阅读说明：本技术 一种无轴承永磁同步电机转子转速和位移软测量方法 (A kind of bearing-free permanent magnet synchronous motor rotor speed and displacement flexible measurement method ) 是由 郝正强 杨阳 徐世文 于 2019-07-12 设计创作，主要内容包括：本发明提供一种无轴承永磁同步电机转子转速和位移软测量方法,为实现无轴承永磁同步电机转矩和径向悬浮力的实时在线控制创造了条件,适用于无轴承永磁同步电机的高性能控制。为了构建转子转速、位移与其它易测辅助变量之间的关系,提出了一种基于自适应BP神经网络建立的软测量模型,与常规参数自适应不同,该方法是对BP神经网络隐层层数进行自适应控制,利用权值直接确定法直接确定神经网络的最小二乘逼近权值和网络精度,通过与期望精度的比较能够实现对神经网络隐层结构的自适应设计。该神经网络模型不依赖电机参数,提高了检测精度和系统的可靠性。(The present invention provides a kind of bearing-free permanent magnet synchronous motor rotor speed and displacement flexible measurement method, to realize that the real-time online control of bearing-free permanent magnet synchronous motor torque and radial suspension force creates condition, the high performance control suitable for bearing-free permanent magnet synchronous motor.In order to construct rotor speed, displacement and other easy relationships surveyed between auxiliary variable, propose a kind of soft-sensing model established based on self-adaptive BP neural networks, it is adaptively different from conventional parameter, this method is to carry out self adaptive control to BP neural network hidden layers numbers, the least square approximation weight and neural network accuracy that neural network is directly determined using direct weight determination, by can be realized the adaptive design to neural network hidden layer configuration compared with expectation quality.The neural network model does not depend on the parameter of electric machine, improves the reliability of detection accuracy and system.)

1. a kind of bearing-free permanent magnet synchronous motor rotor speed and displacement flexible measurement method, which is characterized in that the described method includes:

Selected characteristic variable；

Collecting sample data are simultaneously pre-processed for sample data, to form training sample set；

Self-adaptive BP neural networks hard measurement module is based on according to characteristic variable design；

The self-adaptive BP neural networks are trained using the sample data that the training sample is concentrated, adjusts and determines defeated Enter node layer to hidden node and and hidden node to export node layer weight；

It will be established based on self-adaptive BP neural networks hard measurement block coupled in series into bearing-free permanent magnet synchronous motor system shaftless Bearing permanent magnet synchronous electric motor revolving speed and displacement hard measurement control system.

2. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 1 and displacement flexible measurement method, special Sign is that selected characteristic variable specifically includes:

The selection of soft-sensing model feature primary variables and the selection of auxiliary variable are with rotor speed ω and radial displacement x, y Primary variables determines that measurable variable suspending power controls electric current i by model analysis_x、i_yAnd its integral i_x1、i_y1, direct torque electricity Flow i_d、i_qAnd its integral i_d1、i_q1For auxiliary variable；The input of the BP neural network hard measurement module is auxiliary variable [i_x, i_x1, i_y, i_y1, i_d, i_d1, i_q, i_q1], output is primary variables [ω, x, y], wherein rotor speed output is bearing-free motor tune The random quantity being distributed in fast range with sinusoidal form, radial displacement output are in bearing-free motor air gap distance with sinusoidal form point The random quantity of cloth.

3. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 1 and displacement flexible measurement method, special Sign is that collecting sample data are simultaneously pre-processed for sample data, to form training sample set, specifically includes:

Input is normalized with output signal, calculation method are as follows:

Wherein D^*Data after indicating normalization, D_minFor the minimum value in corresponding sample data, D_maxFor in corresponding sample data Maximum value, after normalized, all variable datas are all normalized to [0,1]；

Data after normalization are filtered, to obtain accurate inputoutput data, form the self-adaptive BP nerve The training sample set of network；

It is concentrated from the training sample, chooses the representative sample data of N group, wherein every group of sample data includes that 8 inputs become Measure X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈]=[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1], corresponding output variable is Y_j= [Y₁, Y₂, Y₃]=[ω, x, y], formed sampled data, for training and testing.

4. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 1 and displacement flexible measurement method, special Sign is, is based on self-adaptive BP neural networks hard measurement module according to characteristic variable design, specifically includes:

Initialize the self-adaptive BP neural networks；

Mean square deviation function is chosen as performance indicator, to self-adaptive BP neural networks training；

Design BP neural network model；

Directly determine the weight of BP neural network；

Adaptive design BP neural network hidden layer.

5. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 4 and displacement flexible measurement method, special Sign is, initializes the self-adaptive BP neural networks, specifically includes:

Network input layer neuron number is 8, chooses initial hidden layer neuron number m=-1+log₂8=2, self-adaptive BP nerve net Network is before one kind to 3 layers of neural network, and the structure of initial time neural network is the connection type of 8-2-3, and input layer has 8 Node, hidden layer have 2 nodes, and output layer has 3 nodes；The input vector of neural network is X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈]=[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1], output vector Y_j=[Y₁, Y₂, Y₃]=[ω, x, y]；

Hidden layer uses tansig type activation primitive, expression are as follows:

Output layer uses logsig type activation primitive, expression are as follows:

6. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 4 and displacement flexible measurement method, special Sign is that self-adaptive BP neural networks training chooses mean square deviation function as performance indicator, and the mean square deviation function embodies Formula are as follows:

Wherein, Y_diFor the corresponding desired output of i-th of sample, Y_iFor the corresponding network output of i-th of sample, N is to participate in training Number of samples.

7. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 4 and displacement flexible measurement method, special Sign is, designs BP neural network model, specifically includes:

Input component X_k(k=1,2 ..., 8) passes through the weight component w that is connected with it_ik(k=1,2 ..., 8；I=1,2 ..., m；M is the quantity of hidden node) it is multiplied, activation primitive f () is inputted after summation, neuron exports Y_iExcept being influenced by input signal Outside, while also it is influenced by inside neurons other factors, an additional input signal b is added in neural Meta Model_i As its threshold value, output variable Y_iExpression formula are as follows:

And value after neuron is to input signal weighted sum is greater than b_i, Y_iGreater than zero；

Self-adaptive BP neural networks share three layers, and first layer is that input layer number is 8, and the second layer is that node in hidden layer is m, Third layer is that output layer number of nodes is 3；

First layer is input layer, and each input node respectively represents input vector X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈]= [i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1] one-component；

The second layer is hidden layer, and the activation primitive of hidden layer uses tansig type activation primitive, the output of i-th of hidden layer node Are as follows:

Wherein w_ikFor the connection weight of input layer to hidden layer node, b_iFor the threshold value of hidden layer；

Third layer is output layer, and the activation primitive of output layer uses logsig type activation primitive, output vector Y_j=[Y₁, Y₂, Y₃] =[ω, x, y], the output of j-th of output node layer are as follows:

Wherein w_jiFor hidden layer node to the connection weight of output node layer, b_jFor the threshold value of output layer.

8. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 4 and displacement flexible measurement method, special Sign is, directly determines the weight of BP neural network, specifically includes:

Define weight column vector w, input excited target matrix X₀It is denoted as respectively with target output vector γ:

W=[w₁ w₂ ... w_n]^T∈Rⁿ,

γ=[φ₁ φ₂ ... φ_m]^T∈R^m,

Then the best initial weights of neural network directly determine are as follows:

Wherein,For matrix X₀PseudoinverseIt is abbreviated as

Expression formula is exported according to designed neural network model, it can be deduced that weight calculation formula:

Wherein X is the input vector of neural network,For the connection weight vector of neural network input layer to hidden layer, For the connection weight vector of hidden layer to output layer, α is the desired value of hidden layer, Y_dFor the desired output vector of neural network；

Adaptive design BP neural network hidden layer, specifically includes:

Using direct weight determination calculate network input layer to hidden layer connection weight vectorWith hidden layer to defeated The connection weight vector of layer outAnd calculate e-learning error E corresponding with the number of hidden nodes_m, i.e. its performance refers to Mark；Enable E_expertIt is expected the study precision reached, if there is E_m>E_expert, then increase a hidden node, i.e. m=m+1, and Recalculate connection weight vector w and e-learning error E_m, continue to compare, until meeting E_m<E_expert, stop operation, Complete adaptive design BP neural network hidden layer.

9. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 1 and displacement flexible measurement method, special Sign is, is trained using the sample data that the training sample is concentrated to the self-adaptive BP neural networks, adjusts and true Determine input layer to hidden node and and hidden node to export node layer weight, specifically include:

Off-line training is carried out to self-adaptive BP neural networks using least square method, 3/4 sample is chosen and neural network is carried out Training, in addition 1/4 test data for being used for neural network is less than neural network output mean square error after n times training 0.001, and determine the weight of self-adaptive BP neural networksWithHidden nodes m and threshold value b_i、b_j。

10. a kind of bearing-free permanent magnet synchronous motor rotor speed according to claim 1 and displacement flexible measurement method, special Sign is, will be based on self-adaptive BP neural networks hard measurement block coupled in series into bearing-free permanent magnet synchronous motor system to establish nothing Bearing permanent magnet synchronous motor revolving speed and displacement hard measurement control system, specifically include:

Trained self-adaptive BP neural networks hard measurement module is accessed in bearing-free permanent magnet synchronous motor control system, is constituted Rotor speed forecast module and radial displacement prediction module realize that rotor speed and radial displacement are predicted in real time online；i_x*、i_yIt * is position The suspending power control electric current that deviation is obtained by PID regulator one and the conversion of PID regulator two is moved, through Park inverse transform module one Obtain the suspending power control electric current i under two-phase stationary coordinate system_1αAnd i *_1β*, then through Clark inverse transform module one obtain three-phase electricity Flow i_1U*、i_1V*、i_1W*, the three-phase current i of real-time control is obtained by current track inverter one_1U、i_1V、i_1W；i_d*、i_q* it is Revolving speed deviation passes through the direct torque electric current that pi regulator converts, and enables i_d*=0, two are obtained through Park inverse transform module two Suspending power under phase rest frame controls electric current i_2αAnd i *_2β*, then through Clark inverse transform module two obtain three-phase current i_2U*、 i_2V*、i_2W*, the three-phase current i of real-time control is obtained by current track inverter two_2U、i_2V、i_2W；Three-phase current i_1U、i_1V、 i_1WIt is input in levitation force winding subsystem, three-phase current i_2U、i_2V、i_2WIt is input in torque winding subsystem, the bearing-free Permanent magnet synchronous motor is made of levitation force winding subsystem and torque winding subsystem；i_1αAnd i_1β、i_2αAnd i_2βRespectively i_1U、 i_1V、i_1WAnd i_2U、i_2V、i_2WThe levitation force winding and torque winding obtained through Clark conversion module one and Clark conversion module two Feedback current under two-phase stationary coordinate system, i_xAnd i_y、i_dAnd i_qRespectively i_1αAnd i_1β、i_2αAnd i_2βThrough one He of Park conversion module Feedback current under levitation force winding and torque winding two-phase rotating coordinate system that Park conversion module two obtains, i_x1And i_y1、i_d1With i_q1Respectively i_xAnd i_y、i_dAnd i_qIntegral, [i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1] it is used as BP neural network hard measurement module Input signal, hard measurement module export feedback signal [ω, x, y].

Technical field

The present invention relates to motor fields more particularly to a kind of bearing-free permanent magnet based on self-adaptive BP neural networks to synchronize electricity Machine rotor revolving speed and displacement flexible measurement method, for the real-time online for realizing bearing-free permanent magnet synchronous motor torque and radial suspension force Control creates condition, the high performance control suitable for bearing-free permanent magnet synchronous motor.

Background technique

Bearing-free permanent magnet synchronous motor is a kind of New-type electric machine for having gathered permanent magnet synchronous motor and magnetic bearing characteristic, it is not Only have the advantages that permanent magnet synchronous motor power factor is high, high-efficient, power density is big, also have magnetic bearing without mechanical friction, The features such as without lubricating and being non-maintaining, it can be achieved that high speed or ultrahigh speed operation, in chemical, life science, energy traffic, navigates The fields such as empty space flight and robot have potential application foreground.

Bearing-free permanent magnet synchronous motor includes torque winding and levitation force winding double winding, and control system is by torque control System processed and suspension Force control system two subsystems composition.For moment controlling system, speed probe needs to detect rotor Revolving speed is sent into moment controlling system and generates direct torque electric current after being compared with given value；For suspension Force control system, position Displacement sensor needs to detect rotor radial displacement, and suspension Force control system is sent into after being compared with given value and generates suspending power control Electric current processed.Traditional bearing-free permanent magnet synchronous motor carries out feedback control using sensor acquisition information, and measurement revolving speed mostly uses The mechanical velocity sensor such as photoelectric coded disk, measurement rotor radial displacement mostly use current vortex sensor, these sensors Motor weight and cost are increased, vulnerable to interference, system reliability is reduced, is not easy to installation and maintenance, is not suitable for severe Environment, and under bearing-free permanent magnet synchronous motor high speed, ultrahigh speed operating status, mechanical sensor has been unable to meet system Performance requirement.Therefore, the research of bearing-free permanent magnet synchronous motor rotor speed and displacement self-test survey technology is particularly important.

Current bearing-free permanent magnet synchronous motor revolving speed self-test survey technology has Based on Back-EMF Method and gray forecast approach.Though Based on Back-EMF Method It is so simple, but low speed bad adaptability can be led to because back-emf is too small in zero-speed or low speed, and quick to the variation of the parameter of electric machine Sense, poor robustness；Gray forecast approach is simple and easy, does not need additional hardware, does not need injection low-and high-frequency signal, but depend on The parameter of electric machine changes in motor temperature, can have an impact when magnetic saturation to Control platform.Rotor radial is displaced self-test skill Art has high-frequency signal injection, model reference adaptive method and support vector machines method.High-frequency signal injection utilizes motor torque control winding The relationship between suspending windings mutual inductance or suspending windings self-induction and displacement, by detection suspending windings both ends differential voltage come Estimate radial displacement, but there are high-frequency signal extract, signal filter process is complicated the defects of；Patent [CN101667799A] The model reference adaptive method based on voltage and current is proposed to detect bearing-free permanent magnet synchronous motor revolving speed, but it is because there are pure Integral element, identification accuracy is poor, and is influenced by stator resistance, and when low speed is unstable；Patent [CN103501148A] proposes Bearing-free permanent magnet synchronous motor based on multicore least square method supporting vector machine is without radial displacement transducer control method, this method Algorithm is simple, and robustness is preferable, but the operation due to there is high level matrix, can expend a large amount of machine memory and operation time.

Summary of the invention

In order to solve the above-mentioned technical problem, the object of the present invention is to provide a kind of based on self-adaptive BP neural networks Bearing-free permanent magnet synchronous motor rotor speed and displacement flexible measurement method, by building rotor speed, displacement with it is other easily survey it is auxiliary The relationship between variable is helped, realizes the real-time detection to rotor speed and displacement.Electric machine control system based on this method building Bearing-free permanent magnet synchronous motor rotor speed and radial displacement can be quick and precisely detected in full speed range, and can be prominent in parameter Become, realize that without sensor stabilization suspension operation, it is same to improve bearing-free permanent magnet for bearing-free permanent magnet synchronous motor under the conditions of load disturbance Working performance when walking motor high speed, ultrahigh speed operation.Compared to other methods, self organizing neural network model does not depend on motor ginseng Number, improves the reliability of detection accuracy and system.

The invention proposes a kind of bearing-free permanent magnet synchronous motor rotor speed and displacement flexible measurement method, the method packets It includes:

Selected characteristic variable；

Collecting sample data are simultaneously pre-processed for sample data, to form training sample set；

Self-adaptive BP neural networks hard measurement module is based on according to characteristic variable design；

The self-adaptive BP neural networks are trained using the sample data that the training sample is concentrated, are adjusted and true Determine input layer to hidden node and and hidden node to export node layer weight；

Self-adaptive BP neural networks hard measurement block coupled in series will be based on into bearing-free permanent magnet synchronous motor system to establish Bearing-free permanent magnet synchronous motor revolving speed and displacement hard measurement control system.

In the present solution, selected characteristic variable, specifically includes:

The selection of soft-sensing model feature primary variables and the selection of auxiliary variable, with rotor speed ω and radial displacement x, Y is primary variables, by model analysis, determines that measurable variable suspending power controls electric current i_x、i_yAnd its integral i_x1、i_y1, torque control Electric current i processed_d、i_qAnd its integral i_d1、i_q1For auxiliary variable；The input of the BP neural network hard measurement module is auxiliary variable [i_x, i_x1, i_y, i_y1, i_d, i_d1, i_q, i_q1], output is primary variables [ω, x, y], wherein rotor speed output is bearing-free electricity The random quantity being distributed in machine speed adjustable range with sinusoidal form, radial displacement output are in bearing-free motor air gap distance with sinusoidal The random quantity of formula distribution.

In the present solution, collecting sample data and pre-processed for sample data, it is specific to wrap to form training sample set It includes:

Input is normalized with output signal, calculation method are as follows:

Wherein D^*Data after indicating normalization, D_minFor the minimum value in corresponding sample data, D_maxFor corresponding sample number Maximum value in, after normalized, all variable datas are all normalized to [0,1]；

Data after normalization are filtered, to obtain accurate inputoutput data, form the self-adaptive BP The training sample set of neural network；

It is concentrated from the training sample, chooses the representative sample data of N group, wherein every group of sample data includes 8 defeated Enter variable X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈]=[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1], corresponding output variable is Y_j=[Y₁, Y₂, Y₃]=[ω, x, y], formed sampled data, for training and testing.

In the present solution, being based on self-adaptive BP neural networks hard measurement module according to characteristic variable design, specifically include:

Initialize the self-adaptive BP neural networks；

Mean square deviation function is chosen as performance indicator, to self-adaptive BP neural networks training；

Design BP neural network model；

Directly determine the weight of BP neural network；

Adaptive design BP neural network hidden layer.

In the present solution, initializing the self-adaptive BP neural networks, specifically include:

Network input layer neuron number is 8, chooses initial hidden layer neuron number m=-1+log₂8=2, self-adaptive BP mind It is before one kind to 3 layers of neural network through network, the structure of initial time neural network is the connection type of 8-2-3, input layer There are 8 nodes, hidden layer there are 2 nodes, and output layer there are 3 nodes；The input vector of neural network is X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈]=[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1], output vector Y_j=[Y₁, Y₂, Y₃]=[ω, x, y]；

Hidden layer uses tansig type activation primitive, expression are as follows:

Output layer uses logsig type activation primitive, expression are as follows:

In the present solution, self-adaptive BP neural networks training chooses mean square deviation function as performance indicator, the mean square deviation letter Number expression are as follows:

Wherein, Y_diFor the corresponding desired output of i-th of sample, Y_iFor the corresponding network output of i-th of sample, N is to participate in Trained number of samples.

In the present solution, design BP neural network model, specifically includes:

Input component X_k(k=1,2 ..., 8) passes through the weight component w that is connected with it_ik(k=1,2 ..., 8；I=1, 2 ..., m；M is the quantity of hidden node) it is multiplied, activation primitive f () is inputted after summation, neuron exports Y_iExcept being believed by input It number influences outer, while also being influenced by inside neurons other factors, an additional input is added in neural Meta Model Signal b_iAs its threshold value, output variable Y_iExpression formula are as follows:

And value after neuron is to input signal weighted sum is greater than b_i, Y_iGreater than zero；

Self-adaptive BP neural networks share three layers, and first layer is that input layer number is 8, and the second layer is node in hidden layer For m, third layer is that output layer number of nodes is 3；

First layer is input layer, and each input node respectively represents input vector X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈] =[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1] one-component；

The second layer is hidden layer, and the activation primitive of hidden layer uses tansig type activation primitive, i-th hidden layer node Output are as follows:

Wherein w_ikFor the connection weight of input layer to hidden layer node, b_iFor the threshold value of hidden layer；

Third layer is output layer, and the activation primitive of output layer uses logsig type activation primitive, output vector Y_j=[Y₁, Y₂, Y₃]=[ω, x, y], j-th output node layer output are as follows:

Wherein w_jiFor hidden layer node to the connection weight of output node layer, b_jFor the threshold value of output layer.

In the present solution, directly determining the weight of BP neural network, specifically include:

Define weight column vector w, input excited target matrix X₀It is denoted as respectively with target output vector γ:

W=[w₁ w₂ ... w_n]^T∈Rⁿ,

γ=[φ₁ φ₂ ... φ_m]^T∈R^m,

Then the best initial weights of neural network directly determine are as follows:

Wherein,For matrix X₀PseudoinverseIt is abbreviated as

Expression formula is exported according to designed neural network model, it can be deduced that weight calculation formula:

Wherein X is the input vector of neural network,For the connection weight vector of neural network input layer to hidden layer,For the connection weight vector of hidden layer to output layer, α is the desired value of hidden layer, Y_dFor neural network desired output to Amount；

Adaptive design BP neural network hidden layer, specifically includes:

Using direct weight determination calculate network input layer to hidden layer connection weight vectorAnd hidden layer To the connection weight vector of output layerAnd calculate e-learning error E corresponding with the number of hidden nodes_m, i.e. its property It can index；Enable E_expertIt is expected the study precision reached, if there is E_m>E_expert, then increase a hidden node, i.e. m=m+1, And recalculate connection weight vector w and e-learning error E_m, continue to compare, until meeting E_m<E_expert, stop fortune It calculates, completes adaptive design BP neural network hidden layer.

In the present solution, being instructed using the sample data that the training sample is concentrated to the self-adaptive BP neural networks Practice, adjust and determine input layer to hidden node and and hidden node to export node layer weight, specifically include:

Off-line training is carried out to self-adaptive BP neural networks using least square method, chooses 3/4 sample to neural network It is trained, in addition 1/4 test data for being used for neural network keeps neural network output mean square error small after n times training In 0.001, and determine the weight of self-adaptive BP neural networksWithHidden nodes m and threshold value b_i、b_j。

In the present solution, will be based on self-adaptive BP neural networks hard measurement block coupled in series to bearing-free permanent magnet synchronous motor system In with establish bearing-free permanent magnet synchronous motor revolving speed and displacement hard measurement control system, specifically include:

Trained self-adaptive BP neural networks hard measurement module is accessed in bearing-free permanent magnet synchronous motor control system, Rotor speed forecast module and radial displacement prediction module are constituted, realizes that rotor speed and radial displacement are predicted in real time online；i_x*、i_y* Convert for offset deviation by PID regulator one and PID regulator two obtained suspending power control electric current, changes the mold through Park inversion Block one obtains the control of the suspending power under two-phase stationary coordinate system electric current i_1αAnd i *_1β*, then through Clark inverse transform module one obtain three Phase current i_1U*、i_1V*、i_1W*, the three-phase current i of real-time control is obtained by current track inverter one_1U、i_1V、i_1W；i_d*、 i_q* pass through the direct torque electric current that pi regulator converts for revolving speed deviation, enable i_d*=0, obtained through Park inverse transform module two Suspending power under to two-phase stationary coordinate system controls electric current i_2αAnd i *_2β*, then through Clark inverse transform module two obtain three-phase current i_2U*、i_2V*、i_2W*, the three-phase current i of real-time control is obtained by current track inverter two_2U、i_2V、i_2W；Three-phase current i_1U、i_1V、i_1WIt is input in levitation force winding subsystem, three-phase current i_2U、i_2V、i_2WIt is input in torque winding subsystem, institute Bearing-free permanent magnet synchronous motor is stated to be made of levitation force winding subsystem and torque winding subsystem；i_1αAnd i_1β、i_2αAnd i_2βRespectively For i_1U、i_1V、i_1WAnd i_2U、i_2V、i_2WThe levitation force winding that is obtained through Clark conversion module one and Clark conversion module two and turn Feedback current under square winding two-phase stationary coordinate system, i_xAnd i_y、i_dAnd i_qRespectively i_1αAnd i_1β、i_2αAnd i_2βThrough Park conversion module Feedback current under levitation force winding and torque winding two-phase rotating coordinate system that one and Park conversion module two obtains, i_x1And i_y1、 i_d1And i_q1Respectively i_xAnd i_y、i_dAnd i_qIntegral, [i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1] it is used as BP neural network hard measurement The input signal of module, hard measurement module export feedback signal [ω, x, y].

The present invention has the characteristics that good non-linear mapping capability according to BP neural network, using self-adaptive BP neural networks It realizes the Nonlinear Mapping between auxiliary variable and rotor revolving speed and radial displacement, establishes bearing-free permanent magnet synchronous motor The soft-sensing model of rotor speed and displacement realizes the hard measurement to rotor speed and displacement, the good, stability with real-time Well, the features such as precision is high.

The present invention is special without the operation mechanism for understanding bearing-free permanent magnet synchronous motor in depth without using Heuristics Property, only the Black-Box identification of radial displacement system need to can be realized using data are output and input.After Black-Box identification, as long as passing through Study to input data can predict rotor speed and radial displacement.

The present invention proposes that direct weight determination directly determines the minimum two of neural network by the pseudoinverse of input vector X Multiply and approaches weight.By direct weight determination, interminable iteration is avoided, there is higher calculating speed, and no longer need Choose the parameters such as learning rate, it is important that weigh by directly calculating the optimal stable state that obtained network weight is the network Value, so as to avoid local minimum point's problem.

Contacting based on Hidden nodes and neural metwork training precision, since direct weight determination quick can must be spellbound Weight and neural network accuracy through network, by being can be realized compared with expectation quality to the adaptive of neural network hidden layer configuration Design, it is determining to solve the problems, such as that BP neural network hidden layer configuration is difficult to, while also ensuring the Generalization Capability at network.

Rotor speed and displacement flexible measurement method needed for input signal be easy to get in Practical Project it is direct Measurable variable, the hard measurement based on self-adaptive BP neural networks can be realized by software programming.Using the solution of the present invention, omit Mechanical sensor and its interface circuit, do not need to carry out other changes to bearing-free permanent magnet synchronous motor system, are easy to work Cheng Shixian.

Additional aspect and advantage of the invention will provide in following description section, will partially become from the following description Obviously, or practice through the invention is recognized.

Detailed description of the invention

Fig. 1 shows the process of a kind of bearing-free permanent magnet synchronous motor rotor speed of the present invention and displacement flexible measurement method Figure；

Fig. 2 shows a kind of structure isoboles of BP neural network hard measurement module of the invention；

Fig. 3 shows a kind of neuron models figure of the present invention；

Fig. 4 shows a kind of self-adaptive BP neural networks illustraton of model of the present invention；

Fig. 5 shows a kind of schematic diagram of direct weight determination of the present invention；

Fig. 6 shows a kind of flow chart of self-adaptive BP neural networks prediction rotor speed and radial displacement of the present invention；

Fig. 7 shows a kind of bearing-free permanent magnet synchronous motor revolving speed using vector controlled of the invention and displacement hard measurement control System block diagram processed；

Main element symbol description:

Bearing-free permanent magnet synchronous motor 1；Levitation force winding subsystem 2；Torque winding subsystem 3；PID regulator 1； PID regulator 25；Park inverse transform module 1；Clark inverse transform module 1；Current track inverter 1；PI is adjusted Device 9；Park inverse transform module 2 10；Clark inverse transform module 2 11；Current track inverter 2 12；BP neural network is soft Measurement module 13；Park conversion module 1；Clark conversion module 1；Park conversion module 2 16；Clark conversion module 2 17.

Specific embodiment

To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real Applying mode, the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application Feature in example and embodiment can be combined with each other.

In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also To be implemented using other than the one described here other modes, therefore, protection scope of the present invention is not by described below Specific embodiment limitation.

Fig. 1 shows the process of a kind of bearing-free permanent magnet synchronous motor rotor speed of the present invention and displacement flexible measurement method Figure.

As shown in Figure 1, the present invention proposes a kind of bearing-free permanent magnet synchronous motor rotor speed and displacement flexible measurement method, institute The method of stating includes:

S102, selected characteristic variable；

S104, collecting sample data are simultaneously pre-processed for sample data, to form training sample set；

S106 is based on self-adaptive BP neural networks hard measurement module according to characteristic variable design；

S108 is trained the self-adaptive BP neural networks using the sample data that the training sample is concentrated, and adjusts It is whole and determine input layer to hidden node and and hidden node to export node layer weight；

S110, will based on self-adaptive BP neural networks hard measurement block coupled in series into bearing-free permanent magnet synchronous motor system with Establish bearing-free permanent magnet synchronous motor revolving speed and displacement hard measurement control system.

Specific implementation process of the invention successively includes following 6 step.

Step 1, selected characteristic variable.

The selection of soft-sensing model characteristic variable is divided into the selection of primary variables and the selection of auxiliary variable, with rotor speed ω and radial displacement x, y are primary variables, by model analysis, determine that measurable variable suspending power controls electric current i_x、i_yAnd its integral i_x1、i_y1, direct torque electric current i_d、i_qAnd its integral i_d1、i_q1For auxiliary variable, as shown in Fig. 2, BP neural network hard measurement mould The input of block 13 is auxiliary variable [i_x, i_x1, i_y, i_y1, i_d, i_d1, i_q, i_q1], output is primary variables [ω, x, y].Wherein, turn Rotor speed output is the random quantity being distributed in bearing-free motor speed adjustable range with sinusoidal form, and radial displacement output is bearing-free electricity The random quantity being distributed in machine air gap distance with sinusoidal form.In order to make sampled data while include dynamic and stable state that system responds Information, it is desirable that the duration long enough of Setting signal value, signal sampling time are sufficiently small.

Step 2, sampled data pre-processes.

Input is normalized with output signal, the method is as follows:

Wherein D^*Data after indicating normalization, D_minFor the minimum value in corresponding sample data, D_maxFor corresponding sample number Maximum value in, after normalized, all variable datas are all normalized to [0,1]；Again to normalization after Data are filtered, to obtain more accurate inputoutput data, form the training sample set of self-adaptive BP neural networks.Choosing The representative sample data of N group is taken, wherein every group of sample data includes 8 input variable X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈]=[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1], corresponding output variable is Y_j=[Y₁, Y₂, Y₃]=[ω, x, y], formation is adopted Sample data, for training and testing.

Step 3, self-adaptive BP neural networks are designed.

Step 3.1, neural network is initialized.

Self-adaptive BP neural networks are initialized first, since network input layer neuron number is 8, choose initial hidden layer mind Through first number m=-1+log₂8=2.Neural network proposed by the present invention for function approximation is before one kind to 3 layers of neural network, The structure of initial time neural network is the connection type of 8-2-3, and input layer has 8 nodes, and hidden layer has 2 nodes, output Layer has 3 nodes.The input vector of neural network is X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈]=[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1], output vector Y_j=[Y₁, Y₂, Y₃]=[ω, x, y].

Hidden layer uses tansig type activation primitive, expression are as follows:

Output layer uses logsig type activation primitive, expression are as follows:

Step 3.2, performance index function.

Self-adaptive BP neural networks training chooses mean square deviation function as performance indicator, and function is as follows:

Wherein, Y_diFor the corresponding desired output of i-th of sample, Y_iFor the corresponding network output of i-th of sample, N is to participate in Trained number of samples.

Step 3.3, neural network model designs.

Fig. 3 shows a kind of neuron models figure of the present invention, inputs component X_k(k=1,2 ..., 8) with it by being connected Weight component w_ik(k=1,2 ..., 8；I=1,2 ..., m；M is the quantity of hidden node) it is multiplied, input activation letter after summation Number f (), neuron export Y_iIt is influenced in addition to being influenced by input signal, while also by inside neurons other factors, because This needs to be added in neural Meta Model an additional input signal b_iAs its threshold value, output variable Y_iExpression formula are as follows:

As can be seen that only value after neuron is to input signal weighted sum is greater than b_i, Y_iJust it is greater than zero.

Fig. 4 shows a kind of self-adaptive BP neural networks model schematic of the present invention, and self-adaptive BP neural networks share three Layer, first layer is that input layer number is 8, and the second layer is that node in hidden layer is m, and third layer is that output layer number of nodes is 3.

First layer is input layer, and each input node respectively represents input vector X_k=[X₁, X₂, X₃, X₄, X₅, X₆, X₇, X₈] =[i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1] one-component.

The second layer is hidden layer, and the activation primitive of hidden layer uses tansig type activation primitive.I-th hidden layer node Output are as follows:

Wherein w_ikFor the connection weight of input layer to hidden layer node, b_iFor the threshold value of hidden layer.

Third layer is output layer, and the activation primitive of output layer uses logsig type activation primitive, output vector Y_j=[Y₁, Y₂, Y₃]=[ω, x, y].The output of j-th of output node layer are as follows:

Wherein w_jiFor hidden layer node to the connection weight of output node layer, b_jFor the threshold value of output layer.

Step 3.4, neural network weight Weigh Direct Determination.

Define weight column vector w, input excited target matrix X₀It can be denoted as respectively with target output vector γ:

W=[w₁ w₂ ... w_n]^T∈Rⁿ；

γ=[φ₁ φ₂ ... φ_m]^T∈R^m；

Then the best initial weights of neural network can directly determine are as follows:

Wherein,For matrix X₀PseudoinverseTherefore it can write a Chinese character in simplified form are as follows:

Expression formula is exported according to designed neural network model, it can be deduced that weight computing formula:

Wherein X is the input vector of neural network,For the connection weight vector of neural network input layer to hidden layer,For the connection weight vector of hidden layer to output layer, α is the desired value of hidden layer, Y_dFor neural network desired output to Amount.It just can be derived that weight by these three equationsWithAs shown in Figure 5.

This weight is the least square approximation weight of the neural network.By direct weight determination, avoid interminable Iteration has higher calculating speed, and no longer needs to choose the parameters such as learning rate, it is important that by directly calculating institute Obtained network weight is the optimal stable state weight of the network, so as to avoid local minimum point's problem.

Step 3.5, BP neural network hidden layer adaptive design.

Using direct weight determination calculate network input layer to hidden layer connection weight vectorAnd hidden layer To the connection weight vector of output layerAnd calculate e-learning error E corresponding with the number of hidden nodes_m, i.e. its property It can index.Enable E_expertIt is expected the study precision reached, if there is E_m>E_expert, then increase a hidden node, i.e. m=m+1, And recalculate connection weight vector w and e-learning error E_m, continue to compare, until meeting E_m<E_expert, stop fortune It calculates, illustrates that the precision of prediction of BP neural network at this time has been able to meet system requirements.

Self-adaptive BP neural networks predict that rotor speed and radial displacement flow chart are as shown in Figure 6.

Step 4, off-line training and test.

Off-line data derives from traditional bearing-free permanent magnet synchronous motor control system, using least square method in PC machine Off-line training is carried out to self-adaptive BP neural networks by Matlab software, 3/4 sample is chosen and neural network is trained, Other 1/4 is used for the test data of neural network.By 2000 times or so training, neural network output mean square error is less than 0.001, it meets the requirements, so that it is determined that the weight of self-adaptive BP neural networksWithHidden nodes m and threshold value b_i、 b_j。

Step 5, bearing-free permanent magnet synchronous motor revolving speed and displacement hard measurement control system are established.

Trained self-adaptive BP neural networks hard measurement module 13 is accessed into bearing-free permanent magnet synchronous motor control system In, rotor speed forecast module and radial displacement prediction module are constituted, realizes that rotor speed and radial displacement are predicted in real time online.Using The bearing-free permanent magnet synchronous motor revolving speed and displacement hard measurement control system block diagram of vector controlled are as shown in Figure 7.Label in Fig. 7: ω *, x*, y* are reference rotor revolving speed and radial displacement；ω, x, y are the rotor using self-adaptive BP neural networks predictive estimation Revolving speed and radial displacement；ε_ω、ε_x、ε_yFor speed error and rotor radial displacement error；i_x*、i_y* pass through PID tune for offset deviation The suspending power control electric current that section device 1 and the conversion of PID regulator 25 obtain, it is static to obtain two-phase through Park inverse transform module 1 Suspending power under coordinate system controls electric current i_1αAnd i *_1β*, then through Clark inverse transform module 1 obtain three-phase current i_1U*、i_1V*、 i_1W*, the three-phase current i of real-time control is obtained by current track inverter 1_1U、i_1V、i_1W；i_d*、i_qIt * is revolving speed deviation By the direct torque electric current that the conversion of pi regulator 9 obtains, i is enabled_d*=0, to obtain two-phase through Park inverse transform module 2 10 static Suspending power under coordinate system controls electric current i_2αAnd i *_2β*, then through Clark inverse transform module 2 11 obtain three-phase current i_2U*、i_2V*、 i_2W*, the three-phase current i of real-time control is obtained by current track inverter 2 12_2U、i_2V、i_2W；Three-phase current i_1U、i_1V、i_1W It is input in levitation force winding subsystem 2, three-phase current i_2U、i_2V、i_2WIt is input in torque winding subsystem 3, bearing-free permanent magnet Synchronous motor 1 is made of levitation force winding subsystem 2 and torque winding subsystem 3；i_1αAnd i_1β、i_2αAnd i_2βRespectively i_1U、i_1V、 i_1WAnd i_2U、i_2V、i_2WThe levitation force winding and torque winding obtained through Clark conversion module 1 and Clark conversion module 2 17 Feedback current under two-phase stationary coordinate system, i_xAnd i_y、i_dAnd i_qRespectively i_1αAnd i_1β、i_2αAnd i_2βThrough one 14 He of Park conversion module Feedback current under levitation force winding and torque winding two-phase rotating coordinate system that Park conversion module 2 16 obtains, i_x1And i_y1、i_d1 And i_q1Respectively i_xAnd i_y、i_dAnd i_qIntegral, [i_x, i_y, i_x1, i_y1, i_d, i_q, i_d1, i_q1] it is used as BP neural network hard measurement mould The input signal of block, hard measurement module export feedback signal [ω, x, y].

Present invention design is based on self-adaptive BP neural networks hard measurement module, is with rotor speed ω and radial displacement x, y Primary variables controls electric current i with suspending power_x、i_yAnd its integral i_x1、i_y1, direct torque electric current i_d、i_qAnd its integral i_d1、i_q1For Auxiliary variable；Acquire input data [i_x、i_y、i_x1、i_y1、i_d、i_q、i_d1、i_q1] with output data [ω, x, y] carry out pretreatment with Normalization, form training sample, self-adaptive BP neural networks are trained, adjust and determine input layer to hidden node with Weight of the hidden node to output node layer；It will be synchronous to bearing-free permanent magnet based on self-adaptive BP neural networks hard measurement block coupled in series In electric system, rotor speed ω and radial displacement x, y are predicted；By above-mentioned signal ω, x, y respectively with given bearing-free permanent magnet Synchronous electric motor rotor reference rotation velocity ω * and reference displacement signal x*, y* are compared；Its difference is adjusted by PID regulator, PI Device, Park inverse transform module, Clark inverse transform module and current track inverter adjust the suspension for respectively obtaining real-time control Power three-phase current i_1U、i_1V、i_1W, torque three-phase current i_2U、i_2V、i_2W。

Further, design self-adaptive BP neural networks need to initialize neural network, set initial time neural network Structure is the connection type of 8-2-3, and input layer has 8 nodes, and hidden layer has 2 nodes, and output layer has 3 nodes；Adaptively BP neural network training chooses mean square deviation function and as performance indicator and carries out neural network model design, and is based on neural network Direct weight determination calculates the optimal stable state weight of network；Adaptive design is carried out to meet to BP neural network hidden layer System requirements finally carries out off-line training and test to neural network, by trained self-adaptive BP neural networks hard measurement mould Block accesses in bearing-free permanent magnet synchronous motor control system.

The present invention has the following advantages that.

1. the present invention has the characteristics that good non-linear mapping capability according to BP neural network, using self-adaptive BP nerve net Network realizes the Nonlinear Mapping between auxiliary variable and rotor revolving speed and radial displacement, establishes bearing-free permanent magnet and synchronizes electricity The soft-sensing model of machine rotor revolving speed and displacement realizes the hard measurement to rotor speed and displacement, the good, stability with real-time Well, the features such as precision is high.

2. the present invention is without using Heuristics, special without the operation mechanism for understanding bearing-free permanent magnet synchronous motor in depth Property, only the Black-Box identification of radial displacement system need to can be realized using data are output and input.After Black-Box identification, as long as passing through Study to input data can predict rotor speed and radial displacement.

3. the present invention proposes that direct weight determination directly determines the minimum of neural network by the pseudoinverse of input vector X Two multiply and approach weight.By direct weight determination, interminable iteration is avoided, there is higher calculating speed, and no longer Need to choose the parameters such as learning rate, it is important that by directly calculating the optimal stable state that obtained network weight is the network Weight, so as to avoid local minimum point's problem.

4. contacting based on Hidden nodes and neural metwork training precision, since direct weight determination can be obtained quickly The weight and neural network accuracy of neural network, by being can be realized compared with expectation quality to the adaptive of neural network hidden layer configuration It should design, it is determining solve the problems, such as that BP neural network hidden layer configuration is difficult to, while also ensuring the Generalization Capability at network.

5. rotor speed and displacement flexible measurement method needed for input signal be easy to get in Practical Project it is straight Measurable variable is connect, the hard measurement based on self-adaptive BP neural networks can be realized by software programming.Using the solution of the present invention, save Mechanical sensor and its interface circuit have been omited, has not needed to carry out other changes to bearing-free permanent magnet synchronous motor system, be easy to Project Realization.

In several embodiments provided herein, it should be understood that disclosed device and method can pass through it Its mode is realized.Apparatus embodiments described above are merely indicative, for example, the division of the unit, only A kind of logical function partition, there may be another division manner in actual implementation, such as: multiple units or components can combine, or It is desirably integrated into another system, or some features can be ignored or not executed.In addition, shown or discussed each composition portion Mutual coupling or direct-coupling or communication connection is divided to can be through some interfaces, the INDIRECT COUPLING of equipment or unit Or communication connection, it can be electrical, mechanical or other forms.

Above-mentioned unit as illustrated by the separation member, which can be or may not be, to be physically separated, aobvious as unit The component shown can be or may not be physical unit；Both it can be located in one place, and may be distributed over multiple network lists In member；Some or all of units can be selected to achieve the purpose of the solution of this embodiment according to the actual needs.

In addition, each functional unit in various embodiments of the present invention can be fully integrated in one processing unit, it can also To be each unit individually as a unit, can also be integrated in one unit with two or more units；It is above-mentioned Integrated unit both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.

Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through The relevant hardware of program instruction is completed, and program above-mentioned can store in computer-readable storage medium, which exists When execution, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: movable storage device, read-only deposits Reservoir (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or The various media that can store program code such as CD.

If alternatively, the above-mentioned integrated unit of the present invention is realized in the form of software function module and as independent product When selling or using, it also can store in a computer readable storage medium.Based on this understanding, the present invention is implemented Substantially the part that contributes to existing technology can be embodied in the form of software products the technical solution of example in other words, The computer software product is stored in a storage medium, including some instructions are used so that computer equipment (can be with It is personal computer, server or network equipment etc.) execute all or part of each embodiment the method for the present invention. And storage medium above-mentioned includes: that movable storage device, ROM, RAM, magnetic or disk etc. are various can store program code Medium.

The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.