along the sliding mode surface.

[0056] The quasi-sliding mode control law is a special control strategy designed to improve the chattering problem in traditional sliding mode control. The core idea of quasi-sliding mode control is to limit the motion trajectory of the system within a certain neighborhood of the ideal sliding mode, and this neighborhood is called the boundary layer of the sliding mode switching surface. Within the boundary layer, the quasi-sliding mode control does not require the existence conditions of the sliding mode to be satisfied, so the chattering phenomenon can be avoided or weakened to a certain extent.

[0057] The weight factor is dynamically adjusted according to the system working state. The PID dominant mode is adopted in the steady state stage, and it is switched to the MPC dominant mode in the transient stage. The smooth switching is achieved through the S-shaped function in the transition region; among them, the stage with an error less than 1% is set as the steady state stage, the stage with an error greater than or equal to 5% is set as the transient stage, and the stage with an error greater than or equal to 1% and less than 5% is set as the transition region.

[0058] The method of online adjusting PID parameters by the fuzzy logic layer according to the error and the rate of change of the error, namely fuzzy adaptive PID control, is a technology that combines fuzzy logic and adaptive PID control. Fuzzy adaptive PID control takes the error (e) and the rate of change of the error (ec) as inputs, and dynamically adjusts the three parameters of the PID controller through a fuzzy logic system: the proportional gain Kp, the integral gain Ki, and the derivative gain Kd. The adjustment process is as follows: first calculate the error and the rate of change of the error, then convert the error e and the rate of change of the error ec into fuzzy sets, then use fuzzy rules for inference, then convert the fuzzy output obtained by fuzzy inference into an exact numerical value to adjust the parameters of the PID controller, and then adjust the Kp, Ki, and Kd parameters of the PID controller according to the exact numerical value obtained by defuzzification, then use the adjusted PID parameters to recalculate the control signal and apply it to the controlled object, and finally, at the next sampling moment, repeat the above process to continue to adjust the PID parameters according to the new error and the rate of change of the error to achieve rolling optimization. Among them, the role of Kp is mainly to accelerate the response speed of the system and eliminate the error; the role of Ki is to eliminate the steady-state error of the system; the role of Kd is to improve the dynamic characteristics of the system and suppress the change of the deviation. Fuzzy adaptive PID control can dynamically adjust the PID parameters according to the real-time feedback of the system to adapt to the dynamic changes and uncertainties of the system, and improve the robustness and adaptability of the system, avoiding the limitations of traditional PID controllers in the face of system dynamic changes.

[0059] The process of the predictive control layer generating the feedforward control quantity based on the rolling optimization in the finite time domain mainly involves the core principle of model predictive control (MPC). MPC is an advanced control strategy that combines a predictive model, rolling optimization, and a feedforward-feedback control structure. MPC uses the predictive model to predict the future output of the system and performs rolling optimization based on these prediction results to generate a control sequence. The steps of the finite time domain rolling optimization are as follows:

[0060] B1. Predictive model: MPC first uses the predictive model to predict the output of the system in the future period of time. These predictions are based on the historical information of the system and the current input.

[0061] B2. Rolling optimization: Different from global optimization, MPC performs rolling optimization in a finite time domain. This means that in each control cycle, MPC will define an optimization problem that involves the system output and control input from the current moment to a finite number of future time steps. By solving this optimization problem, MPC obtains a control sequence, but only applies the first element of the sequence (i.e., the control input at the current moment) to the system. In the next control cycle, MPC will repeat this process to update the control input based on the new prediction results and optimization problem.

[0062] B3. Feedforward control quantity generation: During the receding horizon optimization process, MPC takes into account the dynamic characteristics and constraints of the system. By solving the optimization problem, MPC obtains a control sequence that satisfies these constraints and makes the system output as close as possible to the desired output. The first element of this control sequence is the feedforward control quantity at the current moment.

[0063] MPC not only relies on the feedforward control quantity but also combines feedback control to correct the effects caused by model mismatch and external environmental disturbances. In each control cycle, MPC monitors the actual output of the system and uses this information to update the prediction model and the optimization problem. This feedforward-feedback control structure enables MPC to control the system more accurately and improves its robustness and adaptability.

[0064] The above specific embodiments are merely several preferred embodiments of the present invention. Based on the technical solution of the present invention and the relevant revelations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A servo drive control system based on multi-modal fusion, characterized in that: It includes a dynamic parameter identification module, a nonlinear observer module, an adaptive control module, an anti-interference compensation module, and an energy consumption optimization unit. Among them, the dynamic parameter identification module collects motor current, speed, and load torque signals in real time. The nonlinear observer module uses an Extended State Observer (ESO) to estimate the unmodeled dynamics and disturbances. The adaptive control module integrates a hybrid algorithm of fuzzy PID and Model Predictive Control (MPC). The anti-interference compensation module generates a dynamic compensation amount according to the observer output and injects it into the control loop. The energy consumption optimization unit adjusts the PWM modulation strategy through an online efficiency optimization algorithm.

2. The servo driver control system based on multimodal fusion according to claim 1, wherein: The dynamic parameter identification module uses the recursive least squares method to identify the motor electrical parameters online, constructs the relationship between torque and acceleration using a load inertia identification algorithm, and integrates a parameter anomaly detection mechanism. When the identification value fluctuates by more than ±15%, the system protection is triggered.

3. The servo driver control system based on multimodal fusion according to claim 1, wherein: The nonlinear observer module includes a gain adaptive mechanism, a multi-disturbance frequency division estimation unit, and a sliding mode compensation module. Among them, the gain adaptive mechanism dynamically adjusts the ESO bandwidth parameter according to the disturbance estimation error. The multi-disturbance frequency division estimation unit processes low-frequency load disturbances and high-frequency mechanical resonances separately through a parallel ESO structure. The frequency of the low-frequency load disturbance is less than or equal to 100Hz, and the frequency of the high-frequency mechanical resonance is greater than or equal to 1kHz. The sliding mode compensation module injects a sliding mode control term at the observer output to suppress the observation error chattering.

4. The servo driver control system based on multi-modal fusion according to claim 1, characterized in that: The adaptive control module includes a fuzzy logic layer, a predictive control layer, and a dynamic coupling mechanism. Among them, the fuzzy logic layer adjusts the PID parameters online according to the error and the rate of change of the error. The predictive control layer generates a feedforward control amount based on the finite-time domain rolling optimization. The dynamic coupling mechanism realizes the output fusion of fuzzy PID and Model Predictive Control (MPC) through a weight factor.

5. The servo driver control system based on multi-modal fusion according to claim 1, wherein: The anti-interference compensation module includes a high-frequency vibration suppression sub-module, a low-speed pulsation compensation sub-module, and a temperature drift compensation sub-module. Among them, the high-frequency vibration suppression sub-module uses a notch filter bank to eliminate mechanical resonance. The low-speed pulsation compensation sub-module eliminates the cogging effect based on the harmonic injection method. The temperature drift compensation sub-module corrects the gain parameter in real time through a thermal network.

6. The servo driver control system based on multimodal fusion according to claim 1, characterized in that: The energy consumption optimization unit includes an efficiency optimization engine, a sub-region modulation strategy, and a flux compensation unit. Among them, the efficiency optimization engine synchronously optimizes the Pareto front solutions of the switching loss model and the conduction loss model. The sub-region modulation strategy switches the CPWM / DPWM / THI hybrid mode according to the real-time bus voltage and the phase current amplitude. The flux compensation unit injects a third harmonic component into the modulation wave to reduce the current harmonic distortion rate.

7. A servo drive control system based on multimodal fusion according to claim 3, characterized in that: The gain adaptive mechanism establishes a nonlinear mapping relationship between the bandwidth parameter and the system state variables, introduces a gradient descent algorithm to optimize the mapping coefficient online, and sets a parameter adjustment dead zone.

8. The servo driver control system based on multi-modal fusion according to claim 3, wherein: The multi-disturbance frequency division estimation unit includes a low-frequency ESO channel, a high-frequency ESO channel, and a synthesizer. Among them, the low-frequency ESO channel uses a third-order ESO structure, and the observation bandwidth is set to 3-5 times the motor base frequency. The high-frequency ESO channel uses band-pass filtering preprocessing, and the adjustable range of the center frequency is 500Hz - 5kHz. The synthesizer fuses the observation results of the two channels through a wavelet packet reconstruction algorithm.

9. The servo drive control system based on multi-modal fusion according to claim 3, wherein: The sliding mode compensation module adopts a sliding mode surface function and a quasi-sliding mode control law, and the output limit of the compensation amount does not exceed 20% of the total control amount.

10. A servo drive control system based on multimodal fusion according to claim 4, characterized in that: The weight factor is dynamically adjusted according to the system working state. The PID dominant mode is adopted in the steady state stage, and it is switched to the MPC dominant mode in the transient state stage. The smooth switching is realized through the S-shaped function in the transition region. Among them, the stage with an error less than 1% is set as the steady state stage, the stage with an error greater than or equal to 5% is set as the transient state stage, and the stage with an error greater than or equal to 1% and less than 5% is set as the transition region.

Images