CN106357186A - Method and system for controlling constant torque of switched reluctance motor by use of composite control current - Google Patents

Abstract

The invention relates to a method and a system for controlling constant torque of a switched reluctance motor by use of composite control current. According to the method, linear control current of each phase is obtained on the basis of linear inductance model torque distribution function control, and an echo state network is adopted and outputs non-linear current according to current total torque of the switched reluctance motor, given total torque and output non-linear current feedback as well as switched reluctance motor Jacobian information calculated by an RBF (radial basis function) neural network through parameter learning; non-linear control current of each phase and the linear control current are superposed, and the composite control current is obtained and taken as a set value for a current hysteresis loop controller. Current, torque and positon sensors of the system are connected with a signal processor, the signal processor executes the module of the method and outputs the composite control current, a power converter of the motor is controlled through the current hysteresis loop controller, and torque ripple of the switched reluctance motor is remarkably and effectively inhibited.

Description

A kind of switched reluctance machines permanent torque control method of complex controll electric current and system

Technical field

The present invention relates to switched Reluctance Motor Control technical field, the switching magnetic-resistance electricity of specially a kind of complex controll electric current Machine permanent torque control method and system.

Background technology

Switched reluctance machines (switched reluctance motor, abbreviation srm) are firm reliable, without rare earth material And speed governing is flexibly, thus become the main flow of new energy electric automobile motor.But, switched reluctance machines are special double-salient-pole knot Structure, takes the power supply mode of switching regulator, and magnetic circuit is in strong nonlinearity and saturability, and therefore switched reluctance machines have torque when operating The defects such as pulsation, noise, vibration, especially in low regime, torque ripple is more strong, hinders its new forms of energy in higher performance More common application in electrocar drive system.The method of therefore research suppression switched reluctance machines torque pulsation has extremely Important meaning.

Chinese scholars, from improving motor control strategy, the torque pulsation of research suppression switched reluctance machines, take Obtained many achievements in research.Traditional control method is difficult to meet strong nonlinearity and the switched Reluctance Motor Control of saturability will Ask it is impossible to obtain preferable effect.Intelligent control method need not obtain the accurate mathematical model of controlled device, in nonlinear system Control in there is superiority.Especially neural network control method, has the adaptive ability of on-line study, can infinitely force Nearly Any Nonlinear Function, is very suitable for solving the control problem of nonlinear system.See that relevant ANN Control is opened Close the report of reluctance motor, such as fuzzy neural network is combined with Strategy of Direct Torque Control, devises mixture control, by turning Square and magnetic linkage error signal Training Fuzzy Neural Networks, realize minimum torque ripple control.Each also by neutral net conversion The torque component of phase partitioning obtains preferable control electric current.But using classical bp (back in these ANN Control Propagation) neural network algorithm, convergence rate is slow.In addition existing neural network control method, switched reluctance machines Linear processes characteristic be considered as entirety, make control design case complicated.But it is based on linear inductance model tsf (torque Distribution function) control the Linear Control electric current obtaining to be difficult in adapt to the nonlinear characteristic of reluctance motor, cause Larger torque pulsation.

Content of the invention

The purpose of the present invention is to design a kind of switched reluctance machines permanent torque control method of complex controll electric current, this method Controlled on the Linear Control electric current obtaining based on linear inductance model torque partition function (tsf), devising based on echo shape The non-linear current of state network (echo state network, esn) controls, the linear control that its output is exported with tsf controller Electric current superposition processed obtains complex controll electric current, in two current control collective effects, improves dynamic performance, effectively suppression turns Square is pulsed.

It is another object of the present invention to the switched reluctance machines permanent torque controlling party according to a kind of above-mentioned complex controll electric current Method designs a kind of switched reluctance machines Constant Torque Holding System of complex controll electric current.

A kind of switched reluctance machines permanent torque control method of complex controll electric current of present invention design, key step is such as Under:

, the Linear Control electric current Ji Yu linear inductance model torque partition function (tsf)

Switched reluctance machines adopt ideal linearity each phase inductance l_kzzWhen, torque-current transformed representation is:

$i_{kzz}=\sqrt{\frac{2t_{kzz}}{{\mathrm{dl}}_{kzz}\left(\mathrm{\θ}\right)/d\mathrm{\θ}}}---\left(1\right)$

Wherein, zz=1, the three-phase of 2,3 expression motors, i_kzzFor the current value of each phase, l_kzzFor each phase inductance value, t_kzzFor The torque of each phase partitioning, θ is motor rotor position angle.

Make k_lzz=dl_kzz(θ)/d θ, k_lzzRepresent the rate of change of phase inductance, for the linear inductance of trapezoidal curvilinear motion Model, k_lzzFor constant, span is 0.1～0.3.

The torque partition function that the present invention adopts is cube partition function, and this is current best torque partition function, its table Reaching formula is:

$f\left(\mathrm{\θ}\right)=\frac{3}{{\mathrm{\θ}}_{ov}^2}{(\mathrm{\θ}-{\mathrm{\θ}}_{on})}^2-\frac{2}{{\mathrm{\θ}}_{ov}^3}{(\mathrm{\θ}-{\mathrm{\θ}}_{on})}^3---\left(2\right)$

Wherein θ_onFor turn-on angle, 10～15 degree of span, θ_offFor switch off angle, 25～30 degree of span, θ_ovFor turning Sub- position angle, 2～5 degree of span.

Given total torque isDistribute to each phase torque reference t_kzzIn adjacent biphase be:

$t_{up}=t_{ref}^{*}f\left(\mathrm{\θ}\right)---\left(3\right)$

$t_{dn}=t_{ref}^{*}(1-f(\mathrm{\θ}\left)\right)---\left(4\right)$

T in formula_up、t_dnRepresent respectively adjacent biphase change from small to large open and shutoff phase from big to small.

The each phase torque reference being obtained according to torque partition function, using the torque-current conversion formula of formula (1), you can Quickly it is calculated each accordingly phase linear control electric current i_kzz.

, based on echo state network non-linear current control

The parameter learning of -1 echo state network

The present invention adopts novel artificial neutral net echo state network esn.The generation of dynamic inductance is main and electric Stream is related, and the input choosing echo state network is:

Wherein i_error(k-1) it is echo state network current The real-time output nonlinear electric current of the previous moment in moment, i.e. i_error(k-1)=z^-1i_error(k), z^-1Represent that in time domain, variable exists Conversion during one sampling period of signal lag.Sampling period is 0.1～2 second.T (k) is switched reluctance machines current time Instantaneous total torque.It is the given total torque of switched reluctance machines current time.The present invention passes through the non-linear current of output Feedback, strengthen the dynamic memory ability of echo state network and stability.

The state representation of echo state network is:

Q (k+1)=f (w_inu(k)+wq(k)+w_backi_error(k)) (5)

i_error(k+1)=w_outs(k) (6)

Wherein, q (k)=[q₁(k),q₂(k),…,q_n(k)]^t∈rⁿFor internal state variable；S (k)=[u (k), q^t(k), i_error(k)]^t∈r^p+n+m, []^tThe transposition computing of representing matrix.w_in∈r^n×pRepresent input connection weight matrix, w ∈ r^n×nRepresent Internal connection weight matrix, w_back∈r^n×mRepresent feedback link weight matrix, echo state network initial phase with dimension n=72, Degree of rarefication 1～5%, connection weight spectral radius are 0.1～1, produce internal connection weight w.Need the only reserve pool of training to being Connection weight w of system output_out∈r^p+n+m.F=[f₁,f₂,…,f_n] represent neuron activation functions, f_r(r=1,2 ... n) adopt Hyperbolic tangent function, output layer adopts linear function.Reserve pool dimension d=30～100, internal state variable n=72, p is back The input vector u dimension of sound state network, number m of the output variable of p=3, esn, m=1.

The reserve pool of echo state network is the core of information processing, and its scale impact network approaches energy to nonlinear Power.If its performance index function is:

$e\left(k\right)=\frac{1}{2}{\[t_{ref}\left(k\right)-t\left(k\right)\]}^2=\frac{1}{2}e{\left(k\right)}^2---\left(7\right)$

In formula, t_refK () is the current k moment to give torque；T (k) is current k moment switched reluctance machines reality output Instantaneous total torque.

According to gradient descent method, the output weighed value adjusting algorithm of echo state network is:

$\left\{\begin{array}{c}{w}_{lj}\left(k\right)=w_{lj}(k-1)+{\mathrm{\δw}}_{lj}\left(k\right)\\ {\mathrm{\δw}}_{lj}\left(k\right)=-\mathrm{\η}e\left(k\right)\frac{\∂t\left(k\right)}{\∂ii\left(k\right)}{q}_n\left(k\right)\end{array}\right.---\left(8\right)$

W in formula_ljFor output layer weighting parameter, η is parameter adjustment learning rate scope 0～1.Typically take 0.6；L=1,2 ..., 72, j=1.(8) in formula, ii (k) is the complex controll electric current of three-phaseSum, q_nK () is the State- output of echo state network,Jacobian information for switched reluctance machines.- 2 switched reluctance machines based on rbf neutral net Jacobian information calculates

This invention adopt simple for structure, rbf (the radial basis function) neutral net of fast convergence rate is to formula (8) inCarry out on-line identification；The excitation function of output layer neuron is taken as linear function, and hidden layer is neural The excitation function of unit is taken as Gaussian bases.

Choose x=[ii (k), ii (k-1), t (k-1), t (k-2)]^tInput vector for rbf, wherein；Ii (k) and ii (k-1) total current of current time and previous moment, ii (k-1)=z are represented respectively^-1Ii (k), t (k) and t (k-1) represent respectively Switched reluctance machines are in the instantaneous total torque of current time and previous moment, wherein t (k-1)=z-¹T (k), t (k-2)=z-¹t (k-1).H=[h₁,h₂,…,h_j…h_m]^tRadial direction base vector for rbf, wherein h_jFor Gaussian bases:

$h_j=\exp \left(\frac{\left|\right|x-c_j|{|}^2}{2b_j^2}\right),j=1,2,...m---\left(9\right)$

In formula, c_jFor the center vector of j-th node, c_j=[c_j1,c_j2,…c_ji,…c_jn]^t, wherein b_jBase for node Wide parameter, and be the number more than zero, c_jAnd b_jIt is all to be obtained by study, node m scope 5～20, n=4 is input variable Number.

Jacobian information is:

$\frac{\∂t\left(k\right)}{\∂ii\left(k\right)}\≈\frac{\∂y_m\left(k\right)}{\∂ii\left(k\right)}=\underset{j=1}{\overset{m}{\σ}}{w}_j{h}_j\frac{c_{js}-x_1}{b_j^2}---\left(10\right)$

In formula, x₁=ii (k), s=1,2 ..., n；

According to gradient descent method, output power w_j, node center c_jsAnd node sound stage width parameter b_jIteration as follows:

$\begin{array}{c}{w}_j\left(k\right)=w_j\left(k-1\right)+\mathrm{\η}\left(t\left(k\right)-y_m\left(k\right)\right)h_j\\ +\mathrm{\α}\left(w_j\left(k-1\right)-w_j\left(k-2\right)\right)\end{array}---\left(11\right)$

$\begin{array}{c}{b}_j\left(k\right)=b_j(k-1)+{\mathrm{\η\δb}}_j+\mathrm{\α}\left(b_j\right(k-1)\\ -b_j\left(k-2\right))\end{array}---\left(12\right)$

Wherein

${\mathrm{\δb}}_j=\left(t\right(k)-y_m(k\left)\right)w_j{h}_j\frac{\left|\right|x-c_j|{|}^2}{b_j^3}$

$\begin{array}{c}{c}_{js}\left(k\right)=c_{js}(k-1)+{\mathrm{\η\δc}}_{js}+\mathrm{\α}\left(c_{js}\right(k-1)\\ -c_{js}\left(k-2\right))\end{array}---\left(13\right)$

Wherein

${\mathrm{\δc}}_{js}=\left(t\right(k)-y_m(k\left)\right)w_j{h}_j\frac{x_j-c_{js}}{b_j^2}$

In formula, y_mK () is current k moment rbf neutral net output, η is learning rate, and α is factor of momentum, its scope 0 ～1, often take 0.3；Node m span 5～20, n=4 is input variable number.

- 3 nonlinear Control electric currents

Esn real-time output nonlinear electric current i_error, obtain each phase nonlinear Control electric current i through electric current distribution_kzzBy riser portions Point, sloping portion and constant value segments i_errorComposition, adjacent in same time period biphase is respectively at change from small to large open-minded Phase and shutoff phase from big to small.Taking adjacent a phase, b phase as a example, non-linear current is assigned as:

In formulaRepresent respectively and be in a phase opening phase and the nonlinear Control electric current of the b phase being off phase.

, complex controll electric current switched reluctance machines permanent torque control

Step obtains each phase linear control electric current i by each phase torque distribution_kzz, by rising part, sloping portion and constant Part t_kzzComposition.In same time period adjacent biphase be respectively at change from small to large open mutually and shutoff from big to small Phase.Taking adjacent a phase, b phase as a example, linear current is assigned as:

In formulaRepresent a phase and b phase；

In formulaRepresent and be in a phase opening phase and the Linear Control electric current of the b phase being off phase.

Step gained each phase nonlinear Control electric current i_kzzLinear Control electric current i with step gained_kzzSuperposition, obtains Each phase complex controll electric currentAs the setting value of the current hysteresis-band control device of technology maturation, current hysteresis-band control device defeated Go out, through power inverter driving switch reluctance motor, carry out permanent torque control.

The present invention is a kind of multiple according to a kind of switched reluctance machines permanent torque design of control method of above-mentioned complex controll electric current Close the switched reluctance machines Constant Torque Holding System of control electric current, including signal processor, analog-to-digital conversion module, Hysteresis Current control Device processed, power inverter, current sensor, torque sensor and rotor-position sensor.

Three current sensors are respectively arranged in the three-phase power line of switched reluctance machines, detect each phase current, torque passes Sensor is installed on the output shaft of switched reluctance machines, the output torque of detection motor, and rotor-position sensor is installed on switch magnetic The rotor of resistance motor, detects rotor position angle.Position sensor, torque sensor and current sensor are through analog-to-digital conversion module even Connect signal processor (as dsp chip).

Signal processor contains linear current control module, rbf neural network module and the non-linear electricity of echo state network Flow control module.

Described linear current controller includes torque divider and torque-current transducer, and torque divider includes torque Partition function, the reference that given total torque is assigned as each phase is turned by the current rotor position angle according to rotor-position sensor detection Square, torque-current transducer obtains according to currently each phase torque reference according to the torque-current partition function of linear inductance model Each phase linear control electric current i_kzz.

The total current of described rbf neural network module foundation current time and previous moment and instantaneous total torque, reckoning is opened Close the jacobian information of reluctance motor, be supplied to echo state network module.

Described echo state network non-linear current controller includes echo state network module and electric current distribute module, returns Sound state network module according to the instantaneous total torque of switched reluctance machines current time, switched reluctance machines current time given The feedback of the non-linear current of total torque and output, and the jacobian information of switched reluctance machines, through parameter learning, export Non-linear current i_error.i_errorObtain each phase nonlinear Control electric current i through electric current distribute module_kzz.

Each phase linear control electric current i_kzzWith each phase nonlinear Control electric current i_kzzSuperposition, each phase complex controll electric current obtainingAs its setting value, the output access power changer of current hysteresis-band control device, power becomes input current hystersis controller Parallel operation is output as driving switch reluctance motor three-phase drive power supply.

Signal processor is connected with display screen, the current torque of Real time displaying and current information.

Signal processor is furnished with can interface, can be connected with the can controller of automobile and other electrical communications of automobile.

Rotor-position sensor, three current sensors and torque sensor detection obtain current switch magnetic resistance motor rotor The analogue signal of position angle, the electric current of power supply three-phase and instantaneous total torque is converted to corresponding numeral letter through analog-to-digital conversion module Number, send into signal processor, the linear current control module in signal processor, the non-linear current of echo state network controls Module, according to the current instantaneous total torque of actual measurement and three-phase electricity flow valuve, tries to achieve complex controll electric current, controlling switch reluctance motor, Effectively suppress its torque pulsation, realize permanent torque control.

Compared with prior art, the switched reluctance machines Constant Torque Holding System method of a kind of complex controll of present invention electric current With system advantage it is: 1, fully take into account the nonlinear characteristic of switched reluctance machines, using improved neutral net echo State network, obtains each phase nonlinear Control electric current, it is superimposed with Linear Control electric current, tries to achieve complex controll electric current, effectively The torque pulsation of suppression switched reluctance machines；2nd, compared with the conventional linear current control method of switched reluctance machines, this is compound Torque pulsation rate is reduced to 1.75% by 57.2% by the switched Reluctance Motor Control of control electric current.

Brief description

Fig. 1 be this complex controll electric current the step of switched reluctance machines permanent torque control method embodiment described in linear The linear current control system block diagram of inductor models torque partition function；

Fig. 2 is returning described in the step of switched reluctance machines permanent torque control method embodiment of this complex controll electric current Sound state network structural representation；

Fig. 3 is answering described in the step of switched reluctance machines permanent torque control method embodiment of this complex controll electric current Close control electric current and produce schematic flow sheet；

Fig. 4 is the structural representation of the switched reluctance machines Constant Torque Holding System embodiment of complex controll electric current；

Fig. 5 is the operational flow diagram of the switched reluctance machines Constant Torque Holding System embodiment of complex controll electric current.

Specific embodiment

The switched reluctance machines permanent torque control method embodiment of complex controll electric current

The switched reluctance machines permanent torque control method embodiment of this complex controll electric current, key step is as follows:

, the Linear Control electric current Ji Yu linear inductance model torque partition function (tsf)

As shown in figure 1, switched reluctance machines linear inductance model control includes torque distribution, torque-current change and Three major parts of current hysteresis-band control device.

Switched reluctance machines adopt ideal linearity each phase inductance l_kzzWhen, torque-current transformed representation is:

$i_{kzz}=\sqrt{\frac{2t_{kzz}}{{\mathrm{dl}}_{kzz}\left(\mathrm{\θ}\right)/d\mathrm{\θ}}}---\left(1\right)$

Wherein, zz=1, the three-phase of 2,3 expression motors, i_kzzFor the current value of each phase, l_kzzFor each phase inductance value, t_kzzFor The torque of each phase partitioning, θ is motor rotor position angle.

Make k_lzz=dl_kzz(θ)/d θ, k_lzzRepresent the rate of change of phase inductance, for the linear inductance of trapezoidal curvilinear motion Model, k_lzzFor constant, this example k value is 0.2.

The torque partition function of this example is cube partition function, and its expression formula is:

$f\left(\mathrm{\θ}\right)=\frac{3}{{\mathrm{\θ}}_{ov}^2}{(\mathrm{\θ}-{\mathrm{\θ}}_{on})}^2-\frac{2}{{\mathrm{\θ}}_{ov}^3}{(\mathrm{\θ}-{\mathrm{\θ}}_{on})}^3---\left(2\right)$

Wherein θ_onFor turn-on angle, 11.5 degree of this example value, θ_offFor switch off angle, 30 degree of this example value, θ_ovFor rotor-position Angle, 3.5 degree of this example value.

Given total torque isDistribute to each phase torque reference t_kzzIn adjacent biphase be:

$t_{up}=t_{ref}^{*}f\left(\mathrm{\θ}\right)---\left(3\right)$

$t_{dn}=t_{ref}^{*}(1-f(\mathrm{\θ}\left)\right)---\left(4\right)$

T in formula_up、t_dnRepresent respectively adjacent biphase change from small to large open and shutoff phase from big to small.

The each phase torque reference being obtained according to torque partition function, using the torque-current conversion formula of formula (1), you can Quickly it is calculated each accordingly phase linear control electric current i_kzz.

, based on echo state network non-linear current control

The parameter learning of -1 echo state network

This example echo state network esn is as shown in Fig. 2 the input of echo state network is:

$u\left(k\right)={\[t\left(k\right),t_{ref}^{*}\left(k\right),i_{error}(k-1)\]}^t,$

Wherein i_error(k-1) for the real-time output nonlinear electric current of the previous moment in current time for the echo state network, I.e. i_error(k-1)=z^-1i_error(k), z^-1Represent the conversion when one sampling period of signal lag for the variable in time domain.This example is adopted The sample cycle takes 0.2 second.T (k) is the instantaneous total torque of switched reluctance machines current time.It is that switched reluctance machines are current The given total torque in moment.

The state representation of echo state network is:

Q (k+1)=f (w_inu(k)+wq(k)+w_backi_error(k)) (5)

i_error(k+1)=w_outs(k) (6)

Wherein, q (k)=[q₁(k),q₂(k),…,q_n(k)]^t∈rⁿFor internal state variable；S (k)=[u (k), q^t(k), i_error(k)]^t∈r^p+n+m, []^tThe transposition computing of representing matrix.w_in∈r^n×pRepresent input connection weight matrix, w ∈ r^n×nRepresent Internal connection weight matrix, w_back∈r^n×mExpression feedback link weight matrix, this example echo state network initial phase dimension n= 72, degree of rarefication is 2%, and connection weight spectral radius is 0.5, produces internal connection weight w.The only reserve pool needing training is to system Connection weight w of output_out∈r^p+n+m.F=[f₁,f₂,…,f_n] represent neuron activation functions, f_r(r=1,2 ... n) using double Bent tan, output layer adopts linear function.This example reserve pool dimension d=60, internal state variable n=72, p is echo shape The input vector u dimension of state network, number m of the output variable of p=3, esn, m=1.

The reserve pool of echo state network is that the performance index function of the core of information processing is:

$e\left(k\right)=\frac{1}{2}{\[t_{ref}\left(k\right)-t\left(k\right)\]}^2=\frac{1}{2}e{\left(k\right)}^2---\left(7\right)$

In formula, t_refK () is the current k moment to give torque；T (k) is current k moment switched reluctance machines reality output Instantaneous total torque.

According to gradient descent method, the output weighed value adjusting algorithm of echo state network is:

$\left\{\begin{array}{c}{w}_{lj}\left(k\right)=w_{lj}(k-1)+{\mathrm{\δw}}_{lj}\left(k\right)\\ {\mathrm{\δw}}_{lj}\left(k\right)=-\mathrm{\η}e\left(k\right)\frac{\∂t\left(k\right)}{\∂ii\left(k\right)}{q}_n\left(k\right)\end{array}\right.---\left(8\right)$

W in formula_ljFor output layer weighting parameter, η is parameter adjustment learning rate scope 0～1.Typically take 0.6；L=1,2 ..., 72, j=1.(8) in formula, ii (k) is the complex controll electric current of three-phaseSum, q_nK () is that the state of echo state network is defeated Go out,Jacobian information for switched reluctance machines.- 2 switched reluctance machines based on rbf neutral net Jacobian information calculates

This example adopts rbf neutral net in formula (8)Carry out on-line identification；Output layer neuron swash Number of sending a letter is taken as linear function, and the excitation function of hidden layer neuron is taken as Gaussian bases.

Choose x=[ii (k), ii (k-1), t (k-1), t (k-2)]^tInput vector for rbf, wherein；Ii (k) and ii (k-1) total current of current time and previous moment, ii (k-1)=z are represented respectively^-1Ii (k), t (k) and t (k-1) represent respectively Switched reluctance machines are in the instantaneous total torque of current time and previous moment, wherein t (k-1)=z^-1T (k), t (k-2)=z^-1t (k-1).H=[h₁,h₂,…,h_j…h_m]^tRadial direction base vector for rbf, wherein h_jFor Gaussian bases:

$h_j=\exp \left(\frac{\left|\right|x-c_j|{|}^2}{2b_j^2}\right),j=1,2,...m---\left(9\right)$

In formula, c_jFor the center vector of j-th node, c_j=[c_j1,c_j2,…c_ji,…c_jn]^t, wherein b_jBase for node Wide parameter, and be the number more than zero, c_jAnd b_jIt is all to be obtained by study, this example node m=8, n=4 are input variable number.

Jacobian information is:

$\frac{\∂t\left(k\right)}{\∂ii\left(k\right)}\≈\frac{\∂y_m\left(k\right)}{\∂ii\left(k\right)}=\underset{j=1}{\overset{m}{\σ}}{w}_j{h}_j\frac{c_{js}-x_1}{b_j^2}---\left(10\right)$

In formula, x₁=ii (k), s=1,2 ..., n；

According to gradient descent method, output power w_j, node center c_jsAnd node sound stage width parameter b_jIteration as follows:

$\begin{array}{c}{w}_j\left(k\right)=w_j\left(k-1\right)+\mathrm{\η}\left(t\left(k\right)-y_m\left(k\right)\right)h_j\\ +\mathrm{\α}\left(w_j\left(k-1\right)-w_j\left(k-2\right)\right)\end{array}---\left(11\right)$

$\begin{array}{c}{b}_j\left(k\right)=b_j(k-1)+{\mathrm{\η\δb}}_j+\mathrm{\α}\left(b_j\right(k-1)\\ -b_j\left(k-2\right))\end{array}---\left(12\right)$

Wherein

${\mathrm{\δb}}_j=\left(t\right(k)-y_m(k\left)\right)w_j{h}_j\frac{\left|\right|x-c_j|{|}^2}{b_j^3}$

$\begin{array}{c}{c}_{js}\left(k\right)=c_{js}(k-1)+{\mathrm{\η\δc}}_{js}+\mathrm{\α}\left(c_{js}\right(k-1)\\ -c_{js}\left(k-2\right))\end{array}---\left(13\right)$

Wherein

${\mathrm{\δc}}_{js}=\left(t\right(k)-y_m(k\left)\right)w_j{h}_j\frac{x_j-c_{js}}{b_j^2}$

In formula, y_mK () is current k moment rbf neutral net output, η is learning rate, and α is factor of momentum, and this example takes 0.3.

- 3 nonlinear Control electric currents

Esn real-time output nonlinear electric current i_error, obtain each phase nonlinear Control electric current i through electric current distribution_kzzBy riser portions Point, sloping portion and constant value segments i_errorComposition, adjacent in same time period biphase is respectively at change from small to large open-minded Phase and shutoff phase from big to small.Taking adjacent a phase, b phase as a example, non-linear current is assigned as:

In formulaRepresent respectively and be in a phase opening phase and the nonlinear Control electric current of the b phase being off phase.

, complex controll electric current switched reluctance machines permanent torque control

Step obtains each phase linear control electric current i by each phase torque distribution_kzz, by rising part, sloping portion and constant Part t_kzzComposition.In same time period adjacent biphase be respectively at change from small to large open mutually and shutoff from big to small Phase.Taking adjacent a phase, b phase as a example, linear current is assigned as:

In formulaRepresent a phase and b phase；

In formulaRepresent respectively and be in a phase opening phase and the Linear Control electric current of the b phase being off phase.

As shown in figure 3, step gained each phase nonlinear Control electric current i_kzzLinear Control electric current i with step gained_kzz Superposition, each phase complex controll electric current obtainingAs the setting value of current hysteresis-band control device, current hysteresis-band control device defeated Go out, through power inverter driving switch reluctance motor, carry out permanent torque control.

The switched reluctance machines Constant Torque Holding System embodiment of complex controll electric current

The structure of the switched reluctance machines Constant Torque Holding System embodiment of this complex controll electric current is as shown in figure 4, include Signal processor, analog-to-digital conversion module, current hysteresis-band control device, power inverter, current sensor, torque sensor and rotor Position sensor.

Three current sensors are respectively arranged in the three-phase power line of switched reluctance machines, detect each phase current, torque passes Sensor is installed on the output shaft of switched reluctance machines, the output torque of detection motor, and rotor-position sensor is installed on switch magnetic The rotor of resistance motor, detects rotor position angle.Position sensor, torque sensor and current sensor are through analog-to-digital conversion module even Connect signal processor, this example signal processor is dsp chip.

Signal processor contains linear current control module, rbf neural network module and the non-linear electricity of echo state network Flow control module.

Described linear current controller includes torque divider and torque-current transducer, and torque divider includes torque Partition function, the reference that given total torque is assigned as each phase is turned by the current rotor position angle according to rotor-position sensor detection Square, torque-current transducer obtains according to currently each phase torque reference according to the torque-current partition function of linear inductance model Each phase linear control electric current i_kzz.

The total current of described rbf neural network module foundation current time and previous moment and instantaneous total torque, reckoning is opened Close the jacobian information of reluctance motor, be supplied to echo state network module.

Described echo state network non-linear current controller includes echo state network module and electric current distribute module, returns Sound state network module according to the instantaneous total torque of switched reluctance machines current time, switched reluctance machines current time given The feedback of the non-linear current of total torque and output, and the jacobian information of switched reluctance machines, through parameter learning, export Non-linear current i_error.i_errorObtain each phase nonlinear Control electric current i through electric current distribute module_kzz.

Each phase linear control electric current i_kzzWith each phase nonlinear Control electric current i_kzzSuperposition, each phase complex controll electric current obtainingAs its setting value, the output access power changer of current hysteresis-band control device, power becomes input current hystersis controller Parallel operation is output as driving switch reluctance motor three-phase drive power supply.

This example signal processor is connected with display screen, the current torque of Real time displaying and current information.

This example signal processor is furnished with can interface, to be connected with the can controller of automobile and other electrical communications of automobile.

The flow process that the system is run is as shown in figure 5, rotor-position sensor, three current sensors and torque sensor are examined The analogue signal recording current switch magnetic resistance motor rotor position angle, the electric current of power supply three-phase and instantaneous total torque is through modulus Modular converter is converted to corresponding digital signal, sends into signal processor, the linear current control module in signal processor, returns The non-linear current control module of sound state network, according to the current instantaneous total torque of actual measurement and three-phase electricity flow valuve, tries to achieve respectively Each phase linear control electric current and nonlinear Control electric current, superposition obtains each phase complex controll electric current, as current hysteresis-band control device Setting value, the output access power changer of current hysteresis-band control device, power inverter be output as driving switch magnetic resistance electricity Machine three-phase drive power supply, controlling switch reluctance motor, effectively suppresses its torque pulsation, realizes permanent torque control.

Above-described embodiment, only the purpose of the present invention, technical scheme and beneficial effect are further described is concrete Individual example, the present invention is not limited to this.All any modifications made within the scope of disclosure of the invention, equivalent, change Enter, be all contained within protection scope of the present invention.

Claims (4)

1. the switched reluctance machines permanent torque control method of a kind of complex controll electric current, key step is as follows:

, the Linear Control electric current based on linear inductance model torque partition function

Switched reluctance machines adopt ideal linearity each phase inductance l_kzzWhen, torque-current transformed representation is:

$i_{kzz}=\sqrt{\frac{2t_{kzz}}{{\mathrm{dl}}_{kzz}\left(\mathrm{\θ}\right)/d\mathrm{\θ}}}---\left(1\right)$

Wherein, zz=1, the three-phase of 2,3 expression motors, i_kzzFor the current value of each phase, l_kzzFor each phase inductance value, t_kzzFor each phase The torque of distribution, θ is motor rotor position angle；

Make k_lzz=dl_kzz(θ)/d θ, k_lzzRepresent the rate of change of phase inductance, for the linear inductance model of trapezoidal curvilinear motion, k_lzzFor constant, span is 0.1～0.3；

Torque partition function is cube partition function, and its expression formula is:

$f\left(\mathrm{\θ}\right)=\frac{3}{{\mathrm{\θ}}_{ov}^2}{(\mathrm{\θ}-{\mathrm{\θ}}_{on})}^2-\frac{2}{{\mathrm{\θ}}_{ov}^3}{(\mathrm{\θ}-{\mathrm{\θ}}_{on})}^3---\left(2\right)$

Wherein θ_onFor turn-on angle, 10～15 degree of span, θ_offFor switch off angle, 25～30 degree of span, θ_ovFor rotor-position Angle, 2～5 degree of span；

Given total torque isDistribute to each phase torque reference t_kzzIn adjacent biphase be:

$t_{up}=t_{ref}^{*}f\left(\mathrm{\θ}\right)---\left(3\right)$

$t_{dn}=t_{ref}^{*}(1-f(\mathrm{\θ}\left)\right)---\left(4\right)$

T in formula_up、t_dnRepresent respectively adjacent biphase change from small to large open and shutoff phase from big to small；

The each phase torque reference being obtained according to torque partition function, with the torque-current conversion formula of formula (1), is calculated phase Each phase linear control electric current i answered_kzz；

, based on echo state network non-linear current control

The parameter learning of -1 echo state network

Using echo state network, it inputs and is:

$u\left(k\right)={\[t\left(k\right),t_{ref}^{*}\left(k\right),i_{error}(k-1)\]}^t,$

Wherein i_error(k-1) for the real-time output nonlinear electric current of the previous moment in current time for the echo state network, that is, i_error(k-1)=z^-1i_error(k), z^-1Represent the conversion when one sampling period of signal lag for the variable in time domain；Sampling period For 0.1～2 second；T (k) is the instantaneous total torque of switched reluctance machines current time；When being that switched reluctance machines m is current The given total torque carved；

The state representation of echo state network is:

Q (k+1)=f (w_inu(k)+wq(k)+w_backi_error(k)) (5)

i_error(k+1)=w_outs(k) (6)

Wherein, q (k)=[q₁(k),q₂(k),…,q_n(k)]^t∈rⁿFor internal state variable；S (k)=[u (k), q^t(k),i_error (k)]^t∈r^p+n+m, []^tThe transposition computing of representing matrix；w_in∈r^n×pRepresent input connection weight matrix, w ∈ r^n×nRepresent internal Connection weight matrix, w_back∈r^n×mRepresent feedback link weight matrix, echo state network initial phase dimension n=72, degree of rarefication For 1～5% generation, power spectral radius is internal connection weight w of 0.1～1 generation；

Need connection weight w training reserve pool to export to system_out∈r^p+n+m；F=[f₁,f₂,…,f_n] represent neuronal activation letter Number, f_r(r=1,2 ... n) adopt hyperbolic tangent function, output layer adopts linear function；Reserve pool dimension d=30～100, interior Portion's state variable n=72, p is the input vector u dimension of echo state network, number m of the output variable of p=3, esn, m= 1；

The performance index function of the reserve pool of echo state network is:

$e\left(k\right)=\frac{1}{2}{\[t_{ref}\left(k\right)-t\left(k\right)\]}^2=\frac{1}{2}e{\left(k\right)}^2---\left(7\right)$

In formula, t_refK () is the current k moment to give torque；T (k) is the instantaneous of current k moment switched reluctance machines reality output Total torque；

According to gradient descent method, the output weighed value adjusting algorithm of echo state network is:

$\left\{\begin{array}{c}{w}_{lj}\left(k\right)=w_{lj}(k-1)+{\mathrm{\δw}}_{lj}\left(k\right)\\ {\mathrm{\δw}}_{lj}\left(k\right)=-\mathrm{\η}e\left(k\right)\frac{\∂t\left(k\right)}{\∂ii\left(k\right)}{q}_n\left(k\right)\end{array}\right.---\left(8\right)$

W in formula_ljFor output layer weighting parameter；η is parameter, regularized learning algorithm rate, span 0～1；L=1,2 ..., 72, j=1； (8) in formula, ii (k) is the complex controll electric current of three-phaseSum, q_nK () is the State- output of echo state network,Jacobian information for switched reluctance machines；- 2 switched reluctance machines based on rbf neutral net Jacobian information calculates

Using rbf neutral net in formula (8)Carry out on-line identification；The excitation function of output layer neuron takes For linear function, and the excitation function of hidden layer neuron is taken as Gaussian bases；

Choose x=[ii (k), ii (k-1), t (k-1), t (k-2)]^tInput vector for rbf, wherein；Ii (k) and ii (k-1) point Not Biao Shi current time and previous moment total current, ii (k-1)=z^-1Ii (k), t (k) and t (k-1) represent switch magnetic respectively Resistance motor is in the instantaneous total torque of current time and previous moment, wherein t (k-1)=z^-1T (k), t (k-2)=z^-1t(k-1)；h =[h₁,h₂,…,h_j…h_m]^tRadial direction base vector for rbf, wherein h_jFor Gaussian bases:

$h_j=\exp \left(\frac{\left|\right|x-c_j|{|}^2}{2b_j^2}\right),j=1,2,...m---\left(9\right)$

In formula, c_jFor the center vector of j-th node, c_j=[c_j1,c_j2,…c_ji,…c_jn]^t, wherein b_jSound stage width ginseng for node Number, and be the number more than zero, c_jAnd b_jIt is all to be obtained by study, node m span 5～20, n=4 is input variable Number；

Jacobian information is:

$\frac{\∂t\left(k\right)}{\∂ii\left(k\right)}\≈\frac{\∂y_m\left(k\right)}{\∂ii\left(k\right)}=\underset{j=1}{\overset{m}{\σ}}{w}_j{k}_j\frac{c_{js}-x_1}{b_j^2}---\left(10\right)$

In formula, x₁=ii (k), s=1,2 ..., n；

According to gradient descent method, output power w_j, node center c_jsAnd node sound stage width parameter b_jIteration as follows:

$\begin{array}{c}{w}_j\left(k\right)=w_j(k-1)+\mathrm{\η}\left(t\right(k)-y_m(k\left)\right)h_j\\ +\mathrm{\α}\left(w_j\right(k-1)-w_j(k-2\left)\right)\end{array}---\left(11\right)$

$\begin{array}{c}{b}_j\left(k\right)=b_j(k-1)+{\mathrm{\η\δb}}_j+\mathrm{\α}\left(b_j\right(k-1)\\ -b_j(k-2))\end{array}---\left(12\right)$

Wherein

${\mathrm{\δb}}_j=\left(t\right(k)-y_m(k\left)\right)w_j{h}_j\frac{\left|\right|x-c_j|{|}^2}{b_j^3}$

$\begin{array}{c}{c}_{js}\left(k\right)=c_{js}(k-1)+{\mathrm{\η\δc}}_{js}+\mathrm{\α}\left(c_{js}\right(k-1)\\ -c_{js}(k-2))\end{array}---\left(13\right)$

Wherein

${\mathrm{\δc}}_{js}=\left(t\right(k)-y_m(k\left)\right)w_j{h}_j\frac{x_j-c_{js}}{b_j^2}$

In formula, y_mK () is the output of current k moment rbf neutral net, η is learning rate, and α is factor of momentum, its span 0～ 1；

- 3 nonlinear Control electric currents

Esn real-time output nonlinear electric current i_error, obtain each phase nonlinear Control electric current i through electric current distribution_kzz, by rising part, Sloping portion and constant value segments i_errorComposition, in same time period adjacent biphase be respectively at change from small to large open mutually and Shutoff phase from big to small；Taking adjacent a phase, b phase as a example, non-linear current is allocated as follows:

In formulaRepresent respectively and be in a phase opening phase and the nonlinear Control electric current of the b phase being off phase；

, complex controll electric current switched reluctance machines permanent torque control

Step obtains each phase linear control electric current i by each phase torque distribution_kzz, by rising part, sloping portion and constant value segments t_kzzComposition；Taking adjacent a phase, b phase as a example, Linear Control electric current is assigned as:

In formulaRepresent respectively and be in a phase opening phase and the Linear Control electric current of the b phase being off phase；

Step gained each phase nonlinear Control electric current i_kzzLinear Control electric current i with step gained_kzzSuperposition, obtain is each Phase complex controll electric currentAs the setting value of current hysteresis-band control device, the output of current hysteresis-band control device, through power conversion Device driving switch reluctance motor, carries out permanent torque control.

2. one kind of the switched reluctance machines permanent torque design of control method of complex controll electric current according to claim 1 is multiple Close control electric current switched reluctance machines Constant Torque Holding System it is characterised in that:

Including signal processor, analog-to-digital conversion module, current hysteresis-band control device, power inverter, current sensor, torque sensing Device and rotor-position sensor；

Three current sensors are respectively arranged in the three-phase power line of switched reluctance machines, detect each phase current, torque sensor It is installed on output shaft, the output torque of detection motor of switched reluctance machines, rotor-position sensor is installed on switching magnetic-resistance electricity The rotor of machine, detects rotor position angle；Position sensor, torque sensor and current sensor connect letter through analog-to-digital conversion module Number processor；

Signal processor contains linear current control module, rbf neural network module and echo state network non-linear current control Molding block；

Described linear current controller includes torque divider and torque-current transducer, and torque divider includes torque distribution Given total torque is assigned as the torque reference of each phase by function according to the current rotor position angle of rotor-position sensor detection, Torque-current transducer obtains respectively according to currently each phase torque reference according to the torque-current partition function of linear inductance model Phase linear control electric current i_kzz；

The total current of described rbf neural network module foundation current time and previous moment and instantaneous total torque, calculate switch magnetic The jacobian information of resistance motor, is supplied to echo state network module；

Described echo state network non-linear current controller includes echo state network module and electric current distribute module, echo shape State mixed-media network modules mixed-media turns according to the instantaneous total torque of switched reluctance machines current time, the given total of switched reluctance machines current time The feedback of the non-linear current of square and output, and the jacobian information of switched reluctance machines, through parameter learning, export non-thread Property electric current i_error；i_errorObtain each phase nonlinear Control electric current i through electric current distribute module_kzz；

Each phase linear control electric current i_kzzWith each phase nonlinear Control electric current i_kzzSuperposition, each phase complex controll electric current obtaining Input current hystersis controller is as its setting value, the output access power changer of current hysteresis-band control device, power inverter It is output as driving switch reluctance motor three-phase drive power supply；

Rotor-position sensor, three current sensors and torque sensor detection obtain current switch magnetic resistance motor rotor position The analogue signal at angle, the electric current of power supply three-phase and instantaneous total torque is converted to corresponding digital signal through analog-to-digital conversion module, Send into signal processor, the linear current control module in signal processor, the non-linear current of echo state network controls mould Block, according to the current instantaneous total torque of actual measurement and three-phase electricity flow valuve, tries to achieve complex controll electric current, controlling switch reluctance motor.

3. complex controll electric current according to claim 2 switched reluctance machines Constant Torque Holding System it is characterised in that:

Described signal processor is connected with display screen.

4. complex controll electric current according to claim 2 switched reluctance machines Constant Torque Holding System it is characterised in that:

Described signal processor is furnished with can interface.