CN107276465B - A kind of torque-current neural network switch reluctance motor control method and system - Google Patents

A kind of torque-current neural network switch reluctance motor control method and system Download PDF

Info

Publication number
CN107276465B
CN107276465B CN201710494319.8A CN201710494319A CN107276465B CN 107276465 B CN107276465 B CN 107276465B CN 201710494319 A CN201710494319 A CN 201710494319A CN 107276465 B CN107276465 B CN 107276465B
Authority
CN
China
Prior art keywords
torque
current
neural network
srm
phase
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201710494319.8A
Other languages
Chinese (zh)
Other versions
CN107276465A (en
Inventor
党选举
王土央
李珊
姜辉
伍锡如
张向文
蔡春晓
朱国魂
莫太平
司亚
张堡森
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guilin University of Electronic Technology
Original Assignee
Guilin University of Electronic Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guilin University of Electronic Technology filed Critical Guilin University of Electronic Technology
Priority to CN201710494319.8A priority Critical patent/CN107276465B/en
Publication of CN107276465A publication Critical patent/CN107276465A/en
Application granted granted Critical
Publication of CN107276465B publication Critical patent/CN107276465B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
    • H02P6/00—Arrangements for controlling synchronous motors or other dynamo-electric motors using electronic commutation dependent on the rotor position; Electronic commutators therefor
    • H02P6/34—Modelling or simulation for control purposes
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
    • H02P23/00—Arrangements or methods for the control of AC motors characterised by a control method other than vector control
    • H02P23/0004—Control strategies in general, e.g. linear type, e.g. P, PI, PID, using robust control
    • H02P23/0018—Control strategies in general, e.g. linear type, e.g. P, PI, PID, using robust control using neural networks
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
    • H02P25/00—Arrangements or methods for the control of AC motors characterised by the kind of AC motor or by structural details
    • H02P25/02—Arrangements or methods for the control of AC motors characterised by the kind of AC motor or by structural details characterised by the kind of motor
    • H02P25/08—Reluctance motors
    • H02P25/098—Arrangements for reducing torque ripple
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
    • H02P6/00—Arrangements for controlling synchronous motors or other dynamo-electric motors using electronic commutation dependent on the rotor position; Electronic commutators therefor
    • H02P6/10—Arrangements for controlling torque ripple, e.g. providing reduced torque ripple

Landscapes

  • Engineering & Computer Science (AREA)
  • Power Engineering (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Pharmaceuticals Containing Other Organic And Inorganic Compounds (AREA)
  • Control Of Electric Motors In General (AREA)

Abstract

The present invention is a kind of torque-current neural network switch reluctance motor control method and system, this method is one neural network feedforward controller of each phase configuration of SRM, with torque-current inversion model for its activation primitive, it is input to give each phase torque reference and the rotor position angle of total torque distribution, Feedback error learning is realized with the output of PID controller.The output of neural network feedforward controller is superimposed with the output of PID controller is sent into current hysteresis-band control device as reference current, in conjunction with the control SRM operation of current flow feedback signal.This system SRM installs electric current, position and torque sensor, and signal processor is containing there are three neural network feedforward controller, torque distribution module, pid control module, electric current hysteresis loop control modules.Inner ring current hysteresis-band control device track reference electric current, control SRM operation, has fully considered that SRM has special strong nonlinearity, has effectively reduced the torque pulsation of SRM.

Description

A kind of torque-current neural network switch reluctance motor control method and system
Technical field
The present invention relates to the control technology field of new-energy automobile driving switched reluctance machines, specially a kind of torque- Electric current neural network switch reluctance motor control method and system.
Background technique
Currently, energy and environmental problem becomes increasingly conspicuous, and electric car is widely paid attention to, before having very big development Scape.For other motors, SRM (the SRM table in switched reluctance machines Switched Reluctance Motor, this paper Show switched reluctance machines) there is structure simple rigid, be not necessarily to rare earth material, be suitable for many good characteristics such as frequent start and stop, In New-energy electric vehicle field most application potential.But the characteristics of SRM itself double-salient-pole structure, magnetic circuit are in strong nonlinearity And saturability, cause motor when operating there are torque pulsation, noise, vibration it is big the disadvantages of, hinder SRM in New energy electric sedan-chair Application in motor vehicle drive system.In SRM control, in order to inhibit SRM torque pulsation, researcher is in Direct Torque Control and indirectly Many solutions are proposed on Stator-Quantities Control.
Direct Torque Control DTC (Direct Torque Control, DTC) method instantaneous torque, magnetic linkage determine switch State, realization directly control SRM torque, achieve the purpose that Torque Ripple Reduction.Direct Instantaneous torque control DITC (Direct Instantaneous Torque Control, DITC) generates the switch control of every phase using torque hysteresis-controller Signal processed, structure is simple, is not required to commutation strategy, can inhibit SRM torque pulsation in wide velocity interval.Torque partition function (Torque-Sharing Function, TSF) control strategy, solves in commutation process by torque distribution mechanism, opens phase The problem of with shutdown phase output torque smooth transition, reduce torque pulsation intensity caused by being operated by direct commutation；Torque point Method of completing the square realizes SRM Torque Ripple Reduction usually in conjunction with other control strategies.
In the control problem for solving nonlinear system, intelligent control method has compared to conventional control method to be protruded Advantage, neural network intelligent control especially therein.ANN Control have powerful None-linear approximation ability, at For the research hotspot of intelligent control.Document report is to control target, based on BP (back propagation, BP) nerve with revolving speed The SRM speed control of network and the design of RBF (Radio Basis Function, RBF) neural network, overcomes Traditional control The limitation of device processing nonlinear system.Researcher also by DTC control strategy in conjunction with neural network, with neural network generation It is selected for traditional status switch.And has and utilize collected SRM magnetic linkage, electric current and rotor-position angular data, off-line training mind Through network, magnetic linkage, electric current and rotor position angle relational model are established, realizes the rotor position angle estimation that can be applied to control.Have It is reported in TSF distribution basis, using BP neural network, optimizes SRM conducting and turn-off time, reduce SRM torque pulsation, propose The neural network model of torque and electric current inhibits SRM torque pulsation.Correlative study proposes SNN (Spiking Neural Networks, SNN) realize SRM modeling and control, wherein SNN structure is very similar to RBF neural.It is neural network based SRM control, stresses the modeling of torque, the Modeling Research of electric current.It according to the literature, is knot with the Taylor series of Current Position function Structure builds ANN Control current model, learns Taylor series parameter by Neural Network Online.It also has been reported that for SRM Nonlinear characteristic proposes chaos principle, improves neural network Gradient learning method, excellent by heredity for SRM nonlinear characteristic Change algorithm, using combined training wavelet neural network HTWNN (Hybrid Training Wavelets Neural Network, HTWNN the nonlinear models such as the magnetic linkage torque of SRM) are established, the inevitable small echo labyrinth of the method and macrooperation amount lack It falls into.Adaptive network inference system and fuzzy neural network is respectively adopted to model SRM inductance characteristic and flux linkage characteristic, than There is more preferable performance compared with analysis fuzzy neural network SRM modeling.Double back propagation neural network is such as designed, SRM electric current and torque are carried out Modeling.
In existing document report, model in SRM with control, neural network used is all general neural network mould Type is not associated with the special nonlinear characteristic of SRM and carries out neural network model design, therefore is difficult to control effectively to SRM torque.
Summary of the invention
The purpose of the present invention is designing a kind of torque-current neural network switch reluctance motor control method, this method is One neural network feedforward controller based on torque-current relationship of each phase configuration in SRM three-phase, it is inverse with torque-current Activation primitive of the model expression as feedforward controller neural network hidden layer presses cube partition function for given total torqueDistribution is used as each phase torque reference Tkk(θ), torque reference TkkThe input of (θ) and rotor position angle θ as neural network.Mind Feedback error learning is realized based on the output of outer ring PID controller through network feedforward controller.Neural network feedforward controller The output for compensating PID controller is exported, the two superposition is sent into current hysteresis-band control device as reference current, in conjunction with current electricity Flow feedback signal control SRM operation.This method has fully considered that SRM has special strong nonlinearity, effectively reduces the torque of motor Pulsation.
It is another object of the present invention to design a kind of torque-current neural network switched Reluctance Motor Control System.SRM peace Dress current sensor, position sensor and torque sensor, the signal processor of this system contain three nerve nets of torque-current Network feedforward control module device, torque distribution module, pid control module, electric current hysteresis loop control module.The neural network feedforward Controller is torque-current inversion model neural network.Torque distribution module is connected with neural network feedforward controller；Outer ring PID The output of control module is as neural network feedforward controller learning signal, inner ring current hysteresis-band control device track reference electric current, Through analog line driver control SRM operation, realizes SRM direct torque, effectively reduce the torque pulsation of SRM.
A kind of torque-current neural network switch reluctance motor control method that the present invention designs, comprising the following steps:
Each phase torque distribution of step I SRM
The output of torque reference distribution is as one of neural network feedforward controller input signal.Torque partition function TSF, Using optimal allocation function: cube partition function, expression formula are as follows:
F (θ) indicates cube partition function in formula；θ is rotor position angle；θonFor turn-on angle；θovFor commutation overlap angle；
According to formula (1), by given total torqueDistribution is used as each phase torque reference Tkk(θ), torque distribute formula such as Under:
In formulaTo set torque reference；Tkk(θ) is the phase torque reference changed with θ；F (θ) indicates a cube distribution letter Number；θ is rotor position angle；θonFor turn-on angle；θovFor commutation overlap angle；θoffTo turn off angle；A, B of kk=1,2,3 expression SRM And C three-phase.
Step II, the neural network feedforward controller based on torque-current relationship construction
The torque-current inversion model and neural network feedforward controller excitation function of II -1 SRM
Practical SRM phase inductance has strong nonlinearity characteristic, and accurate SRM mathematical model is difficult to obtain.If ignoring magnetic saturation With edge effect to get simplified linear inductance model is arrived, by using the torque-current conversion formula of the linear inductance model It calculates and obtains each phase reference current.Because simplifying the non-thread sex differernce of linear inductance model and actual inductance model, resulting each phase Reference current is difficult to accurately control motor torque, and the torque pulsation of its control output is caused to deteriorate.
Comprehensively consider SRM nonlinear characteristic and magnetic saturation characteristic, analysis torque and phase current relationship, the present invention is using calibrated True torque-current inversion model expression formula:
T in formula (3)e(θ) is the phase torque reference changed with θ, and value is that torque partition function exports Tkk；a(θ),b (θ) parameter is respectively by rotor position angle θ and fixed weight vector Wa、WbIt determines；I (θ) is the phase current changed with θ.
General hidden layer excitation function of formula (3) the torque-current inversion model expression formula as neural network feedforward controller, structure The neural network of torque-current inversion model is made, to obtain each phase optimal reference electric current.
II -2, neural network feedforward controller
The present invention is one neural network feedforward controller of each phase configuration in SRM three-phase, is torque-current against mould Type neural network.The neural network feedforward controller of a certain phase is the neural network for including input layer, hidden layer and output layer.Ginseng Examine torque TkkThe input of (θ) and rotor position angle θ as neural network feedforward controller, Tkk(θ) is phase torque reference by step The torque partition function TSF of I is obtained；θ is SRM rotor position angle, as measured by the position sensor on SRM；Before neural network Present the hidden layer H=[O of controller1,O2,…Oj], OjFor the output of hidden layer node, j is node in hidden layer, j=1,2, The maximum value of 3 ..., j is 6~200；Output layer is summation linear function.
The neural network weighting coefficient study of neural network feedforward controller uses well known momentum gradient science of law learning method.
Neural network feedforward controller operational process is as follows:
II -21, hidden layer weight vector
Parameter determines before neural network learning in general hidden layer excitation function: fixed weight vector Wa=[wa1,wa2,… waj]T、Wb=[wb1,wb2,…wbj]TRandom initializtion is any number between 10~30；Output layer weight vector Wout= [w1,w2,…wj]TInitialization, value are greater than zero, and it is 0.1 which, which takes initial value,.
II -22, rotor position angle θ
Rotor position angle θ measured by position sensor on SRM is normalized to period change between 0 to 1 through trigonometric function The numerical value of change.
II -23, the output of each node of hidden layer
It is calculated by formula (3), obtains output valve O of i (θ) value as each node of hidden layerj。
II -24, neural network feedforward controller exports
In formulaIndicating the output electric current of some neural network feedforward controller, referred to as certain phase neural network exports electric current, wjFor weighting coefficient, j=1,2,3 ... m, m=40, kk=1,2,3.
II -25, according to learning signal ierrorAdjustment output weight
Current torque value Tc and given total torque are obtained by the torque sensor installed on SRMDifference it is inclined as torque Poor Δ T, as the defeated of outer ring PID controller (Proportion Integral Derivative proportional plus integral plus derivative controller) Enter, outer ring PID controller obtains exporting electric current i with the PID that θ changeserror, ierrorIt distributes to obtain the PID control of certain phase through electric current Electric current For certain resulting phase reference current of this method；
II -3, the Feedback error learning of neural network feedforward controller
For neural network feedforward controller of the present invention using the on-line study for having supervision, supervised learning signal is PID control The control electric current i that device obtainserror.The input layer of neural network need to only adjust the weight of output layer using fixed weight.
The performance index function of the weight of output layer are as follows:
Step II -21 determines neural network feedforward controller hidden layer weight vector WaAnd WbNumerical value after, according under gradient Drop method adjusts the output weight of neural network feedforward controller；
W in formulajIt (k) is the output layer weight at k moment, wj(k-1) be the k moment previous moment output layer weight, wj It (k-2) is the output layer weight of the previous moment at (k-1) moment, Δ wjIt (k) is adjustment value increase；ierrorIt (k) is the k moment PID exports electric current；OjIt (k) is the output quantity at hidden layer k moment；J is node in hidden layer (j=1...m), and α is factor of momentum, Range, 0.01~0.1, η are parameter regularized learning algorithm rate, range 0~1.
Step III, current hysteresis-band control
The output of neural network feedforward controllerWith PID controller outputThe reference of this method is obtained after superposition Electric currentAs the reference current of interior circular current hysteresis loop controller, while the current sensor installed on SRM also will be current The current value I of three-phasekk.mearCircular current hysteresis loop controller in being sent into, interior circular current hysteresis loop controller is through analog line driver control SRM operation processed, realizes direct torque, effectively inhibits the torque pulsation of SRM.
In SRM starting, neural network feedforward controller and PID controller collective effect obtain reference currentMake System steady operation；For neural network feedforward controller after sufficiently learning, the PID of PID output exports electric current ierrorIt goes to zero, The PID control electric current of each phaseIt goes to zero, the ANN Control electric current of each phaseTend to reference currentAnd make defeated Torque T outcTend to given total torqueTorque deviation Δ T goes to zero, and PID controller maintains to stablize output, and SRM is in nerve net Dynamic balance state is under the control of network feedforward controller and PID controller.
A kind of torque-current neural network switch reluctance motor control method according to the present invention, designs a kind of torque-electricity Flow neural network switched Reluctance Motor Control System, including signal processor, analog line driver, current sensor, position sensing Device, torque sensor, display and SRM install torque sensor on the output shaft of switched reluctance machines, export the current of SRM Dtc signal Tc；Installation site sensor on magnetic resistance motor rotor exports current rotor angular position theta；The three-phase input end of SRM point Not An Zhuan current sensor, export the current value of current each phase.
The signal processor contains the three neural network feedforward controllers matched with SRM tri-, is torque-current Inversion model neural network, there are also torque distribution module, pid control module, electric current hysteresis loop control modules.
The torque distribution module is by given total torqueIt is assigned as each phase torque reference Tkk(θ) and rotor position angle θ Together, three neural network feedforward controllers are inputted respectively, and three neural network feedforward controllers export the neural network of each phase Control electric currentIt is respectively fed to three current adders.
The current torque value Tc and given total torque that the torque sensor installed on SRM obtainsTorque subtracter is accessed, The torque deviation Δ T access external circulation PID controller of output, outer ring PID controller obtain PID output electric current i accordinglyerror, ierror Electric current distribution module is accessed, electric current distribution module obtains the PID control electric current of each phase according to current rotor position angleAccess Three current adders.The i of PID output simultaneouslyerrorNeural network as three torque-current neural network feedforward controllers Learning signal.
Three current adders realize each phase neural network output electric current respectivelyWith PID control electric currentSuperposition, obtains The ideal reference current of three-phaseElectric current hysteresis loop control module is accessed, while the current sensor installed on SRM will also work as The current value of three-phase is sent into interior circular current hysteresis loop controller as current feedback signal, interior circular current hysteresis loop controller Output is through analog line driver control SRM operation.
Signal processor is connect with display, display control state and control result.
Signal processor configure CAN (controller local area network Controller Area Network) interface, provide with outside If communication interface.
Compared with prior art, a kind of torque-current neural network switch reluctance motor control method of the present invention and system The advantages of are as follows: 1, reference RBF neural network structure, using SRM torque-current reverse-power function as neural network hidden layer Activation primitive constructs a kind of neural network feedforward controller based on torque-current relationship, for the special non-thread of SRM Property is modeled, and the ANN Control electric current of output can compensate for the PID control electric current of the output of outer ring PID controller, the two Superposition obtains each phase reference current；2, the PID output electric current of outer ring PID controller output is as neural network feedforward controller Learning signal establishes neural network feedforward controller using Feedback error learning method, with on-line study ability and certainly Adapt to adjustment capability；3, under the cooperation control of interior circular current hystersis controller, outer ring PID controller and neural network feedforward control The control strategy that device combines can effectively realize direct torque, and effectively inhibit the torque pulsation of SRM；It feedovers with impassivity network Controller conventional method compares, and the present invention realizes effective control of SRM torque, torque pulsation rate low 50%；4, designed Neural network feedforward controller is simple for structure, and pace of learning is fast, it is easy to accomplish.
Detailed description of the invention
Fig. 1 is that certain phase neural network of this torque-current neural network switch reluctance motor control method embodiment feedovers Controller architecture schematic diagram；
Fig. 2 is this torque-current neural network switched Reluctance Motor Control System embodiment overall structure diagram.
Specific embodiment
Torque-current neural network switch reluctance motor control method embodiment
This torque-current neural network switch reluctance motor control method embodiment the following steps are included:
Each phase torque distribution of step I SRM
This example torque partition function TSF uses optimal allocation function cube partition function, expression formula are as follows:
F (θ) indicates cube partition function in formula；
According to formula (1), by given total torqueDistribution is used as each phase torque reference Tkk(θ), torque distribute formula such as Under:
In formulaTo set torque reference；Tkk(θ) is the phase torque reference changed with θ；F (θ) indicates a cube distribution letter Number；θ is rotor position angle；θonFor turn-on angle；θovFor commutation overlap angle；θoffTo turn off angle；A, B of kk=1,2,3 expression SRM And C three-phase.
Step II, the neural network feedforward controller based on torque-current relationship construction
The torque-current inversion model and neural network feedforward controller excitation function of II -1 SRM
This example uses accurate torque-current inversion model expression formula:
T in formula (3) formulae(θ) is phase torque reference, and value is that torque partition function exports Tkk(θ), a (θ), b (θ) ginseng Number is respectively by rotor position angle θ and fixed weight vector Wa、WbIt determines.Mind of the formula (3) as neural network feedforward controller Through network general hidden layer excitation function.
II -2, neural network feedforward controller
This example is one neural network feedforward controller of each phase configuration in SRM motor three-phase, is that torque-current is inverse Model Neural.The neural network feedforward controller of a certain phase is as shown in Figure 1, being includes input layer, hidden layer and output layer Neural network.Its torque reference TkkThe input of (θ) and rotor position angle θ as neural network feedforward controller, Tkk(θ) is Phase torque reference is obtained by the torque partition function TSF of step I.θ is SRM rotor position angle, by the position sensor institute on SRM It measures；Hidden layer H=[O1,O2,…Oj], OjFor the output of hidden layer node, j is node in hidden layer, j=1,2,3 ..., The maximum value of this example j is 40；The SRM torque-current inversion model of step II -1, i.e. formula (3) are used as general hidden layer excitation function；It is defeated Layer is summation linear function out.
The neural network weighting coefficient study of this example neural network feedforward controller uses well known momentum gradient calligraphy learning.
This example neural network feedforward controller operational process is as follows:
II -21, hidden layer weight vector
Parameter determines before neural network learning in general hidden layer excitation function: fixed weight vector Wa=[wa1,wa2,… waj]T、Wb=[wb1,wb2,…wbj]TRandom initializtion is any number between 10~30；Output layer weight vector Wout= [w1,w2,…wj]TInitialization, this example WoutTaking initial value is 0.1.
II -22, rotor position angle θ
Rotor position angle θ measured by position sensor on SRM is normalized to period change between 0 to 1 through trigonometric function The numerical value of change.
II -23, the output of each node of hidden layer
It is calculated by formula (3), obtains output valve O of i (θ) value as each node of hidden layerj。
II -24, neural network feedforward controller exports
In formulaIndicating the output electric current of some neural network feedforward controller, referred to as certain phase neural network exports electric current, wjFor weighting coefficient, the maximum value of j=1,2,3 ..., this example j are 40.
II -25, according to learning signal ierrorAdjustment output weight
Current torque value Tc and given total torque are obtained by the torque sensor installed on SRMDifference it is inclined as torque Poor Δ T, as the input of outer ring PID controller, outer ring PID controller obtains PID output electric current ierror, ierrorThrough electric current point With obtaining the PID control electric current of three-phase For certain resulting phase reference current of this method；
II -3, the Feedback error learning of neural network feedforward controller
For this example neural network feedforward controller using the on-line study for having supervision, supervised learning signal is PID controller Obtained control electric current ierror.The input layer of neural network need to only adjust the weight of output layer using fixed weight.
The performance index function of the weight of output layer are as follows:
Step II -21 determines neural network feedforward controller hidden layer weight vector WaAnd WbNumerical value after, according under gradient Drop method adjusts the output weight of neural network feedforward controller；
W in formulajIt (k) is the output layer weight at k moment, wj(k-1) be the k moment previous moment output layer weight, wj It (k-2) is the output layer weight of the previous moment at (k-1) moment, Δ wjIt (k) is adjustment value increase；ierrorIt (k) is the k moment PID exports electric current；OjIt (k) is the output quantity at hidden layer k moment；α is factor of momentum, and η is parameter regularized learning algorithm rate, this example α= 0.01, η=0.1.
In SRM starting, neural network feedforward controller and PID controller collective effect obtain reference currentMake System steady operation；Neural network feedforward controller is after sufficiently learning, ierrorIt goes to zero,It goes to zero,Tend toAnd make TcTend to T* ref, torque deviation Δ T goes to zero, and PID controller maintains to stablize output, and SRM feedovers in neural network Dynamic balance state is under the control of controller and PID controller.
Step III, current hysteresis-band control
The output of neural network feedforward controllerIt is exported with PID controllerReference current is obtained after superpositionIts For the reference current of interior circular current hysteresis loop controller, while the current sensor installed on SRM will also work as the electric current of three-phase Value Ikk.mearCircular current hysteresis loop controller in being sent into, interior circular current hysteresis loop controller are run through analog line driver control SRM, It realizes direct torque, effectively inhibits the torque pulsation of SRM.
Torque-current neural network switched Reluctance Motor Control System embodiment
According to above-mentioned torque-current neural network switch reluctance motor control method embodiment, this torque-current is designed Neural network switched Reluctance Motor Control System embodiment includes signal processor, analog line driver, electric current biography as shown in Figure 2 Sensor, position sensor, torque sensor, display and switched reluctance machines SRM are installed on the output shaft of switched reluctance machines Torque sensor exports the current torque signal T of SRMc；Installation site sensor on magnetic resistance motor rotor exports current rotor Angular position theta；The three-phase input end of SRM installs current sensor respectively, exports the current value of current each phase.
This example signal processor contains the three neural network feedforward controllers matched with SRM tri-, is torque-current Inversion model neural network, there are also torque distribution module, pid control module, electric current hysteresis loop control modules；Torque distribution module will Given total torque Tr*efIt is assigned as each phase torque reference Tkk(θ) and rotor position angle θ together, input three neural networks respectively Feedforward controller exports the ANN Control electric current of each phaseIt is respectively fed to three current adders；
The current torque value Tc and given total torque that the torque sensor installed on SRM obtainsTorque subtracter is accessed, The torque deviation Δ T access external circulation PID controller of output, outer ring PID controller obtain control electric current i accordinglyerror, ierrorIt connects Enter electric current distribution module, electric current distribution module obtains the PID control electric current of three-phase according to current rotor position angleAccess three A current adder.The i of PID output simultaneouslyerrorLearning signal as three neural network feedforward controller neural networks.
Three current adders realize ANN Control electric current respectivelyWith PID control electric currentSuperposition, obtains three-phase Ideal reference currentElectric current hysteresis loop control module is accessed, while the current sensor installed on SRM will also work as first three The current value of phase is sent into interior circular current hysteresis loop controller, the output of interior circular current hysteresis loop controller as current feedback signal Through analog line driver control SRM operation.
This example signal processor is connect with display, display control state and control result.
This example signal processor configures CAN interface, provides and peripheral communication interface.
Above-described embodiment is only further described the purpose of the present invention, technical scheme and beneficial effects specific A example, present invention is not limited to this.All any modifications made within the scope of disclosure of the invention, change equivalent replacement Into etc., it is all included in the scope of protection of the present invention.

Claims (6)

1. a kind of torque-current neural network switch reluctance motor control method, comprising the following steps:
Each phase torque distribution of step I SRM
Using optimal allocation function: cube partition function, expression formula are as follows:
F (θ) indicates cube partition function in formula；θ is rotor position angle；θonFor turn-on angle；θovFor commutation overlap angle；
According to formula (1), by given total torqueDistribution is used as each phase torque reference Tkk(θ), it is as follows that torque distributes formula:
In formulaTo set torque reference；Tkk(θ) is the phase torque reference changed with θ；θoffTo turn off angle；Kk=1,2,3 table Show A, B and C three-phase of SRM；
Step II, the neural network feedforward controller based on torque-current relationship construction
The torque-current inversion model and neural network feedforward controller excitation function of II -1 SRM
Using torque-current inversion model expression formula:
T in formula (3)e(θ) is the phase torque reference changed with θ, and value is that torque partition function exports Tkk(θ)；a(θ),b(θ) Parameter is respectively by rotor position angle θ and fixed weight vector Wa、WbIt determines；I (θ) is to change phase current with θ；
Torque-current inversion model expression formula is the general hidden layer excitation function of neural network feedforward controller；
II -2, neural network feedforward controller
One neural network feedforward controller of each phase configuration in SRM three-phase is torque-current inversion model neural network； The neural network feedforward controller of a certain phase is the neural network for including input layer, hidden layer and output layer；Torque reference Tkk(θ) Input with rotor position angle θ as neural network feedforward controller, Tkk(θ) is that phase torque reference is distributed by the torque of step I Function TSF is obtained；θ is SRM rotor position angle, as measured by the position sensor on SRM；Neural network feedforward controller it is hidden [the O of H=containing layer1,O2,…Oj], OjFor the output of hidden layer node, j is node in hidden layer, j=1,2,3 ... the maximum of, j Value is 6~200；Output layer is summation linear function；
Neural network feedforward controller operational process is as follows:
II -21, hidden layer weight vector
Parameter determines before neural network learning in general hidden layer excitation function: fixed weight vector Wa=[wa1,wa2,…waj]T、Wb =[wb1,wb2,…wbj]TRandom initializtion is any number between 10~30；Output layer weight vector Wout=[w1,w2,… wj]TInitialization, value are greater than zero；
II -22, rotor position angle θ
Rotor position angle θ measured by position sensor on SRM is normalized to mechanical periodicity between 0 to 1 through trigonometric function Numerical value；
II -23, the output of each node of hidden layer
It is calculated by torque-current inversion model expression formula, obtains output valve O of i (θ) value as each node of hidden layerj；
II -24, neural network feedforward controller exports
In formulaIndicate the output electric current of some neural network feedforward controller, referred to as certain phase neural network exports electric current, wjFor Weighting coefficient；
II -25, according to learning signal ierrorAdjustment output weight
Current torque value Tc and given total torque are obtained by the torque sensor installed on SRMDifference be torque deviation Δ T, As the input of outer ring PID controller, outer ring PID controller obtains exporting electric current i with the PID that θ changeserror, ierrorThrough electric current Distribution obtains the PID control electric current of certain phase For certain resulting phase reference current of this method；
II -3, the Feedback error learning of neural network feedforward controller
Neural network feedforward controller input layer of the present invention is using fixed weight, the performance index function of the weight of output layer Are as follows:
Step II -21 determines neural network feedforward controller hidden layer weight vector WaAnd WbNumerical value after, declined according to gradient Method adjusts the output weight of neural network feedforward controller；
W in formulajIt (k) is the output layer weight at k moment, wj(k-1) be the k moment previous moment output layer weight, wj(k-2) It is the output layer weight of the previous moment at (k-1) moment, Δ wjIt (k) is adjustment value increase；ierror(k) defeated for the PID at k moment Current value out；OjIt (k) is the output quantity at hidden layer k moment；α is factor of momentum, and α value is that 0.01~0.1, η is parameter adjustment Learning rate, η value 0~1；
Step III, current hysteresis-band control
The ANN Control electric current and PID controller of the output of neural network feedforward controller export PID control electric currentIt is folded The reference control electric current obtained after addingAs the input current of interior circular current hysteresis loop controller, while being installed on SRM Current sensor will also work as the current value I of three-phasekk.mearCircular current hysteresis loop controller, interior circular current hysteresis loop control in being sent into Device processed realizes direct torque, inhibits the torque pulsation of SRM through analog line driver control SRM operation.
2. a kind of torque-current neural network switch reluctance motor control method according to claim 1, feature exist In:
The neural network weighting coefficient study of neural network feedforward controller uses momentum gradient calligraphy learning in the step II -2.
3. a kind of torque-current neural network switch reluctance motor control method according to claim 1, feature exist In:
In SRM starting, neural network feedforward controller and PID controller collective effect are obtained with reference to control electric currentMake System steady operation；Neural network feedforward controller is after sufficiently learning, ierrorIt goes to zero,It goes to zero,Tend toAnd make TcTend to T* ref, torque deviation Δ T goes to zero, and PID controller maintains to stablize output, and SRM feedovers in neural network Dynamic balance state is under the control of controller and PID controller.
4. a kind of torque-current neural network switch reluctance motor control method according to any one of claim 1 to 3 A kind of torque-current neural network switched Reluctance Motor Control System of design, including signal processor, analog line driver, electricity Flow sensor, position sensor, torque sensor, display and switched reluctance machines SRM, on the output shaft of switched reluctance machines Torque sensor is installed, the current torque signal T of SRM is exportedc；Installation site sensor on magnetic resistance motor rotor, output are current Rotor position angle θ；The three-phase input end of SRM installs current sensor respectively, exports the current value of current each phase；Its feature exists In:
The signal processor contains the three neural network feedforward controllers matched with SRM tri-, is torque-current against mould Type neural network, there are also torque distribution module, pid control module, electric current hysteresis loop control modules；Torque distribution module will give Total torqueIt is assigned as each phase torque reference Tkk(θ) and rotor position angle θ together, input three neural network feedforwards respectively Controller exports the ANN Control electric current of each phaseIt is respectively fed to three current adders；
The current torque value Tc and given total torque that the torque sensor installed on SRM obtainsAccess torque subtracter, output Torque deviation Δ T access external circulation PID controller, outer ring PID controller obtain accordingly PID output electric current ierror, ierrorAccess Electric current distribution module, electric current distribution module obtain the PID control electric current of each phase according to current rotor angular position thetaIt accesses above-mentioned Three current adders；The i of PID output simultaneouslyerrorThe study of neural network as three neural network feedforward controllers Signal；
Three current adders realize each phase ANN Control electric current respectivelyWith PID control electric currentSuperposition, obtains three-phase Reference currentElectric current hysteresis loop control module is accessed, while the current sensor installed on SRM will also work as the electricity of three-phase Flow valuve is sent into interior circular current hysteresis loop controller as current feedback signal, and the output of interior circular current hysteresis loop controller is through power Driver control SRM operation.
5. a kind of torque-current neural network switched Reluctance Motor Control System according to claim 4, feature exist In:
The signal processor is connect with display, display control state and control result.
6. a kind of torque-current neural network switched Reluctance Motor Control System according to claim 4, feature exist In:
The signal processor configures CAN interface.
CN201710494319.8A 2017-06-26 2017-06-26 A kind of torque-current neural network switch reluctance motor control method and system Active CN107276465B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710494319.8A CN107276465B (en) 2017-06-26 2017-06-26 A kind of torque-current neural network switch reluctance motor control method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710494319.8A CN107276465B (en) 2017-06-26 2017-06-26 A kind of torque-current neural network switch reluctance motor control method and system

Publications (2)

Publication Number Publication Date
CN107276465A CN107276465A (en) 2017-10-20
CN107276465B true CN107276465B (en) 2019-06-28

Family

ID=60068896

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710494319.8A Active CN107276465B (en) 2017-06-26 2017-06-26 A kind of torque-current neural network switch reluctance motor control method and system

Country Status (1)

Country Link
CN (1) CN107276465B (en)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107659238A (en) * 2017-11-13 2018-02-02 中国科学院宁波材料技术与工程研究所 A kind of Direct Torque Control device of switched reluctance machines
JP6748135B2 (en) 2018-03-19 2020-08-26 ファナック株式会社 Machine learning device, servo control device, servo control system, and machine learning method
CN109435705B (en) * 2018-10-29 2020-09-11 北京新能源汽车股份有限公司 Electric automobile and damping control method, device, equipment and medium thereof
CN109742999B (en) * 2019-01-17 2020-07-10 桂林电子科技大学 Dynamic neural network adaptive inverse SRM torque control method and system
CN109802612B (en) * 2019-03-28 2021-02-26 重庆长安新能源汽车科技有限公司 Vehicle and driving motor control system and method thereof
CN110022109B (en) * 2019-04-17 2021-01-05 桂林电子科技大学 Method and system for controlling torque ripple of SRM (sequence-derived minimum-mean-square) of torque-current neural network model
CN110647078B (en) * 2019-09-26 2020-10-16 西安科技大学 Underground unattended drainage system for coal mine and control method
CN112117947B (en) * 2020-09-30 2022-03-11 桂林电子科技大学 SRM torque ripple suppression control system and method based on current injection method
CN112886893B (en) * 2021-01-11 2022-11-04 山东省科学院自动化研究所 Switched reluctance motor torque control method and system based on turn-off angle optimization
CN112865657A (en) * 2021-01-11 2021-05-28 山东中环联科机电科技有限公司 Control method of disk type switched reluctance motor
CN112928965B (en) * 2021-03-29 2022-09-13 桂林电子科技大学 Flux linkage based torque ripple suppression control system and method for switched reluctance motor
CN113765449B (en) * 2021-07-29 2023-07-07 桂林电子科技大学 SRM flux linkage control system and method based on double Sigmod neural network inductance model
CN114397808A (en) * 2021-12-09 2022-04-26 北京航空航天大学 High-precision control system and method for proportional valve of breathing machine

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8618761B2 (en) * 2004-10-29 2013-12-31 Virginia Tech Intellectual Properties, Inc. Method for controlling motor operation using information characterizing regions of motor operation
CN106357186A (en) * 2016-09-05 2017-01-25 桂林电子科技大学 Method and system for controlling constant torque of switched reluctance motor by use of composite control current
CN106357192A (en) * 2016-09-05 2017-01-25 桂林电子科技大学 Method and system for lowering torque pulsation of switched reluctance motor by current adaptive control

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015104240A (en) * 2013-11-26 2015-06-04 株式会社デンソー Rotary electric machine driving device, and electric power steering device using the same

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8618761B2 (en) * 2004-10-29 2013-12-31 Virginia Tech Intellectual Properties, Inc. Method for controlling motor operation using information characterizing regions of motor operation
CN106357186A (en) * 2016-09-05 2017-01-25 桂林电子科技大学 Method and system for controlling constant torque of switched reluctance motor by use of composite control current
CN106357192A (en) * 2016-09-05 2017-01-25 桂林电子科技大学 Method and system for lowering torque pulsation of switched reluctance motor by current adaptive control

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Sensorless Torque and Thrust Estimation of a Rotational/Linear Two Degrees-of-Freedom Switched Reluctance Motor;Yasukazu Sato, Keita Murakami, and Yugo Tsuboi;《IEEE TRANSACTIONS ON MAGNETICS》;20160731;第52卷(第7期);全文
基于柔性神经网络的SRM直接瞬时转矩控制;党选举，袁小唐，胡景佳;《电气传动》;20140531;第44卷(第5期);全文
开关磁阻电机转矩脉动抑制方法研究与实现;杨茂骑;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20170315(第03期);全文

Also Published As

Publication number Publication date
CN107276465A (en) 2017-10-20

Similar Documents

Publication Publication Date Title
CN107276465B (en) A kind of torque-current neural network switch reluctance motor control method and system
CN110022109B (en) Method and system for controlling torque ripple of SRM (sequence-derived minimum-mean-square) of torque-current neural network model
CN104533716B (en) A kind of independent pitch load controlled method based on Kalman filtering
CN103051274A (en) Variable damping-based passive control method for two-degree-of-freedom permanent magnetic synchronous motor
CN109787531A (en) A kind of switching magnetic-resistance hub motor forecast Control Algorithm
CN110429895A (en) The building method of hybrid vehicle switching magnetic-resistance BSG optimization linear controller
CN105720879B (en) A kind of novel doubly fed induction generator control method based on direct character control
CN111293952B (en) Switched reluctance motor torque control system and method based on inductance rate model
CN107294453B (en) Magnetic linkage combines the method and system for inhibiting switched reluctance machines torque pulsation with electric current
CN103647493B (en) A kind of infinite method for estimating rotating speed of H of permagnetic synchronous motor
CN104167968B (en) A kind of Vector Control System of Induction Motor method
CN205051611U (en) Motor drive system that restraines switched reluctance motor torque ripple
CN110247586B (en) Efficiency optimization-based permanent magnet synchronous motor torque distribution method for electric bus
CN112886893A (en) Switched reluctance motor torque control method and system based on turn-off angle optimization
CN107317477A (en) A kind of control method, control system and the control device of DC/DC converters
CN105871278A (en) Direct feature control-based novel salient-pole permanent magnet synchronous motor control method
CN102522948A (en) Hybrid intelligent adjusting method of torque hysteresis width in DTC (Direct Torque Control) system
CN104935231A (en) Current control method of induction motor based on forecast mode, and current controller of induction motor
CN105099263B (en) Supersonic motor Position And Velocity control system based on generalized regression nerve networks identifier
CN103337997B (en) A kind of Over Electric Motor with PMSM vector control system and method
Song et al. Review on control of permanent magnet brushless DC motor for electric vehicle
CN109861618A (en) The building method of the anti-interference composite controller of Hybrid Vehicle BSG alternating current generator
Xi et al. Sliding mode control of an outer-rotor magnetic-gear-integrated motor with a Halbach array
Jin-Feng et al. Control of switched reluctance motors based on improved BP neural networks
Cao et al. Neural network self-adaptive PID control for driving and regenerative braking of electric vehicle

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant