Driving optimization method of permanent magnet synchronous motor for radar servo system

Abstract

The invention provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system, and belongs to the technical field of radar servo control, and the method comprises the steps: a motor parameter online identification module based on Kalman filtering identifies and constructs a state vector, determines a current working mode of the motor according to the state vector, determines a cooperation parameter, and carries out the optimization of the current working mode according to the cooperation parameter; the motor is preliminarily adjusted by combining feedforward current and variable bias anti-backlash current, a motor differential error of the radar servo system is determined by adopting load weighted differential negative feedback, a current reference value of the motor after preliminary adjustment is determined, and a voltage vector duty ratio of adaptive space vector pulse width modulation is adjusted in real time based on output of a BP neural network; the composite control signal is obtained to drive the permanent magnet synchronous motor, and the precise driving requirement of the permanent magnet synchronous motor is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar servo control, and in particular to a driving optimization method for a permanent magnet synchronous motor of a radar servo system. BACKGROUND

[0002] The radar servo system is the core execution unit for realizing accurate pointing of the radar beam and completing target detection and tracking, and its driving performance directly determines the detection accuracy and response speed of the radar. The permanent magnet synchronous motor (PMSM) is widely used in radar servo driving due to its high power density and wide speed regulation range, and the space vector pulse width modulation (SVPWM) is the core technology of motor driving, and the optimization of its duty cycle directly determines the driving performance. However, the existing SVPWM optimization method lacks intelligence in the radar scene and is difficult to adapt to complex working conditions, and the specific defects are as follows:

[0003] The traditional SVPWM relies on a fixed PI controller or a simple model prediction, and its proportional / integral coefficient is a fixed value that is offline adjusted, which cannot be adaptively adjusted with mode switching, resulting in a mismatch between the duty cycle and the current demand. For example, during fine scanning, current chattering may be caused by overly sensitive parameters, and during fast scanning, the current tracking error increases due to response lag, ultimately affecting the radar beam pointing accuracy or dynamic response speed. The existing technology is mostly single-dimensional decision making (such as only based on current error), and does not intelligently fuse and weight multiple sources of information. In complex working conditions such as multi-motor synchronization, it cannot output accurate duty cycles, limiting the overall performance of the radar servo. It is difficult to meet the precise driving requirements of the permanent magnet synchronous motor.

[0004] Therefore, the present application proposes a driving optimization method for a permanent magnet synchronous motor of a radar servo system. SUMMARY

[0005] The present application provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system to solve the technical problems mentioned above.

[0006] The present application provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system, comprising:

[0007] Step 1: The motor parameter online identification module based on Kalman filtering identifies and constructs a state vector, and determines the current working mode of the motor according to the state vector, wherein the state vector includes: motor speed, direct-axis current, cross-axis current and load torque;

[0008] Step 2: Estimate the load torque observation value based on the load torque observer and combining motor direct-axis and quadrature-axis coupling characteristics to calculate feedforward current, designing variable bias and clearance elimination current to compensate gear clearance, and determining cooperative parameters based on the state vector and current working mode, and combining the feedforward current and variable bias and clearance elimination current to preliminarily adjust the motor, wherein the cooperative parameters include proportional coefficient and integral coefficient of a load-weighted differential negative feedback PI regulator based on the current working mode;

[0009] Step 3: determining motor differential error of the radar servo system by using load-weighted differential negative feedback , wherein N is the total number of permanent magnet synchronous motors participating in synchronization in the radar servo system; is the load weight coefficient of the i-th motor; is the average speed of the i-th motor, and the unit is rad / s; is the reference speed, and the unit is rad / s;

[0010] Step 4: determining the direct-axis current reference value of the motor after preliminary adjustment , the quadrature-axis current reference value , and based on the output of the BP neural network, adjusting the duty cycle of the adaptive space vector pulse width modulation voltage vector in real time to obtain a composite control signal to drive the permanent magnet synchronous motor, wherein the input of the BP neural network is: the direct-axis current reference value after preliminary adjustment , the quadrature-axis current reference value , the motor differential error , the motor current error based on preliminary adjustment , and the load torque observation value , and the output of the BP neural network is: the predicted direct-axis current value and the predicted quadrature-axis current value.

[0011] Preferably, the current working mode of the motor is determined according to the state vector, including:

[0012] Each vector element in the state vector is matched with a mode reference table in sequence to determine the current working mode;

[0013] , wherein the current working mode includes: radar fine scanning mode, radar regular scanning mode, and radar fast scanning mode;

[0014] When in the radar fine scanning mode, the modified sliding mode control and the maximum torque current ratio control are adopted;

[0015] When in the radar regular scanning mode, the model predictive control and the active disturbance rejection control are adopted;

[0016] When in the radar fast scanning mode, the model predictive control and the dynamic field weakening control are adopted;

[0017] And the working mode switching condition is: when the motor speed , the radar is switched from the normal scanning mode to the fast scanning mode, when the motor speed , and the beam pointing error , the radar is switched from the normal scanning mode to the fine scanning mode.

[0018] Preferably, the change rate of the modified sliding mode control sliding surface is:

[0019] ;

[0020] Wherein, is the sliding surface, unit: rad / s; is the sliding mode control coefficient, unit: ; , n is the shape parameter; is the decay coefficient of the exponential term, unit: ; is the coefficient of the saturation function, unit: ; is the saturation function; is the boundary coefficient of the saturation function, unit: rad / s;

[0021] Wherein, the control law of the modified sliding mode control , wherein K is the control gain.

[0022] Preferably, , wherein, is the real-time load of the i-th motor, unit: , is the rated load of the motor, unit: , and The value range of is 1 to 1.1;

[0023] The motor speed is corrected by a PI regulator, the proportional coefficient and the integral coefficient of the PI regulator, wherein the multi-motor synchronous starting process is: optimizing the motor in descending order according to the load weight coefficient , and the adjacent motor optimization delay is 0.1s.

[0024] Preferably, the variable bias anti-pumping current satisfies:

[0025] ;

[0026] Wherein, is the zero point bias current, is the rated current of the motor; is a current compensation for the inflection point; is a current threshold value; is an actual current signal of the motor.

[0027] Preferably, the reference speed The radar beam pointing error feedback correction specifically includes:

[0028] The deviation angle between the actual pointing and the target pointing is obtained by the radar antenna based beam pointing measurement device

[0029] If , according to The reference speed is adjusted in real time, wherein is the initial reference speed; .

[0030] Preferably, the output of the BP neural network is used to adjust the voltage vector duty ratio of the adaptive space vector pulse width modulation in real time, including:

[0031] The direct-axis current reference value of the motor after preliminary adjustment , the quadrature-axis current reference value , the basic duty ratio is determined;

[0032] The prediction compensation amount is determined, and , , is the output direct-axis current value and the output quadrature-axis current value based on the current prediction model at the kth moment; , is the predicted direct-axis current value and the predicted quadrature-axis current value output by the BP neural network; is the error between the BP neural network output and the current prediction model; is a prediction proportional coefficient, with a unit of , is a prediction integral coefficient, with a unit of ;

[0033] The basic compensation amount is determined, wherein is the error between the BP neural network output and the measured current; , , is the measured direct-axis current and the measured quadrature-axis current at the kth moment; is a compensation proportional coefficient, with a unit of , is a compensation integral coefficient, with a unit of ;

[0034] The voltage vector duty ratio is calculated .

[0035] Preferably, the load torque observation value satisfy:

[0036] ,in, The base load torque observation value output by the extended stater; For adaptive gain coefficients; This is the error in the estimation of the load torque; This is the estimated load torque obtained based on the mechanical motion equations of the electric motor; It is a symbolic function; It is a non-linear adjustment index;

[0037] Generate feedforward current At that time, considering the direct-axis coupling relationship of the motor, the following conditions are met:

[0038] ,in, This represents the number of pole pairs of the motor. For permanent magnet flux linkage; It is a direct-axis inductor; It is a quadrature axis inductor; It is the direct-axis current.

[0039] Preferably, the BP neural network adopts a dual-branch structure enhanced by multi-scale feature fusion and attention mechanism, specifically including:

[0040] Input direct-axis current reference value Quadrature axis current reference value Based on a single fully connected layer, the static current demand is feature-encoded and a static feature vector is output.

[0041] Input motor differential error Based on the initial adjustment of motor current error and load torque observations Based on two fully connected layers, dynamic perturbation and observation information are feature encoded and input as dynamic feature vectors. The first layer of the two fully connected layers contains 24 neurons and the activation function is LeakyReLU, and the second layer contains 16 neurons and the activation function is LeakyReLU.

[0042] The static and dynamic feature vectors are concatenated, and channel attention weights A are generated through global average pooling, global max pooling, and a shared connection layer, satisfying the following:

[0043] ,in, For the Sigmoid function; This is for fully connected layer operations; is a global average pooling operation; is a global max pooling operation; is a concatenation feature vector;

[0044] channel-weighted features , the spatial attention weight is generated by a 1x1 convolution layer and a Sigmoid function , and the final fusion feature is obtained wherein, is an element-wise multiplication, is a convolution operation using a convolution kernel size of 1x1;

[0045] The 2-layer fully connected layer is used to output a predicted direct-axis current value and a predicted quadrature-axis current value, wherein the loss function is a multi-objective loss with dynamic weights and regularization: wherein, is a regularization coefficient; is an L2 regularization term of the network weight W; , , is a reference speed error amplitude; is a reference current error amplitude, is an L2 regularization function.

[0046] Compared with the prior art, the beneficial effects of the present application are as follows:

[0047] By Kalman filtering, the motor state is identified online, the load torque observation and feedforward compensation, the variable bias current is used to eliminate the gear gap, the load weighted differential negative feedback is used to optimize the multi-motor synchronization, and the BP neural network is used to adaptively adjust the SVPWM duty ratio. According to different working modes such as fine scanning, conventional and fast, the control parameters are adaptively matched, the motor current chattering is effectively reduced, the influence of gear gap on precision is compensated, and the multi-motor synchronization precision is improved. Finally, the beam pointing precision and response speed of the radar servo system are improved, and the precise driving demand of the permanent magnet synchronous motor is met.

[0048] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the structures specifically pointed out in the written description and drawings.

[0049] The technical solutions of the present application will be further described in detail below with the help of the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0050] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and are used for explaining the application together with the general description given above and the detailed description given below and are intended to provide further description of subject matter claimed. In the drawings:

[0051] Figure 1 A flow chart of a driving optimization method for a permanent magnet synchronous motor of a radar servo system in an embodiment of the application;

[0052] Figure 2 A structure diagram of a BP neural network in an embodiment of the application. DETAILED DESCRIPTION

[0053] The preferred embodiments of the application are described below in conjunction with the accompanying drawings, which should be understood to be a part of the specification, illustrate embodiments of the application and are used for explaining the application together with the general description given above and the detailed description given below and are intended to provide further description of subject matter claimed.

[0054] The application provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system, comprising:

[0055] Step 1: a motor parameter online identification module based on Kalman filtering identifies and constructs a state vector, and determines the current working mode of the motor according to the state vector, wherein the state vector comprises motor speed, direct-axis current, cross-axis current and load torque;

[0056] In this embodiment, the state equation and the observation equation of the online identification module are as follows:

[0057] State equation

[0058] Observation equation , wherein x is a state vector and is , wherein motor speed, direct-axis current, cross-axis current and load torque, respectively; is a stator resistance, , are direct-axis inductance and cross-axis inductance, respectively; is the number of pole pairs; is a permanent magnet flux linkage; J is a rotational inertia. is a direct-axis voltage input, is a cross-axis voltage input; is a process noise vector, representing random disturbances inside the motor system, such as parameter fluctuations, load sudden changes and other unpredictable factors, which is part of the system's own uncertainty described in the Kalman filtering state equation; is an observation noise vector, representing random errors introduced in the measurement process by sensors such as current and speed sensors, which is part of the measurement uncertainty described in the Kalman filtering observation equation.

[0059] Step 2: Estimate the load torque observation value based on the load torque observer , and calculate the feedforward current in combination with the motor direct-axis and quadrature-axis coupling characteristics, design the variable bias and clearance compensation current to compensate for the gear clearance, and determine the coordination parameters based on the state vector and the current working mode, and combine the feedforward current and the variable bias and clearance compensation current to preliminarily adjust the motor, wherein the coordination parameters include the proportional coefficient and the integral coefficient of the load-weighted differential negative feedback PI regulator based on the current working mode;

[0060] Step 3: Determine the motor differential error of the radar servo system using the load-weighted differential negative feedback , wherein N is the total number of permanent magnet synchronous motors participating in synchronization in the radar servo system; is the load weight coefficient of the i-th motor; is the average speed of the i-th motor, with the unit of rad / s; is the reference speed, with the unit of rad / s;

[0061] Step 4: Determine the direct-axis current reference value of the preliminarily adjusted motor , the quadrature-axis current reference value , and based on the output of the BP neural network, adjust the voltage vector duty cycle of the adaptive space vector pulse width modulation in real time to obtain a composite control signal to drive the permanent magnet synchronous motor, wherein the input of the BP neural network is: the preliminarily adjusted direct-axis current reference value , the quadrature-axis current reference value , the motor differential error , the motor current error based on the preliminarily adjusted motor , and the load torque observation value , and the output of the BP neural network is: the predicted direct-axis current value and the predicted quadrature-axis current value.

[0062] In this embodiment, the Kalman filtering algorithm is used to collect and process motor operating data (such as speed and current) in real time to identify the function module of the key state of the motor, for example, the speed signal of the motor encoder, the direct-axis current and the quadrature-axis current of the Hall current sensor are collected once every 10 milliseconds, and then the load torque is estimated through the Kalman filtering algorithm, and finally a state vector containing the motor speed, direct-axis current, quadrature-axis current and load torque is constructed.

[0063] State vector is a collection of key state variables of motor operation. Motor speed is the angular velocity of motor rotation, with a unit of rad / s, measured by an encoder. For example, the speed is 3 rad / s when the radar servo motor is fine scanning. Direct-axis current is the current of the rotor magnetic pole axis direction (d-axis) of the permanent magnet synchronous motor, detected by a Hall current sensor. For example, the current is 2 A in the normal mode. Quadrature-axis current is the q-axis current perpendicular to the d-axis, also detected by a Hall current sensor. For example, the current is 15 A in the fast mode. Load torque is the resistance torque of the motor driving the radar servo system, estimated by Kalman filtering. For example, the torque caused by air resistance when the radar antenna rotates is 10 N·m.

[0064] Working mode is a motor operation mode divided according to the range of each parameter in the state vector, used to match different control requirements.

[0065] Load torque observer is an algorithm module for estimating the observed value of load torque by combining motor current, speed and other signals with the motor mathematical model (such as voltage equation).

[0066] Motor direct and quadrature axis coupling characteristics are the characteristics that the d-axis (direct axis) and q-axis (quadrature axis) of the permanent magnet synchronous motor are mutually influenced by the difference in inductance and other parameters. When calculating the feedforward current, this characteristic needs to be considered. For example, when the d-axis current increases by 1 A, the q-axis current will change by 0.3 A due to the coupling effect, so the feedforward current needs to compensate for this coupling amount to make the q-axis current more accurate.

[0067] Feedforward current is the current calculated in advance to compensate for disturbances such as load.

[0068] Variable bias anti-backlash current is a bias current dynamically adjusted according to the load torque to compensate for the gear transmission backlash.

[0069] Coordination parameters refer to the proportional and integral coefficients of the load-weighted differential negative feedback PI regulator, and the parameter values change with the working mode. For example, in the fine scanning mode, the proportional coefficient is set to 0.2 and the integral coefficient is set to 0.05 to pursue stable control. In the fast mode, the proportional coefficient is set to 0.8 and the integral coefficient is set to 0.01 to pursue fast response, which are all pre-set.

[0070] Load-weighted differential negative feedback is a feedback method for optimizing synchronization accuracy when multiple motors are synchronized. According to the load weight of each motor, the difference between the actual speed and the reference speed of the motor is weighted and summed.

[0071] In this embodiment, the motor control expects the current target value of the d-axis and q-axis. For example, in the fine scanning mode, the d-axis current target value is set to 0 A and the q-axis current target value is set to 8 A to pursue low torque fluctuation. In the fast mode, the d-axis current target value is set to 0 A and the q-axis current target value is set to 15 A to pursue high torque. ​= -1A (field weakening control), = 0A (normal operation), = 20A. = 20A.

[0072] The BP neural network is a kind of multilayer feedforward neural network trained by back propagation algorithm, which is used to learn the mapping relationship between input and output.

[0073] The adaptive space vector pulse width modulation (SVPWM) is a modulation method for controlling motor voltage and current by adjusting the duty cycle of voltage vector, and here the duty cycle is adjusted in real time based on the BP neural network output. For example, if the BP neural network predicts that the quadrature axis current needs to increase by 0.2A, the duty cycle of the corresponding voltage vector in SVPWM is adjusted to make the actual quadrature axis current close to the predicted value, thereby improving the current control accuracy.

[0074] The composite control signal is a control signal fused with feedforward current, variable bias current for eliminating backlash, load weighted differential negative feedback adjustment and BP neural network optimization, which is used to drive the permanent magnet synchronous motor. For example, the final output control signal contains 1A bias current feedforward 2A current BP neural network optimized duty cycle adjustment, so that the motor can run stably under complex working conditions.

[0075] The beneficial effects of the above technical solutions are: through Kalman filtering online identification of motor state, load torque observation and feedforward compensation, variable bias current for eliminating backlash, load weighted differential negative feedback optimization of multiple motor synchronization, combined with BP neural network adaptive adjustment of SVPWM duty cycle, adaptive matching of control parameters according to different working modes such as precision scanning, conventional scanning and fast scanning, the influence of motor current chattering and gear backlash on precision is effectively reduced, the precision of multiple motor synchronization is improved, and finally the beam pointing precision and response speed of the radar servo system are improved, meeting the precise driving demand of the permanent magnet synchronous motor.

[0076] The application provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system, which determines the current working mode of the motor according to a state vector, including:

[0077] Each vector element in the state vector is matched with a mode comparison table in sequence to determine the current working mode;

[0078] The current working mode includes: radar precision scanning mode, radar conventional scanning mode and radar fast scanning mode.

[0079] When in the radar precision scanning mode, the modified sliding mode control and maximum torque current ratio control are adopted.

[0080] When in the radar conventional scanning mode, the model predictive control and active disturbance rejection control are adopted.

[0081] When in the radar fast scanning mode, the model predictive control and dynamic field weakening control are adopted.

[0082] and the working mode switching condition is: when the motor speed switches from the radar normal scanning mode to the radar fast scanning mode when the motor speed and the beam pointing error switches from the radar normal scanning mode to the radar fine scanning mode.

[0083] In this embodiment, the motor speed is measured in real time by the encoder, the direct axis / quadrature axis current is collected by the current sensor, the load torque is estimated by the algorithm, and then these parameters are compared with the range of each mode in the comparison table one by one to determine the current working mode. For example, if the motor speed collected is 4 rad / s, the beam pointing error (measured by the radar position sensor) is 0.0008 rad, and the comparison table requires that the radar fine scanning mode requires a speed of ≤5 rad / s and a beam error of ≤0.001 rad, therefore it is matched to the fine scanning mode. It should be noted that the mode comparison table is a table of state parameter ranges corresponding to different working modes preset in advance.

[0084] In this embodiment, the radar fine scanning mode is used for scenarios requiring high precision and small range scanning of the radar, and uses reconstructed sliding mode control and maximum torque current ratio control. The reconstructed sliding mode control is an improvement on the traditional sliding mode control, which adjusts the shape parameters and decay coefficients of the sliding surface to make the control process smoother and reduce the chattering of current and speed; the maximum torque current ratio control (MTPA) allows the motor to generate the maximum torque with the minimum current, improving energy efficiency while ensuring scanning accuracy. For example, when fine scanning, the motor drives the radar antenna to make a small angle adjustment, the reconstructed sliding mode control can control the speed fluctuation within 0.1 rad / s, and the MTPA control optimizes the quadrature axis current to maximize the torque to current ratio, ensuring high precision pointing.

[0085] The radar normal scanning mode is adapted to the scanning needs of the radar at medium speed, balancing accuracy and response, and uses model predictive control and active disturbance rejection control. Model predictive control (MPC) is based on the mathematical model of the motor, which predicts the current, speed and other states in the future for several periods, and selects the optimal voltage vector in advance; active disturbance rejection control (ADRC) estimates and compensates for load changes (such as wind load disturbance of the radar antenna) and internal system disturbances in real time through an extended state observer. For example, in the normal mode, the motor speed is maintained at about 15 rad / s, the MPC predicts the current demand every 50 microseconds, and the ADRC compensates for a wind load disturbance of ±0.5 Nm, so that the current control response time is within 50 milliseconds, balancing accuracy and response speed.

[0086] The radar fast scanning mode is to meet the demand of large range and high speed scanning of the radar, and model predictive control and dynamic field weakening control are adopted. The model predictive control calculates the optimal voltage vector quickly to ensure high response speed. The dynamic field weakening control adjusts the direct axis current to a negative value to weaken the permanent magnet field when the motor runs at high speed, so that the motor breaks through the base speed (such as 30 rad / s) and runs. For example, in the fast mode, the motor needs to accelerate to 40 rad / s, the model predictive control updates the voltage vector hundreds of times per second, and the dynamic field weakening control adjusts the direct axis current to -2A to weaken the magnetic field, so that the motor can still output enough torque at high speed, while maintaining millisecond-level response speed.

[0087] The working mode switching condition is that when the motor speed reaches the fast mode threshold (such as ≥ 30 rad / s), the system is switched from the normal mode to the fast mode; and when the motor speed drops to the fine scanning mode threshold (such as ≤ 5 rad / s) and the beam pointing error meets the high precision requirement (such as ≤ 0.001 rad), the system is switched from the normal mode to the fine scanning mode. When switching, the parameters of the control algorithm (such as controller gain, algorithm internal coefficient) are smoothly transitioned to avoid current or speed mutation. For example, when the motor speed is accelerated from 25 rad / s to 31 rad / s in the normal mode, the system detects and smoothly switches to the fast mode, and the prediction step of the model predictive control and the current adjustment rate of the dynamic field weakening control are synchronized to adapt to high speed running. If the speed drops to 4 rad / s and the beam error is 0.0009 rad in the normal mode, the system is smoothly switched to the fine scanning mode, and the parameters of the improved sliding mode control and MTPA control are gradually adjusted to the fine scanning adaptation value.

[0088] The beneficial effects of the above technical solutions are: through matching of the state vector and the mode comparison table, adaptive switching of the fine scanning, normal and fast three radar scanning modes is realized, and the control strategies suitable for different modes are adopted to ensure extremely high pointing accuracy in fine scanning, balance performance in normal mode and high response speed in fast mode, while avoiding control impact through smooth mode switching, and comprehensively improving the scanning performance of the radar servo system in different scenes.

[0089] The application provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system. The change rate of the sliding surface of the improved sliding mode control is

[0090] ;

[0091] Wherein, is the sliding surface, and the unit is rad / s; is the sliding mode control coefficient, and the unit is ; , n is a shape parameter; is the attenuation coefficient of the exponential term, and the unit is ; is the coefficient of the saturation function, unit is ; is the saturation function; is the boundary coefficient of the saturation function, unit is rad / s;

[0092] wherein the control law of the reconstruction sliding mode control is wherein K is the control gain.

[0093] Preferably, wherein, is the real-time load of the i-th motor, unit is , is the rated load of the motor, unit is , and the value range of is 1 to 1.1;

[0094] corrected from the motor speed by a PI regulator, the proportional coefficient of the PI regulator is , and the integral coefficient is wherein the multi-motor synchronous starting process is: optimizing the motor in descending order according to the load weight coefficient , and the adjacent motor optimization delay is 0.1s.

[0095] Preferably, the variable biasing and gap elimination current satisfies:

[0096] ;

[0097] wherein, is the zero point biasing current, is the rated current of the motor; is the inflection point compensation current; is the current threshold value; is the actual current signal of the motor.

[0098] Preferably, the reference speed is corrected by the radar beam pointing error feedback, specifically including:

[0099] the deviation angle between the actual pointing and the target pointing is obtained by the beam pointing measurement device based on the radar antenna

[0100] if , the reference speed is adjusted in real time according to , wherein, is the initial reference speed; .

[0101] In this embodiment, the sliding mode surface s is a physical quantity (unit rad / s) that synthesizes the motor speed, current and other states and target state deviation, and is used to reflect the overall situation of control deviation; the sliding mode surface change rate The dynamic change speed is described. The sliding mode control coefficient q is 5 rad / s2, which is determined by simulation to make the sliding mode surface tend to be stable quickly, the shape parameters m=1.0 and n=0.2, which are verified by MATLAB simulation, the chattering is smallest under this value, which meets the smoothness requirement of the fine scanning mode, the exponential term decay coefficient =0.8s / rad (control exponential term decay speed, make the sliding mode surface change more smoothly), the saturation function coefficient =8 rad / s2 (balance control speed and smoothness), the saturation function boundary coefficient =0.2 rad / s, which defines the linear interval boundary of the saturation function, when ∣ ∣≤1, sat( )= ; otherwise, sign( ).

[0102] According to the sliding mode surface s, the control amount u is calculated to adjust the motor voltage / current command, and the control gain K is 10, so that the control amount changes linearly with the sliding mode surface deviation and quickly corrects the motor state.

[0103] In this embodiment, the gear backlash is eliminated by segmented current compensation, and the radar servo transmission precision is improved.

[0104] In this embodiment, the beam pointing error is quickly corrected by adjusting the reference speed, and the radar pointing precision is improved. The initial reference speed is 10 rad / s, and the initial reference speed is set in advance and directly used.

[0105] The beneficial effects of the above technical scheme are: by optimizing the load weight coefficient of the smoothness and rapidity of the motor state adjustment through the reconstructed sliding mode control, the PI regulator realizes the cooperation of multiple motor synchronous starting and speed accurate correction variable bias, gear backlash elimination current compensation, radar beam pointing error feedback correction reference speed and other links, improves the control precision of the permanent magnet synchronous motor and the multi-motor synchronization, eliminates the influence of the mechanical transmission backlash, and finally guarantees the beam pointing precision and response speed of the radar in the fine scanning, conventional scanning and fast scanning modes.

[0106] The application provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system, which adjusts the duty cycle of a voltage vector of adaptive space vector pulse width modulation in real time based on the output of a BP neural network, and includes the following steps:

[0107] Based on the preliminary adjusted direct-axis current reference value and quadrature-axis current reference value , determine the basic duty cycle ;

[0108] determine the prediction compensation amount , and , , is the output direct-axis current value and the output quadrature-axis current value based on the current prediction model at the kth moment; , is the predicted direct-axis current value and the predicted quadrature-axis current value output by the BP neural network; is the error between the BP neural network output and the current prediction model; is the prediction proportional coefficient, with the unit of , is the prediction integral coefficient, with the unit of ;

[0109] In this embodiment, simulation verification shows that the value of is , the prediction integral coefficient is , which will make the prediction error convergence speed increase by 20%.

[0110] In this embodiment, test verification shows that the value of is , the compensation integral coefficient is , which will make the measured and predicted current error decrease by 15%.

[0111] determine the basic compensation amount , wherein is the error between the BP neural network output and the measured current; , , is the measured direct-axis current and the measured quadrature-axis current at the kth moment; is the compensation proportional coefficient, with the unit of , is the compensation integral coefficient, with the unit of ;

[0112] calculate the voltage vector duty cycle .

[0113] Preferably, the load torque observation value satisfies:

[0114] , wherein is the basic load torque observation value output by the extended state observer; is the adaptive gain coefficient; is the estimation error of load torque; is the estimation load torque based on motor mechanical equation; is the sign function; is the nonlinear adjustment index;

[0115] generating feedforward current , combined with the motor d-q axis coupling relationship, meet:

[0116] wherein, is the motor pole pair number; is the permanent magnet flux linkage; is the direct axis inductance; is the cross axis inductance; is the direct axis current.

[0117] In this embodiment, is the basic load torque estimated by the extended state observer after collecting motor speed, current and other signals. Experimental verification shows that when the load fluctuation is less than or equal to 0.2 times the rated load, the observation error is minimum, and at this time, The value of is 0.2.

[0118] In this embodiment, the motor mechanical equation is: wherein, is the moment of inertia, B is the damping coefficient, is the electromagnetic torque, is the load torque. is the angular acceleration, is the angular velocity, .

[0119] In this embodiment, the observer is adapted to adjust the rate for different error sizes, and small errors are fine-tuned and large errors are quickly corrected, and at this time, The value of is 1.3.

[0120] In this embodiment, the adaptive SVPWM basic duty ratio is:

[0121] Based on the d-axis current reference value and the q-axis current reference value of the motor after preliminary adjustment, the voltage u1 and u2 are determined, the sector angle is determined by unit, wherein 0 to 60° is sector 1 and 60° to 120° is sector 2.

[0122] For example, in sector 1, , wherein, is the switching period, is the DC bus voltage;

[0123] Z1 (000) and Z2 (111) are adopted alternately, each accounting for .

[0124] In this embodiment, the working mode switching transition strategy is: adopting exponential weighted average switching g step parameter , wherein, is a time constant, and the value is 0.1; is the dynamic parameter value of the g step after switching; is the final value of the target parameter in the new mode after mode switching, is the parameter value in the original mode before mode switching, is the time after mode switching, is the time constant of exponential weighted average, such as switching to fast mode after the rotation speed is greater than or equal to 20 rad / s for 50 ms, and switching to fine scanning mode after the rotation speed is less than or equal to 5 rad / s and the beam error is less than or equal to 0.001 rad for 100 ms.

[0125] In this embodiment, the current prediction model is an analytical model based on the Park equation , which is implemented as follows: .

[0126] In this embodiment, the q-axis current reference value is superimposed with the anti-backlash current as the anti-backlash current superposition mode, and the anti-backlash effect is best when the rotation speed is 0.1 rad~0.3 rad.

[0127] The beneficial effects of the above technical solutions are: through the double closed loop compensation of the BP neural network output to the space vector pulse width modulation duty cycle prediction error and measured error, the voltage vector accurately matches the motor current demand, the adaptive load torque observation based on the extended state observer and the feedforward current calculation of the cross-axis coupling characteristics accurately compensate the load disturbance and the influence of axis coupling, improve the current control precision, dynamic response speed and anti-interference ability of the permanent magnet synchronous motor, and provide reliable driving protection for stable and efficient operation of the radar servo system.

[0128] The application provides a driving optimization method for a permanent magnet synchronous motor of a radar servo system, the BP neural network adopts a double-branch structure enhanced by multi-scale feature fusion and attention mechanism, as shown in Figure 2 , and specifically includes:

[0129] Input the d-axis current reference value , the q-axis current reference value , encode the static current demand based on a 1-layer fully connected layer, and output a static feature vector;

[0130] Input the motor differential error , based on the motor current error after preliminary adjustment and load torque observation value , the dynamic disturbance is encoded based on a two-layer full connection layer on observation information, and a dynamic feature vector is input, wherein the first layer of the two-layer full connection layer contains 24 neurons and the activation function is LeakyReLU, and the second layer contains 16 neurons and the activation function is LeakyReLU;

[0131] The static feature vector and the dynamic feature vector are spliced, and a channel attention weight A is generated through global average pooling, global maximum pooling and a shared connection layer, satisfying:

[0132] , wherein, is a Sigmoid function; is a full connection layer operation; is a global average pooling operation; is a global maximum pooling operation; is a spliced feature vector;

[0133] The features after channel weighting , a spatial attention weight is generated through a 1x1 convolution layer and a Sigmoid function , and the final fusion feature is obtained , wherein, is an element-wise multiplication, is a convolution operation using a convolution kernel size of 1x1;

[0134] The two-layer full connection layer outputs a predicted direct-axis current value and a predicted quadrature-axis current value, wherein the loss function is a multi-objective loss with dynamic weight and regularization: , wherein, is a regularization coefficient; is an L2 regularization term of network weight W; , , is a reference speed error amplitude; is a reference current error amplitude, is an L2 regularization function.

[0135] In this embodiment, during training, 10000 sets of working condition data containing speeds of 5-50 rad / s and loads of 0.5-1.1 times the rated load are used, the batch size is set to 64, and the network is trained for 1000 rounds until the loss function converges to ≤0.0005, ensuring the network generalization ability and convergence stability.

[0136] In this embodiment, the first layer of the two-layer full connection layer contains 16 neurons and the second layer contains 2 neurons, corresponding to the predicted direct-axis current value and the predicted quadrature-axis current value, respectively.

[0137] In this embodiment, the loss function simultaneously constrains the prediction error of the direct-axis current and the quadrature-axis current, and adds regularization to prevent overfitting.

[0138] In this embodiment, through experiments, the value of the regularization coefficient is 0.001, which can effectively prevent network overfitting.

[0139] In this embodiment, after the Fch is processed by a 1x1 convolutional layer (the number of convolutional kernels is 8, and the feature dimension is adjusted), spatial attention weights As (the dimension is consistent with Fch) are generated by a Sigmoid function; then Fch and As are multiplied element by element to obtain the final fusion feature Ffu, which further highlights the key feature region in the spatial dimension and improves the accuracy of feature expression.

[0140] In this embodiment, global average pooling is performed on Fconcat, which compresses 32-dimensional features into 1-dimensional features, and the average value of each channel feature is taken; global maximum pooling takes the maximum value of each channel feature to capture global statistical information of the feature.

[0141] In this embodiment, the static branch includes a 1-layer fully connected layer containing 32 neurons, which adopts a ReLU activation function.

[0142] In this embodiment, the shared connection layer includes 16 neurons (ReLU)→8 neurons (Sigmoid).

[0143] In this embodiment, the optimizer is Adam, the learning rate is 1e-3, the batch size is 64, and the loss ≤5e−5.

[0144] In this embodiment, The value of is 0.5 rad / s, The value of is 0.2 .

[0145] The beneficial effects of the above technical solutions are: through the static and dynamic double branches to extract multi-scale features, combined with channel and spatial attention mechanisms to accurately highlight key features, through the loss function with dynamic weights and regularization to efficiently train, accurately predict the direct-axis current and the quadrature-axis current under different working conditions, and improve the stability and precision of motor driving.

[0146] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and variations.

Claims

1. A drive optimization method for a permanent magnet synchronous motor used in a radar servo system, characterized in that, include: Step 1: The online motor parameter identification module based on Kalman filtering identifies and constructs a state vector, and determines the current operating mode of the motor based on the state vector. The state vector includes: motor speed, direct-axis current, quadrature-axis current, and load torque. Step 2: Estimate the load torque observations based on the load torque observer. Furthermore, the feedforward current is calculated based on the AC-DC coupling characteristics of the motor, the backlash compensation gear is designed using the variable bias backlash elimination current, and the coordination parameters are determined based on the state vector and the current working mode. The motor is then preliminarily adjusted using the feedforward current and the variable bias backlash elimination current. The coordination parameters include the proportional coefficient and integral coefficient of the load-weighted differential negative feedback PI regulator based on the current working mode. Step 3: Determine the motor differential speed error of the radar servo system using load-weighted differential speed negative feedback. Where N is the total number of permanent magnet synchronous motors participating in synchronization in the radar servo system; Let be the load weighting coefficient for the i-th motor; Let be the average speed of the i-th motor, in rad / s; Reference speed, unit: rad / s; Step 4: Determine the reference value of the direct-axis current of the motor after preliminary adjustment. Quadrature axis current reference value The voltage vector duty cycle of the adaptive space vector pulse width modulation is adjusted in real time based on the output of the BP neural network to obtain a composite control signal to drive the permanent magnet synchronous motor. The input of the BP neural network is the initially adjusted direct-axis current reference value. Quadrature axis current reference value Motor differential speed error Based on the initial adjustment of motor current error and load torque observations The output of the BP neural network is: predicted direct-axis current value and predicted quadrature-axis current value.

2. The drive optimization method for a permanent magnet synchronous motor used in a radar servo system according to claim 1, characterized in that, The current operating mode of the motor is determined based on the state vector, including: Each vector element in the state vector is matched sequentially with the pattern lookup table to determine the current working mode; The current operating modes include: radar fine scan mode, radar conventional scan mode, and radar fast scan mode; When in the radar fine scan mode, modified sliding mode control and maximum torque-current ratio control are used; When in the radar's normal scanning mode, model predictive control and active disturbance rejection control are employed. When in radar fast scanning mode, model predictive control and dynamic field weakening control are used; The working mode switching condition is: when the motor speed... At that time, the radar switches from the normal scanning mode to the rapid scanning mode, when the motor speed... And beam pointing error At that time, the radar switches from the conventional scanning mode to the fine scanning mode.

3. The drive optimization method for a permanent magnet synchronous motor for a radar servo system according to claim 2, characterized in that, The rate of change of the sliding surface controlled by the modified sliding form for: ； in, For sliding surfaces, the unit is rad / s; This is the sliding mode control coefficient, in units of... ; , n are morphological parameters; The decay coefficient of the exponential term, in units of ; The coefficients of the saturation function are in units of . ; It is a saturation function; is the boundary coefficient of the saturation function, in rad / s; Among them, the control law for modified sliding mode control , where K is the control gain.

4. The drive optimization method for a permanent magnet synchronous motor for a radar servo system according to claim 1, characterized in that, ,in, The real-time load of the i-th motor is given by . , Rated load of the motor, in units of ,and The value range is from 1 to 1.1; The motor speed is corrected by a PI controller, the proportional coefficient of which is... Integral coefficient The multi-motor synchronous starting process is as follows: based on the load weighting coefficient... The motors are optimized in descending order, with an optimization delay of 0.1s for adjacent motors.

5. The drive optimization method for a permanent magnet synchronous motor for a radar servo system according to claim 1, characterized in that, The variable bias gap-eliminating current satisfy: ； in, This is the zero-point bias current. This is the rated current of the motor; Compensation current for inflection point; The current threshold; This is the actual current signal of the motor.

6. The drive optimization method for a permanent magnet synchronous motor for a radar servo system according to claim 1, characterized in that, The reference speed Correction based on radar beam pointing error feedback includes: A beam pointing measurement device based on a radar antenna obtains the deviation angle between the actual pointing and the target pointing. ; like ,according to The reference speed is adjusted in real time, among which, Initial reference speed; .

7. The drive optimization method for a permanent magnet synchronous motor for a radar servo system according to claim 1, characterized in that, The voltage vector duty cycle of adaptive space vector pulse width modulation is adjusted in real time based on the output of a BP neural network, including: Based on the initially adjusted reference value of the motor's direct-axis current Quadrature axis current reference value Determine the base duty cycle ; Determine the predicted compensation amount ,and , , For the k-th time step, the output direct-axis current value and the output quadrature-axis current value are based on the current prediction model. , These are the predicted direct-axis current values ​​and predicted quadrature-axis current values ​​output by the BP neural network. The error between the BP neural network output and the current prediction model; This is the prediction scale factor, in units of... , For predicting integral coefficients, the unit is... ; Determine the basic compensation amount ,in, The error between the BP neural network output and the measured current; , , Let be the measured direct-axis current and the measured quadrature-axis current at the k-th time. This is the compensation ratio coefficient, in units of... , To compensate for the integral coefficient, the unit is... ; Calculate the voltage vector duty cycle .

8. The drive optimization method for a permanent magnet synchronous motor for a radar servo system according to claim 1, characterized in that, The load torque observation value satisfy: ,in, The base load torque observation value output by the extended stater; For adaptive gain coefficients; This is the error in the estimation of the load torque; This is the estimated load torque obtained based on the mechanical motion equations of the electric motor; It is a symbolic function; It is a non-linear adjustment index; Generate feedforward current At that time, considering the direct-axis coupling relationship of the motor, the following conditions are met: ,in, This represents the number of pole pairs of the motor. For permanent magnet flux linkage; It is a direct-axis inductor; It is a quadrature axis inductor; It is the direct-axis current.

9. The drive optimization method for a permanent magnet synchronous motor for a radar servo system according to claim 7, characterized in that, The BP neural network employs a dual-branch structure enhanced by multi-scale feature fusion and attention mechanisms, specifically including: Input direct-axis current reference value Quadrature axis current reference value Based on a single fully connected layer, the static current demand is feature-encoded and a static feature vector is output. Input motor differential error Based on the initial adjustment of motor current error and load torque observations Based on two fully connected layers, dynamic perturbation and observation information are feature encoded and input as dynamic feature vectors. The first layer of the two fully connected layers contains 24 neurons and the activation function is LeakyReLU, and the second layer contains 16 neurons and the activation function is LeakyReLU. Static and dynamic feature vectors are concatenated, and channel attention weights A are generated through global average pooling, global max pooling, and shared connection layers, satisfying the following: ,in, For the Sigmoid function; This is for fully connected layer operations; This is a global average pooling operation; This is a global max pooling operation; To concatenate feature vectors; Channel-weighted features Spatial attention weights are generated using a 1×1 convolutional layer and a sigmoid function. And obtain the final fusion features. ,in, For element-wise multiplication, To perform a convolution operation using a kernel size of 1×1; Based on the output of two fully connected layers, predict the direct-axis current value and the quadrature-axis current value, where the loss function... For multi-objective loss with dynamic weights and regularization: ,in, The regularization coefficient is used. For the network weight W, it is the L2 regularization term; , , The reference rotational speed error amplitude; For reference current error amplitude, This is the L2 regularization function.

Images