CN106357192A - Method and system for lowering torque pulsation of switched reluctance motor by current adaptive control - Google Patents

Abstract

The invention relates to a method and system for lowering torque pulsation of a switched reluctance motor by current adaptive control; the method comprises: preprocessing a deviation: subjecting the torque deviation to nonlinear conversion; solving torque estimated output and adaptive PID (proportion integration differentiation) control coefficient by using double-weight neural network; acquiring current set total current by PID control calculation, and acquiring each phase control current via current allocation; predicting current feed-forward compensation control by finite difference extended Kalman filter, and effectively inhibiting and lowering torque pulsation of the switched reluctance motor by joint action of adaptive PID control and prediction-based current feed-forward compensation control. Current, torque and rotor position sensors are connected with a signal processor, the signal processor executes modules of the method, compensated three-phase reference current is output to control a power converter of the motor via a current hysteresis controller, and torque pulsation of the switched reluctance motor is significantly and effective inhibited.

Description

Current automatic adaptation controls the method and system reducing switched reluctance machines torque pulsation

Technical field

The present invention relates to the motor power of electric automobile drives field, specially a kind of current automatic adaptation controls and reduces switch The method and system of reluctance motor torque pulsation.

Background technology

Switched reluctance machines (switched reluctance motor, abbreviation srm) both no winding also no permanent magnets, tool Have that structure is simple, efficiency high, stable and reliable operation, with low cost, easy to maintain many advantages, such as.But, switched reluctance machines For special double-salient-pole structure, take the power supply mode of switching regulator, electromagnetic property is in nonlinearity and strong coupling, leads to it During operating, torque pulsation is larger, and the noise being led to by torque pulsation and vibration problem seriously restrict its application and develop, therefore The research method reducing switched reluctance machines torque pulsation is extremely important.

In order to suppress and reducing the torque pulsation of switched reluctance machines srm, Chinese scholars have carried out many research, Also achieved with many achievements in research.

Torque partition function method is to propose a kind of control strategy with reference to srm feature, is applied to suppress more and reduces srm torque Pulsation, this strategy passes through suitable partition function, so that torque is seamlessly transitted during commutation, is then further added by suitably controlling plan Motor total output torque steady tracking is slightly made to give torque reference.

A kind of offline torque distribution control strategy that another document proposes, and with traditional line style, sinusoidal pattern, Cubic, Exponential type torque distribution control strategy compares.It is the accuracy improving model, proposes on the basis of linear model The nonlinear model of torque-current, but do not provide the concrete identification process of unknown parameter.

The document also having proposes electric current distribution on the basis of torque distribution and magnetic linkage distributes control strategy.Both modes Avoid directly finding the non-linear relation of torque-current, control sensitivity higher, but be only limited to the theoretical research emulating at present.

Research is separately had to introduce iterative learning controller on the basis of traditional torque distribution control strategy, with torque deviation Little reference current is iterated learn control, output current variable quantity compensates reference current for control targe, and in tradition only Have 1 and -1 two states hystersis controller on the basis of, devise a subregion with 1,0 and -1 three kind of working condition thin Change hysteresis current controller, it is torque pulsation inhibited more effective for simulation results show.

Add speed ring in addition with traditional Direct Torque Control, and allow little fluctuating error for Hysteresis control Shortcoming, proposes Direct Torque pid (proportion integral derivative) and controls to adjust voltage duty cycle, realize The direct torque of srm.

But, due to the nonlinear characteristic that switched reluctance machines are extremely strong, the above-mentioned existing electricity based on simplification mathematical model Machine method for controlling torque, is all easily caused violent torque pulsation it is impossible to practical application.

Content of the invention

The purpose of the present invention is that a kind of current automatic adaptation of design controls the method reducing switched reluctance machines torque pulsation, first Carry out deviation pretreatment, torque deviation is carried out non-linear conversion；Try to achieve the ginseng of self adaptation pid control with two-weight neural network Number；The current feedforward compensa-tion adopting the prediction of finite difference extended Kalman filter again controls, and improves control system one-step prediction Ability.Self adaptation pid controls and controls collective effect with the current feedforward compensa-tion based on prediction, effectively suppresses and reduces srm torque Pulsation.

It is another object of the present invention to control according to a kind of current automatic adaptation of the invention described above reducing switched reluctance machines The method of torque pulsation, a kind of current automatic adaptation of design controls the system reducing switched reluctance machines torque pulsation, by self adaptation Pid controls and the current feedforward compensa-tion based on prediction controls and melts and be harmonious, and constitutes the control system of switched reluctance machines.

The current automatic adaptation of present invention design controls the method reducing switched reluctance machines torque pulsation, and key step is such as Under:

, torque deviation pretreatment

Present invention introduces nonlinear function carries out pretreatment to torque deviation, carry out little error large gain, the little increasing of big error The pretreatment of benefit.Concrete pretreatment is as follows:

$fal\left(\mathrm{\&delta;}t\right)=\left\{\begin{array}{cc}|\mathrm{\&delta;}t{|}^{\mathrm{\&alpha;}}sign\left(\mathrm{\&delta;}t\right),& \left|\mathrm{\&delta;}t\right|\&greaterequal;\mathrm{\&delta;}\\ \frac{\mathrm{\&delta;}t}{{\mathrm{\&delta;}}^{(1-\mathrm{\&alpha;})}},& \left|\mathrm{\&delta;}t\right|<\mathrm{\&delta;}\end{array}\right.---\left(1\right)$

Wherein δ represents the transformation range of feedback deviation, span 0.01t_d～0.1t_d, t_dFor setting torque, fal represents Preconditioned functions, sign represents sign function, and that is, δ t is more than zero, and value is 1, and less than zero, value is -1.δ t is to set torque t_dTransient torque t with actual measurement_eBetween deviation.α is regulation coefficient scope 0～1.

Self adaptation pid direct torque based on two-weight neural network

- 1 two-weight neural network

Srm is the strongly non-linear system of structure changes, variable element it is difficult to obtain accurate mathematical model.The present invention is led to Cross two-weight neural network (double weight neural network, dwnn) to pid control (ratio, integration and differential Control) three parameter on-line tuning, realize srm self adaptation pid control.

Two-weight neural network dwnn includes input layer, hidden layer and output layer.The torque installed on switched reluctance machines Instantaneous torque t of sensor detection_e, instantaneous torque t of previous moment_{e_1}And setting electric current i_dIncrement △ i_dPrevious moment valueAs the input quantity of dwnn, set torque t_dAs the desired value of dwnn, after its study, dwnn predicts that obtaining torque estimates Meter output t_outAnd the proportionality coefficient k that pid controls_p, integral coefficient k_iiAnd differential coefficient k_d.

Pid controls torque deviation pre-processed results fal (δ t) obtaining according to step i and two-weight neural network dwnn The proportionality coefficient k obtaining_p, integral coefficient k_iiAnd differential coefficient k_dIt is calculated current setting total current i through pid_d, divide through electric current Join and obtain each phase control electric currentWith

The present invention adds the self-adaptative adjustment of power, and the dwnn after improvement is described as follows:

$\left\{\begin{array}{c}h\left(j\right)=f\left(\underset{z=1}{\overset{3}{\&sigma;}}{w_{zj}}^m{\left(x_z-q_{zj}\right)}^{b_{zj}}\right)\\ f\left(x\right)=\frac{1}{1+e^{-ax}}\\ t_{out}=\underset{j=1}{\overset{3}{\&sigma;}}{v}_j*h\left(j\right)\end{array}\right.---\left(2\right)$

Wherein h (j) is hidden layer output function, and f (x) is activation primitive, t_outTorque estimation output for dwnn, w_zjFor Directional weighting, q_zjFor core weights, v_jFor exporting weights, 0 < a < 1, m span 1～10, b_zjFor self-adaptative adjustment power, e Bottom for natural logrithm.Hidden layer is z layer, z=1,2,3；Output layer is j layer, j=1,2,3.

In order to obtain the parameter in formula (2), the performance index function is taken to be:

$\left\{\begin{array}{c}{\mathrm{\&delta;t}}_1\left(k\right)=t_e\left(k\right)-t_{out}\left(k\right)\\ j=\frac{1}{2}{\left({\mathrm{\&delta;t}}_1\right(k\left)\right)}^2\end{array}\right.---\left(3\right)$

Wherein, △ t₁It is defined as the transient torque t surveying_eExport t with two-weight neural network dwnn torque estimation_outBetween Deviation.

According to gradient descent method, directional weighting w_zj, core weights q_zj, output weights v_jAnd power b_zjIncrement iteration Algorithm is as follows:

$\left\{\begin{array}{c}{\mathrm{\&delta;v}}_j\left(k\right)=-\mathrm{\&eta;}\frac{\&part;j}{\&part;v_j}={\mathrm{\&eta;\&delta;t}}_1\left(k\right)h_j\left(k\right)\\ {\mathrm{\&delta;w}}_{zj}=-\mathrm{\&eta;}\frac{\&part;j}{\&part;w_{zj}}={\mathrm{\&eta;ma\&delta;t}}_1\left(k\right)v_j{h}_j(1-h_j){w_{zj}}^{m-1}{(x_z-q_{zj})}^{b_{zj}}\\ {\mathrm{\&delta;q}}_{zj}=-\mathrm{\&eta;}\frac{\&part;j}{q_{zj}}={\mathrm{\&eta;ab}}_{zj}{\mathrm{\&delta;t}}_1\left(k\right)v_j{h}_j(1-h_j){w_{zj}}^m{(x_z-q_{zj})}^{(b_{zj}-1)}\\ {\mathrm{\&delta;b}}_{zj}=-{\mathrm{\&eta;}}_b\frac{\&part;j}{\&part;b_{zj}}={\mathrm{\&eta;}}_b{\mathrm{a\&delta;t}}_1\left(k\right)v_j{h}_j(1-h_j){w_{zj}}^m{(x_z-q_{zj})}^{b_{zj}}\ln (x_z-q_{zj})\end{array}\right.---\left(4\right)$

η is the learning rate of weights, value 0～1；η_bFor power learning rate, scope 0～1.

Self adaptation pid of -2 two-weight neural networks controls

Self adaptation pid controls the setting electric current of output as follows:

i_d(k)=i_d(k-1)+k_pxc(1)+k_iixc(2)+k_dxc(3) (5)

Wherein,

Choose performance index function as follows,

$e\left(k\right)=\frac{1}{2}{\left(t_d\right(k)-t_e(k\left)\right)}^2=\frac{1}{2}\mathrm{\&delta;}t{\left(k\right)}^2---\left(6\right)$

According to gradient descent method, k_p,k_ii,k_dIterative algorithm as follows:

$\left\{\begin{array}{c}{\mathrm{\&delta;k}}_p\left(k\right)=-{\mathrm{\&eta;}}_{k_p}\frac{\&part;e}{\&part;k_p}={\mathrm{\&eta;}}_{k_p}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(1\right)\\ {\mathrm{\&delta;k}}_{ii}\left(k\right)=-{\mathrm{\&eta;}}_{k_{ii}}\frac{\&part;e}{\&part;k_{ii}}={\mathrm{\&eta;}}_{k_{ii}}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(2\right)\\ {\mathrm{\&delta;k}}_d\left(k\right)=-{\mathrm{\&eta;}}_{k_d}\frac{\&part;e}{\&part;k_d}={\mathrm{\&eta;}}_{k_d}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(3\right)\end{array}\right.---\left(7\right)$

η_kp, η_kz, η_kdIt is the learning coefficient of ratio, integration and differential coefficient respectively, span 0～1, all it is taken as 0.2.Its value is obtained by two-weight neural network identification.

It is i_dIncrement △ i_dPrevious moment value, approximately take:

$\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d^{*}}\&ap;\frac{\&part;t_{out}}{\&part;{\mathrm{\&delta;i}}_d}=\underset{j=1}{\overset{3}{\&sigma;}}{\mathrm{ab}}_{1j}{v}_j{h}_j(1-h_j){w_{1j}}^m{({\mathrm{\&delta;i}}_d-q_{1j})}^{(b_{1j}-1)}---\left(8\right)$

Wherein t_outFor the torque estimation output of two-weight neural network identification, w_1jIt is to be connected to hidden layer from jth output layer In z=1 layer neuron directional weighting, q_1jFor its core weights, b_1jAdjust power for it.These parameters pass through formula (4) Obtain.

Pid is output as setting total current i_d.Three-phase control electric current is respectively i_a ^*And i_b ^*And i_c ^*, below with b phase for turning off Phase, c phase is for, as a example opening phase, the electric current of biphase commutation is allocated as follows:

$\left\{\begin{array}{c}{i_c}^{*}={i_d}^{*}f\left(\mathrm{\&theta;}\right)\\ {i_b}^{*}={i_d}^{*}(1-f(\mathrm{\&theta;}\left)\right)\end{array}\right.---\left(9\right)$

Wherein: partition function:

$f\left(\mathrm{\&theta;}\right)=\frac{3}{{\mathrm{\&theta;}}_{ov}^2}\&centerdot;{(\mathrm{\&theta;}-{\mathrm{\&theta;}}_{on})}^2-\frac{2}{{\mathrm{\&theta;}}_{ov}^3}\&centerdot;{(\mathrm{\&theta;}-{\mathrm{\&theta;}}_{on})}^3,$

Wherein θ is motor rotor position angle, θ_onFor turn-on angle, 10～15 degree of span；θ_ovOverlapping for rotor-position Angle, 2～5 degree of span.

Current feedforward compensa-tion based on Kalman prediction controls

The current forecasting of -1 finite difference EKF

The current feedforward compensa-tion of the present invention controls, and first passes through finite difference EKF one-step prediction motor Output current, then by the difference real-Time Compensation reference current of current forecasting value and reference current, makes reference current arrive in error Before, carry back and make correction, indirectly suppress the torque pulsation of srm, the pre- electric current of finite difference EKF of the present invention Compensate and include two parts, finite difference extended Kalman filter fdekf and relative error processing module.Switched reluctance machines n =1,2,3 phases, i.e. n-th voltage u in a, b, c three-phase_n, rotor position angle θ and compensate after output three-phase current i_a、i_b And i_cInput for fdekf, the three-phase current i of fdekf prediction_a ^*、i_b ^*And i_c ^*Input relative error processing module, double power simultaneously Value neutral net dwnn self adaptation pid controls the three-phase current i of output_a ^*、i_b ^*And i_c ^*Also relative error processing module, phase are inputted Three-phase current deviation to be compensated e is exported to Error processing module_ia ^*、e_ib ^*And e_ic ^*, three-phase current deviation to be compensated e_ia ^*、e_ib ^*With e_ic ^*Control the three-phase current i of output with dwnn self adaptation pid_a ^*、i_b ^*And i_c ^*Sum is the three-phase reference current i after compensating_a、i_b And i_c.

The n-th electric current i of switched reluctance machines_n, rotating speed w and rotor position angle θ state equation as follows:

$\begin{array}{c}\left[\begin{array}{c}{i}_n\left(k\right)\\ w\left(k\right)\\ \mathrm{\&theta;}\left(k\right)\end{array}\right]=\left[\begin{array}{ccc}1-aart& -aact& 0\\ 0& 1& 0\\ 0& t& 1\end{array}\right]\left[\begin{array}{c}{i}_n(k-1)\\ w(k-1)\\ \mathrm{\&theta;}(k-1)\end{array}\right]\\ \left[\begin{array}{cc}aa& 0\\ 0& \frac{t}{j}(t_e-t_l)\\ 0& 0\end{array}\right]\left[\begin{array}{c}{u}_n(k-1)\\ 1\end{array}\right]\end{array}---\left(9\right)$

Wherein,ψ is switched reluctance motor flux linkage, and t is the sampling period, and scope is 1～3 Second, value is 1 second.J is to select rotary inertia, t_eFor transient torque, t_lFor load torque, u_nFor n-th voltage, r is switch magnetic The winding resistance value of resistance motor.

The present invention adopts the fdekf of technology maturation to realize to the output three-phase current i after compensating_a、i_bAnd i_cNext step pre- Survey, obtain i_a ^*、i_b ^*And i_c ^*.Resistance r in formula (9), parameter aa and c are to estimate by fdekf to obtain.

The Front feedback control of -2 electric currents

The three-phase current i of fdekf prediction_a ^*And i_b ^*And i_c ^*Control gained three-phase reference current i with dwnn self adaptation pid_a ^*With i_b ^*And i_c ^*, obtaining three-phase current error to be compensated through relative error after processing is e_ia ^*And e_ib ^*And e_ic ^*, for feedforward compensation, obtain Three-phase reference current i to after compensate_aAnd i_bAnd i_c.

Its relative error processing procedure is as follows:

$\left\{\begin{array}{c}{e_{i_a}}^{*}=({i_a}^{*}-{i_a}^{*})*\frac{{i_a}^{*}}{{i_a}^{*}}\\ {e_{i_b}}^{*}=({i_b}^{*}-{i_b}^{*})*\frac{{i_b}^{*}}{{i_b}^{*}}\\ {e_{i_c}}^{*}=({i_c}^{*}-{i_c}^{*})*\frac{{i_c}^{*}}{{i_c}^{*}}\end{array}\right.---\left(10\right)$

Three-phase reference current i after feedforward compensation_a、i_bAnd i_cAs follows:

$\left\{\begin{array}{c}{i}_a={i_a}^{*}+{e_{i_a}}^{*}\\ i_b={i_b}^{*}+{e_{i_b}}^{*}\\ i_c={i_c}^{*}+{e_{i_c}}^{*}\end{array}\right.---\left(11\right)$

Three-phase reference current after feedforward compensation setting as the current hysteresis-band control device of technology maturation is obtained by formula (11) Definite value, the output of current hysteresis-band control device, through power inverter driving switch reluctance motor, effectively suppresses switched reluctance machines Torque pulsation.

The present invention controls, according to a kind of above-mentioned current automatic adaptation, the method reducing switched reluctance machines torque pulsation, designs one Plant current automatic adaptation and control the system reducing switched reluctance machines torque pulsation, including signal processor, analog-to-digital conversion module, electricity Stream hystersis controller, power inverter, three-phase 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.

Signal processor contains torque deviation pretreatment module, the pid self-adaptive control module of two-weight neural network and Current feedforward compensa-tion module based on limited spread Kalman prediction.

Torque deviation pretreatment module is to setting torque t_dInstantaneous torque t with torque sensor detection_eTorque deviation enter Row nonlinear function pretreatment, its result fal (δ t) is as the input of pid self-adaptive control module.

Two-weight neural network dwnn in the pid self-adaptive control module of two-weight neural network is examined with torque sensor Instantaneous torque t surveyed_e, instantaneous torque t of previous moment_{e_1}And setting electric current i_dIncrement △ i_dPrevious moment valueAs Input quantity, sets torque t_dAs the desired value of dwnn, prediction after its study obtains torque estimation output t_outAnd pid The proportionality coefficient k controlling_p, integral coefficient k_iiAnd differential coefficient k_d.

Pid controls the ratio obtaining according to result fal (δ t) and the two-weight neural network of torque deviation pretreatment module Coefficient k_p, integral coefficient k_iiAnd differential coefficient k_dIt is calculated current setting total current i through pid_d, obtain each phase through electric current distribution Control electric currentWith

The filter of finite difference spreading kalman is included based on the current feedforward compensa-tion module of limited spread Kalman prediction Ripple device fdekf and relative error processing module.Fdekf is with phase voltage u_n, rotor position angle θ and compensate after output three-phase electricity Stream i_a、i_bAnd i_cFor input, the three-phase current i of prediction_a ^*、i_b ^*And i_c ^*Input relative error processing module, current forecasting value and pid The difference of the control electric current of output is as three-phase current deviation to be compensated e_ia ^*、e_ib ^*And e_ic ^*, three-phase current deviation to be compensated e_ia ^*、 e_ib ^*And e_ic ^*Control the three-phase current i of output with dwnn self adaptation pid_a ^*、i_b ^*And i_c ^*Sum is the three-phase reference current after compensating i_a、i_bAnd i_c.Three-phase reference current input current hystersis controller accesses as its setting value, the output of current hysteresis-band control device Power inverter, power inverter is output as driving switch reluctance motor three phase mains, effectively suppresses its torque pulsation.

The system, according to current torque and current value, suppresses switched reluctance machines torque by the current automatic adaptation of the present invention Pulse methods, real-Time Compensation reference current, before error arrives, carry back and reference current is made with correction, indirectly suppress srm Torque pulsation.

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

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

Compared with prior art, a kind of current automatic adaptation of the present invention controls the method reducing switched reluctance machines torque pulsation Advantage with system is: 1, carries out the pretreatment of nonlinear transformation to the torque deviation of switched reluctance machines, to adapt to the non-of srm Linear characteristic；2nd, during self adaptation pid controls, parameter passes through improved two-weight neural network on-line tuning；3rd, pass through limited spread The predicted current of Kalman filtering, realizes current feedforward compensa-tion control, improves the predictive ability of control system；4th, dwnn self adaptation The direct torque of pid with reference to fdekf prediction transient current, realize current feedforward compensa-tion, two kinds of controls act on, effectively simultaneously The torque pulsation of suppression suppression switched reluctance machines；Distributed using traditional torque open loop and control, the pulsation rate of switched reluctance machines reaches 46.42%, and drop to 1.65% using the pulsation rate that the present invention suppresses switched reluctance machines.

Brief description

Fig. 1 is the step -1 that this current automatic adaptation controls the embodiment of the method reducing switched reluctance machines torque pulsation Structural representation is adjusted based on two-weight neural network current automatic adaptation；

Fig. 2 is the step -1 that this current automatic adaptation controls the embodiment of the method reducing switched reluctance machines torque pulsation Two-weight neural network learning structure schematic diagram；

Fig. 3 is the step -1 that this current automatic adaptation controls the embodiment of the method reducing switched reluctance machines torque pulsation Current feedforward compensa-tion structure chart based on limited spread Kalman prediction；

Fig. 4 is the structural representation that this current automatic adaptation controls the system embodiment reducing switched reluctance machines torque pulsation Figure；

Fig. 5 is the operational flow diagram that this current automatic adaptation controls the system reducing switched reluctance machines torque pulsation.

Specific embodiment

Current automatic adaptation controls the embodiment of the method reducing switched reluctance machines torque pulsation

This current automatic adaptation controls the embodiment of the method reducing switched reluctance machines torque pulsation, and key step is as follows:

, torque deviation pretreatment

It is as follows that the introducing nonlinear function of this example carries out pretreatment to torque deviation:

$fal\left(\mathrm{\&delta;}t\right)=\left\{\begin{array}{cc}|\mathrm{\&delta;}t{|}^{\mathrm{\&alpha;}}sign\left(\mathrm{\&delta;}t\right),& \left|\mathrm{\&delta;}t\right|\&greaterequal;\mathrm{\&delta;}\\ \frac{\mathrm{\&delta;}t}{{\mathrm{\&delta;}}^{(1-\mathrm{\&alpha;})}},& \left|\mathrm{\&delta;}t\right|<\mathrm{\&delta;}\end{array}\right.---\left(1\right)$

Wherein δ represents the transformation range of feedback deviation, span 0.01t_d～0.1t_d, t_dFor setting torque, fal represents Preconditioned functions, sign represents sign function, and that is, δ t is more than zero, and value is 1, and less than zero, value is -1.δ t is to set torque t_dTransient torque t with actual measurement_eBetween deviation.α is regulation coefficient, and this example takes 1.

Self adaptation pid direct torque based on two-weight neural network

- 1 two-weight neural network

As shown in Fig. 2 two-weight neural network dwnn includes input layer, hidden layer and output layer.As shown in figure 1, switch Instantaneous torque t of the torque sensor detection installed on reluctance motor_e, instantaneous torque t of previous moment_{e_1}And setting electric current i_d Increment △ i_dPrevious moment valueAs the input quantity of dwnn, set torque t_dAs the desired value of dwnn, through its study Dwnn prediction obtains torque estimation output t afterwards_outAnd the proportionality coefficient k that pid controls_p, integral coefficient k_iiAnd differential coefficient k_d.

Pid controls torque deviation pre-processed results fal (δ t) obtaining according to step i and two-weight neural network dwnn The proportionality coefficient k obtaining_p, integral coefficient k_iiAnd differential coefficient k_dIt is calculated current setting total current i through pid_d, divide through electric current Join and obtain each phase control electric currentWith

Dwnn after adding the self-adaptative adjustment of power to improve is described as follows:

$\left\{\begin{array}{c}h\left(j\right)=f\left(\underset{z=1}{\overset{3}{\&sigma;}}{w_{zj}}^m{\left(x_z-q_{zj}\right)}^{b_{zj}}\right)\\ f\left(x\right)=\frac{1}{1+e^{-ax}}\\ t_{out}=\underset{j=1}{\overset{3}{\&sigma;}}{v}_j*h\left(j\right)\end{array}\right.---\left(2\right)$

Wherein h (j) is hidden layer output function, and f (x) is activation primitive, t_outTorque estimation output for dwnn, w_zjFor Directional weighting, q_zjFor core weights, v_jFor exporting weights, this example a=0.1, m=1, b_zjFor self-adaptative adjustment power, e is nature The bottom of logarithm.Hidden layer is z layer, z=1,2,3；Output layer is j layer, j=1,2,3.

In order to obtain the parameter in formula (2), the performance index function is taken to be:

$\left\{\begin{array}{c}{\mathrm{\&delta;t}}_1\left(k\right)=t_e\left(k\right)-t_{out}\left(k\right)\\ j=\frac{1}{2}{\left({\mathrm{\&delta;t}}_1\right(k\left)\right)}^2\end{array}\right.---\left(3\right)$

Wherein, △ t₁It is defined as the transient torque t surveying_eExport t with two-weight neural network dwnn torque estimation_outBetween Deviation.

According to gradient descent method, directional weighting w_zj, core weights q_zj, output weights v_jAnd power b_zjIncrement iteration Algorithm is as follows:

$\left\{\begin{array}{c}{\mathrm{\&delta;v}}_j\left(k\right)=-\mathrm{\&eta;}\frac{\&part;j}{\&part;v_j}={\mathrm{\&eta;\&delta;t}}_1\left(k\right)h_j\left(k\right)\\ {\mathrm{\&delta;w}}_{zj}=-\mathrm{\&eta;}\frac{\&part;j}{\&part;w_{zj}}={\mathrm{\&eta;ma\&delta;t}}_1\left(k\right)v_j{h}_j(1-h_j){w_{zj}}^{m-1}{(x_z-q_{zj})}^{b_{zj}}\\ {\mathrm{\&delta;q}}_{zj}=-\mathrm{\&eta;}\frac{\&part;j}{q_{zj}}={\mathrm{\&eta;ab}}_{zj}{\mathrm{\&delta;t}}_1\left(k\right)v_j{h}_j(1-h_j){w_{zj}}^m{(x_z-q_{zj})}^{(b_{zj}-1)}\\ {\mathrm{\&delta;b}}_{zj}=-{\mathrm{\&eta;}}_b\frac{\&part;j}{\&part;b_{zj}}={\mathrm{\&eta;}}_b{\mathrm{a\&delta;t}}_1\left(k\right)v_j{h}_j(1-h_j){w_{zj}}^m{(x_z-q_{zj})}^{b_{zj}}\ln (x_z-q_{zj})\end{array}\right.---\left(4\right)$

η is the learning rate of weights, and this example takes η=0.2, η_bFor power learning rate, this example takes η_b=0.4.

Self adaptation pid of -2 two-weight neural networks controls

Self adaptation pid controls the setting electric current of output as follows:

i_d(k)=i_d(k-1)+k_pxc(1)+k_iixc(2)+k_dxc(3) (5)

Wherein,

Choose performance index function as follows,

$e\left(k\right)=\frac{1}{2}{\left(t_d\right(k)-t_e(k\left)\right)}^2=\frac{1}{2}\mathrm{\&delta;}t{\left(k\right)}^2---\left(6\right)$

According to gradient descent method, k_p,k_ii,k_dIterative algorithm as follows:

$\left\{\begin{array}{c}{\mathrm{\&delta;k}}_p\left(k\right)=-{\mathrm{\&eta;}}_{k_p}\frac{\&part;e}{\&part;k_p}={\mathrm{\&eta;}}_{k_p}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(1\right)\\ {\mathrm{\&delta;k}}_{ii}\left(k\right)=-{\mathrm{\&eta;}}_{k_{ii}}\frac{\&part;e}{\&part;k_{ii}}={\mathrm{\&eta;}}_{k_{ii}}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(2\right)\\ {\mathrm{\&delta;k}}_d\left(k\right)=-{\mathrm{\&eta;}}_{k_d}\frac{\&part;e}{\&part;k_d}={\mathrm{\&eta;}}_{k_d}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(3\right)\end{array}\right.---\left(7\right)$

η_kp, η_kz, η_kdIt is the learning coefficient of ratio, integration and differential coefficient respectively, this example is all taken as 0.2.Its value Obtained by two-weight neural network identification.

It is i_dIncrement △ i_dPrevious moment value, approximately take:

$\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d^{*}}\&ap;\frac{\&part;t_{out}}{\&part;{\mathrm{\&delta;i}}_d}=\underset{j=1}{\overset{3}{\&sigma;}}{\mathrm{ab}}_{1j}{v}_j{h}_j(1-h_j){w_{1j}}^m{({\mathrm{\&delta;i}}_d-q_{1j})}^{(b_{1j}-1)}---\left(8\right)$

Wherein t_outFor the torque estimation output of two-weight neural network identification, w_1jIt is to be connected to hidden layer from jth output layer In z=1 layer neuron directional weighting, q_1jFor its core weights, b_1jAdjust power for it.These parameters pass through formula (4) Obtain.

Pid is output as setting total current i_d.Three-phase control electric current is respectively i_a ^*And i_b ^*And i_c ^*, below with b phase for turning off Phase, c phase is for, as a example opening phase, the electric current of biphase commutation is allocated as follows:

$\left\{\begin{array}{c}{i_c}^{*}=i_d*f\left(\mathrm{\&theta;}\right)\\ {i_b}^{*}=i_d*(1-f(\mathrm{\&theta;}\left)\right)\end{array}\right.---\left(9\right)$

Wherein: partition function:

$f\left(\mathrm{\&theta;}\right)=\frac{3}{{\mathrm{\&theta;}}_{ov}^2}\&centerdot;{(\mathrm{\&theta;}-{\mathrm{\&theta;}}_{on})}^2-\frac{2}{{\mathrm{\&theta;}}_{ov}^3}\&centerdot;{(\mathrm{\&theta;}-{\mathrm{\&theta;}}_{on})}^3,$

Wherein θ is motor rotor position angle, θ_onFor turn-on angle, this example θ_on=11.25 degree, θ_ovFor rotor-position angle overlap, This example θ_ov=3.5 degree.

Current feedforward compensa-tion based on Kalman prediction controls

The current forecasting of -1 finite difference EKF

As shown in figure 3, the pre- current compensation of this example finite difference EKF includes two parts, finite difference extends Kalman filter fdekf and relative error processing module.Switched reluctance machines n=1,2,3 phases, i.e. n-th in a, b, c three-phase Phase voltage u_n, rotor position angle θ and compensate after output three-phase current i_a、i_bAnd i_cInput for fdekf, fdekf prediction Three-phase current i_a ^*、i_b ^*And i_c ^*Input relative error processing module, two-weight neural network dwnn self adaptation pid control simultaneously is defeated The three-phase current i going out_a ^*、i_b ^*And i_c ^*Also input relative error processing module, relative error processing module exports three-phase electricity to be compensated Stream deviation e_ia ^*、e_ib ^*And e_ic ^*, three-phase current deviation to be compensated e_ia ^*、e_ib ^*And e_ic ^*Control the three of output with dwnn self adaptation pid Phase current i_a ^*、i_b ^*And i_c ^*Sum is the three-phase reference current i after compensating_a、i_bAnd i_c.

The n-th electric current i of switched reluctance machines_n, rotating speed w and rotor position angle θ state equation as follows:

$\begin{array}{c}\left[\begin{array}{c}{i}_n\left(k\right)\\ w\left(k\right)\\ \mathrm{\&theta;}\left(k\right)\end{array}\right]=\left[\begin{array}{ccc}1-aart& -aact& 0\\ 0& 1& 0\\ 0& t& 1\end{array}\right]\left[\begin{array}{c}{i}_n(k-1)\\ w(k-1)\\ \mathrm{\&theta;}(k-1)\end{array}\right]\\ +\left[\begin{array}{cc}aa& 0\\ 0& \frac{t}{j}(t_e-t_l)\\ 0& 0\end{array}\right]\left[\begin{array}{c}{u}_n(k-1)\\ 1\end{array}\right]\end{array}---\left(9\right)$

Wherein,ψ is switched reluctance motor flux linkage, and t is the sampling period, this example t=1 second.J is choosing Select rotary inertia, t_eFor transient torque, t_lFor load torque, u_nFor n-th voltage, r is the winding resistance value of switched reluctance machines.

This example adopts fdekf to realize to the output three-phase current i after compensating_a、i_bAnd i_cNext step prediction, obtain i_a ^*、 i_b ^*And i_c ^*.Resistance r in formula (9), parameter aa and c are estimated by fdekf and obtain.

The Front feedback control of -2 electric currents

The three-phase current i of fdekf prediction_a ^*And i_b ^*And i_c ^*Control gained three-phase reference current i with dwnn self adaptation pid_a ^*With i_b ^*And i_c ^*, obtaining three-phase current error to be compensated through relative error after processing is e_ia ^*And e_ib ^*And e_ic ^*, for feedforward compensation, obtain Three-phase reference current i to after compensate_aAnd i_bAnd i_c.

Its relative error processing procedure is as follows:

$\left\{\begin{array}{c}{e_{i_a}}^{*}=({i_a}^{*}-{i_a}^{*})*\frac{{i_a}^{*}}{{i_a}^{*}}\\ {e_{i_b}}^{*}=({i_b}^{*}-{i_b}^{*})*\frac{{i_b}^{*}}{{i_b}^{*}}\\ {e_{i_c}}^{*}=({i_c}^{*}-{i_c}^{*})*\frac{{i_c}^{*}}{{i_c}^{*}}\end{array}\right.---\left(10\right)$

Three-phase reference current i after feedforward compensation_a、i_bAnd i_cAs follows:

$\left\{\begin{array}{c}{i}_a={i_a}^{*}+{e_{i_a}}^{*}\\ i_b={i_b}^{*}+{e_{i_b}}^{*}\\ i_c={i_c}^{*}+{e_{i_c}}^{*}\end{array}\right.---\left(11\right)$

Three-phase reference current after feedforward compensation setting as the current hysteresis-band control device of technology maturation is obtained by formula (11) Definite value, the output of current hysteresis-band control device, through power inverter driving switch reluctance motor, effectively suppresses switched reluctance machines Torque pulsation.

Current automatic adaptation controls the system embodiment reducing switched reluctance machines torque pulsation

This controls the embodiment of the method reducing switched reluctance machines torque pulsation, design current according to above-mentioned current automatic adaptation Self Adaptive Control reduces the system embodiment of switched reluctance machines torque pulsation, as shown in Figures 4 and 5, including signal processor, mould Number modular converter, current hysteresis-band control device, power inverter, three-phase current sensor, torque sensor and rotor position sensing Device.

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 transient torque of detection motor, and rotor-position sensor is installed on switch magnetic The rotor of resistance motor, detects rotor position angle.Rotor-position sensor, torque sensor and current sensor are through analog digital conversion mould Block connects signal processor.

Signal processor contains torque deviation pretreatment module, the pid self-adaptive control module of two-weight neural network and Current feedforward compensa-tion module based on limited spread Kalman prediction.

Torque deviation pretreatment module is to setting torque t_dInstantaneous torque t with torque sensor detection_eTorque deviation enter Row nonlinear function pretreatment, its result fal (δ t) is as the input of pid self-adaptive control module.

Two-weight neural network dwnn in the pid self-adaptive control module of two-weight neural network is examined with torque sensor Instantaneous torque t surveyed_e, instantaneous torque t of previous moment_{e_1}And setting electric current i_dIncrement △ i_dPrevious moment valueAs defeated Enter amount, set torque t_dAs the desired value of dwnn, prediction after its study obtains torque estimation output t_outAnd pid control The proportionality coefficient k of system_p, integral coefficient k_iiAnd differential coefficient k_d.

Pid controls the ratio obtaining according to result fal (δ t) and the two-weight neural network of torque deviation pretreatment module Coefficient k_p, integral coefficient k_iiAnd differential coefficient k_dIt is calculated current setting total current i through pid_d, obtain each phase through electric current distribution Control electric currentWith

The filter of finite difference spreading kalman is included based on the current feedforward compensa-tion module of limited spread Kalman prediction Ripple device fdekf and relative error processing module.Fdekf is with phase voltage u_n, rotor position angle θ and compensate after output three-phase electricity Stream i_a、i_bAnd i_cFor input, the three-phase current i of prediction_a ^*、i_b ^*And i_c ^*Input relative error processing module, current forecasting value and pid The difference of the control electric current of output is as three-phase current deviation to be compensated e_ia ^*、e_ib ^*And e_ic ^*, three-phase current deviation to be compensated e_ia ^*、 e_ib ^*And e_ic ^*Control the three-phase current i of output with dwnn self adaptation pid_a ^*、i_b ^*And i_c ^*Sum is the three-phase reference current after compensating i_a、i_bAnd i_c.Three-phase reference current input current hystersis controller accesses as its setting value, the output of current hysteresis-band control device Power inverter, power inverter is output as driving switch reluctance motor three phase mains, effectively suppresses its torque pulsation.

The system, according to current torque and current value, suppresses switched reluctance machines torque by the current automatic adaptation of the present invention Pulse methods, real-Time Compensation reference current, before error arrives, carry back and reference current is made with correction, indirectly suppress srm Torque pulsation.

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

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

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. a kind of current automatic adaptation controls the method reducing switched reluctance machines torque pulsation, and key step is as follows:

, torque deviation pretreatment

Introduce nonlinear function and torque deviation carried out with pretreatment:

$fal\left(\mathrm{\&delta;}t\right)=\left\{\begin{array}{cc}|\mathrm{\&delta;}t{|}^{\mathrm{\&alpha;}}sign\left(\mathrm{\&delta;}t\right),& \left|\mathrm{\&delta;}t\right|\&greaterequal;\mathrm{\&delta;}\\ \frac{\mathrm{\&delta;}t}{{\mathrm{\&delta;}}^{(1-\mathrm{\&alpha;})}},& \left|\mathrm{\&delta;}t\right|<\mathrm{\&delta;}\end{array}\right.---\left(1\right)$

Wherein δ represents the transformation range of feedback deviation, span 0.01t_d～0.1t_d, t_dFor setting torque, fal represents pre- place Reason function, sign represents sign function, and that is, δ t is more than zero, and value is 1, and less than zero, value is -1；δ t is to set torque t_dWith The transient torque t of actual measurement_eBetween deviation；α is regulation coefficient scope 0～1；

Self adaptation pid direct torque based on two-weight neural network

- 1 two-weight neural network

Two-weight neural network dwnn includes input layer, hidden layer and output layer；The torque sensing installed on switched reluctance machines Instantaneous torque t of device detection_e, instantaneous torque t of previous moment_{e_1}And setting electric current i_dIncrement △ i_dPrevious moment valueMake Input quantity for dwnn, sets torque t_dAs the desired value of dwnn, after its study, to obtain torque estimation defeated for dwnn prediction Go out t_outAnd the proportionality coefficient k that pid controls_p, integral coefficient k_iiAnd differential coefficient k_d；

Torque deviation pre-processed results fal (δ t) that pid control foundation step i obtains and two-weight neural network dwnn obtain Proportionality coefficient k_p, integral coefficient k_iiAnd differential coefficient k_dIt is calculated current setting total current i through pid_d, distribute through electric current To each phase control electric currentWith

The present invention adds the self-adaptative adjustment of power, and the two-weight neural network dwnn after improvement is described as follows:

$\left\{\begin{array}{c}h\left(j\right)=f\left(\underset{z=1}{\overset{3}{\mathrm{\&sigma;}}}{w_{zj}}^m{\left(x_z-q_{zj}\right)}^{b_{zj}}\right)\\ f\left(x\right)=\frac{1}{1+e^{-ax}}\\ t_{out}=\underset{j=1}{\overset{3}{\mathrm{\&sigma;}}}{v}_j*h\left(j\right)\end{array}\right.---\left(2\right)$

Wherein h (j) is hidden layer output function, and f (x) is activation primitive, t_outTorque estimation output for dwnn, w_zjFor direction Weights, q_zjFor core weights, v_jFor export weights, 0 < a < 1, take a=0.1, m span 1～10, take m=1, b_zjFor adaptive Power should be adjusted, e is the bottom of natural logrithm；Hidden layer is z layer, z=1,2,3；Output layer is j layer, j=1,2,3；

The performance index function is taken to be:

$\left\{\begin{array}{c}{\mathrm{\&delta;t}}_1\left(k\right)=t_e\left(k\right)-t_{out}\left(k\right)\\ j=\frac{1}{2}{\left({\mathrm{\&delta;t}}_1\left(k\right)\right)}^2\end{array}\right.---\left(3\right)$

Wherein, △ t₁It is defined as the transient torque t surveying_eExport t with two-weight neural network dwnn torque estimation_outBetween inclined Difference；

According to gradient descent method, directional weighting w_zj, core weights q_zj, output weights v_jAnd power b_zjIncrement iterative algorithm such as Under:

$\left\{\begin{array}{c}{\mathrm{\&delta;v}}_j\left(k\right)=-\mathrm{\&eta;}\frac{\&part;j}{\&part;v_j}={\mathrm{\&eta;\&delta;t}}_1\left(k\right)h_j\left(k\right)\\ {\mathrm{\&delta;w}}_{zj}=-\mathrm{\&eta;}\frac{\&part;j}{\&part;w_{zj}}={\mathrm{\&eta;ma\&delta;t}}_1\left(k\right)v_j{h}_j\left(1-h_j\right){w_{zj}}^{m-1}{\left(x_z-q_{zj}\right)}^{b_{zj}}\\ {\mathrm{\&delta;q}}_{zj}=-\mathrm{\&eta;}\frac{\&part;j}{q_{zj}}={\mathrm{\&eta;ab}}_{zj}{\mathrm{\&delta;t}}_1\left(k\right)v_j{h}_j\left(1-h_j\right){w_{zj}}^m{\left(x_z-q_{zj}\right)}^{\left(b_{zj}-1\right)}\\ {\mathrm{\&delta;b}}_{zj}=-{\mathrm{\&eta;}}_b\frac{\&part;j}{\&part;b_{zj}}={\mathrm{\&eta;}}_b{\mathrm{a\&delta;t}}_1\left(k\right)v_j{h}_j\left(1-h_j\right){w_{zj}}^m{\left(x_z-q_{zj}\right)}^{b_{zj}}\ln \left(x_z-q_{zj}\right)\end{array}\right.---\left(4\right)$

η is the learning rate of weights, value 0～1；η_bFor power learning rate, span 0～1；

Self adaptation pid of -2 two-weight neural networks controls

Self adaptation pid controls the setting electric current of output as follows:

i_d(k)=i_d(k-1)+k_pxc(1)+k_iixc(2)+k_dxc(3) (5)

Wherein,

Choose performance index function as follows,

$e\left(k\right)=\frac{1}{2}{\left(t_d\right(k)-t_e(k\left)\right)}^2=\frac{1}{2}\mathrm{\&delta;}t{\left(k\right)}^2---\left(6\right)$

According to gradient descent method, k_p,k_ii,k_dIterative algorithm as follows:

$\left\{\begin{array}{c}{\mathrm{\&delta;k}}_p\left(k\right)=-{\mathrm{\&eta;}}_{k_p}\frac{\&part;e}{\&part;k_p}={\mathrm{\&eta;}}_{k_p}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(1\right)\\ {\mathrm{\&delta;k}}_{ii}\left(k\right)=-{\mathrm{\&eta;}}_{k_{ii}}\frac{\&part;e}{\&part;k_{ii}}={\mathrm{\&eta;}}_{k_{ii}}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(2\right)\\ {\mathrm{\&delta;k}}_d\left(k\right)=-{\mathrm{\&eta;}}_{k_d}\frac{\&part;e}{\&part;k_d}={\mathrm{\&eta;}}_{k_d}\mathrm{\&delta;}t\left(k\right)\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d}xc\left(3\right)\end{array}\right.---\left(7\right)$

η_kp, η_kz, η_kdIt is the learning coefficient of ratio, integration and differential coefficient respectively, span 0～1,Its value is by double power Value neural network identification obtains；

It is i_dIncrement △ i_dPrevious moment value, approximately take:

$\frac{\&part;t_e}{\&part;{\mathrm{\&delta;i}}_d^{*}}\&ap;\frac{\&part;t_{out}}{\&part;{\mathrm{\&delta;i}}_d}=\underset{j=1}{\overset{3}{\&sigma;}}{\mathrm{ab}}_{1j}{v}_j{h}_j(1-h_j){w_{1j}}^m{({\mathrm{\&delta;i}}_d-q_{1j})}^{(b_{1j}-1)}---\left(8\right)$

Wherein t_outFor the torque estimation output of two-weight neural network identification, w_1jIt is to be connected to hidden layer from jth output layer The directional weighting of z=1 layer neuron, q_1jFor its core weights, b_1jAdjust power for it；These parameters are passed through formula (4) and are obtained Arrive；

Pid is output as setting total current i_d；Three-phase control electric current is respectively i_a ^*And i_b ^*And i_c ^*, below with b phase for turning off phase, c Mutually for, as a example opening phase, the electric current of biphase commutation is allocated as follows:

$\left\{\begin{array}{c}{i_c}^{*}=i_d*f\left(\mathrm{\&theta;}\right)\\ {i_b}^{*}=i_d*(1-f(\mathrm{\&theta;}\left)\right)\end{array}\right.---\left(9\right)$

Wherein: partition function:

$f\left(\mathrm{\&theta;}\right)=\frac{3}{{\mathrm{\&theta;}}_{ov}^2}\&centerdot;{(\mathrm{\&theta;}-{\mathrm{\&theta;}}_{on})}^2-\frac{2}{{\mathrm{\&theta;}}_{ov}^3}\&centerdot;{(\mathrm{\&theta;}-{\mathrm{\&theta;}}_{on})}^3,$

Wherein θ is motor rotor position angle, θ_onFor turn-on angle, 10～15 degree of span, θ_ovFor rotor-position angle overlap, value 2～5 degree of scope；

Current feedforward compensa-tion based on Kalman prediction controls

The current forecasting of -1 finite difference EKF

The pre- current compensation of described finite difference EKF includes two parts, finite difference extended Kalman filter Fdekf and relative error processing module；Switched reluctance machines n=1,2,3 phases, i.e. n-th voltage u in a, b, c three-phase_n, turn Output three-phase current i after sub- angular position theta and compensation_a、i_bAnd i_cInput for fdekf, the three-phase current i of fdekf prediction_a ^*、 i_b ^*And i_c ^*Input relative error processing module, the three-phase current of the output of two-weight neural network dwnn self adaptation pid control simultaneously i_a ^*、i_b ^*And i_c ^*Also input relative error processing module, relative error processing module exports three-phase current deviation to be compensated e_ia ^*、 e_ib ^*And e_ic ^*, three-phase current deviation to be compensated e_ia ^*、e_ib ^*And e_ic ^*Control the three-phase current i of output with dwnn self adaptation pid_a ^*、 i_b ^*And i_c ^*Sum is the three-phase reference current i after compensating_a、i_bAnd i_c；

The n-th electric current i of switched reluctance machines_n, rotating speed w and rotor position angle θ state equation as follows:

$\begin{array}{c}\left[\begin{array}{c}{i}_n\left(k\right)\\ w\left(k\right)\\ \mathrm{\&theta;}\left(k\right)\end{array}\right]=\left[\begin{array}{ccc}1-aart& -aact& 0\\ 0& 1& 0\\ 0& t& 1\end{array}\right]\left[\begin{array}{c}{i}_n(k-1)\\ w(k-1)\\ \mathrm{\&theta;}(k-1)\end{array}\right]\\ +\left[\begin{array}{cc}aa& 0\\ 0& \frac{t}{j}(t_e-t_l)\\ 0& 0\end{array}\right]\left[\begin{array}{c}{u}_n(k-1)\\ 1\end{array}\right]\end{array}---\left(9\right)$

Wherein,ψ is switched reluctance motor flux linkage, and t is the sampling period, and value is 1～3 second, and j is to select Rotary inertia, t_eFor transient torque, t_lFor load torque, u_nFor n-th voltage, r is the winding resistance value of switched reluctance machines；Electricity Resistance r, parameter aa and c are to estimate by fdekf to obtain；

The Front feedback control of -2 electric currents

The three-phase current i of fdekf prediction_a ^*And i_b ^*And i_c ^*Control gained three-phase reference current i with dwnn self adaptation pid_a ^*And i_b ^*And i_c ^*, obtaining three-phase current error to be compensated through relative error after processing is e_ia ^*And e_ib ^*And e_ic ^*, for feedforward compensation, mended Three-phase reference current i after repaying_aAnd i_bAnd i_c；

Its relative error processing procedure is as follows:

$\left\{\begin{array}{c}{e_{i_a}}^{*}=({i_a}^{*}-{i_a}^{*})*\frac{{i_a}^{*}}{{i_a}^{*}}\\ {e_{i_b}}^{*}=({i_b}^{*}-{i_b}^{*})*\frac{{i_b}^{*}}{{i_b}^{*}}\\ {e_{i_c}}^{*}=({i_c}^{*}-{i_c}^{*})*\frac{{i_c}^{*}}{{i_c}^{*}}\end{array}\right.---\left(10\right)$

Three-phase reference current i after feedforward compensation_a、i_bAnd i_cAs follows:

$\left\{\begin{array}{c}{i}_a={i_a}^{*}+{e_{i_a}}^{*}\\ i_b={i_b}^{*}+{e_{i_b}}^{*}\\ i_c={i_c}^{*}+{e_{i_c}}^{*}\end{array}\right.---\left(11\right)$

Three-phase reference current after feedforward compensation is obtained as the setting value of current hysteresis-band control device, Hysteresis Current control by formula (11) The output of device processed, through power inverter driving switch reluctance motor, the torque pulsation of effect suppression switched reluctance machines.

2. current automatic adaptation according to claim 1 controls the method reducing switched reluctance machines torque pulsation, design A kind of current automatic adaptation controls the system reducing switched reluctance machines torque pulsation, including signal processor, analog-to-digital conversion module, Current hysteresis-band control device, power inverter, three-phase current sensor, torque sensor and rotor-position sensor；Its feature exists In:

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 torque deviation pretreatment module, the pid self-adaptive control module of two-weight neural network and being based on The current feedforward compensa-tion module of limited spread Kalman prediction；

Torque deviation pretreatment module is to setting torque t_dInstantaneous torque t with torque sensor detection_eTorque deviation carry out non- Linear function pretreatment, its result fal (δ t) is as the input of pid self-adaptive control module；

Two-weight neural network dwnn in the pid self-adaptive control module of two-weight neural network is detected with torque sensor Instantaneous torque t_e, instantaneous torque t of previous moment_{e_1}And setting electric current i_dIncrement △ i_dPrevious moment valueAs input Amount, sets torque t_dAs the desired value of dwnn, prediction after its study obtains torque estimation output t_outAnd pid controls Proportionality coefficient k_p, integral coefficient k_iiAnd differential coefficient k_d；

Pid controls the proportionality coefficient obtaining according to result fal (δ t) and the two-weight neural network of torque deviation pretreatment module k_p, integral coefficient k_iiAnd differential coefficient k_dIt is calculated current setting total current i through pid_d, obtain each phase control through electric current distribution Electric currentWith

Finite difference extended Kalman filter is included based on the current feedforward compensa-tion module of limited spread Kalman prediction Fdekf and relative error processing module；Fdekf is with phase voltage u_n, rotor position angle θ and compensate after output three-phase current i_a、 i_bAnd i_cFor input, the three-phase current i of prediction_a ^*、i_b ^*And i_c ^*Input relative error processing module, current forecasting value is exported with pid Control electric current difference as three-phase current deviation to be compensated e_ia ^*、e_ib ^*And e_ic ^*, three-phase current deviation to be compensated e_ia ^*、e_ib ^*With e_ic ^*Control the three-phase current i of output with dwnn self adaptation pid_a ^*、i_b ^*And i_c ^*Sum is the three-phase reference current i after compensating_a、i_b And i_c；Three-phase reference current input current hystersis controller is as its setting value, the output access power of current hysteresis-band control device Changer, power inverter is output as driving switch reluctance motor three phase mains.

3. current automatic adaptation according to claim 2 controls the system reducing switched reluctance machines torque pulsation, its feature It is:

Described signal processor is connected with display screen, the current electric current of Real time displaying, moment information and torque pulsation rate information.

4. current automatic adaptation according to claim 2 controls the system reducing switched reluctance machines torque pulsation, its feature It is:

Described signal processor is furnished with can interface.