CN-116455279-B - Motor rotation angle self-adaptive prediction compensation method based on neural network

CN116455279BCN 116455279 BCN116455279 BCN 116455279BCN-116455279-B

Abstract

The invention discloses a motor rotation angle self-adaptive prediction compensation method based on a neural network, which belongs to the technical field of motor control, and solves the technical problems of long time and poor precision of the traditional rotation sensor zero offset angle calibration mode; S2, controlling a given d-axis current Id through a motor, S3, calibrating and modifying a zero offset angle delta of rotation, S4, setting a q-axis current Iq given by a program, S5, reading an output torque of the motor to be calibrated, marking as T+, S6, setting a q-axis current-Iq given by the program, S7, reading the output torque of the motor to be calibrated again, marking as T-, S8, cycling the operation of S4 to S7 for N times, adding preset precision to the zero offset angle of rotation each time, S9, setting the input end of the neural network model as the sum value of positive torque and negative torque as |T + +T ‑ |, setting the output end as the zero offset angle delta, inputting a zero value, and predicting the real zero offset angle delta of rotation.

Inventors

PAN MINGZHANG
SU TIECHENG
WANG YUPENG
GUAN WEI
LIANG LU
LIANG KE
TANG YU

Assignees

广西大学

Dates

Publication Date: 20260512
Application Date: 20230330

Claims (5)

1. The motor rotation angle self-adaptive prediction compensation method based on the neural network is characterized by determining the calibration of the rotation zero offset angle by utilizing a controller of the motor and a dynamic test rack, and comprises the following steps of: S1, controlling a DC rated voltage for a motor, and controlling the motor in a torque mode, wherein the motor to be marked is dragged to a set rotating speed by a measurement and control machine, the set rotating speed cannot be located in a field-weakening rotating speed area, and the measurement and control machine records the output torque of the motor to be marked; s2, controlling a given d-axis current Id through a motor; S3, calibrating and modifying the zero deflection angle of the rotary transformer, and marking the zero deflection angle as delta; S4, setting a program to give a q-axis current Iq; s5, reading the output torque of the motor to be marked, and marking as ; S6, setting a program to give a q-axis current-Iq; s7, again reading the output torque of the motor to be marked, and marking as ; S8, cycling the operations from the step S4 to the step S7 for N times, and correcting the rotation zero offset angle each time, wherein the rotation zero offset angle is self-added with preset precision; S9, building neural network model prediction, wherein the input end is the sum value of positive torque and negative torque recorded N times The zero offset angle delta is recorded for N times at the output end, zero values are input into the established neural network model, and the real rotation zero offset angle delta can be predicted, wherein the precision is preset.
2. The motor rotation angle adaptive prediction compensation method based on the neural network according to claim 1, further comprising step S10, modifying the preset precision of step S8 to a specified high precision, and then cycling the operations of step S3 to step S9 to advance the prediction precision to the specified high precision.
3. The motor rotation angle adaptive prediction compensation method based on the neural network according to claim 2, wherein the specified high precision is 0.01.
4. The motor rotation angle self-adaptive prediction compensation method based on the neural network according to claim 1, wherein the preset precision is 0.1.
5. The motor rotation angle adaptive prediction compensation method based on the neural network according to claim 1, wherein N in the step S8 is not less than 100.

Description

Motor rotation angle self-adaptive prediction compensation method based on neural network Technical Field The invention relates to the technical field of motor control, in particular to a motor rotation angle self-adaptive prediction compensation method and device based on a neural network. Background According to vector control of a permanent magnet synchronous motor, in order to maximize torque output by the motor, an electromagnetic field generated by a stator winding is always orthogonal to a permanent magnet field of a rotor, and therefore the position angle of the rotor needs to be accurately obtained. In the ideal state, the development and design stage of the motor can ensure that the zero position of the rotation sensor is overlapped with the A axis, but in the actual production process of the motor, the installation and positioning of the rotation sensor are inconsistent due to machining deviation and installation deviation, so that the deflection angles of the rotation sensors of all motors are inconsistent. Therefore, when the motor is in offline detection, zero offset angle calibration of the rotary sensor is needed, manual calibration is mainly relied on at present, and the problems of long time and low precision exist. When the motor operating frequency is high, time delay of each part of the motor control system related to current sampling and rotor position sampling can cause that the actual feedback current of the motor cannot track the given current after the vector control reaches a steady state, and the calculated PWM duty ratio cannot accurately and equivalently act on the next PWM period, and as the motor operating frequency is improved, the error between the feedback current and the given current is increased, so that the control precision and the operating performance of the motor under the high-frequency operating condition are greatly influenced. Disclosure of Invention The invention aims to solve the technical problem of the prior art, and aims to provide a motor rotation angle self-adaptive prediction compensation method based on a neural network. The technical scheme of the invention is that the motor rotation angle self-adaptive prediction compensation method based on the neural network is characterized in that a controller of the motor and a dynamic test bench are utilized to determine the calibration of the rotation zero offset angle, and the method comprises the following steps: S1, controlling a DC rated voltage for a motor, and controlling the motor in a torque mode, wherein the motor to be marked is dragged to a set rotating speed by a measurement and control machine, the set rotating speed cannot be located in a field-weakening rotating speed area, and the measurement and control machine records the output torque of the motor to be marked; s2, controlling a given d-axis current Id through a motor; S3, calibrating and modifying the zero deflection angle of the rotary transformer, and marking the zero deflection angle as delta; S4, setting a program to give a q-axis current Iq; s5, reading an output torque of the motor to be marked, and marking the output torque as T+; s6, setting a program to give a q-axis current-Iq; s7, reading the output torque of the motor to be marked again, and marking the output torque as T-; S8, cycling the operations from the step S4 to the step S7 for N times, and correcting the rotation zero offset angle each time, wherein the rotation zero offset angle is self-added with preset precision; s9, building a neural network model prediction, wherein the input end is the sum value |T ++T- | of positive torque and negative torque recorded for N times, the output end is zero offset angle delta recorded for N times, zero value is input into the built neural network model, and the real rotation zero offset angle delta can be predicted, and the precision is preset. As a further improvement, step S10 is further included, the preset precision of step S8 is modified to the specified high precision, and then the operations of steps S3 to S9 are looped, so that the prediction precision can be advanced to the specified high precision. Further, the specified high precision is 0.01. Further, the preset precision is 0.1. Further, N in step S8 is not less than 100. Advantageous effects Compared with the prior art, the invention has the advantages that: 1. The invention can automatically run after the algorithm model is built, does not need personnel to participate, has high degree of automation, and saves labor and time cost compared with the traditional method. 2. The prediction accuracy in the invention can increase the circulation times of the program to achieve higher prediction accuracy than manual operation. Drawings FIG. 1 is a schematic diagram of a stationary coordinate system ABC; FIG. 2 is a schematic diagram of a stationary coordinate system αβ; FIG. 3 is a schematic diagram of a rotor synchronous rotation coordinate system dq; FIG. 4 is a schematic il