An axial flux permanent magnet synchronous motor non-inductive control method, device and storage medium suitable for low-altitude aircraft

Abstract

The application discloses a kind of axial flux permanent magnet synchronous motor non-inductive control methods, equipment and storage medium suitable for low altitude aircraft, belong to permanent magnet synchronous motor control technical field.The application does not depend on physical position / speed sensor, through physical perception sampling and non-equal interval time coding adaptation flight dynamic sampling;Adopt TCN and Mamba-SSM parallel structure to extract multi-scale electromagnetic physical characteristics;Rely on phi-NN architecture and global-local GLAR strategy, learn and solidify motor dq axis electromagnetic physical topology;Through physical sparse pruning and working condition prototype matching, high-precision estimation of electric angle and speed under small sample is realized;Finally access field-oriented control system to form full closed-loop control.The application solves the problems of model mismatch, poor physical consistency, weak non-equal interval sampling adaptation and difficult small sample generalization of traditional scheme, and is suitable for non-inductive control of axial flux permanent magnet synchronous motor of low altitude aircraft propulsion system.

Claims

1. A sensorless control method for an axial flux permanent magnet synchronous motor suitable for low-altitude aircraft, characterized in that, Includes the following steps: S1. The sampling frequency of the stator three-phase current is dynamically adjusted according to the real-time speed of the motor. The stator three-phase current is collected and transformed into dq axis current. The precise physical timestamp of each sampling is recorded synchronously. The time interval is mapped to the attention bias through the non-equal interval time attention mechanism and embedded into the neural network input layer. S2. Using the dq-axis current carrying a precise physical timestamp as input, the high-frequency transient features of the current are extracted by using the temporal convolutional network branch, and the global long-range temporal dependence features of the motor are extracted by using the Mamba state space model branch. The two types of branch features are fused to obtain multi-scale electromagnetic physical features. S3. Using multi-scale electromagnetic physical characteristics as input, and based on the physical structure embedded neural network and global-local adaptive resampling strategy, the physical information neural network is used as the teacher network to learn the electromagnetic operation law of the motor, and a lightweight student network with embedded motor dq axis electromagnetic physical topology is obtained through knowledge distillation. S4. Perform physical sparse pruning optimization on the output features of the lightweight student network, and combine it with the working condition prototype matching to realize the estimation output of rotor electrical angle and speed. S5. The estimated rotor electrical angle and speed are connected to the field-oriented control system to complete the coordinate transformation and closed-loop speed adjustment. At the same time, the real-time speed is transmitted back to the sampling end to dynamically adjust the acquisition frequency, forming a closed-loop sensorless control throughout the entire process.

2. The sensorless control method for axial flux permanent magnet synchronous motors suitable for low-altitude aircraft according to claim 1, characterized in that, In S1, differentiated sampling control is achieved based on the motor speed. High-frequency sampling is used for low-speed operation and low-frequency sampling is used for high-speed operation. No data interpolation is performed during the sampling process, and the original timing characteristics of non-equal interval sampling are fully preserved.

3. The sensorless control method for axial flux permanent magnet synchronous motors suitable for low-altitude aircraft according to claim 1, characterized in that, In S2, the temporal convolutional network uses an expanded convolutional structure to extract high-frequency transient features of the current and information on the rate of change of the current; the Mamba state-space model uses a selective scanning mechanism to model the global long-range temporal dependence of the motor with linear complexity. The global long-range temporal dependence features of the motor are the parameter fluctuation features of the motor, including stator resistance drift, permanent magnet flux decay, and device aging parameter features.

4. The sensorless control method for axial flux permanent magnet synchronous motors suitable for low-altitude aircraft according to claim 1, characterized in that, In S3, the teacher network adopts a physical information neural network, using the extended back EMF state equation of the motor in the full speed domain as a physical constraint, and combines automatic differentiation technology to construct a joint loss function of data-driven loss and physical residual loss to fit the electromagnetic state evolution law of the motor.

5. The sensorless control method for axial flux permanent magnet synchronous motors suitable for low-altitude aircraft according to claim 1, characterized in that, In S3, the global-local adaptive resampling strategy combines Monte Carlo integration to complete the model accuracy evaluation, and simultaneously performs global random sampling and local residual hotspot region resampling to optimize the spatial distribution of training points.

6. The sensorless control method for axial flux permanent magnet synchronous motors suitable for low-altitude aircraft according to claim 1, characterized in that, In S3, the student network uses the continuous state field output by the teacher network as a soft label and learns the physical prediction trajectory through knowledge distillation. During the training process, L2 regularization constraints are added, and the physical topology relationship matrix of the motor dq axis is extracted through hierarchical agglomerative clustering and hard-embedded into the student network architecture.

7. The sensorless control method for axial flux permanent magnet synchronous motors suitable for low-altitude aircraft according to claim 1, characterized in that, In S4, a physical dimension mask is constructed based on the electromagnetic coupling mechanism of the motor's dq axis. The redundancy features are pruned by combining physical sparse attention and prototype attention dual mechanisms. The global-local adaptive resampling strategy is reused to complete physical topology locking under small sample conditions.

8. The sensorless control method for axial flux permanent magnet synchronous motors suitable for low-altitude aircraft according to claim 1, characterized in that, In S5, the estimated electrical angle is used to complete the Park transformation and inverse Park transformation, thereby decoupling the stator current excitation component and the torque component; the estimated speed is fed back to the speed loop PI regulator to dynamically adjust the torque current and achieve stable control of the motor speed without steady-state error.

9. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the sensorless control method for an axial flux permanent magnet synchronous motor suitable for low-altitude aircraft as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the sensorless control method for an axial flux permanent magnet synchronous motor suitable for low-altitude aircraft as described in any one of claims 1 to 8.

Images