CN102842909B - Method for controlling power electronic hybrid system - Google Patents

Abstract

The invention discloses a method for controlling a power electronic hybrid system. The method comprises a power electronic hybrid control system comprising an active power filter and a static var compensator, wherein the static var compensator comprises a passive power filter group and a thyristor control reactor (TCR); and the active power filter, the passive power filter group and the TCR access between a power grid and loads connected with the power grid in sequence. The passive power filter and the active power filter are used jointly to perform harmonic suppression on high-voltage busbars; and the static var compensator is used to perform reactive compensation, so that delay compensation and online control are realized, control accuracy is enhanced, tracking performance is improved, electric energy loss of the power grid is reduced, and pollution of the power grid is purified.

Description

A kind of power electronics hybrid system control method

Technical field

The present invention relates to power electronic system, particularly a kind of power electronics hybrid system control method.

Background technology

Along with the development of power electronic technology, network load is that nonlinear user and reactive load get more and more, such as rolling mill, arc furnace, electric railway etc.Increasing of this kind of load makes harmonic components in electrical network more and more serious, and cause electric network element loss to increase, the equipment that affects normally runs, and causes electrical network particular resonance to cause serious accident what is more; And the increase of load or burden without work can cause Voltage Drop, the problems such as loss increase.Therefore the power grid environment of pure a, high-quality and power supply quality is had to be that to operate department be also that load user expects jointly to electrical network simultaneously.But also there is a lot of deficiency in current product major part:

1, Active Power Filter-APF and Static Var Compensator SVC are used alone, function singleness, can not realize the dual-use function of filtering and reactive power compensation simultaneously.

2, the control method that Active Power Filter-APF is common has PI to control, Hysteresis control etc.; Adopt PI control time control precision poor, easily vibrate when compensation capacity is low and PI parameter be comparatively difficult to regulate; During Hysteresis control, the setting of ring width is comparatively large to systematic influence, and the frequency requirement of ring width hour switch tube is also high, causes application not to be very extensive.

3, Active Power Filter-APF uses the control precision of PI controller own just not high, once there is load changing, when harmonic wave situation changes, filter effect is especially undesirable, can not realize On-line Control.

4, Static Var Compensator SVC uses PI controller, once reactive load change, compensation effect is undesirable, can not realize On-line Control.

5, general first-harmonic resonance injection active power filter, its first-harmonic resonance injection branch is when being applied to high pressure, and first-harmonic resonance part will bear very high voltage, and capacity is comparatively large, and during engineer applied, cost is too high.

Summary of the invention

Technical problem to be solved by this invention is, not enough for prior art, provides a kind of power electronics hybrid system control method, realize compensation of delay and On-line Control, improve control precision, and improve tracking performance, reduce the electric energy loss of electrical network, purifying electric network pollution.

For solving the problems of the technologies described above, the technical solution adopted in the present invention is: a kind of power electronics hybrid system control method, comprise power electronics hybrid control system, power electronics hybrid control system includes active power filter, Static Var Compensator, described Static Var Compensator comprises passive power filter group and thyristor-controlled reactor, Active Power Filter-APF, passive power filter group, thyristor-controlled reactor successively and access electrical network and between the load be connected with electrical network, it is characterized in that, described power electronics hybrid system control method comprises Control Method of Active Power Filter and Static Var Compensator control method:

Described Control Method of Active Power Filter comprises the following steps:

1) load current I is detected _la, I _lb, I _lc, the actual current i be injected in three phase network of Active Power Filter-APF _fha, i _fhb, i _fhc, computational load harmonic current i _lha, i _lhb, i _lhc;

2) π-Smith Compensator is utilized to carry out e to system ^{π-τ t}compensation of delay, make i _fha, i _fhb, i _fhcwith i _lha, i _lhb, i _lhcbetween phase difference be 180 °;

3) PI controller regulating load harmonic current i is used _lha, i _lhb, i _lhcthe current i that be injected in three phase network actual in Active Power Filter-APF _fha, i _fhb, i _fhcbetween error, produce the drive singal of IGBT in Active Power Filter-APF with triangle wave;

4) signal after PI adjustment and Active Power Filter-APF are injected the electric current of three phase network as the input of prediction neural network, obtain predicted value i _n-inst(s); i _fh(s-i) represent that s-i moment filter is injected into the electric current in electrical network, i _n-insts () represents the output that neural network prediction obtains;

5) according to load harmonic current i _lha, i _lhb, i _lhcthe current i injected actual in Active Power Filter-APF _fha, i _fhb, i _fhcbetween error i _error, PI regulate after the variable quantity of signal u (s) and the output i of neural network prediction _n-insts (), utilizes wide area PREDICTIVE CONTROL criterion, calculate the updated value K of PI controller parameter _p, K _i, realize the on-line control of PI controller;

Described Static Var Compensator control method comprises the following steps:

1) load current i is detected _la, i _lb, i _lcwith three-phase power grid voltage u _a, u _b, u _c, calculate the reactive current i of three phase network _lqa, i _lqb, i _lqc;

2) reference voltage u is set as by the rated voltage that the regulating load of PI controller normally works _ref, reference voltage u _refwith the three-phase power grid voltage u detected _a, u _b, u _cbetween error regulate with PI controller, after linearized link, produce the control signal of thyristor in thyristor-controlled reactor;

3) signal u (s) after PI being regulated and the three-phase power grid voltage u detected _a, u _b, u _cas the input of prediction neural network, obtain the three-phase power grid voltage output valve predicted;

4) variable quantity, the reference voltage u of signal u (s) after regulating according to PI _refwith the output of neural network prediction, utilize wide area PREDICTIVE CONTROL criterion, calculate the updated value K of PI controller _p, K _i, realize the on-line control of PI controller.

Operation principle of the present invention is: described blended electric power electronics hybrid system is by injection active power filter, and passive power filter group PPF, thyristor-controlled reactor TCR form.Passive filter group assume responsibility for main harmonics filtering, and injection active power filter carries out the suppression of all the other harmonic waves, PPF and TCR formation Static Var Compensator SVC carries out the reactive power compensation from perception to capacitive.The filter of described blended electric power electronic system carries out compensation of delay by π-Smith Compensator, and by the output of neural network prediction Active Power Filter-APF, wide area PREDICTIVE CONTROL criterion calculates the updated value controling parameters of PI controller, realizes On-line Control; Its Static Var Compensator is by neural network prediction line voltage, and wide area PREDICTIVE CONTROL criterion calculates the updated value of PI controller, realizes On-line Control.

Compared with prior art, the beneficial effect that the present invention has is: power electronic system of the present invention is mixed with filter and reactive-load compensator, realizes the dual-use function of harmonics restraint and reactive power compensation; The Active Power Filter-APF of this blended electric power electronic system adopts π-Smith Compensator structure to carry out filtering delay-time compensation, and filter effect is desirable; The Active Power Filter-APF of this blended electric power electronic system adopts neural net to carry out prediction and exports, and realize online modification PI parameter by wide area PREDICTIVE CONTROL criterion, control precision promotes greatly, realizes accurate filtering; The Static Var Compensator SVC of this blended electric power electronic system adopts neural net to carry out prediction line voltage, and realize online modification PI parameter by wide area PREDICTIVE CONTROL criterion, control precision promotes greatly, realizes fine compensation.

Accompanying drawing explanation

Fig. 1 is one embodiment of the invention power electronics hybrid system structural representation;

Fig. 2 is one embodiment of the invention injection active filter system control principle drawing;

Fig. 3 is one embodiment of the invention reactive-load compensator Systematical control schematic diagram;

Fig. 4 is one embodiment of the invention prediction neural network training schematic diagram;

Fig. 5 is one embodiment of the invention prediction neural network structure chart;

Fig. 6 is that one embodiment of the invention PI parameter upgrades flow chart.

Embodiment

As shown in Figure 1-Figure 3, power electronics hybrid control system of the present invention includes active power filter, Static Var Compensator, described Static Var Compensator comprises passive power filter group and thyristor-controlled reactor, Active Power Filter-APF, passive power filter group, thyristor-controlled reactor are successively and access electrical network and between the load be connected with electrical network, Static Var Compensator SVC is made up of thyristor-controlled reactor TCR and passive power filter PPF.Passive power filter PPF is made up of several groups of single tuned filters.Injection active power filter is made up of injection branch and Active Power Filter-APF, and wherein injection branch comprises two groups of single tuned filters, one group of high pass filter and point voltage inductance.This injection branch had both met offset current can inject electrical network smoothly, again reduces the cost of injection branch.Because under 35Kv high pressure, when adopting first-harmonic resonance pouring-in, its injection branch first-harmonic capacity is very large, and cost is very high.Injection active power filter is by can the PI controller of online modification parameter regulate, IGBT control signal is formed after triangular modulation, the output of neural network prediction Active Power Filter-APF, the on-line amending of PI parameter is carried out by wide area PREDICTIVE CONTROL criterion, realize accurately controlling, carry out compensation of delay by π-Smith's prediction simultaneously.Static Var Compensator SVC is controlled by voltage close loop, uses PI controller to regulate, neural network prediction, the parameter of wide area PREDICTIVE CONTROL criterion online modification PI controller.

Power electronics hybrid system control method, comprises the control to Active Power Filter-APF and the control to Static Var Compensator:

Wherein to the control of Active Power Filter-APF, comprise and specifically comprising the following steps:

1) load current i is detected _la, i _lb, i _lc, the actual current i be injected in electrical network of injection active power filter _fha, i _fhb, i _fhc, computational load harmonic current i _lha, i _lhb, i _lhc.

2) control objectives realizes i _fha, i _fhb, i _fhc=-i _lha,-i _lhb,-i _lhc, namely the offset current that exports of injection active power filter is with the load harmonic current equal and opposite in direction that obtains of detection and direction is contrary, and system itself exists time delay e ^{-τ t}, then by π-Smith's prediction, e is carried out to system ^{π-τ t}compensation of delay, thus i _fha, i _fhb, i _fhcwith i _lha, i _lhb, i _lhcbetween phase difference be 180 °, need not carry out oppositely detecting the harmonic wave that obtains in instruction current, thus reach the function of compensation of delay.

3) error between load-side harmonic wave and the actual electric current injected of Active Power Filter-APF uses PI to regulate, and produces the drive singal of IGBT in filter with triangle wave.

4) neural net carries out Model Distinguish, and its identification process to export with neural net and sampling obtains filter and injects error between power network current for according to the training carrying out weights amendment, till error is in the scope of setting, complete identification process.The input of neural net has unit time delay process, and the signal after PI regulates and filter inject the electric current of electrical network as the input of prediction neural network, obtain predicted value i _n-inst(s).

5) load harmonic i _lhthe current i injected actual in filter _fhbetween error i _error, PI regulate after the variable quantity of signal u (s) and the output i of neural network prediction _n-insts (), as the input of wide area PREDICTIVE CONTROL criterion, calculates the updated value K of the parameter of PI controller _p, K _i, realize the on-line control of PI controller.

Static Var Compensator SVC control method key step comprises:

1) load current i is detected _la, i _lb, i _lcwith line voltage u _a, u _b, u _c, calculate reactive current i in electrical network _lqa, i _lqb, i _lqc.

2) rated voltage that user load normally works is set as reference voltage u _ref, reference voltage and detect that the error between line voltage uses PI to regulate, after linearized link, forms the control signal of thyristor.

3) signal u (s) after PI adjustment and the input of the line voltage detected as prediction neural network, obtain the line voltage output valve predicted.

4) variable quantity, the reference voltage u of signal u (s) after regulating according to PI _refwith the output of neural network prediction, utilize wide area PREDICTIVE CONTROL criterion, calculate the updated value K of PI controller _p, K _i, realize the on-line control of PI controller.

See Fig. 3, the realization of compensation of delay function: the principle of filtering is exactly detect the harmonic wave existed in load-side to make the null process of power network current by injecting the electric current contrary with its equal and opposite in direction direction.I _sh=i _lh+ i _fh≈ 0, i.e. i _fh=-i _lh, i _shrepresent power network current, i _lhrepresent load-side harmonic current, i _fhrepresent that Active Power Filter-APF is injected into the electric current in electrical network.But owing to there is time delay in system, if its total time delay process is e ^{-τ s}if, compensated in advance time link e ^{(τ-π) s}, then total time delay will become e ^{-τ s}+ e ^{(τ-π) s}=e ^{-π s}, i.e. i _fh=i _lh* e ^{-π s}=-i _lh, realize i _fhwith i _lhbetween reverse.Specific implementation step is as follows:

1) first G is supposed _inv(s), G _outs () is all there is not time delay, carry out CONTROLLER DESIGN G with this situation _c(s).Then G _cs the closed loop transfer function, of () is as follows

$G_{\mathrm{closed}}=\frac{G_c\left(s\right)*G_{\mathrm{inv}}\left(s\right)*G_{\mathrm{out}}\left(s\right)}{1+G_c\left(s\right)*G_{\mathrm{inv}}\left(s\right)*G_{\mathrm{out}}\left(s\right)}$

G _cs () represents the transfer function of PI controller;

G _invs () represents the transfer function of inverter;

G _outs () represents that electric current outputs to the transfer function injected between electrical network from inverter;

2) our target is that design obtains controller make G ^* _closedcontain G _inv(s) and G _outthe time delay process e of (s) ^{-τ s}, such to closed loop transfer function, be

G ^* _closed=G _closed ^*e ^-τs

Namely $\frac{G_c^{*}\left(s\right).G_{\mathrm{inv}}\left(s\right).G_{\mathrm{out}}\left(s\right).e^{-\mathrm{\τs}}}{1+G_c^{*}\left(s\right).G_{\mathrm{inv}}\left(s\right).G_{\mathrm{out}}\left(s\right).e^{-\mathrm{\τs}}}=\frac{G_c\left(s\right).G_{\mathrm{inv}}\left(s\right).G_{\mathrm{out}}\left(s\right)}{1+G_c\left(s\right).G_{\mathrm{inv}}\left(s\right).G_{\mathrm{out}}\left(s\right)}{e}^{-\mathrm{\τs}}$

Solve $G_c^{*}\left(s\right)=\frac{G_c\left(s\right)}{1+G_c\left(s\right).G_{\mathrm{inv}}\left(s\right).G_{\mathrm{out}}\left(s\right)(1-e^{-\mathrm{\τs}})}=\frac{G_c\left(s\right)}{1+G_c\left(s\right).G_p\left(s\right)(1-e^{-\mathrm{\τs}})}$

Wherein G _p(s)=G _inv(s) .G _out(s)

3) according to analysis above, i _fhand i _lhbetween direction contrary, namely phase difference is π therebetween, and this just means

G _p(s)e ^-τs＝G _p(s)e ^-πs

$G_c^{*}\left(s\right)=\frac{G_c\left(s\right)}{1+G_c\left(s\right).G_p\left(s\right)(1-e^{-\mathrm{\πs}})}$

4) according to the circuit analysis to model, the transfer function of closed-loop system is

$G_{\mathrm{closed}}=\frac{I_{\mathrm{cfh}}\left(s\right)}{-I_{\mathrm{Lh}}\left(s\right)}=\frac{G_c^{*}\left(s\right)*G_{\mathrm{inv}}\left(s\right)*G_{\mathrm{out}}\left(s\right)}{1+G_c^{*}\left(s\right)*G_{\mathrm{inv}}\left(s\right)*G_{\mathrm{out}}\left(s\right)}*e^{-\mathrm{\πs}}$

Can find out that active power filter system eliminates the impact of time delay by this closed loop transfer function, and achieve control objectives i _fh=-i _lh, this process is also π-Smith's forecasting process.

The neural network prediction model of Active Power Filter-APF and Static Var Compensator, the process that wide area PREDICTIVE CONTROL realizes On-line Control is identical, therefore concentrates explanation at this.

See Fig. 4, the Model Distinguish process of neural net and the training of neural net, until output error stops within the scope of setting.For filter be regulated by PI after signal u (s) and the actual output i of filter _fhs () is as input, for Static Var Compensator be regulated by PI after signal u (s) and actual electric network voltage as input, two models, all by the foundation that the error between predicted value and actual value is revised as weights, adopt Levenberg-Marquardt gradient descent algorithm to carry out weights amendment.Constantly repeat this process, until error terminates training after arriving the scope of setting, become forecast model.

See Fig. 5, the forecast model of neural net adopts the three layer feedforward neural networks with time delay.Wherein the activation primitive of hidden layer is hyperbolic tangent function, and the activation primitive of output layer is linear activation primitive.

$y_n\left(s\right)=\underset{j=1}{\overset{\mathrm{hid}}{\mathrm{\Σ}}}{w}_{\mathrm{jk}}*f_j\left({\mathrm{net}}_j\left(s\right)\right)+b_k$

${\mathrm{net}}_j\left(s\right)=\underset{i=1}{\overset{n_u}{\mathrm{\Σ}}}{w}_{j,i}*u(s-i)+\underset{i=1}{\overset{n_{I_{\mathrm{Fh}}}}{\mathrm{\Σ}}}{w}_{j,n_u+i}*I_{\mathrm{Fh}}(s-i)+b_j$

F _jrepresent the output activation primitive of hidden layer, w _jk, b _krepresent that hidden layer is to the weights of output layer and deviation respectively, w _{j, i}, represent the weights of input layer to hidden layer, b _jrepresent the deviation of input layer to hidden layer, net _jrepresent the long-pending summation of the weights that all inputs are corresponding with it.

See Fig. 2, Fig. 3, realize being reached by wide area PREDICTIVE CONTROL criterion to the on-line control of PI.Traditional its parameters of PI controller is constant when being, but is all non linear system in Practical Project, becomes and can realize online modification in whole control procedure when usually all requiring that PI parameter is.Wide-area control criterion GPCcriterion is applied for realizing this function.It is as follows that it shifts process onto:

1) traditional PI controller can be written as:

Wherein K _p, K _irepresent scale amplifying multiple and integration time constant respectively.

The Discrete PI of incremental form controls to be expressed as:

Δu(s)=u(s)-u(s-1)=(K _p+K _i)e(s)+(-K _p)e(s-1)=k ₀e(s)+k ₁e(s-1)=k(s).e ^T(s)

Wherein k (s)=[k ₀k ₁], e (s)=[e (s) e (s-1)], k ₀=K _p+ K _i; k ₁=-K _p

2) target function of wide-area control criterion applied of this control system is as follows:

$J=\underset{k=N_1}{\overset{N_2}{\mathrm{\Σ}}}{[y_r(s+k)-y_{n-\mathrm{inst}}(s+k)]}^2+\underset{k=1}{\overset{N_u}{\mathrm{\Σ}}}{\mathrm{\λ}}_k{\left[\mathrm{\Δu}(s+k-1)\right]}^2$

Wherein, N ₁: minimum prediction time domain length

N ₂: maximum predicted time domain length

N _u: control time domain length

Y _r(s+k): the Expected Response in (s+k) moment

Y _n-inst(s+k): the output of (s+k) moment neural network prediction

Δ u (.): the control variables minimizing target function J

λ _k: weight

3) by known Δ u (s+k-1)=k (s) .e of analysis above ^t(s); K (s)=[k ₀k ₁], consider that k (s) is for vector, order $\frac{\∂J}{\∂k\left(s\right)}=0,$ Just obtain $k\left(s\right)={\left[e^T(a_1^T{a}_1+\mathrm{\λ})e\right]}^{-1}\left({\mathrm{Ca}}_1{e}^T\right),$ Namely K is obtained _p, K _iupdated value.Which achieves the On-line Control ability of PI controller.

See Fig. 6, whole computational process can be described below:

The first step: utilize Ziegler-Nichols setting method to determine the parameter K of PI controller _p, K _i.

Second step: establish s=1, determines the output of neural net.S represents current time.

3rd step: the error between calculation expectation value and predicted value, if error is zero just keep adjuster K _p, K _iparameter constant; If error is non-vanishing, carry out the 4th step.Desired value is the harmonic signal that compensates of the needs that calculate and without function signal.Predicted value is the output valve of neural network prediction.

4th step: utilize determine k (s).

5th step: parameter k (s) being converted to PI controller.

6th step: make s=s+1, jumps to the 3rd step.

Claims (5)

1. a power electronics hybrid system control method, utilize power electronics hybrid control system, power electronics hybrid control system includes active power filter, Static Var Compensator, described Static Var Compensator comprises passive power filter group and thyristor-controlled reactor, Active Power Filter-APF, passive power filter group, thyristor-controlled reactor successively and access electrical network and between the load be connected with electrical network, it is characterized in that, described power electronics hybrid system control method comprises Control Method of Active Power Filter and Static Var Compensator control method:

Described Control Method of Active Power Filter comprises the following steps:

Detect load current , , , the actual electric current be injected in three phase network of Active Power Filter-APF , , , computational load harmonic current , , ;

Utilize -Smith Compensator is carried out system compensation of delay, make , , with , , between phase difference be ;

Use PI controller regulating load harmonic current , , the electric current that be injected in three phase network actual in Active Power Filter-APF , , between error, produce the drive singal of IGBT in Active Power Filter-APF with triangle wave;

Using the signal after PI regulates and the actual electric current of three phase network of injecting of Active Power Filter-APF as the input of prediction neural network, obtain predicted value ; represent that s-i moment Active Power Filter-APF is injected into the electric current in electrical network, represent the output that neural network prediction obtains;

According to load harmonic current , , the electric current that inject actual in Active Power Filter-APF , , between error , PI regulate after signal variable quantity and the output of neural network prediction , utilize wide area PREDICTIVE CONTROL criterion, calculate the updated value of PI controller parameter , , realize the on-line control of PI controller;

Described Static Var Compensator control method comprises the following steps:

Detect load current , , and three-phase power grid voltage , , , calculate the reactive current of three phase network , , ;

The rated voltage that load normally works is set as reference voltage , reference voltage with the three-phase power grid voltage detected , , between error regulate with PI controller, after linearized link, produce the control signal of thyristor in thyristor-controlled reactor;

Signal after PI is regulated with the three-phase power grid voltage detected , , as the input of prediction neural network, obtain the three-phase power grid voltage output valve predicted;

Signal after regulating according to PI variable quantity, reference voltage with the output of neural network prediction, utilize wide area PREDICTIVE CONTROL criterion, calculate the updated value of PI controller , , realize the on-line control of PI controller.

2. power electronics hybrid system control method according to claim 1, is characterized in that, described Active Power Filter-APF is injection active power filter.

3. power electronics hybrid system control method according to claim 1, is characterized in that, the target function of described wide area PREDICTIVE CONTROL criterion is:

，

Wherein, for minimum prediction time domain length, for maximum predicted time domain length, for controlling time domain length, for the Expected Response in (s+k) moment, for the output of (s+k) moment neural network prediction, for the control variables of J, for weight.

4. power electronics hybrid system control method according to claim 1, is characterized in that, described passive power filter group is connected to form by some groups of single tuned filters.

5. power electronics hybrid system control method according to claim 1, is characterized in that, the updated value of described calculating PI controller , concrete steps be:

1): utilize Ziegler-Nichols setting method to determine the parameter of PI controller , ;

2) output of neural net: establish s=1, is determined; Wherein s represents current time;

3): the error between calculation expectation value and predicted value, if error is zero just keep PI controller , parameter constant; If error is non-vanishing, enter 4); Its expected value is the harmonic signal that compensates of the needs that calculate and without function signal; Predicted value is the output valve of neural network prediction;

4): utilize , determine ; Wherein , ; J is the target function of wide-area control criterion;

5): will convert the parameter of PI controller to , ;

6): make s=s+1,3 are jumped to).