A Multimodal Control Method and System for Brushless Doubly-Fed Machines

Abstract

The present invention provides a multi-modal control method and system for a brushless doubly-fed motor, which relates to the technical field of motor control. The method includes: extracting the torque-speed slope and the dynamic response time constant of two points based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, and constructing a linear correction function; fusing the output value of the linear correction function with the optimization result to generate a modal transition compensation coefficient, and injecting the compensation coefficient into the rotor-side voltage control loop based on the feed-forward control channel to generate an anti-disturbance dynamic compensation control strategy; for the non-smooth adjustment term in the dynamic compensation control strategy, using the subgradient projection algorithm to smooth the discontinuous boundary of the control quantity, and verifying the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate a final control instruction. The present invention can achieve adaptive smooth switching control of a brushless doubly-fed motor.

Description

Technical Field

[0001] The present invention relates to the technical field of motor control, and particularly to a multi-modal control method and system for a brushless doubly-fed motor. Background Art

[0002] Brushless doubly-fed motors (BDFMs) are widely used in fields such as wind power generation, ship propulsion, and industrial drives due to their advantages of brushless structure, high reliability, and small inverter capacity. However, their complex non-linear coupling characteristics and multi-modal operation requirements (such as constant torque, field-weakening speed regulation, regenerative braking, etc.) pose challenges to traditional control methods. The existing technologies mainly have the following defects:

[0003] For example, existing control methods (such as extended state observers, exogenous autoregressive models) rely on motor inductance parameters or recursive least squares methods for online calculation, resulting in high system sensitivity to parameters and high computational complexity, making it difficult to adapt to real-time dynamic working conditions. In addition, traditional PI controllers are prone to overshoot and oscillation during dynamic processes, affecting switching stability.

[0004] Although sliding mode control has strong robustness, the chattering phenomenon caused by its switching function will exacerbate torque ripple, especially leading to power winding voltage fluctuations and motor overheating under unbalanced loads. In addition, when dealing with multi-modal switching in existing methods, the non-smooth boundaries of control strategies are prone to cause discontinuous jumps in current or torque, and additional smoothing algorithms need to be designed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a multi-modal control method and system for a brushless doubly-fed motor, which can achieve adaptive smooth switching control of the brushless doubly-fed motor, effectively suppress chattering and non-smooth boundary problems, and improve robustness and dynamic response in multi-modal scenarios.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows:

[0007] In a first aspect, a multi-modal control method for a brushless doubly-fed motor, the method includes:

[0008] Step S1: Construct a modal recognition model including a bidirectional long short-term memory network, input a standardized operation data set, extract the dynamic characteristics of the motor speed-torque curve, and identify the current operation mode; the operation modes include constant torque mode, field-weakening speed regulation mode, regenerative braking mode, and transient overload mode;

[0009] Step S2: According to the operation mode, match the initial control parameters from a preset control strategy library, dynamically iterate and optimize the parameters to obtain an optimization result; based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extract the torque-speed slope and dynamic response time constant of the two points, and construct a linear correction function;

[0010] Step S3: Fuse the output value of the linear correction function with the optimization result to generate a modal transition compensation coefficient, and inject the compensation coefficient into the rotor-side voltage control loop based on the feedforward control channel to generate a dynamic compensation control strategy for disturbance rejection;

[0011] Step S4: For the non-smooth adjustment term in the dynamic compensation control strategy, use the subgradient projection algorithm to smooth the discontinuous boundary of the control quantity, and verify the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate the final control instruction;

[0012] Step S5: According to the final control instruction, realize the shock-free switching between different operating modes by adjusting the cooperative relationship between the stator winding frequency and the rotor-side inverter duty ratio, and complete the closed-loop control of the multi-modal operation of the motor.

[0013] In a second aspect, a multi-modal control system for a brushless doubly-fed motor includes:

[0014] A construction module for constructing a modal recognition model including a bidirectional long short-term memory network, inputting a standardized operation data set, extracting the dynamic characteristics of the motor speed-torque curve, and identifying the current operating mode; the operating modes include a constant torque mode, a field-weakening speed regulation mode, a regenerative braking mode, and a transient overload mode;

[0015] A correction module for matching initial control parameters from a preset control strategy library according to the operating mode, dynamically iteratively optimizing the parameters to obtain an optimization result; based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extracting the torque-speed slope and the dynamic response time constant of the two points, and constructing a linear correction function;

[0016] A generation module for fusing the output value of the linear correction function with the optimization result to generate a modal transition compensation coefficient, and injecting the compensation coefficient into the rotor-side voltage control loop based on the feedforward control channel to generate a dynamic compensation control strategy for disturbance rejection;

[0017] A control module for using the subgradient projection algorithm to smooth the discontinuous boundary of the control quantity for the non-smooth adjustment term in the dynamic compensation control strategy, and verifying the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate the final control instruction;

[0018] A switching module for realizing shock-free switching between different operating modes according to the final control instruction by adjusting the cooperative relationship between the stator winding frequency and the rotor-side inverter duty ratio, and completing the closed-loop control of the multi-modal operation of the motor.

[0019] In a third aspect, a computing device includes:

[0020] One or more processors;

[0021] A storage device for storing one or more programs, which when executed by the one or more processors cause the one or more processors to implement the method.

[0022] In a fourth aspect, a computer-readable storage medium stores a program that, when executed by a processor, implements the method.

[0023] The above solution of the present invention has at least the following beneficial effects:

[0024] By dynamically screening key parameters and extracting features, redundant data interference is effectively reduced. The recurrent neural network (RNN) is good at processing time series data and capturing dynamic changes during the operation of the motor, improving the real-time performance of modal recognition. Through the trained RNN architecture, it can quickly adapt to different operating conditions. By weighing multiple objectives such as power generation cost, electricity purchase price, load power shortage penalty, and voltage deviation, the operating cost is reduced and the energy utilization efficiency is improved. By calculating the subgradient of the non-smooth function, the non-smooth optimization problem existing in the control strategy is solved. Through data-driven, intelligent recognition, multi-objective optimization, and non-smooth problem handling, the brushless doubly-fed motor is efficient and stable under multi-modal operation, enhancing the adaptability of the motor system. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a schematic flowchart of a multi-modal control method for a brushless doubly-fed motor provided by an embodiment of the present invention.

[0026] Figure 2 is a schematic diagram of a multi-modal control system for a brushless doubly-fed motor provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0028] As Figure 1 shown, an embodiment of the present invention proposes a multi-modal control method for a brushless doubly-fed motor, and the method includes the following steps:

[0029] Step S1: Construct a modal recognition model including a bidirectional long short-term memory network, input a standardized operation data set, extract the dynamic characteristics of the motor speed-torque curve, and identify the current operation mode; the operation modes include a constant torque mode, a field weakening speed regulation mode, a regenerative braking mode, and a transient overload mode;

[0030] Step S2: According to the operation mode, match the initial control parameters from a preset control strategy library, and perform dynamic iterative optimization on the parameters to obtain an optimization result; based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extract the torque-speed slope and the dynamic response time constant of the two points, and construct a linear correction function;

[0031] Step S3: Integrate the output value of the linear correction function with the optimization result to generate a modal transition compensation coefficient, and inject the compensation coefficient into the rotor-side voltage control loop based on the feedforward control channel to generate a dynamic compensation control strategy for anti-disturbance;

[0032] Step S4: For the non-smooth adjustment term in the dynamic compensation control strategy, use the subgradient projection algorithm to smooth the discontinuous boundary of the control quantity, and verify the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate a final control instruction;

[0033] Step S5: According to the final control instruction, realize the shockless switching between different operation modes by adjusting the cooperative relationship between the stator winding frequency and the rotor-side inverter duty ratio, and complete the closed-loop control of the multi-modal operation of the motor.

[0034] In the embodiment of the present invention, a modal recognition model is constructed through a bidirectional long short-term memory network, which can effectively extract the dynamic characteristics of the motor speed-torque curve, and accurately identify multiple operation modes such as constant torque, field weakening speed regulation, regenerative braking, and transient overload in real time; based on the modal matching of the initial control parameters and iterative optimization, combined with a linear correction function to compensate for the dynamic differences between the steady-state and target operating points, improve the adaptability of the control parameters to different modes, and enhance the dynamic response ability of the system; inject a modal transition compensation coefficient through the feedforward control channel to suppress the influence of external disturbances on the voltage control loop; use the subgradient projection algorithm to smooth the non-smooth adjustment term, and verify the control continuity in combination with the Lyapunov stability criterion to ensure the stability and robustness of the system during modal switching; through the coordinated adjustment of the stator winding frequency and the rotor-side inverter duty ratio, realize a smooth transition between multiple modes, eliminate electrical and mechanical shocks during the switching process, improve the smoothness and reliability of the motor operation, and the full-process closed-loop control design, combined with data-driven modal recognition and model-driven control strategies, takes into account both the dynamic response speed and the steady-state accuracy.

[0035] In a preferred embodiment of the present invention, before step S1, it further includes:

[0036] The rotational speed pulse signal is collected in real time by an encoder installed at the end of the motor rotor shaft, and the instantaneous values of the three-phase current and the output torque waveform are synchronously obtained by using the stator-side current sensor and the torque sensor, generating an original operation data set;

[0037] Perform multi-source heterogeneous data alignment on the original operation data set, adopt the Kalman filtering algorithm to attenuate the high-frequency noise in the rotational speed signal in the frequency domain, and at the same time suppress the electromagnetic harmonic interference of the current signal through sliding window mean filtering, generating a denoised primary data set;

[0038] Calculate the dynamic load change rate based on the rated parameters of the motor and the real-time working conditions, calibrate the dynamic range of the current and torque values in the primary data set according to the dynamic load change rate, and use the piecewise linear interpolation method to correct the normalization scale factor online, generating dynamic normalization parameters;

[0039] According to the dynamic normalization parameters, perform amplitude normalization and time sequence synchronization processing on the denoised primary data set, generating a standardized operation data set containing multi-dimensional correlation features of rotational speed-current-torque.

[0040] In the embodiment of the present invention, when the above steps are specifically applied, they can be specifically implemented through the following steps:

[0041] The rotational speed pulse signal is collected in real time by an encoder installed at the end of the motor rotor shaft. Synchronously, the instantaneous values of the three-phase current are obtained by using the stator-side current sensor, and the output torque waveform is collected by the torque sensor. The three types of signals are aligned according to the time stamp, generating an original operation data set containing rotational speed, current, and torque; perform time synchronization calibration on the multi-source heterogeneous signals (rotational speed pulse, current, torque) in the original data set to ensure that the data sampling points are aligned; for the rotational speed signal, adopt the Kalman filtering algorithm to attenuate the high-frequency noise in the frequency domain and retain the true rotational speed trend; for the current signal, through the sliding window mean filtering method, perform sliding average processing on the periodic electromagnetic harmonic interference to suppress the random pulse noise; finally, generate a denoised primary data set to ensure that the signal is smooth and the time sequence is consistent.

[0042] Based on motor parameters such as rated power, speed, and torque, and combined with real-time operating conditions (such as current speed and output torque), the dynamic load change rate (i.e., the amplitude of load fluctuation over time) is calculated. The current and torque values in the primary dataset are dynamically range-calibrated based on the load change rate (for example, expanding the effective value range of the current signal under high load). Using piecewise linear interpolation, the normalization scale factor is corrected online according to different load variation intervals (such as light load, rated load, and overload) (for example, mapping the current value to the range of 0-1 according to the current load range) to generate dynamic normalization parameters that adapt to real-time operating conditions. The generated dynamic normalization parameters are used to normalize the amplitude of the speed, current, and torque signals in the denoised primary dataset (for example, converting the speed to a percentage of the rated speed and the current to a proportion of the rated current). Timestamp alignment and linear interpolation are used to ensure that the three types of signals are fully synchronized in the time series. Ultimately, a standardized operating dataset containing multidimensional correlation features between speed, current, and torque is generated for input into the subsequent modal identification model. Through synchronized acquisition and time alignment of encoders, current sensors, and torque sensors, the spatiotemporal consistency of speed, current, and torque signals is ensured. Kalman filtering is combined with sliding window mean filtering to specifically eliminate high-frequency noise in the speed signal and electromagnetic harmonics in the current signal, improving signal quality and preventing noise from interfering with modal recognition accuracy. Normalization parameters are adjusted based on the real-time load change rate, allowing the amplitude range of signals such as current and torque to dynamically adapt to the operating conditions, avoiding the feature distortion caused by fixed normalization methods during sudden load changes. Through time synchronization and dynamic normalization, the physical coupling relationship between speed, current, and torque (such as the linkage between current and torque during speed changes) is preserved, providing high-quality input for the bidirectional long-short-term memory network to extract deep dynamic features. The standardized dataset takes into account both steady-state and dynamic operating conditions, adapting to motor operation scenarios under different loads and speeds.

[0043] In a preferred embodiment of the present invention, step S1: constructing a modal recognition model including a bidirectional long short-term memory network, inputting a standardized operating data set, extracting dynamic features of the motor speed-torque curve, and identifying the current operating mode; the operating modes include constant torque mode, weak magnetic speed regulation mode, regenerative braking mode and transient overload mode, including:

[0044] Step S1.1: Constructing a bidirectional long short-term memory network modal recognition model, wherein the input layer receives the time-series speed-current-torque sequence in the standardized operating data set, extracts dynamic features through a bidirectional time window, and outputs a feature vector containing the transient response characteristics of the motor;

[0045] Step S1.2: In the fully connected layer of the bidirectional long short-term memory network, the probability distribution of each operating mode is calculated based on the eigenvector, and the mode corresponding to the maximum probability is selected as the current operating mode;

[0046] Step S1.3: When the transient overload mode is recognized, trigger a modal switching warning signal, and verify the persistence of the overload state based on the second derivative change rate of the speed-torque curve;

[0047] Among them, the dynamic characteristics include the amplitude of speed fluctuation, the gradient of torque change, and the current harmonic distortion rate.

[0048] In the embodiment of the present invention, when the above steps are specifically applied, they can be specifically implemented through the following steps:

[0049] Construct a multi-layer bidirectional LSTM network. The input layer receives the time-series data (speed-current-torque sequence) in the standardized operation dataset. The middle layer extracts features from the sequence through a bidirectional time window (such as 5 forward time steps + 5 backward time steps) to capture the time-series dependence and dynamic change rules of the motor operation state; through the forget gate, input gate, and output gate mechanisms of the LSTM unit, automatically extract dynamic characteristics such as the amplitude of speed fluctuation (reflecting load stability), the gradient of torque change (characterizing acceleration and deceleration characteristics), and the current harmonic distortion rate (reflecting the degree of electromagnetic interference), and output a high-dimensional feature vector containing the transient response characteristics of the motor; input the feature vector output by the bidirectional LSTM into the fully connected layer, calculate the probability distribution of the current state belonging to the four modes of constant torque, field weakening speed regulation, regenerative braking, and transient overload through the Softmax activation function, select the mode corresponding to the maximum probability value as the current operation mode, and output a confidence score (i.e., the maximum probability value). When the transient overload mode is recognized (the probability value exceeds the set threshold, such as 0.8), immediately trigger a modal switching warning signal to prompt the control system to prepare for load mutation, calculate the second derivative of the current speed-torque curve (i.e., the change rate of the change rate). If this value continuously exceeds the critical value (such as for 3 consecutive sampling periods), it is confirmed that the overload state actually exists; if the second derivative quickly drops, it is determined as a short-term disturbance to avoid false triggering.

[0050] In the embodiment of the present invention, the bidirectional LSTM network captures both historical and future time-series information, effectively extracts the dynamic characteristics of the motor operation state (such as speed fluctuation and torque mutation). Compared with traditional machine learning methods, it has stronger representation ability for non-stationary and non-linear motor operation data. Through probability distribution calculation and confidence evaluation, high-precision classification of the four modes of constant torque, field weakening speed regulation, regenerative braking, and transient overload is achieved. Combining the probability threshold and the second derivative verification of the curve, false judgment caused by short-term disturbance is avoided, ensuring timely warning when an overload actually occurs, improving the system safety and anti-interference ability. The extracted dynamic characteristics (such as the current harmonic distortion rate) are directly related to the electromagnetic characteristics of the motor, which helps to quickly adjust control parameters during modal switching and improve the system dynamic response speed. The time-series modeling ability of the bidirectional LSTM enables it to adapt to the operation mode recognition under different load and speed conditions and has stronger robustness to complex working conditions.

[0051] In a preferred embodiment of the present invention, step S2: According to the operating mode, match the initial control parameters from the preset control strategy library, and dynamically iterate and optimize the parameters to obtain an optimization result; based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extract the torque-speed slope and the dynamic response time constant of the two points, and construct a linear correction function, including:

[0052] Step S2.1: According to the identified operating mode, match the corresponding initial proportional-integral parameters, rotor-side voltage limit amplitude, and stator frequency adjustment step size from the preset control strategy library;

[0053] Step S2.2: Use the RMSProp optimization algorithm to dynamically iterate and optimize the initial parameters to minimize the weighted loss function of the torque tracking error and the speed overshoot in the current mode, and generate an optimized set of control parameters;

[0054] Step S2.3: Based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extract the difference in torque-speed slope and the difference in dynamic response time constant at the two points respectively, and construct a linear correction function.

[0055] In the embodiment of the present invention, when the above steps are specifically applied, they can be specifically implemented through the following steps: Four typical control parameter groups corresponding to four operating modes (constant torque, field-weakening speed regulation, regenerative braking, transient overload) are pre-stored in the control strategy library, including proportional-integral (PI) controller parameters, rotor-side voltage limit amplitude, stator frequency adjustment step size, etc. According to the current operating mode identified in step S1, retrieve the initial parameter group corresponding to the mode from the strategy library, for example:

[0056] Constant torque mode: Adopt a larger torque PI gain and a smaller voltage limit;

[0057] Field-weakening speed regulation mode: Reduce the stator magnetic flux and increase the voltage limit;

[0058] Regenerative braking mode: Adjust the PI parameters to achieve energy feedback and set a negative torque limit;

[0059] Transient overload mode: Increase the current limit and improve the dynamic response speed.

[0060] Define a weighted loss function, which includes two indicators: torque tracking error (the deviation between the actual torque and the reference torque) and speed overshoot (the difference between the peak speed and the steady-state value during the dynamic process). Allocate weights according to the importance of the working conditions, and use the RMSProp optimization algorithm to iteratively update the initial parameters. By calculating the gradient information of the loss function, adaptively adjust the learning rate, gradually reduce the loss value, and finally generate an optimized set of control parameters. During the operation of the motor, recalculate the loss function and update the parameters every fixed period (such as 10 ms) to achieve the dynamic optimization of the control parameters.

[0061] Extract the slopes of the torque-speed curve at these two points (representing the system stiffness) from the steady-state operating points of the current mode (such as the current speed and torque values) and the expected operating points of the target switching mode (such as the target speed and torque values) respectively, calculate the dynamic response time constants corresponding to the two points (reflecting the speed of the system transitioning from one steady state to another), calculate the slope difference and the time constant difference, which are used as key characteristic quantities during the mode switching process. Based on the above differences, construct a linear correction function, with the input being the distance between the current operating point and the target operating point, and the output being the correction coefficient used to compensate the control parameters.

[0062] By combining the preset strategy library with dynamic optimization, the control parameters can quickly adapt to the characteristic requirements of different operating modes (such as the constant torque mode emphasizes torque accuracy, and the field-weakening speed regulation mode focuses on the speed range), improving the steady-state performance of the system in each mode. The RMSProp algorithm iteratively optimizes the parameters, effectively reducing the torque tracking error and speed overshoot, enhancing the dynamic response speed and stability of the system, especially performing better during load mutations. The linear correction function provides precise parameter compensation for the mode switching process based on the differences in the working point characteristics, avoiding current shocks and torque fluctuations caused by parameter mutations and achieving a smooth transition. The dynamic iteration mechanism enables the control parameters to adapt to changes in motor parameters (such as resistance changes caused by temperature) and external disturbances in real time, improving the robustness of the system. The RMSProp algorithm has a faster convergence speed compared to the traditional gradient descent method, reducing the parameter optimization time.

[0063] In a preferred embodiment of the present invention, step S3: Fuse the output value of the linear correction function with the optimization result to generate a mode transition compensation coefficient, and inject the compensation coefficient into the rotor-side voltage control loop based on the feedforward control channel to generate an anti-disturbance dynamic compensation control strategy, including:

[0064] Step S3.1: Dynamically weight and fuse the output value of the linear correction function with the control parameters optimized by RMSProp, where the weight is adjusted in real time according to the mode switching transition time, to generate a mode transition compensation coefficient;

[0065] Step S3.2: Inject the compensation coefficient into the reference signal terminal of the rotor-side voltage control loop through the feedforward control channel, and superimpose it on the closed-loop feedback signal to generate a composite control quantity;

[0066] Step S3.3: According to the switching frequency limit of the rotor-side inverter, dynamically allocate the duty cycle of the pulse width modulation for the composite control quantity to generate a dynamic compensation control strategy against load disturbances and voltage fluctuations.

[0067] In the embodiment of the present invention, when the above steps are specifically applied, they can be specifically implemented through the following steps: Dynamically adjust the weight according to the modal switching transition time (such as the switching duration from constant torque to field-weakening speed regulation):

[0068] At the initial stage of switching: Increase the weight of the output value of the linear correction function to quickly respond to the change of the operating point;

[0069] In the middle stage of switching: Balance the weights of the correction function and the optimized parameters, taking into account both dynamic response and stability;

[0070] At the later stage of switching: Decrease the weight of the correction function to make the system gradually transition to the optimized parameters of the target mode.

[0071] Fusion calculation: Weight and fuse the output value of the linear correction function (the compensation amount representing the difference in the operating point) and the control parameters optimized by RMSProp (such as PI gain, voltage limit) according to the real-time weight to generate a modal transition compensation coefficient.

[0072] Introduce a feedforward control channel in the rotor-side voltage control loop, inject the compensation coefficient generated in step S3.1 as a feedforward signal into the reference signal terminal, and superimpose the feedforward compensation signal on the closed-loop feedback signal (such as speed error, torque error) to form a composite control quantity for directly controlling the voltage output of the rotor-side inverter. The feedforward channel can compensate in advance for the system dynamic changes (such as load mutation, inductance parameter change) caused by modal switching, and reduce the influence of disturbances on the control loop.

[0073] According to the maximum switching frequency limit of the rotor-side inverter, perform amplitude limiting processing on the composite control quantity to ensure that the control signal is within the range that can be implemented by the hardware. Convert the amplitude-limited composite control quantity into the duty cycle of a pulse width modulation (PWM) signal, and precisely control the rotor-side voltage by dynamically adjusting the duty cycle of each phase of the PWM wave. During the generation of the PWM, add additional anti-disturbance measures (such as harmonic injection, dead zone compensation) to further suppress the current harmonics and torque pulsation caused by load changes or voltage fluctuations.

[0074] Through dynamic weighted fusion, the control parameters change smoothly during the mode switching process, avoiding current shocks and torque fluctuations caused by parameter mutations, achieving seamless connection between different operating modes. The feedforward control channel compensates in advance for the system dynamic changes caused by mode switching. Compared with pure feedback control, it can respond to disturbances more quickly, reduce dynamic errors. Considering the dynamic duty cycle allocation restricted by the inverter switching frequency, it ensures that the control strategy can be stably implemented on the actual hardware platform, avoiding device losses and electromagnetic interference caused by too high switching frequency. Combining feedforward compensation and closed-loop feedback, the system has stronger resistance to load disturbances and voltage fluctuations, especially performs better under complex working conditions such as regenerative braking and transient overload. The dynamic compensation strategy reduces current harmonics and torque ripples by optimizing the PWM duty cycle allocation, improving the smoothness of the motor output torque and control accuracy.

[0075] In a preferred embodiment of the present invention, step S4: For the non-smooth adjustment term in the dynamic compensation control strategy, the subgradient projection algorithm is used to smooth the discontinuous boundary of the control quantity, and the continuous differentiability of the control surface is verified in combination with the Lyapunov stability criterion to generate the final control instruction, including:

[0076] Step S4.1: For the non-smooth adjustment term caused by the switching logic in the dynamic compensation control strategy, the subgradient projection algorithm is used to construct a smooth transition interval at the discontinuous boundary of the control quantity, and the optimal projection path in the direction of the subgradient within the interval is calculated;

[0077] Step S4.2: Based on the Lyapunov stability criterion, a quadratic function containing the speed error and torque error is constructed to verify the continuous differentiability and stability boundary of the control surface after smoothing;

[0078] Step S4.3: If the stability verification passes, the final control instruction is generated; if not, return to step S2.2 to readjust the control parameters and optimize the correction function weight.

[0079] In the embodiments of the present invention, when the above steps are specifically applied, they can be specifically implemented through the following steps: Locate the non-smooth adjustment terms generated by the mode switching logic (such as voltage limit switching, sudden change of PI parameters) in the dynamic compensation control strategy. These terms appear as discontinuous boundaries in the control quantity space. Expand a certain width (such as ±5% control quantity range) on both sides of the discontinuous boundary to form a smooth transition interval. In the transition interval, use the subgradient projection algorithm to calculate the optimal projection path of the control quantity, so that the control quantity smoothly transitions from one stable region to another along this path, avoiding the jumps caused by the traditional switching logic. Define a quadratic function containing the speed error (the difference between the actual speed and the target speed) and the torque error (the difference between the actual torque and the target torque) as the Lyapunov candidate function, calculate the derivative of the Lyapunov function with respect to time, and analyze its sign characteristics on the control surface (defined by the smoothed control parameters). If the derivative is negative semi-definite on the entire control surface, it proves that the system is stable; at the same time, determine the stability boundary (i.e., the critical condition for the derivative to change from negative to positive) to ensure that the system operates within the boundary. If the Lyapunov stability verification passes, convert the smoothed control parameters into the final control instruction and send it to the motor actuator. If the stability verification fails (such as the derivative is positive in some regions), return to step S2.2, adjust the learning rate of the RMSProp optimization algorithm or correct the weight distribution of the linear correction function, regenerate the control parameters and repeat the verification process.

[0080] The subgradient projection algorithm smooths the non-smooth adjustment terms, eliminates the control chattering caused by sudden parameter changes in traditional switching control, reduces the mechanical vibration of the motor and current harmonics. The Lyapunov stability criterion theoretically ensures the stability of the control strategy in the entire operating region, avoids introducing new unstable factors due to smoothing processing. By verifying the stability boundary, the system has stronger robustness to parameter adjustments (such as changes in motor resistance) and external disturbances (such as sudden load changes), reducing the risk of instability. The iterative adjustment mechanism enables the control strategy to automatically optimize according to the real-time working conditions and still maintain good dynamic performance and steady-state accuracy under complex operating conditions.

[0081] In a preferred embodiment of the present invention, step S5: According to the final control instruction, realize the shockless switching between different operating modes by adjusting the cooperative relationship between the stator winding frequency and the rotor-side inverter duty ratio, and complete the closed-loop control of the motor multi-mode operation, including:

[0082] Step S5.1: According to the stator frequency target value in the final control instruction, adjust the power supply frequency of the stator winding through the space vector modulation algorithm;

[0083] Step S5.2: Synchronously according to the rotor-side inverter duty ratio instruction, adopt the dead-time compensation strategy to eliminate the harmonic components caused by the delay of the switching devices;

[0084] Step S5.3: Monitor the torque ripple and speed jump during the mode switching in real time. If they exceed the preset threshold, trigger the closed-loop iterative optimization process until shock-free switching is achieved.

[0085] In the embodiment of the present invention, when the above steps are specifically applied, they can be specifically implemented through the following steps: Extract the stator frequency target value from the final control instruction, and this value is dynamically determined according to the current operating mode (such as constant torque / field-weakening speed regulation) and load demand.

[0086] Implementation of Space Vector Modulation (SVM):

[0087] Convert the target frequency into the space vector expression of the three-phase stator voltage; calculate the conduction time of each phase switching device through the SVM algorithm to generate a stator voltage waveform close to a sine wave; dynamically adjust the sampling period of the SVM according to the frequency change rate to ensure good voltage waveform quality in the high-frequency field-weakening region.

[0088] Step S5.2: Optimization of the duty cycle of the rotor-side inverter, implementation process:

[0089] Execution of the duty cycle instruction: Generate a PWM control signal according to the rotor-side inverter duty cycle target value in the final control instruction, detect the turn-on / turn-off delay time of the power switching device (such as IGBT / MOSFET), insert a compensation time in the PWM signal to eliminate the voltage waveform distortion caused by the dead time, dynamically adjust the compensation amount according to the switching frequency and load current to improve the compensation accuracy, and add a specific harmonic injection algorithm (such as third harmonic injection) during the duty cycle modulation to reduce the low-order harmonic content in the output voltage.

[0090] Step S5.3: Closed-loop monitoring and iterative optimization, implementation process:

[0091] Collect the motor torque and speed data in real time, calculate the torque ripple amplitude and speed jump during the mode switching, compare them with the preset threshold (such as torque ripple < 5% of the rated torque, speed jump < 3% of the rated speed). If the monitoring index exceeds the threshold, trigger the closed-loop feedback mechanism, adjust the coordinated change rate of the stator frequency and the inverter duty cycle, and dynamically correct the control parameters through a PID controller or a fuzzy algorithm to gradually reduce the dynamic error during the switching process;

[0092] Repeat steps S5.1~S5.3 until shock-free switching is achieved.

[0093] Through the coordinated adjustment of the stator frequency and the inverter duty ratio, smooth transitions among modes such as constant torque, field-weakening speed regulation, and regenerative braking are achieved, eliminating torque mutations and current shocks commonly seen in traditional control methods. The combination of dead-time compensation and the SVM algorithm effectively reduces harmonic components, decreases motor losses and noise, improves system efficiency. The real-time monitoring and iterative optimization mechanism ensures that the system can still maintain stability in the face of parameter perturbations (such as resistance changes caused by temperature) and external disturbances (such as sudden load changes), enhancing control robustness, and maintaining good control performance in both the high-speed field-weakening region and the low-speed constant-torque region, expanding the effective operating range of the motor.

[0094] As Figure 2 shown, a multi-modal control system for a brushless doubly-fed motor includes:

[0095] A construction module for constructing a mode recognition model containing a bidirectional long short-term memory network, inputting a standardized operation data set, extracting the dynamic characteristics of the motor speed-torque curve, and identifying the current operating mode; the operating modes include a constant torque mode, a field-weakening speed regulation mode, a regenerative braking mode, and a transient overload mode;

[0096] A correction module for matching initial control parameters from a preset control strategy library according to the operating mode, dynamically iteratively optimizing the parameters to obtain an optimization result; based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extracting the torque-speed slope and the dynamic response time constant of the two points, and constructing a linear correction function;

[0097] A generation module for fusing the output value of the linear correction function with the optimization result, generating a mode transition compensation coefficient, and injecting the compensation coefficient into the rotor-side voltage control loop based on the feedforward control channel to generate an anti-disturbance dynamic compensation control strategy;

[0098] A control module for smoothing the discontinuous boundary of the control quantity of the non-smooth adjustment term in the dynamic compensation control strategy by using the subgradient projection algorithm, and verifying the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate a final control instruction;

[0099] A switching module for achieving shock-free switching between different operating modes according to the final control instruction by adjusting the coordinated relationship between the stator winding frequency and the rotor-side inverter duty ratio, and completing the closed-loop control of the multi-modal operation of the motor.

[0100] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A multimodal control method for a brushless doubly-fed motor, characterized in that, The method includes: Step S1: Construct a modal recognition model including a bidirectional long short-term memory network, input a standardized operation data set, extract the dynamic characteristics of the motor speed-torque curve, and identify the current operation mode; the operation modes include a constant torque mode, a field weakening speed regulation mode, a regenerative braking mode, and a transient overload mode; Step S2: According to the operation mode, match the initial control parameters from a preset control strategy library, dynamically iterate and optimize the parameters to obtain an optimization result; based on the steady-state operating point of the current mode and the expected operating point of the target switching mode, extract the torque-speed slope and the dynamic response time constant of the two points, and construct a linear correction function; Step S3: Fuse the output value of the linear correction function with the optimization result to generate a modal transition compensation coefficient, and inject the compensation coefficient into the rotor side voltage control loop based on the feedforward control channel to generate a dynamic compensation control strategy for anti-disturbance; Step S4: For the non-smooth adjustment term in the dynamic compensation control strategy, use the subgradient projection algorithm to smooth the discontinuous boundary of the control quantity, and verify the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate a final control instruction; Step S5: According to the final control instruction, realize the shockless switching between different operation modes by adjusting the cooperative relationship between the stator winding frequency and the rotor side inverter duty ratio, and complete the closed-loop control of the motor multi-modal operation.

2. The multimodal control method for a brushless doubly-fed motor according to claim 1, characterized in that Before step S1, it further includes: Real-time collect the speed pulse signal through an encoder installed at the motor rotor shaft end, and synchronously obtain the three-phase current instantaneous value and the output torque waveform by using the stator side current sensor and the torque sensor to generate an original operation data set; Perform multi-source heterogeneous data alignment on the original operation data set, use the Kalman filter algorithm to attenuate the high-frequency noise in the speed signal in the frequency domain, and at the same time suppress the electromagnetic harmonic interference of the current signal through the sliding window mean filter to generate a denoised primary data set; Calculate the dynamic load change rate based on the motor rated parameters and the real-time working conditions, calibrate the dynamic range of the current and torque values in the primary data set according to the dynamic load change rate, and use the piecewise linear interpolation method to online correct the normalization scale factor to generate dynamic normalization parameters; According to the dynamic normalization parameters, perform amplitude normalization and time sequence synchronization processing on the denoised primary data set to generate a standardized operation data set including multi-dimensional correlation characteristics of speed-current-torque.

3. The multimodal control method of the brushless doubly-fed motor according to claim 2, characterized in that Step S1: Construct a modal recognition model including a bidirectional long short-term memory network, input a standardized operation data set, extract the dynamic characteristics of the motor speed-torque curve, and identify the current operation mode; the operation modes include a constant torque mode, a field weakening speed regulation mode, a regenerative braking mode, and a transient overload mode, including: Step S1.1: Construct a modal recognition model of a bidirectional long short-term memory network, the input layer of which receives the time sequence speed-current-torque sequence in the standardized operation data set, extracts dynamic characteristics through a bidirectional time window, and outputs a feature vector including the motor transient response characteristics; Step S1.2: In the fully connected layer of the bidirectional long short-term memory network, calculate the probability distribution of each operating mode according to the feature vector, and select the mode corresponding to the maximum probability as the current operating mode; Step S1.3: When the transient overload mode is recognized, trigger a mode switching warning signal, and verify the persistence of the overload state based on the second derivative change rate of the speed-torque curve; Among them, the dynamic features include the amplitude of speed fluctuation, the gradient of torque change, and the current harmonic distortion rate.

4. The brushless doubly-fed motor multi-modal control method according to claim 3, wherein Step S2: According to the operating mode, match the initial control parameters from the preset control strategy library, and perform dynamic iterative optimization on the parameters to obtain the optimization result; based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extract the torque-speed slope and the dynamic response time constant of the two points, and construct a linear correction function, including: Step S2.1: According to the recognized operating mode, match the corresponding initial proportional-integral parameters, the rotor-side voltage limit amplitude, and the stator frequency adjustment step size from the preset control strategy library; Step S2.2: Use the RMSProp optimization algorithm to perform dynamic iterative optimization on the initial parameters to minimize the weighted loss function of the torque tracking error and the speed overshoot in the current mode, and generate an optimized set of control parameters; Step S2.3: Based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extract the difference in torque-speed slope and the difference in dynamic response time constant at the two points respectively, and construct a linear correction function.

5. The multimodal control method of the brushless doubly-fed motor according to claim 4, characterized in that Step S3: Integrate the output value of the linear correction function with the optimization result to generate a mode transition compensation coefficient, and inject the compensation coefficient into the rotor-side voltage control loop based on the feedforward control channel to generate an anti-disturbance dynamic compensation control strategy, including: Step S3.1: Dynamically weight and integrate the output value of the linear correction function with the control parameters optimized by RMSProp, where the weight is adjusted in real time according to the mode switching transition time, and generate a mode transition compensation coefficient; Step S3.2: Inject the compensation coefficient into the reference signal terminal of the rotor-side voltage control loop through the feedforward control channel, and superimpose it with the closed-loop feedback signal to generate a composite control quantity; Step S3.3: According to the switching frequency limit of the rotor-side inverter, dynamically allocate the pulse width modulation duty cycle of the composite control quantity to generate a dynamic compensation control strategy against load disturbance and voltage fluctuation.

6. The multimodal control method for a brushless doubly-fed motor according to claim 5, characterized in that Step S4: For the non-smooth adjustment term in the dynamic compensation control strategy, use the subgradient projection algorithm to smooth the discontinuous boundary of the control quantity, and verify the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate the final control instruction, including: Step S4.1: For the non-smooth adjustment term caused by the switching logic in the dynamic compensation control strategy, use the subgradient projection algorithm to construct a smooth transition interval at the discontinuous boundary of the control quantity, and calculate the optimal projection path in the direction of the subgradient within the interval; Step S4.2: Based on the Lyapunov stability criterion, construct a quadratic function containing the speed error and the torque error, and verify the continuous differentiability and stability boundary of the control surface after smoothing; Step S4.3: If the stability verification passes, generate the final control instruction; if not, return to Step S2.2 to readjust the control parameters and optimize the correction function weights.

7. A brushless doubly-fed motor multimodal control method according to claim 6, characterized in that Step S5: According to the final control instruction, achieve shockless switching between different operating modes by adjusting the coordination relationship between the stator winding frequency and the rotor-side inverter duty ratio, and complete the closed-loop control of the motor multi-mode operation, including: Step S5.1: According to the stator frequency target value in the final control instruction, adjust the power supply frequency of the stator winding through the space vector modulation algorithm; Step S5.2: Synchronously, according to the rotor-side inverter duty ratio instruction, adopt a dead zone compensation strategy to eliminate the harmonic components caused by the switch device delay; Step S5.3: Real-time monitor the torque pulsation and speed jump during the mode switching process. If they exceed the preset threshold, trigger the closed-loop iterative optimization process until shockless switching is achieved.

8. A system information processing system based on power data, which implements the method described in any one of claims 1 to 7, characterized in that, Including: A construction module for constructing a mode recognition model including a bidirectional long short-term memory network, inputting a standardized operation data set, extracting the dynamic characteristics of the motor speed-torque curve, and identifying the current operating mode; the operating modes include constant torque mode, field weakening speed regulation mode, regenerative braking mode, and transient overload mode; A correction module for matching initial control parameters from a preset control strategy library according to the operating mode, dynamically iteratively optimizing the parameters to obtain an optimization result; based on the steady-state operating point of the current mode and the desired operating point of the target switching mode, extract the torque-speed slope and dynamic response time constant of the two points, and construct a linear correction function; A generation module for fusing the output value of the linear correction function with the optimization result to generate a mode transition compensation coefficient, and injecting the compensation coefficient into the rotor-side voltage control loop based on the feedforward control channel to generate an anti-disturbance dynamic compensation control strategy; A control module for using the subgradient projection algorithm to smooth the discontinuous boundary of the control quantity for the non-smooth adjustment term in the dynamic compensation control strategy, and verifying the continuous differentiability of the control surface in combination with the Lyapunov stability criterion to generate the final control instruction; A switching module for achieving shockless switching between different operating modes according to the final control instruction by adjusting the coordination relationship between the stator winding frequency and the rotor-side inverter duty ratio, and completing the closed-loop control of the motor multi-mode operation.

9. A computing device, characterized in that, Including: One or more processors; A storage device for storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A program is stored in the computer-readable storage medium, and when the program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images