CN105262113A - Photovoltaic power generation system reactive power control method based on probabilistic fuzzy neural network - Google Patents

Abstract

The invention discloses a photovoltaic power generation system reactive power control method based on a probabilistic fuzzy neural network. The method comprises the following steps: S1, a photovoltaic power generation system mathematical model is built, and maximum allowable values of active power and reactive power injected into a power grid by the photovoltaic power generation system are solved; S2, a power grid fault controller model for the photovoltaic power generation system is built; S3, a probabilistic fuzzy neural network controller is built, and reference values of active current and reactive current injected to the power grid by a three-phase inverter are solved; S4, an error back propagation learning algorithm mechanism for the probabilistic fuzzy neural network controller is built; and S5, a Boost chopper circuit inner loop controller model and a three-phase inverter inner loop current control model are built. In conditions of power grid voltage mutation and fall, the working mode of the photovoltaic power generation system can be quickly adjusted so as to be adaptive to limitations of the photovoltaic array maximum output power, grid-connected inverter rated capacity and the maximum output current, and stability is strong, and the tracking speed is quick.

Description

Based on the photovoltaic generating system powerless control method of Probabilistic Fuzzy neural net

Technical field

The invention belongs to solar photovoltaic technology field, relate to a kind of grid-connected photovoltaic power generation system reactive power control method, be specifically related to a kind of stage type grid-connected photovoltaic power generation system reactive power control method based on Probabilistic Fuzzy neural net.

Technical background

A large amount of photovoltaic generating system access electrical network, the method fast photovoltaic generating system being cut out electrical network due to electric network fault can not meet the demands at present, suddenly Large Copacity photovoltaic is cut out system and can cause serious impact to network system, even cause mains breakdown.In order to ensure that photovoltaic generating system does not depart from electrical network when fault, photovoltaic generating system is needed to have certain low voltage crossing (LowVoltageRide-Through, LVRT) ability.Under line voltage normal operational condition, photovoltaic combining inverter adopts conventional voltage/current double closed-loop control strategy to realize being incorporated into the power networks of photovoltaic generating system usually.When line voltage generation three-phase symmetrical falls, traditional double-loop control and LVRT control strategy is adopted mutually to switch to realize the low voltage crossing of photovoltaic generating system.

Because line voltage symmetry is fallen containing negative sequence component, combining inverter only need be suppressed to export electric current, prevent overcurrent protection action, realize its low voltage crossing when line voltage symmetry is fallen; But, in electrical network actual motion, most fault is unbalanced fault, comprise single-phase earthing, phase fault, double earthfault etc., when line voltage occur asymmetric fall fault time, if adopt the low voltage crossing control strategy of traditional three-phase voltage symmetry, because negative-sequence current exists, the grid-connected power of combining inverter and photovoltaic array send unbalanced power, photovoltaic combining inverter DC voltage falls and fluctuates widely, affect the stable operation of photovoltaic generating system, even cause its off-grid.

Patent of invention CN102856916B " a kind of single-phase photovoltaic inverter powerless control method and circuit " proposes a kind of powerless control method of single-phase photovoltaic inverter, this method does not consider the restriction of inverter capacity and output current maximum permissible value, feasibility under grid fault conditions is worth discussion, at electrical network generation single-phase fault, in phase to phase fault and three-phase fault situation, inverter output current can increase considerably, inverter components and parts overcurrent is easily caused to burn on the one hand, another aspect photovoltaic array power output is injected grid power with inverter and is not mated and will cause inverter dc-link capacitance overvoltage, thus cause photovoltaic DC-to-AC converter relaying protection switch trip, affect the safe and stable operation of electrical network.Application number is the patent of invention " a kind of powerless control method of non-isolated single-phase photovoltaic grid-connected inverter " of 201410189148.4, adopt DSP control chip sampling line voltage and inverter output current, realize the function of the idle output of lead and lag, the method does not consider the restriction of inverter output current restriction and photovoltaic array power output under grid fault conditions, is difficult to meet national standard and accesses to photovoltaic DC-to-AC converter the technical requirement that electrical network low voltage passes through.

Summary of the invention

The object of the invention is to overcome the deficiency of traditional photovoltaic generating system at fault ride-through of power grid control strategy, there is provided a kind of in grid voltage mutation and the mode of operation that can adjust photovoltaic generating system under falling situation rapidly, to adapt to the restriction of photovoltaic array peak power output and combining inverter rated capacity and maximum output current, there is the photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net of the advantages such as stability is strong, tracking velocity is fast.

The object of the invention is to be achieved through the following technical solutions: based on the photovoltaic generating system powerless control method of Probabilistic Fuzzy neural net, comprise the following steps:

S1, set up photovoltaic generating system Mathematical Modeling, carry out power calculation and genlock, ask for the maximum permissible value of active power that photovoltaic generating system injects to electrical network and reactive power;

S2, set up the electric network fault controller model of photovoltaic generating system, according to the maximum permissible value of active power, reactive power and three-phase inverter electric current, set up double mode switching control strategy, dynamically adjust the mode of operation of Boost boost chopper and the Probabilistic Fuzzy ANN Control input reference signal of three-phase inverter;

S3, set up Probabilistic Fuzzy nerve network controller, ask for active current and reactive current reference value that three-phase inverter injects electrical network;

S4, set up Probabilistic Fuzzy nerve network controller error back-propagating learning algorithm mechanism, construct a gradient vector, make wherein each element be the first differential of energy function relative to algorithm parameter, thus complete the parameters on line modifying of Probabilistic Fuzzy neural net;

S5, set up Boost boost chopper inner ring controller model, in Boost boost chopper, the output signal of ring controller compares with triangular carrier and forms the pwm pulse control signal of Boost boost chopper switching device; Set up three-phase inverter inner ring current controller model, realize the tracing control of output current to reference signal, the output signal of three-phase inverter inner ring current controller is compared with triangular carrier and forms the pwm pulse control signal of each switching device of three-phase inverting circuit, thus the target that under realizing photovoltaic generating system cutting-in control and grid fault conditions, dynamic reactive supports.

Further, the main circuit of described photovoltaic generating system comprises photovoltaic array, Boost boost chopper, three-phase inverter, grid-connected reactor L, grid-connected switch S SR, grid-connected transformer and three phase network, and photovoltaic array is successively by Boost boost chopper, three-phase inverter, grid-connected reactor L, grid-connected switch S SR and grid-connected transformer access three phase network; C _pVfor photovoltaic array output port shunt capacitance, simultaneously for Boost boost chopper provides input voltage; V _pVand I _pVbe respectively output voltage and the output current of photovoltaic array; DC capacitor C _dcbe connected on Boost boost chopper output, C _dcsimultaneously as the dc-link capacitance of three-phase inverter, its operating voltage is V _dc; Three-phase inverter is by grid-connected reactor L, grid-connected switch S SR and grid-connected transformer access three-phase alternating current electrical network;

Described step S1 specifically comprises following sub-step:

S11, three-phase inverter exchange the relation exporting phase voltage and line voltage and are expressed as:

$\left[\begin{array}{c}{v}_a\\ v_b\\ v_c\end{array}\right]=\frac{1}{3}\left[\begin{array}{ccc}1& 0& -1\\ -1& 1& 0\\ 0& -1& 1\end{array}\right]\left[\begin{array}{c}{v}_{ab}\\ v_{bc}\\ v_{ca}\end{array}\right]---\left(1\right)$

Wherein, [v _ab, v _bc, v _ca] ^tfor three-phase inverter exchanges output line voltage, [v _a, v _b, v _c] ^tphase voltage is exported for three-phase inverter exchanges;

S12, employing Clark conversion, be transformed into static α β coordinate system, be expressed as by the phase voltage of abc coordinate system:

$\left[\begin{array}{c}{v}_{\mathrm{\α}}\\ v_{\mathrm{\β}}\end{array}\right]=\sqrt{\frac{2}{3}}\left[\begin{array}{ccc}1& -\frac{1}{2}& -\frac{1}{2}\\ 0& \frac{\sqrt{3}}{2}& -\frac{\sqrt{3}}{2}\end{array}\right]\left[\begin{array}{c}{v}_a\\ v_b\\ v_c\end{array}\right]---\left(2\right)$

Wherein, v _α, v _βbe respectively the component of three-phase inverter output voltage α axle and β axle under static α β coordinate system;

S13, employing Park conversion, by the output voltage [v of static α β coordinate system _α, v _β] ^tbe transformed into synchronous rotating reference frame, be expressed as:

$\left[\begin{array}{c}{v}_d\\ v_q\end{array}\right]=\left[\begin{array}{cc}cos\left({\mathrm{\θ}}_e\right)& sin\left({\mathrm{\θ}}_e\right)\\ -sin\left({\mathrm{\θ}}_e\right)& cos\left({\mathrm{\θ}}_e\right)\end{array}\right]\left[\begin{array}{c}{v}_{\mathrm{\α}}\\ v_{\mathrm{\β}}\end{array}\right]---\left(3\right)$

Wherein, v _d, v _qbe respectively the component of three-phase inverter output voltage d axle and q axle under dq synchronous coordinate system;

S14, employing Park conversion, by three-phase inverter output current [i _a, i _b, i _c] ^tbe transformed into dq synchronous coordinate system, be expressed as:

$\left[\begin{array}{c}{i}_d\\ i_q\end{array}\right]=\sqrt{\frac{2}{3}}\left[\begin{array}{ccc}\cos \left({\mathrm{\θ}}_e\right)& \cos ({\mathrm{\θ}}_e-\frac{2}{3}\mathrm{\π})& \cos ({\mathrm{\θ}}_e+\frac{2}{3}\mathrm{\π})\\ -\sin \left({\mathrm{\θ}}_e\right)& -\sin ({\mathrm{\θ}}_e-\frac{2}{3}\mathrm{\π})& -\sin ({\mathrm{\θ}}_e+\frac{2}{3}\mathrm{\π})\end{array}\right]\left[\begin{array}{c}{i}_a\\ i_b\\ i_c\end{array}\right]---\left(4\right)$

Wherein, i _d, i _qbe respectively d axle and q axle power network current component under dq synchronous coordinate system, i _qcomponent is used for controlling the real component injected to electrical network of photovoltaic generating system, i _dcomponent is used for controlling the idle component injected to electrical network of photovoltaic generating system;

S15, the data obtained according to step S13 and S14, the instantaneous active power P injected to electrical network by photovoltaic generating system and instantaneous reactive power Q is expressed as:

$P=\frac{3}{2}(v_d{i}_d+v_q{i}_q);Q=\frac{3}{2}(v_q{i}_d-v_d{i}_q)---\left(5\right)$

After genlock (PLL), three-phase inverter output voltage phasor and line voltage phasor same-phase, i.e. v _d=0, then instantaneous active power and instantaneous reactive power are expressed as:

$P=\frac{3}{2}{v}_q{i}_q;Q=\frac{3}{2}{v}_q{i}_d---\left(6\right);$

S16, establish V _sagrepresent the perunit value of Voltage Drop amplitude, its scope is [0,1], then photovoltaic generating system needs the reactive current reference value injecting electrical network to be expressed as:

$I_r^{*}=\{\begin{array}{cc}0\%,& V_{sag}\&le;0.1\\ 200V_{sag}\%,& 0.1<V_{sag}\&le;0.5\\ 100\%,& V_{sag}>0.5\end{array}---\left(7\right)$

$V_{sag}=1-\frac{\min ({\left|v_a\right|}_{rms},{\left|v_b\right|}_{rms},{\left|v_c\right|}_{rms})}{V_{base}}---\left(8\right)$

Wherein, V _baserepresent voltage base value, v _{a, rms}, v _{b, rms}, v _{c, rms}be respectively a, b, c phase voltage v _a, v _b, v _ceffective value;

Work as V _sagwhen being greater than 0.1pu, photovoltaic generating system starts the reactive power support and control pattern of electrical network; Work as V _sagwhen being less than or equal to 0.1, reactive current reference value is 0, and system is in conventional maximum power output pattern, and photovoltaic generating system only injects active current and active power to electrical network;

After reactive power support and control pattern starts, in order to meet low voltage ride through of photovoltaic inverter (LVRT) requirement, and avoid inverter output current to exceed device rated current causing component damage, the apparent power maximum of three-phase inverter is expressed as:

|S|＝(|v _a| _rms+|v _b| _rms+|v _c| _rms)I _max(9)

Wherein, I _maxrepresent that photovoltaic generating system injects the maximum permissible value of power network current;

Therefore, the maximum permissible value of active power and reactive power is expressed as:

$P^{*}=\left|S\right|\sqrt{1-I_r^{*2}};Q^{*}=\left|S\right|I_r^{*}---\left(10\right)$

Above-mentioned active power and reactive power maximum permissible value P ^*, Q ^*using injecting the reference value of active power and reactive power as photovoltaic generating system to electrical network, after asking for deviation respectively with the actual active power of photovoltaic generating system and reactive power P, Q, as the input signal of Probabilistic Fuzzy nerve network controller.

Further, the double mode switching control strategy in described step S2 is specially:

Pattern I: work as P ^*be greater than the active power rate P of photovoltaic array _pVtime, by controlling the DC bus-bar voltage of three-phase inverter, by the active power rate P of photovoltaic array _pVwhole injection electrical network; Three-phase inverter controlled current flow is built in I _maxwithin, Boost boost chopper is operated in MPPT maximum power point tracking pattern (MPPT);

Pattern II: work as P ^*be less than or equal to the active power rate P of photovoltaic array _pVtime, Boost boost chopper suspends MPPT maximum power point tracking, and starts to follow the tracks of power P ^*; The unbalanced problem of power between Boost boost chopper and three-phase inverter solves by controlling three-phase inverter DC bus-bar voltage.

Further, the Probabilistic Fuzzy nerve network controller in described S3 comprises 6 layer network structures: the 1st layer be input layer, the 2nd layer be degree of membership layer, the 3rd layer be probability layer, the 4th layer be TSK indistinct logic computer preparative layer, the 5th layer be rules layer, the 6th layer for output layer; In described degree of membership layer, each node adopts asymmetric Gaussian function to realize obfuscation computing;

The calculation process of described Probabilistic Fuzzy nerve network controller is:

A definition jth fuzzy If-Then Rule Expression is as follows:

Rule j: if $x_1-M_1^j,x_2-M_2^j,$ Then $T_k=\underset{i}{\&Sigma;}{c}_{ik}{x}_i$ (11)

Wherein, x _i, i=1,2, be the input of Probabilistic Fuzzy nerve network controller, with for fuzzy set, T _kfor TSK Fuzzy inferential engine, c _ikfor adjustable weight coefficient;

1st layer (input layer): the node of input layer is by input variable x _i, i=1,2, be delivered to the 2nd layer, the relation between the node input signal of input layer and output signal is as follows:

${\mathrm{net}}_i^1\left(N\right)=x_i^1,y_i^1\left(N\right)=f_i^1\left({\mathrm{net}}_i^1\right(N\left)\right)={\mathrm{net}}_i^1\left(N\right),i=1,2---\left(12\right)$

Wherein, n represents iterations, represent the 1st node layer output signal; Under grid fault conditions, for the active power controller function of pattern I, input variable for the active power controller function of pattern II, input variable e=P ^*-P; For Reactive Power Control function, input variable e=Q ^*-Q;

2nd layer (degree of membership layer): this layer of each node adopts asymmetric Gaussian function to realize obfuscation computing, the relation between node input signal and output signal is expressed as follows:

${\mathrm{net}}_j^2\left(N\right)=\left\{\begin{array}{cc}-\frac{{\left(y_i^1\right(N)-m_j^2(N\left)\right)}^2}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right(N\left)\right)}^2},& -\mathrm{\&infin;}<y_i^1\left(N\right)\&le;m_j^2\\ -\frac{{\left(y_i^1\right(N)-m_j^2(N\left)\right)}^2}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right(N\left)\right)}^2},& m_j^2<y_i^1\left(N\right)<\mathrm{\&infin;}\end{array}\right.---\left(13\right)$

$y_i^2\left(N\right)=f_j^2\left({\mathrm{net}}_j^2\right(N\left)\right)=\exp \left({\mathrm{net}}_j^2\right(N\left)\right)---\left(14\right)$

Wherein, be the average of i-th asymmetric Gaussian function of input variable jth item, with the left standard difference and the right standard that are respectively i-th asymmetric Gaussian function of input variable jth item are poor, it is the output variable of the 2nd node layer;

3rd layer (probability layer): the node input/output relation of the 3rd layer is as follows:

$\begin{array}{c}{P}_{jp}\left(N\right)=f_{jp}\left(y_j^2\right(N\left)\right)=\exp \&lsqb;-\frac{{\left(y_j^2\right(N)-m_{jp}^3)}^2}{{\left({\mathrm{\&sigma;}}_{jp}^3\right)}^2}\&rsqb;\\ j=1,2,\mathrm{...},6;p=1,2,3\end{array}---\left(15\right)$

Wherein, P _jp(N) corresponding to the output signal of a jth input variable p node, with correspond respectively to average and the standard deviation of jth input variable p the asymmetric Gaussian function of node; In order to reduce amount of calculation, can be by with be arranged to constant, as p=1, order as p=2, order as p=3, order

4th layer (TSK indistinct logic computer preparative layer): in this layer, output signal the linear combination into input signal, the output signal of a kth node is:

$T_k\left(N\right)=\underset{i}{\&Sigma;}{c}_{ik}\left(N\right)x_i\left(N\right),i=1,2;k=1,2,\mathrm{...},9---\left(16\right)$

Wherein, c _ikfor adjustable weight coefficient, x _ifor input variable, N is iterations;

5th layer (rules layer): this layer of Part I is layers 2 and 3 output signal node and P _jp(N) product, is designated as ∏, and therefore the output signal of this layer of Part I kth node can be expressed as:

$\begin{array}{c}{y}_k^I\left(N\right)=y_r^2\left(N\right)y_l^2\left(N\right)S_r\left(N\right)S_l\left(N\right),r=1,2,3\\ l=4,5,6;k=3(r-1)+(l-3)\end{array}---\left(17\right)$

$S_j\left(N\right)=\underset{p}{\&Pi;}{P}_{jp}\left(N\right),j=1,\mathrm{2...},6;p=1,2,3---\left(18\right)$

Wherein, S _j(N) be the 3rd node layer output signal P _jp(N) product, corresponding to a jth node of the 2nd layer;

This layer of Part II is the 5th layer of output signal expression formula, is Part I output signal t is outputed signal with the 4th layer _k(N) product, therefore the output signal of a kth node can be expressed as:

$y_k^O\left(N\right)=T_k\left(N\right)y_k^I\left(N\right),k=1,2,\mathrm{...},9---\left(19\right)$

Wherein, represent the output signal of a rules layer kth node;

6th layer (output layer): this layer is made up of a node O, calculate the weighted accumulation effect of all upper layer node output signal, its Mathematical Modeling is expressed as:

${\mathrm{net}}_o^6\left(N\right)=\underset{k}{\&Sigma;}{w}_k^6\left(N\right)y_k^O\left(N\right),o=1;k=1,2,\mathrm{...},9---\left(20\right)$

$y_o^6\left(N\right)=f_o^6\left({\mathrm{net}}_o^6\right(N\left)\right)={\mathrm{net}}_o^6\left(N\right),o=1---\left(21\right)$

Wherein, represent that a kth obfuscation rule is to the weight coefficient of o output signal action intensity, it is a kth input signal of the 6th node layer; for photovoltaic DC-to-AC converter injects the active current of electrical network, for photovoltaic DC-to-AC converter injects the active current of electrical network.

Further, described step S4 concrete methods of realizing is: be defined as follows energy function:

$E\left(N\right)=\frac{1}{2}{\left(y^{*}\right(N)-y(N\left)\right)}^2=\frac{1}{2}{e}^2\left(N\right)---\left(22\right)$

Wherein, the tracking error that E (N) is Probabilistic Fuzzy nerve network controller the N time iteration, y ^*(N) and y (N) desired output that represents the N time iteration controller respectively export with actual;

Described Probabilistic Fuzzy nerve network controller parameters on line modifying comprises the following steps:

S41, for Probabilistic Fuzzy neural network algorithm the 6th layer (output layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_o^6=-\frac{\&part;E}{\&part;y_o^6\left(N\right)}=-\frac{\&part;E}{\&part;y}\frac{\&part;y}{\&part;y_o^6\left(N\right)}---\left(23\right)$

Need the weight coefficient that iteration upgrades be expressed as:

${\mathrm{\&Delta;w}}_k^6=-{\mathrm{\&eta;}}_1\frac{\&part;E}{\&part;w_k^6\left(N\right)}=-{\mathrm{\&eta;}}_1\frac{\&part;E}{\&part;y_o^6\left(N\right)}\frac{\&part;y_o^6\left(N\right)}{\&part;w_k^6\left(N\right)}={\mathrm{\&eta;}}_1{\mathrm{\&delta;}}_o^6{y}_k^O---\left(24\right)$

Wherein, η ₁represent learning rate, the N+1 time and the N time weight coefficient iterative relation as follows:

$w_k^6(N+1)=w_k^6\left(N\right)+{\mathrm{\&Delta;w}}_k^6---\left(25\right);$

S42, for Probabilistic Fuzzy neural network algorithm the 5th layer (rules layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_k^O=\frac{\&part;E}{\&part;y_k^O\left(N\right)}=-\frac{\&part;E}{\&part;y_o^6\left(N\right)}\frac{\&part;y_o^6\left(N\right)}{\&part;y_k^O\left(N\right)}={\mathrm{\&delta;}}_o^6{w}_k^6---\left(26\right)$

${\mathrm{\&delta;}}_k^I=\frac{\&part;E}{\&part;y_k^I\left(N\right)}=-\frac{\&part;E}{\&part;y_k^O\left(N\right)}\frac{\&part;y_k^O\left(N\right)}{\&part;y_k^I\left(N\right)}={\mathrm{\&delta;}}_k^O{T}_k---\left(27\right)$

S43, for Probabilistic Fuzzy neural network algorithm the 4th layer (TSK indistinct logic computer preparative layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_k^4=-\frac{\&part;E}{\&part;T_k\left(N\right)}=-\frac{\&part;E}{\&part;y_k^O\left(N\right)}\frac{\&part;y_k^O\left(N\right)}{\&part;T_k\left(N\right)}={\mathrm{\&delta;}}_k^O{y}_k^I---\left(28\right)$

The weight coefficient c needing iteration to upgrade _ikbe expressed as:

${\mathrm{\&Delta;c}}_{ik}=-{\mathrm{\&eta;}}_2\frac{\&part;E}{\&part;c_{ik}\left(N\right)}=-{\mathrm{\&eta;}}_2\frac{\&part;E}{\&part;T_k\left(N\right)}\frac{\&part;T_k\left(N\right)}{\&part;c_{ik}\left(N\right)}={\mathrm{\&eta;}}_2{\mathrm{\&delta;}}_k^4{x}_i---\left(29\right)$

Wherein, η ₂represent learning rate, the N+1 time and the N time weight coefficient c _ikiterative relation as follows:

c _ik(N+1)＝c _ik(N)+Δc _ik(30)

S44, for Probabilistic Fuzzy neural network algorithm the 2nd layer (degree of membership layer), the error of back-propagating is needed to be expressed as:

$\begin{array}{c}{\mathrm{\&delta;}}_j^2=-\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}=\frac{\&part;E}{\&part;y_k^I\left(N\right)}\frac{\&part;y_k^I\left(N\right)}{\&part;y_j^2\left(N\right)}\frac{\&part;y_j^2\left(N\right)}{\&part;{\mathrm{net}}_j^2\left(N\right)}\\ =\left\{\begin{array}{c}{h}_j\underset{r}{\&Sigma;}{\mathrm{\&delta;}}_k^I{y}_k^I,j=1,2,3;r=1,2,3;k=3(j-1)+r\\ h_j\underset{r}{\&Sigma;}{\mathrm{\&delta;}}_k^I{y}_k^I,j=4,5,6;r=1,2,3;k=j+3(r-2)\end{array}\right.\end{array}---\left(31\right)$

$h_j=1-y_j^2\underset{p}{\&Sigma;}\frac{y_j^2-m_{jp}^3}{{\left({\mathrm{\&sigma;}}_{jp}^3\right)}^2},p=1,2,3---\left(32\right)$

The asymmetric Gaussian function average needing iteration to upgrade be expressed as:

$\begin{array}{c}{\mathrm{\&Delta;m}}_j^2=-{\mathrm{\&eta;}}_3\frac{\&part;E}{\&part;m_j^2\left(N\right)}=-{\mathrm{\&eta;}}_3\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;m_j^2\left(N\right)}\\ =\left\{\begin{array}{cc}{\mathrm{\&eta;}}_3{\mathrm{\&delta;}}_j^2\frac{2(y_i^1-m_j^2)}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right)}^2},& -\mathrm{\&infin;}<y_i^1\&le;m_j^2,j=1,2,\mathrm{...},6\\ {\mathrm{\&eta;}}_3{\mathrm{\&delta;}}_j^2\frac{2(y_i^1-m_j^2)}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right)}^2},& m_j^2<y_i^1<\mathrm{\&infin;},j=1,2,\mathrm{...6}\end{array}\right.\end{array}---\left(33\right)$

Wherein, η ₃represent the learning rate of algorithm; Asymmetric Gaussian function average iterative relation as follows:

$m_j^2(N+1)=m_j^2\left(N\right)+{\mathrm{\&Delta;m}}_j^2---\left(34\right)$

The left standard of the asymmetric Gaussian function needing iteration to upgrade is poor and right standard is poor with be expressed as follows:

$\begin{array}{c}{\mathrm{\&sigma;}}_{L\_j}^2=-{\mathrm{\&eta;}}_4\frac{\&part;E}{\&part;{\mathrm{\&sigma;}}_{L\_j}^2}=-{\mathrm{\&eta;}}_4\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;{\mathrm{\&sigma;}}_{L\_j}^2\left(N\right)}\\ ={\mathrm{\&eta;}}_4{\mathrm{\&delta;}}_j^2\frac{2{(y_i^1-m_j^2)}^2}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right)}^3},j=1,2,\mathrm{...},6\end{array}---\left(35\right)$

$\begin{array}{c}{\mathrm{\&sigma;}}_{R\_j}^2=-{\mathrm{\&eta;}}_5\frac{\&part;E}{\&part;{\mathrm{\&sigma;}}_{R\_j}^2}=-{\mathrm{\&eta;}}_5\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;{\mathrm{\&sigma;}}_{R\_j}^2\left(N\right)}\\ ={\mathrm{\&eta;}}_5{\mathrm{\&delta;}}_j^2\frac{2{(y_i^1-m_j^2)}^2}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right)}^3},j=1,2,\mathrm{...},6\end{array}---\left(36\right)$

Wherein, η ₄, η ₅represent the learning rate of algorithm;

The left standard of asymmetric Gaussian function is poor and right standard is poor with rule of iteration as follows:

${\mathrm{\&sigma;}}_{L\_j}^2(N+1)={\mathrm{\&sigma;}}_{L\_j}^2\left(N\right)+{\mathrm{\&Delta;\&sigma;}}_{L\_j}^2---\left(37\right)$

${\mathrm{\&sigma;}}_{R\_j}^2(N+1)={\mathrm{\&sigma;}}_{R\_j}^2\left(N\right)+{\mathrm{\&Delta;\&sigma;}}_{R\_j}^2---\left(38\right)$

According to above-mentioned steps, by parameter with repetitive exercise is carried out, parameter respectively according to formula (25), (30), (34), (37), (38) with initial value is set to 0 respectively, and 1,1,1; Parameter initial value is set to-1,0 respectively, and 1 ,-1,0,1; Learning rate η ₁~ η ₅initial value is zero.

Further, ring controller adoption rate integral controller in Boost boost chopper in described step S5, the input signal of pi controller selects input signal by switch according to the output mode signal of electric network fault controller, system works when pattern I, by the output signal under MPPT maximum power point tracking pattern (MPPT) with the real output signal V of photovoltaic array _pVask deviation, using the input signal that deviation signal controls as Boost boost chopper inner ring; When system works is at pattern II, by three-phase inverter DC bus-bar voltage reference value with actual value V _dcthe input signal controlled as Boost boost chopper inner ring after asking for deviation; In Boost boost chopper, the output signal of ring controller compares with triangular carrier and forms the pwm pulse control signal of Boost boost chopper switching device.

Three-phase photovoltaic inverter inner ring current controller model in described step S5, adoption rate integral control, realizes output current i _a, i _b, i _cto reference signal tracing control, the output signal of three-phase photovoltaic inverter inner ring current controller is compared with triangular carrier and forms the pwm pulse control signal of each switching device of three-phase inverting circuit, thus realize photovoltaic generating system cutting-in control and under grid fault conditions to the target that electrical network dynamic reactive power supports.

The invention has the beneficial effects as follows: instant invention overcomes the deficiency of traditional photovoltaic generating system at fault ride-through of power grid control strategy, with grid voltage sags degree, inverter output current limits value for constraints, according to the maximum permissible value scope of active power, reactive power and three-phase inverter electric current, set up the double mode switching control strategy of fault ride-through of power grid; Set up the calculation process of Probabilistic Fuzzy nerve network controller, all adopt asymmetric Gaussian function to realize obfuscation computing at each node of degree of membership layer, improve traditional fuzzy neural network to model parameter than more sensitive deficiency; Set up the error back-propagating learning algorithm mechanism of Probabilistic Fuzzy nerve network controller, complete the parameters on line modifying of Probabilistic Fuzzy neural net; Set up Boost boost chopper inner ring controller model and three-phase photovoltaic inverter inner ring current diffusion limited model simultaneously, inner ring controller output signal is compared the pwm pulse control signal forming each switching device with triangular carrier, realize the target that dynamic reactive supports under photovoltaic generating system cutting-in control and grid fault conditions.This control method is in grid voltage mutation and the mode of operation that can adjust photovoltaic generating system under falling situation rapidly, to adapt to the restriction of photovoltaic array peak power output and combining inverter rated capacity and maximum output current, have that stability is strong, tracking velocity be fast, reactive power and the control effects such as inverter DC bus-bar voltage overshoot is little, for photovoltaic generating system cutting-in control and fault traversing Controller gain variations provide feasible technological means.

Accompanying drawing explanation

Fig. 1 is control method flow chart of the present invention;

Fig. 2 is photovoltaic power generation system structure figure of the present invention;

Fig. 3 is that the control model of photovoltaic generating system electric network fault controller of the present invention selects flow chart;

Fig. 4 is Probabilistic Fuzzy nerve network controller Organization Chart of the present invention;

Fig. 5 is asymmetric Gauss's membership function schematic diagram of Probabilistic Fuzzy nerve network controller;

Fig. 6 is the design sketch of the embodiment of the present invention based on the photovoltaic generating system control model I of Probabilistic Fuzzy ANN Control;

Fig. 7 is the design sketch of the embodiment of the present invention based on the photovoltaic generating system control model II of Probabilistic Fuzzy ANN Control.

Embodiment

Technical scheme of the present invention is further illustrated below in conjunction with accompanying drawing.

As shown in Figure 1, a kind of photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net of the present invention, comprises the following steps:

S1, set up photovoltaic generating system Mathematical Modeling, carry out power calculation and genlock, according to constraintss such as Voltage Drop degree, inverter output current limits values, ask for the maximum permissible value of active power that photovoltaic generating system injects to electrical network and reactive power;

As shown in Figure 2, the main circuit of described photovoltaic generating system comprises photovoltaic array, Boost boost chopper, three-phase inverter, grid-connected reactor L, grid-connected switch S SR, grid-connected transformer and three phase network, and photovoltaic array is successively by Boost boost chopper, three-phase inverter, grid-connected reactor L, grid-connected switch S SR and grid-connected transformer access three phase network; C _pVfor photovoltaic array output port shunt capacitance, simultaneously for Boost boost chopper provides input voltage; V _pVand I _pVbe respectively output voltage and the output current of photovoltaic array; DC capacitor C _dcbe connected on Boost boost chopper output, C _dcsimultaneously as the dc-link capacitance of three-phase inverter, its operating voltage is V _dc; I _maxrepresent that photovoltaic generating system injects the maximum permissible value of power network current; Three-phase inverter is by grid-connected reactor L, grid-connected switch S SR and grid-connected transformer access three-phase alternating current electrical network;

Described step S1 specifically comprises following sub-step:

S11, three-phase inverter exchange the relation exporting phase voltage and line voltage and are expressed as:

$\left[\begin{array}{c}{v}_a\\ v_b\\ v_c\end{array}\right]=\frac{1}{3}\left[\begin{array}{ccc}1& 0& -1\\ -1& 1& 0\\ 0& -1& 1\end{array}\right]\left[\begin{array}{c}{v}_{ab}\\ v_{bc}\\ v_{ca}\end{array}\right]---\left(1\right)$

Wherein, [v _ab, v _bc, v _ca] ^tfor three-phase inverter exchanges output line voltage, [v _a, v _b, v _c] ^tphase voltage is exported for three-phase inverter exchanges;

S12, employing Clark conversion, be transformed into static α β coordinate system, be expressed as by the phase voltage of abc coordinate system:

$\left[\begin{array}{c}{v}_{\mathrm{\&alpha;}}\\ v_{\mathrm{\&beta;}}\end{array}\right]=\sqrt{\frac{2}{3}}\left[\begin{array}{ccc}1& -\frac{1}{2}& -\frac{1}{2}\\ 0& \frac{\sqrt{3}}{2}& -\frac{\sqrt{3}}{2}\end{array}\right]\left[\begin{array}{c}{v}_a\\ v_b\\ v_c\end{array}\right]---\left(2\right)$

Wherein, v _α, v _βbe respectively the component of three-phase inverter output voltage α axle and β axle under static α β coordinate system;

S13, employing Park conversion, by the output voltage [v of static α β coordinate system _α, v _β] ^tbe transformed into synchronous rotating reference frame, be expressed as:

$\left[\begin{array}{c}{v}_d\\ v_q\end{array}\right]=\left[\begin{array}{cc}cos\left({\mathrm{\&theta;}}_e\right)& sin\left({\mathrm{\&theta;}}_e\right)\\ -sin\left({\mathrm{\&theta;}}_e\right)& cos\left({\mathrm{\&theta;}}_e\right)\end{array}\right]\left[\begin{array}{c}{v}_{\mathrm{\&alpha;}}\\ v_{\mathrm{\&beta;}}\end{array}\right]---\left(3\right)$

Wherein, v _d, v _qbe respectively the component of three-phase inverter output voltage d axle and q axle under dq synchronous coordinate system;

S14, employing Park conversion, by three-phase inverter output current [i _a, i _b, i _c] ^tbe transformed into dq synchronous coordinate system, be expressed as:

$\left[\begin{array}{c}{i}_d\\ i_q\end{array}\right]=\sqrt{\frac{2}{3}}\left[\begin{array}{ccc}\cos \left({\mathrm{\&theta;}}_e\right)& \cos ({\mathrm{\&theta;}}_e-\frac{2}{3}\mathrm{\&pi;})& \cos ({\mathrm{\&theta;}}_e+\frac{2}{3}\mathrm{\&pi;})\\ -\sin \left({\mathrm{\&theta;}}_e\right)& -\sin ({\mathrm{\&theta;}}_e-\frac{2}{3}\mathrm{\&pi;})& -\sin ({\mathrm{\&theta;}}_e+\frac{2}{3}\mathrm{\&pi;})\end{array}\right]\left[\begin{array}{c}{i}_a\\ i_b\\ i_c\end{array}\right]---\left(4\right)$

Wherein, i _d, i _qbe respectively d axle and q axle power network current component under dq synchronous coordinate system, i _qcomponent is used for controlling the real component injected to electrical network of photovoltaic generating system, i _dcomponent is used for controlling the idle component injected to electrical network of photovoltaic generating system;

S15, the data obtained according to step S13 and S14, the instantaneous active power P injected to electrical network by photovoltaic generating system and instantaneous reactive power Q is expressed as:

$P=\frac{3}{2}(v_d{i}_d+v_q{i}_q);Q=\frac{3}{2}(v_q{i}_d-v_d{i}_q)---\left(5\right)$

After genlock (PLL), three-phase inverter output voltage phasor and line voltage phasor same-phase, i.e. v _d=0, then instantaneous active power and instantaneous reactive power are expressed as:

$P=\frac{3}{2}{v}_q{i}_q;Q=\frac{3}{2}{v}_q{i}_d---\left(6\right);$

S16, establish V _sagrepresent the perunit value of Voltage Drop amplitude, its scope is [0,1], then photovoltaic generating system needs the reactive current reference value injecting electrical network to be expressed as:

$I_r^{*}=\{\begin{array}{cc}0\%,& V_{sag}\&le;0.1\\ 200V_{sag}\%,& 0.1<V_{sag}\&le;0.5\\ 100\%,& V_{sag}>0.5\end{array}---\left(7\right)$

$V_{sag}=1-\frac{\min ({\left|v_a\right|}_{rms},{\left|v_b\right|}_{rms},{\left|v_c\right|}_{rms})}{V_{base}}---\left(8\right)$

Wherein, V _baserepresent voltage base value, v _{a, rms}, v _{b, rms}, v _{c, rms}be respectively a, b, c phase voltage v _a, v _b, v _ceffective value;

Work as V _sagwhen being greater than 0.1pu, photovoltaic generating system starts the reactive power support and control pattern of electrical network; Work as V _sagwhen being less than or equal to 0.1, reactive current reference value is 0, and system is in conventional maximum power output pattern, and photovoltaic generating system only injects active current and active power to electrical network;

After reactive power support and control pattern starts, in order to meet low voltage ride through of photovoltaic inverter (LVRT) requirement, and avoid inverter output current to exceed device rated current causing component damage, the apparent power maximum of three-phase inverter is expressed as:

|S|＝(|v _a| _rms+|v _b| _rms+|v _c| _rms)I _max(9)

Wherein, I _maxrepresent that photovoltaic generating system injects the maximum permissible value of power network current;

Therefore, the maximum permissible value of active power and reactive power is expressed as:

$P^{*}=\left|S\right|\sqrt{1-I_r^{*2}};Q^{*}=\left|S\right|I_r^{*}---\left(10\right)$

Above-mentioned active power and reactive power maximum permissible value P ^*, Q ^*using injecting the reference value of active power and reactive power as photovoltaic generating system to electrical network, after asking for deviation respectively with the actual active power of photovoltaic generating system and reactive power P, Q, as the input signal of Probabilistic Fuzzy nerve network controller.

S2, set up the electric network fault controller model of photovoltaic generating system, according to the maximum permissible value of active power, reactive power and three-phase inverter electric current, set up double mode switching control strategy, dynamically adjust the mode of operation of Boost boost chopper and the Probabilistic Fuzzy ANN Control input reference signal of three-phase inverter; Described double mode switching control strategy is specially:

Pattern I: work as P ^*be greater than the active power rate P of photovoltaic array _pVtime, by controlling the DC bus-bar voltage of three-phase inverter, by the active power rate P of photovoltaic array _pVwhole injection electrical network; Three-phase inverter controlled current flow is built in I _maxwithin, Boost boost chopper is operated in MPPT maximum power point tracking pattern (MPPT);

Pattern II: work as P ^*be less than or equal to the active power rate P of photovoltaic array _pVtime, Boost boost chopper suspends MPPT maximum power point tracking, and starts to follow the tracks of power P ^*; The unbalanced problem of power between Boost boost chopper and three-phase inverter solves by controlling three-phase inverter DC bus-bar voltage.

Electric network fault controller selects mode of operation by interrupteur SW 3 (as shown in Figure 2).As shown in Figure 3, the concrete operations flow process of double mode switching control strategy of the present invention comprises the following steps:

S21, reading three-phase inverter exchange output line voltage v _ab, v _bc, v _ca, photovoltaic generating system injects the maximum permissible value I of power network current _max, calculate the active power rate P of photovoltaic array _pV, and calculate V _sag;

S22, judge V _sagwhether be greater than 0.1, if so, photovoltaic generating system starts the reactive power support and control pattern of electrical network, carries out next step, otherwise inoperation;

S23, calculating | S|, P ^*, Q ^*;

S24, judge P _pVwhether be less than or equal to P ^*if, then Dietary behavior I, otherwise Dietary behavior II.

S3, set up Probabilistic Fuzzy nerve network controller, ask for active current and reactive current reference value that three-phase inverter injects electrical network; As shown in Figure 4, described Probabilistic Fuzzy nerve network controller comprises 6 layer network structures: the 1st layer be input layer, the 2nd layer be degree of membership layer, the 3rd layer be probability layer, the 4th layer be TSK indistinct logic computer preparative layer, the 5th layer be rules layer, the 6th layer for output layer; In described degree of membership layer, each node adopts asymmetric Gaussian function to realize obfuscation computing, and Fig. 5 is asymmetric Gauss's membership function schematic diagram of Probabilistic Fuzzy nerve network controller;

The calculation process of described Probabilistic Fuzzy nerve network controller is:

A definition jth fuzzy If-Then Rule Expression is as follows:

Rule j: if $x_1-M_1^j,x_2-M_2^j,$ Then $T_k=\underset{i}{\&Sigma;}{c}_{ik}{x}_i$ (11)

Wherein, x _i, i=1,2, be the input of Probabilistic Fuzzy nerve network controller, with for fuzzy set, T _kfor TSK Fuzzy inferential engine, c _ikfor adjustable weight coefficient;

1st layer (input layer): the node of input layer is by input variable x _i, i=1,2, be delivered to the 2nd layer, the relation between the node input signal of input layer and output signal is as follows:

${\mathrm{net}}_i^1\left(N\right)=x_i^1,y_i^1\left(N\right)=f_i^1\left({\mathrm{net}}_i^1\right(N\left)\right)={\mathrm{net}}_i^1\left(N\right),i=1,2---\left(12\right)$

Wherein, n represents iterations, represent the 1st node layer output signal; Under grid fault conditions, for the active power controller function of pattern I, input variable for the active power controller function of pattern II, input variable e=P ^*-P; For Reactive Power Control function, input variable e=Q ^*-Q;

2nd layer (degree of membership layer): this layer of each node adopts asymmetric Gaussian function to realize obfuscation computing, the relation between node input signal and output signal is expressed as follows:

${\mathrm{net}}_j^2\left(N\right)=\left\{\begin{array}{cc}-\frac{{\left(y_i^1\right(N)-m_j^2(N\left)\right)}^2}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right(N\left)\right)}^2},& -\mathrm{\&infin;}<y_i^1\left(N\right)\&le;m_j^2\\ -\frac{{\left(y_i^1\right(N)-m_j^2(N\left)\right)}^2}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right(N\left)\right)}^2},& m_j^2<y_i^1\left(N\right)<\mathrm{\&infin;}\end{array}\right.---\left(13\right)$

$y_i^2\left(N\right)=f_j^2\left({\mathrm{net}}_j^2\right(N\left)\right)=\exp \left({\mathrm{net}}_j^2\right(N\left)\right)---\left(14\right)$

Wherein, be the average of i-th asymmetric Gaussian function of input variable jth item, with the left standard difference and the right standard that are respectively i-th asymmetric Gaussian function of input variable jth item are poor, it is the output variable of the 2nd node layer;

3rd layer (probability layer): the node input/output relation of the 3rd layer is as follows:

$\begin{array}{c}{P}_{jp}\left(N\right)=f_{jp}\left(y_j^2\right(N\left)\right)=\exp \&lsqb;-\frac{{\left(y_j^2\right(N)-m_{jp}^3)}^2}{{\left({\mathrm{\&sigma;}}_{jp}^3\right)}^2}\&rsqb;\\ j=1,2,\mathrm{...},6;p=1,2,3\end{array}---\left(15\right)$

Wherein, P _jp(N) corresponding to the output signal of a jth input variable p node, with correspond respectively to average and the standard deviation of jth input variable p the asymmetric Gaussian function of node; In order to reduce amount of calculation, can be by with be arranged to constant, as p=1, order as p=2, order as p=3, order

4th layer (TSK indistinct logic computer preparative layer): in this layer, output signal the linear combination into input signal, the output signal of a kth node is:

$T_k\left(N\right)=\underset{i}{\&Sigma;}{c}_{ik}\left(N\right)x_i\left(N\right),i=1,2;k=1,2,\mathrm{...},9---\left(16\right)$

Wherein, c _ikfor adjustable weight coefficient, x _ifor input variable, N is iterations;

5th layer (rules layer): this layer of Part I is layers 2 and 3 output signal node with product, be designated as ∏, therefore the output signal of this layer of Part I kth node can be expressed as:

$\begin{array}{c}{y}_k^I\left(N\right)=y_r^2\left(N\right)y_l^2\left(N\right)S_r\left(N\right)S_l\left(N\right),r=1,2,3\\ l=4,5,6;k=3(r-1)+(l-3)\end{array}---\left(17\right)$

$S_j\left(N\right)=\underset{p}{\&Pi;}{P}_{jp}\left(N\right),j=1,2,\mathrm{...},6;p=1,2,3---\left(18\right)$

Wherein, S _j(N) be the 3rd node layer output signal P _jp(N) product, corresponding to a jth node of the 2nd layer;

This layer of Part II is the 5th layer of output signal expression formula, is Part I output signal t is outputed signal with the 4th layer _k(N) product, therefore the output signal of a kth node can be expressed as:

$y_k^O\left(N\right)=T_k\left(N\right)y_k^I\left(N\right),k=1,2,\mathrm{...},9---\left(19\right)$

Wherein, represent the output signal of a rules layer kth node;

6th layer (output layer): this layer is made up of a node O, calculate the weighted accumulation effect of all upper layer node output signal, its Mathematical Modeling is expressed as:

${\mathrm{net}}_o^6\left(N\right)=\underset{k}{\&Sigma;}{w}_k^6\left(N\right)y_k^O\left(N\right),o=1;k1,2,\mathrm{...},9---\left(20\right)$

$y_o^6\left(N\right)=f_o^6\left({\mathrm{net}}_o^6\right(N\left)\right)={\mathrm{net}}_o^6\left(N\right),o=1---\left(21\right)$

Wherein, represent that a kth obfuscation rule is to the weight coefficient of o output signal action intensity, it is a kth input signal of the 6th node layer; for photovoltaic DC-to-AC converter injects the active current of electrical network, for photovoltaic DC-to-AC converter injects the active current of electrical network.

S4, set up Probabilistic Fuzzy nerve network controller error back-propagating learning algorithm mechanism, construct a gradient vector, make wherein each element be the first differential of energy function relative to algorithm parameter, thus complete the parameters on line modifying of Probabilistic Fuzzy neural net;

The concrete methods of realizing of this step is: be defined as follows energy function:

$E\left(N\right)=\frac{1}{2}{\left(y^{*}\right(N)-y(N\left)\right)}^2=\frac{1}{2}{e}^2\left(N\right)---\left(22\right)$

Wherein, the tracking error that E (N) is Probabilistic Fuzzy nerve network controller the N time iteration, y ^*(N) and y (N) desired output that represents the N time iteration controller respectively export with actual;

Described Probabilistic Fuzzy nerve network controller parameters on line modifying comprises the following steps:

S41, for Probabilistic Fuzzy neural network algorithm the 6th layer (output layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_o^6=-\frac{\&part;E}{\&part;y_o^6\left(N\right)}=-\frac{\&part;E}{\&part;y}\frac{\&part;y}{2y_o^6\left(N\right)}---\left(23\right)$

Need the weight coefficient that iteration upgrades be expressed as:

${\mathrm{\&Delta;w}}_k^6=-{\mathrm{\&eta;}}_1\frac{\&part;E}{\&part;w_k^6\left(N\right)}=-{\mathrm{\&eta;}}_1\frac{\&part;E}{\&part;y_o^6\left(N\right)}\frac{\&part;y_o^6\left(N\right)}{\&part;w_k^6\left(N\right)}={\mathrm{\&eta;}}_1{\mathrm{\&delta;}}_o^6{y}_k^O---\left(24\right)$

Wherein, η ₁represent learning rate, the N+1 time and the N time weight coefficient iterative relation as follows:

$w_k^6(N+1)=w_k^6\left(N\right)+{\mathrm{\&Delta;w}}_k^6---\left(25\right);$

S42, for Probabilistic Fuzzy neural network algorithm the 5th layer (rules layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_k^O=\frac{\&part;E}{\&part;y_k^O\left(N\right)}=-\frac{\&part;E}{\&part;y_o^6\left(N\right)}\frac{\&part;y_o^6\left(N\right)}{\&part;y_k^O\left(N\right)}={\mathrm{\&delta;}}_o^6{w}_k^6---\left(26\right)$

${\mathrm{\&delta;}}_k^I=\frac{\&part;E}{\&part;y_k^I\left(N\right)}=-\frac{\&part;E}{\&part;y_k^O\left(N\right)}\frac{\&part;y_k^O\left(N\right)}{\&part;y_k^I\left(N\right)}={\mathrm{\&delta;}}_k^O{T}_k---\left(27\right)$

S43, for Probabilistic Fuzzy neural network algorithm the 4th layer (TSK indistinct logic computer preparative layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_k^4=-\frac{\&part;E}{\&part;T_k\left(N\right)}=-\frac{\&part;E}{\&part;y_k^O\left(N\right)}\frac{\&part;y_k^O\left(N\right)}{\&part;T_k\left(N\right)}={\mathrm{\&delta;}}_k^O{y}_k^I---\left(28\right)$

The weight coefficient c needing iteration to upgrade _ikbe expressed as:

${\mathrm{\&Delta;c}}_{ik}=-{\mathrm{\&eta;}}_2\frac{\&part;E}{\&part;c_{ik}\left(N\right)}=-{\mathrm{\&eta;}}_2\frac{\&part;E}{\&part;T_k\left(N\right)}\frac{\&part;T_k\left(N\right)}{\&part;c_{ik}\left(N\right)}={\mathrm{\&eta;}}_2{\mathrm{\&delta;}}_k^4{x}_i---\left(29\right)$

Wherein, η ₂represent learning rate, the N+1 time and the N time weight coefficient c _ikiterative relation as follows:

c _ik(N+1)＝c _ik(N)+Δc _ik(30)

S44, for Probabilistic Fuzzy neural network algorithm the 2nd layer (degree of membership layer), the error of back-propagating is needed to be expressed as:

$\begin{array}{c}{\mathrm{\&delta;}}_j^2=-\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}=\frac{\&part;E}{\&part;y_k^I\left(N\right)}\frac{\&part;y_k^I\left(N\right)}{\&part;y_j^2\left(N\right)}\frac{\&part;y_j^2\left(N\right)}{\&part;{\mathrm{net}}_j^2\left(N\right)}\\ =\left\{\begin{array}{c}{h}_j\underset{r}{\&Sigma;}{\mathrm{\&delta;}}_k^I{y}_k^I,j=1,2,3;r=1,2,3;k=3(j-1)+r\\ h_j\underset{r}{\&Sigma;}{\mathrm{\&delta;}}_k^I{y}_k^I,j=4,5,6;r=1,2,3;k=j+3(r-2)\end{array}\right.\end{array}---\left(31\right)$

$h_j=1-y_j^2\underset{p}{\&Sigma;}\frac{y_j^2-m_{jp}^3}{{\left({\mathrm{\&sigma;}}_{jp}^3\right)}^2},p=1,2,3---\left(32\right)$

The asymmetric Gaussian function average needing iteration to upgrade be expressed as:

$\begin{array}{c}{\mathrm{\&Delta;m}}_j^2=-{\mathrm{\&eta;}}_3\frac{\&part;E}{\&part;m_j^2\left(N\right)}=-{\mathrm{\&eta;}}_3\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;m_j^2\left(N\right)}\\ =\left\{\begin{array}{cc}{\mathrm{\&eta;}}_3{\mathrm{\&delta;}}_j^2\frac{2(y_i^1-m_j^2)}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right)}^2},& -\mathrm{\&infin;}<y_i^1\&le;m_j^2,j=1,2,\mathrm{...},6\\ {\mathrm{\&eta;}}_3{\mathrm{\&delta;}}_j^2\frac{2(y_i^1-m_j^2)}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right)}^2},& m_j^2<y_i^1<\mathrm{\&infin;},j=1,2,\mathrm{...6}\end{array}\right.\end{array}---\left(33\right)$

Wherein, η ₃represent the learning rate of algorithm; Asymmetric Gaussian function average iterative relation as follows:

$m_j^2(N+1)=m_j^2\left(N\right)+{\mathrm{\&Delta;m}}_j^2---\left(34\right)$

The left standard of the asymmetric Gaussian function needing iteration to upgrade is poor and right standard is poor with be expressed as follows:

$\begin{array}{c}{\mathrm{\&sigma;}}_{L\_j}^2=-{\mathrm{\&eta;}}_4\frac{\&part;E}{\&part;{\mathrm{\&sigma;}}_{L\_j}^2}=-{\mathrm{\&eta;}}_4\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;{\mathrm{\&sigma;}}_{L\_j}^2\left(N\right)}\\ ={\mathrm{\&eta;}}_4{\mathrm{\&delta;}}_j^2\frac{2{(y_i^1-m_j^2)}^2}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right)}^3},j=1,2,\mathrm{...},6\end{array}---\left(35\right)$

$\begin{array}{c}{\mathrm{\&sigma;}}_{R\_j}^2=-{\mathrm{\&eta;}}_5\frac{\&part;E}{\&part;{\mathrm{\&sigma;}}_{R\_j}^2}=-{\mathrm{\&eta;}}_5\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;{\mathrm{\&sigma;}}_{R\_j}^2\left(N\right)}\\ ={\mathrm{\&eta;}}_5{\mathrm{\&delta;}}_j^2\frac{2{(y_i^1-m_j^2)}^2}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right)}^3},j=1,2,\mathrm{...},6\end{array}---\left(36\right)$

Wherein, η ₄, η ₅represent the learning rate of algorithm;

The left standard of asymmetric Gaussian function is poor and right standard is poor with rule of iteration as follows:

${\mathrm{\&sigma;}}_{L\_j}^2(N+1)={\mathrm{\&sigma;}}_{L\_j}^2\left(N\right)+{\mathrm{\&Delta;\&sigma;}}_{L\_j}^2---\left(37\right)$

${\mathrm{\&sigma;}}_{R\_j}^2(N+1)={\mathrm{\&sigma;}}_{R\_j}^2\left(N\right)+{\mathrm{\&Delta;\&sigma;}}_{R\_j}^2---\left(38\right)$

In above-mentioned steps, formula (23), (24), (26), (27), (28), (29), (31), (33), (35), the computing of (36) isostructure single order partial differential, all belong to the gradient vector of structure.According to above-mentioned steps, by parameter c _ik, with repetitive exercise is carried out, parameter respectively according to formula (25), (30), (34), (37), (38) c _ik, with initial value is set to 0 respectively, and 1,1,1; Parameter initial value is set to-1,0 respectively, and 1 ,-1,0,1; Learning rate η ₁~ η ₅initial value is zero.

S5, set up Boost boost chopper inner ring controller model, in Boost boost chopper, the output signal of ring controller compares with triangular carrier and forms the pwm pulse control signal of Boost boost chopper switching device; Set up three-phase inverter inner ring current controller model, realize the tracing control of output current to reference signal, the output signal of three-phase inverter inner ring current controller is compared with triangular carrier and forms the pwm pulse control signal of each switching device of three-phase inverting circuit, thus the target that under realizing photovoltaic generating system cutting-in control and grid fault conditions, dynamic reactive supports.Ring controller adoption rate integral controller in described Boost boost chopper, the input signal of pi controller selects input signal by interrupteur SW 1 according to the output mode signal of electric network fault controller, system works when pattern I, by the output signal under MPPT maximum power point tracking pattern (MPPT) with the real output signal V of photovoltaic array _pVask deviation, using the input signal that deviation signal controls as Boost boost chopper inner ring; When system works is at pattern II, by three-phase inverter DC bus-bar voltage reference value with actual value V _dcthe input signal controlled as Boost boost chopper inner ring after asking for deviation; In Boost boost chopper, the output signal of ring controller compares with triangular carrier and forms the pwm pulse control signal of Boost boost chopper switching device.Described three-phase photovoltaic inverter inner ring current controller model, adoption rate integral control, realizes output current i _a, i _b, i _cto reference signal tracing control, the output signal of three-phase photovoltaic inverter inner ring current controller is compared with triangular carrier and forms the pwm pulse control signal of each switching device of three-phase inverting circuit, thus realize photovoltaic generating system cutting-in control and under grid fault conditions to the target that electrical network dynamic reactive power supports.

Below in conjunction with specific embodiment, control method of the present invention is described:

Fig. 6 gives the design sketch of the photovoltaic generating system control model I based on Probabilistic Fuzzy ANN Control.Fig. 6 (a) gives photovoltaic array power P _pV, photovoltaic generating system injects the active-power P of electrical network and reactive power Q, reactive power maximum permissible value Q ^*oscillogram; Fig. 6 (b) gives power network current d axle and q axle component reference value under dq synchronous coordinate system with oscillogram; Fig. 6 (c) gives photovoltaic DC-to-AC converter DC bus-bar voltage V _dcand reference value photovoltaic array output voltage V _pVwith output current I _pV, electrical network A phase, B phase, C phase voltage effective value | v _a| _rms, | v _b| _rms, | v _c| _rmsoscillogram; Fig. 6 (d) gives photovoltaic DC-to-AC converter A phase, B phase, C phase output current i _a, i _b, i _coscillogram.

As shown in Figure 6, in an initial condition, photovoltaic array power P _pVfor 612W, it is 524W that photovoltaic generating system injects the active-power P of electrical network, and difference power is Boost boost chopper and three-phase inverting circuit switching loss and conduction loss, and three-phase inverter output current peak value is 4.1A.As t=0.1s, there are two relative ground circuits in electrical network, causes the Voltage Drop of 0.3pu.Now, Boost boost chopper is operated in maximal power tracing pattern (pattern I), photovoltaic array output voltage V _pVwith output current I _pVremain unchanged, be respectively 151V and 4.05A, intensity of illumination is 600W/m ², therefore the active-power P of photovoltaic generating system injection electrical network keeps constant, and three-phase inverter output current peak value increases to 6.4A.Because now inverter output current does not exceed maximum permissible current, photovoltaic generating system possesses the reactive power support function to electrical network under pattern I condition, and now photovoltaic system is 456Var to the reactive power Q that electrical network injects, to the reactive current component that electrical network injects 2.9A is increased to, to the active current that electrical network injects from 0A 4.9A is increased to from 4.1A.Inverter side three-phase voltage effective value is respectively 0.7pu, 0.87pu, 0.87pu, the reactive power Q that photovoltaic DC-to-AC converter exports and photovoltaic DC-to-AC converter DC bus-bar voltage V _dcdynamic response time be respectively 0.3s and 1.45%, reduce more than 50% than adopting traditional proportional integral (PI) controller.

Fig. 7 gives the design sketch of the photovoltaic generating system control model II based on Probabilistic Fuzzy ANN Control.Fig. 7 (a) gives photovoltaic array power P _pV, photovoltaic generating system injects the active-power P of electrical network and reactive power Q, reactive power maximum permissible value Q ^*oscillogram; Fig. 7 (b) gives power network current d axle and q axle component reference value under dq synchronous coordinate system with oscillogram; Fig. 7 (c) gives photovoltaic DC-to-AC converter DC bus-bar voltage V _dcand reference value photovoltaic array output voltage V _pVwith output current I _pV, electrical network A phase, B phase, C phase voltage effective value | v _a| _rms, | v _b| _rms, | v _c| _rmsoscillogram; Fig. 7 (d) gives photovoltaic DC-to-AC converter A phase, B phase, C phase output current i _a, i _b, i _coscillogram.

As shown in Figure 7, in an initial condition, photovoltaic array power P _pVfor 1005W, it is 882W that photovoltaic generating system injects the active-power P of electrical network, and difference power is Boost boost chopper and three-phase inverting circuit switching loss and conduction loss, and as t=0.1s, two relative ground circuits occur electrical network, cause the Voltage Drop of 0.7pu.Be no more than maximum permissible value to limit photovoltaic DC-to-AC converter output current, Boost boost chopper operating condition is transitioned into pattern II from pattern I, and the criterion of its handoff procedure as shown in Figure 3.By changing the operational mode of Boost circuit, the power output P of photovoltaic array _pVbe reduced to 102W, the active-power P that photovoltaic DC-to-AC converter injects electrical network is reduced to 21W, and reactive power Q is increased to 527W simultaneously, to the reactive current component that electrical network injects 5.9A is increased to, to the active current that electrical network injects from 0A be reduced to 1.3A from 6.2A, play the function that power system reactive power is supported.Inverter side three-phase voltage effective value is respectively 0.3pu, 0.67pu, 0.68pu, photovoltaic array output voltage V _pV173.0V is elevated to, photovoltaic array output current I from 150.4V _pV0.52A is reduced to, the reactive power Q that photovoltaic DC-to-AC converter exports and photovoltaic DC-to-AC converter DC bus-bar voltage V from 6.6A _dcdynamic response time be respectively 0.15s and 7.1%, reduce more than 60% than adopting traditional proportional integral (PI) controller.

From Fig. 6, the dynamic response waveform of Fig. 7 can be found out, Probabilistic Fuzzy neural net method is applied in photovoltaic generating system, the grid voltage mutation caused at electric network fault and the mode of operation of photovoltaic generating system can be adjusted under falling situation fast, to adapt to the restriction of photovoltaic array peak power output and combining inverter rated capacity and maximum output current, there is stability strong, tracking velocity is fast, reactive power and the control effects such as inverter DC bus-bar voltage overshoot is little, the feasibility of this control method is not limited to the operating mode and circuit topology mentioned in the embodiment of the present invention simultaneously, extend in the grid-connected photovoltaic power generation system of other topological structures and electric network fault operating mode.

Those of ordinary skill in the art will appreciate that, embodiment described here is to help reader understanding's principle of the present invention, should be understood to that protection scope of the present invention is not limited to so special statement and embodiment.Those of ordinary skill in the art can make various other various concrete distortion and combination of not departing from essence of the present invention according to these technology enlightenment disclosed by the invention, and these distortion and combination are still in protection scope of the present invention.

Claims (7)

1., based on the photovoltaic generating system powerless control method of Probabilistic Fuzzy neural net, it is characterized in that, comprise the following steps:

S1, set up photovoltaic generating system Mathematical Modeling, carry out power calculation and genlock, ask for the maximum permissible value of active power that photovoltaic generating system injects to electrical network and reactive power;

S2, set up the electric network fault controller model of photovoltaic generating system, according to the maximum permissible value of active power, reactive power and three-phase inverter electric current, set up double mode switching control strategy, dynamically adjust the mode of operation of Boost boost chopper and the Probabilistic Fuzzy ANN Control input reference signal of three-phase inverter;

S3, set up Probabilistic Fuzzy nerve network controller, ask for active current and reactive current reference value that three-phase inverter injects electrical network;

S4, set up Probabilistic Fuzzy nerve network controller error back-propagating learning algorithm mechanism, construct a gradient vector, make wherein each element be the first differential of energy function relative to algorithm parameter, thus complete the parameters on line modifying of Probabilistic Fuzzy neural net;

S5, set up Boost boost chopper inner ring controller model, in Boost boost chopper, the output signal of ring controller compares with triangular carrier and forms the pwm pulse control signal of Boost boost chopper switching device; Set up three-phase inverter inner ring current controller model, realize the tracing control of output current to reference signal, the output signal of three-phase inverter inner ring current controller is compared with triangular carrier and forms the pwm pulse control signal of each switching device of three-phase inverting circuit, thus the target that under realizing photovoltaic generating system cutting-in control and grid fault conditions, dynamic reactive supports.

2. the photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net according to claim 1, it is characterized in that, the main circuit of described photovoltaic generating system comprises photovoltaic array, Boost boost chopper, three-phase inverter, grid-connected reactor L, grid-connected switch S SR, grid-connected transformer and three phase network, and photovoltaic array is successively by Boost boost chopper, three-phase inverter, grid-connected reactor L, grid-connected switch S SR and grid-connected transformer access three phase network; C _pVfor photovoltaic array output port shunt capacitance, simultaneously for Boost boost chopper provides input voltage; V _pVand I _pVbe respectively output voltage and the output current of photovoltaic array; DC capacitor C _dcbe connected on Boost boost chopper output, C _dcsimultaneously as the dc-link capacitance of three-phase inverter, its operating voltage is V _dc; Three-phase inverter is by grid-connected reactor L, grid-connected switch S SR and grid-connected transformer access three-phase alternating current electrical network;

Described step S1 specifically comprises following sub-step:

S11, three-phase inverter exchange the relation exporting phase voltage and line voltage and are expressed as:

$\left[\begin{array}{c}{v}_a\\ v_b\\ v_c\end{array}\right]=\frac{1}{3}\left[\begin{array}{ccc}1& 0& -1\\ -1& 1& 0\\ 0& -1& 1\end{array}\right]\left[\begin{array}{c}{v}_{ab}\\ v_{bc}\\ v_{ca}\end{array}\right]---\left(1\right)$

Wherein, [v _ab, v _bc, v _ca] ^tfor three-phase inverter exchanges output line voltage, [v _a, v _b, v _c] ^tphase voltage is exported for three-phase inverter exchanges;

S12, employing Clark conversion, be transformed into static α β coordinate system, be expressed as by the phase voltage of abc coordinate system:

$\left[\begin{array}{c}{v}_{\mathrm{\&alpha;}}\\ v_{\mathrm{\&beta;}}\end{array}\right]=\sqrt{\frac{2}{3}}\left[\begin{array}{ccc}1& -\frac{1}{2}& -\frac{1}{2}\\ 0& \frac{\sqrt{3}}{2}& -\frac{\sqrt{3}}{2}\end{array}\right]\left[\begin{array}{c}{v}_a\\ v_b\\ v_c\end{array}\right]---\left(2\right)$

Wherein, v _α, v _βbe respectively the component of three-phase inverter output voltage α axle and β axle under static α β coordinate system;

S13, employing Park conversion, by the output voltage [v of static α β coordinate system _α, v _β] ^tbe transformed into synchronous rotating reference frame, be expressed as:

$\left[\begin{array}{c}{v}_d\\ v_q\end{array}\right]=\left[\begin{array}{cc}cos\left({\mathrm{\&theta;}}_e\right)& \sin \left({\mathrm{\&theta;}}_e\right)\\ -sin\left({\mathrm{\&theta;}}_e\right)& cos\left({\mathrm{\&theta;}}_e\right)\end{array}\right]\left[\begin{array}{c}{v}_{\mathrm{\&alpha;}}\\ v_{\mathrm{\&beta;}}\end{array}\right]---\left(3\right)$

Wherein, v _d, v _qbe respectively the component of three-phase inverter output voltage d axle and q axle under dq synchronous coordinate system;

S14, employing Park conversion, by three-phase inverter output current [i _a, i _b, i _c] ^tbe transformed into dq synchronous coordinate system, be expressed as:

$\left[\begin{array}{c}{i}_d\\ i_q\end{array}\right]=\sqrt{\frac{2}{3}}\left[\begin{array}{ccc}cos\left({\mathrm{\&theta;}}_e\right)& cos({\mathrm{\&theta;}}_e-\frac{2}{3}\mathrm{\&pi;})& cos({\mathrm{\&theta;}}_e+\frac{2}{3}\mathrm{\&pi;})\\ -sin\left({\mathrm{\&theta;}}_e\right)& -\sin ({\mathrm{\&theta;}}_e-\frac{2}{3}\mathrm{\&pi;})& -\sin ({\mathrm{\&theta;}}_e+\frac{2}{3}\mathrm{\&pi;})\end{array}\right]\left[\begin{array}{c}{i}_a\\ i_b\\ i_c\end{array}\right]---\left(4\right)$

Wherein, i _d, i _qbe respectively d axle and q axle power network current component under dq synchronous coordinate system, i _qcomponent is used for controlling the real component injected to electrical network of photovoltaic generating system, i _dcomponent is used for controlling the idle component injected to electrical network of photovoltaic generating system;

S15, the data obtained according to step S13 and S14, the instantaneous active power P injected to electrical network by photovoltaic generating system and instantaneous reactive power Q is expressed as:

$P=\frac{3}{2}(v_d{i}_d+v_q{i}_q);Q=\frac{3}{2}(v_q{i}_d-v_d{i}_q)---\left(5\right)$

After genlock (PLL), three-phase inverter output voltage phasor and line voltage phasor same-phase, i.e. v _d=0, then instantaneous active power and instantaneous reactive power are expressed as:

$P=\frac{3}{2}{v}_q{i}_q;Q=\frac{3}{2}{v}_q{i}_d---\left(6\right);$

S16, establish V _sagrepresent the perunit value of Voltage Drop amplitude, its scope is [0,1], then photovoltaic generating system needs the reactive current reference value injecting electrical network to be expressed as:

$I_r^{*}=\left\{\begin{array}{cc}0\%,& V_{sag}\&le;0.1\\ 200V_{sag}\%,& 0.1<V_{sag}\&le;0.5\\ 100\%,& V_{sag}>0.5\end{array}\right.---\left(7\right)$

$V_{sag}=1-\frac{min\left(\right|v_a{|}_{rms},|v_b{|}_{rms},|v_c{|}_{rms})}{V_{base}}---\left(8\right)$

Wherein, V _baserepresent voltage base value, v _{a, rms}, v _{b, rms}, v _{c, rms}be respectively a, b, c phase voltage v _a, v _b, v _ceffective value;

Work as V _sagwhen being greater than 0.1pu, photovoltaic generating system starts the reactive power support and control pattern of electrical network; Work as V _sagwhen being less than or equal to 0.1, reactive current reference value is 0, and system is in conventional maximum power output pattern, and photovoltaic generating system only injects active current and active power to electrical network;

After reactive power support and control pattern starts, in order to meet low voltage ride through of photovoltaic inverter (LVRT) requirement, and avoid inverter output current to exceed device rated current causing component damage, the apparent power maximum of three-phase inverter is expressed as:

|S|＝(|v _a| _rms+|v _b| _rms+|v _c| _rms)I _max(9)

Wherein, I _maxrepresent that photovoltaic generating system injects the maximum permissible value of power network current;

Therefore, the maximum permissible value of active power and reactive power is expressed as:

$P^{*}=\left|S\right|\sqrt{1-I_r^{*2}};Q^{*}=\left|S\right|I_r^{*}---\left(10\right)$

Above-mentioned active power and reactive power maximum permissible value P ^*, Q ^*using injecting the reference value of active power and reactive power as photovoltaic generating system to electrical network, after asking for deviation respectively with the actual active power of photovoltaic generating system and reactive power P, Q, as the input signal of Probabilistic Fuzzy nerve network controller.

3. the photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net according to claim 2, is characterized in that, the double mode switching control strategy in described step S2 is specially:

Pattern I: work as P ^*be greater than the active power rate P of photovoltaic array _pVtime, by controlling the DC bus-bar voltage of three-phase inverter, by the active power rate P of photovoltaic array _pVwhole injection electrical network; Three-phase inverter controlled current flow is built in I _maxwithin, Boost boost chopper is operated in MPPT maximum power point tracking pattern (MPPT);

Pattern II: work as P ^*be less than or equal to the active power rate P of photovoltaic array _pVtime, Boost boost chopper suspends MPPT maximum power point tracking, and starts to follow the tracks of power P ^*; The unbalanced problem of power between Boost boost chopper and three-phase inverter solves by controlling three-phase inverter DC bus-bar voltage.

4. the photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net according to claim 3, it is characterized in that, the Probabilistic Fuzzy nerve network controller in described S3 comprises 6 layer network structures: the 1st layer be input layer, the 2nd layer be degree of membership layer, the 3rd layer be probability layer, the 4th layer be TSK indistinct logic computer preparative layer, the 5th layer be rules layer, the 6th layer for output layer; In described degree of membership layer, each node adopts asymmetric Gaussian function to realize obfuscation computing;

The calculation process of described Probabilistic Fuzzy nerve network controller is:

A definition jth fuzzy If-Then Rule Expression is as follows:

Rule j: if $x_1=M_1^j,x_2=M_2^j,$ Then $T_k=\underset{i}{\&Sigma;}{c}_{ik}{x}_i---\left(11\right)$

Wherein, x _i, i=1,2, be the input of Probabilistic Fuzzy nerve network controller, with for fuzzy set, T _kfor TSK Fuzzy inferential engine, c _ikfor adjustable weight coefficient;

1st layer (input layer): the node of input layer is by input variable x _i, i=1,2, be delivered to the 2nd layer, the relation between the node input signal of input layer and output signal is as follows:

${\mathrm{net}}_i^1\left(N\right)=x_i^1,y_i^1\left(N\right)=f_i^1\left({\mathrm{net}}_i^1\right(N\left)\right)={\mathrm{net}}_i^1\left(N\right),i=1,2---\left(12\right)$

Wherein, n represents iterations, represent the 1st node layer output signal; Under grid fault conditions, for the active power controller function of pattern I, input variable for the active power controller function of pattern II, input variable e=P ^*-P; For Reactive Power Control function, input variable e=Q ^*-Q;

2nd layer (degree of membership layer): this layer of each node adopts asymmetric Gaussian function to realize obfuscation computing, the relation between node input signal and output signal is expressed as follows:

${\mathrm{net}}_j^2\left(N\right)=\left\{\begin{array}{cc}-\frac{{\left(y_i^1\right(N)-m_j^2(N\left)\right)}^2}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right(N\left)\right)}^2},& -\mathrm{\&infin;}<y_i^1\left(N\right)\&le;m_j^2\\ -\frac{{\left(y_i^1\right(N)-m_j^2(N\left)\right)}^2}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right(N\left)\right)}^2},& m_j^2<y_i^1\left(N\right)<\mathrm{\&infin;}\end{array}\right.---\left(13\right)$

$y_i^2\left(N\right)=f_j^2\left({\mathrm{net}}_j^2\right(N\left)\right)=\exp \left({\mathrm{net}}_j^2\right(N\left)\right)---\left(14\right)$

Wherein, be the average of i-th asymmetric Gaussian function of input variable jth item, with the left standard difference and the right standard that are respectively i-th asymmetric Gaussian function of input variable jth item are poor, it is the output variable of the 2nd node layer;

3rd layer (probability layer): the node input/output relation of the 3rd layer is as follows:

$P_{jp}\left(N\right)=f_{jp}\left(y_j^2\right(N\left)\right)=\exp \&lsqb;-\frac{{\left(y_j^2\right(N)-m_{jp}^3)}^2}{{\left({\mathrm{\&sigma;}}_{jp}^3\right)}^2}\&rsqb;---\left(15\right)$

j＝1,2,...,6；p＝1,2,3

Wherein, P _jp(N) corresponding to the output signal of a jth input variable p node, with correspond respectively to average and the standard deviation of jth input variable p the asymmetric Gaussian function of node; In order to reduce amount of calculation, can be by with be arranged to constant, as p=1, order as p=2, order as p=3, order

4th layer (TSK indistinct logic computer preparative layer): in this layer, output signal the linear combination into input signal, the output signal of a kth node is:

$T_k\left(N\right)=\underset{i}{\&Sigma;}{c}_{ik}\left(N\right)x_i\left(N\right),i=1,2;k=1,2,...,9---\left(16\right)$

Wherein, c _ikfor adjustable weight coefficient, x _ifor input variable, N is iterations;

5th layer (rules layer): this layer of Part I is layers 2 and 3 output signal node and P _jp(N) product, is designated as Π, and therefore the output signal of this layer of Part I kth node can be expressed as:

$\begin{array}{c}{y}_k^I\left(N\right)=y_r^2\left(N\right)y_l^2\left(N\right)S_r\left(N\right)S_l\left(N\right),r=1,2,3;\\ l=4,5,6;K=3(r-1)+(l-3)\end{array}---\left(17\right)$

$S_j\left(N\right)=\underset{p}{\&Pi;}{P}_{jp}\left(N\right),j=1,2,...,6;p=1,2,3---\left(18\right)$

Wherein, S _j(N) be the 3rd node layer output signal P _jp(N) product, corresponding to a jth node of the 2nd layer;

This layer of Part II is the 5th layer of output signal expression formula, is Part I output signal t is outputed signal with the 4th layer _k(N) product, therefore the output signal of a kth node can be expressed as:

$y_k^O\left(N\right)=T_k\left(N\right)y_k^I\left(N\right),k=1,2,...,9---\left(19\right)$

Wherein, represent the output signal of a rules layer kth node;

6th layer (output layer): this layer is made up of a node O, calculate the weighted accumulation effect of all upper layer node output signal, its Mathematical Modeling is expressed as:

${\mathrm{net}}_o^6\left(N\right)=\underset{k}{\&Sigma;}{w}_k^6\left(N\right)y_k^O\left(N\right),o=1;k=1,2,...,9---\left(20\right)$

$y_o^6\left(N\right)=f_o^6\left({\mathrm{net}}_o^6\right(N\left)\right)={\mathrm{net}}_o^6\left(N\right),o=1---\left(21\right)$

Wherein, represent that a kth obfuscation rule is to the weight coefficient of o output signal action intensity, it is a kth input signal of the 6th node layer; for photovoltaic DC-to-AC converter injects the active current of electrical network, for photovoltaic DC-to-AC converter injects the active current of electrical network.

5. the photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net according to claim 4, is characterized in that, described step S4 concrete methods of realizing is: be defined as follows energy function:

$E\left(N\right)=\frac{1}{2}{\left(y^{*}\right(N)-y(N\left)\right)}^2=\frac{1}{2}{e}^2\left(N\right)---\left(22\right)$

Wherein, the tracking error that E (N) is Probabilistic Fuzzy nerve network controller the N time iteration, y ^*(N) and y (N) desired output that represents the N time iteration controller respectively export with actual;

Described Probabilistic Fuzzy nerve network controller parameter training process is as follows:

S41, for Probabilistic Fuzzy neural network algorithm the 6th layer (output layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_o^6=-\frac{\&part;E}{\&part;y_o^6\left(N\right)}=-\frac{\&part;E}{\&part;y}\frac{\&part;y}{2y_o^6\left(N\right)}---\left(23\right)$

Need the weight coefficient that iteration upgrades be expressed as:

${\mathrm{\&Delta;w}}_k^6=-{\mathrm{\&eta;}}_1\frac{\&part;E}{\&part;w_k^6\left(N\right)}=-{\mathrm{\&eta;}}_1\frac{\&part;E}{\&part;y_o^6\left(N\right)}\frac{\&part;y_o^6\left(N\right)}{\&part;w_k^6\left(N\right)}={\mathrm{\&eta;}}_1{\mathrm{\&delta;}}_o^6{y}_k^O---\left(24\right)$

Wherein, η ₁represent learning rate, the N+1 time and the N time weight coefficient iterative relation as follows:

$w_k^6(N+1)=w_k^6\left(N\right)+{\mathrm{\&Delta;w}}_k^6---\left(25\right);$

S42, for Probabilistic Fuzzy neural network algorithm the 5th layer (rules layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_k^O=\frac{\&part;E}{\&part;y_k^O\left(N\right)}=-\frac{\&part;E}{\&part;y_o^6\left(N\right)}\frac{\&part;y_o^6\left(N\right)}{\&part;y_k^O\left(N\right)}={\mathrm{\&delta;}}_o^6{w}_k^6---\left(26\right)$

${\mathrm{\&delta;}}_k^I=\frac{\&part;E}{\&part;y_k^I\left(N\right)}=-\frac{\&part;E}{\&part;y_k^O\left(N\right)}\frac{\&part;y_k^O\left(N\right)}{\&part;y_k^I\left(N\right)}={\mathrm{\&delta;}}_k^O{T}_k---\left(27\right)$

S43, for Probabilistic Fuzzy neural network algorithm the 4th layer (TSK indistinct logic computer preparative layer), the error of back-propagating is needed to be expressed as:

${\mathrm{\&delta;}}_k^4=-\frac{\&part;E}{\&part;T_k\left(N\right)}=-\frac{\&part;E}{\&part;y_k^O\left(N\right)}\frac{\&part;y_k^O\left(N\right)}{\&part;T_k\left(N\right)}={\mathrm{\&delta;}}_k^O{y}_k^I---\left(28\right)$

The weight coefficient c needing iteration to upgrade _ikbe expressed as:

${\mathrm{\&Delta;c}}_{ik}=-{\mathrm{\&eta;}}_2\frac{\&part;E}{\&part;c_{ik}\left(N\right)}=-{\mathrm{\&eta;}}_2\frac{\&part;E}{\&part;T_k\left(N\right)}\frac{\&part;T_k\left(N\right)}{\&part;c_{ik}\left(N\right)}={\mathrm{\&eta;}}_2{\mathrm{\&delta;}}_k^4{x}_i---\left(29\right)$

Wherein, η ₂represent learning rate, the N+1 time and the N time weight coefficient c _ikiterative relation as follows:

c _ik(N+1)＝c _ik(N)+Δc _ik(30)

S44, for Probabilistic Fuzzy neural network algorithm the 2nd layer (degree of membership layer), the error of back-propagating is needed to be expressed as:

$\begin{array}{c}{\mathrm{\&delta;}}_j^2=-\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}=\frac{\&part;E}{\&part;y_k^I\left(N\right)}\frac{\&part;y_k^I\left(N\right)}{\&part;y_j^2\left(N\right)}\frac{\&part;y_j^2\left(N\right)}{\&part;{\mathrm{net}}_j^2\left(N\right)}\\ =\left\{\begin{array}{c}{h}_j\underset{r}{\mathrm{\&Sigma;}}{\mathrm{\&delta;}}_k^I{I}_k^I,j=1,2,3;r=1,2,3;k=3(j-1)+r\\ h_j\underset{r}{\mathrm{\&Sigma;}}{\mathrm{\&delta;}}_k^I{y}_k^I,j=4,5,6;r=1,2,3;k=j+3(r-2)\end{array}\right.\end{array}---\left(31\right)$

$h_j=1-y_j^2\underset{p}{\&Sigma;}\frac{y_j^2-m_{jp}^3}{{\left({\mathrm{\&sigma;}}_{jp}^3\right)}^2},p=1,2,3---\left(32\right)$

The asymmetric Gaussian function average needing iteration to upgrade be expressed as:

$\begin{array}{c}{\mathrm{\&Delta;m}}_j^2=-{\mathrm{\&eta;}}_3\frac{\&part;E}{\&part;m_j^2\left(N\right)}=-{\mathrm{\&eta;}}_3\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;m_j^2\left(N\right)}\\ =\left\{\begin{array}{cc}{\mathrm{\&eta;}}_3{\mathrm{\&delta;}}_j^2\frac{2(y_i^1-m_j^2)}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right)}^2},& -\mathrm{\&infin;}<y_i^1\&le;m_j^2,j=1,2,...,6\\ {\mathrm{\&eta;}}_3{\mathrm{\&delta;}}_j^2\frac{2(y_i^1-m_j^2)}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right)}^2},& m_j^2<y_i^1<\mathrm{\&infin;},j=1,2,...,6\end{array}\right.\end{array}---\left(33\right)$

Wherein, η ₃represent the learning rate of algorithm; Asymmetric Gaussian function average iterative relation as follows:

$m_j^2(N+1)=m_j^2\left(N\right)+{\mathrm{\&Delta;m}}_j^2---\left(34\right)$

The left standard of the asymmetric Gaussian function needing iteration to upgrade is poor and right standard is poor with be expressed as follows:

$\begin{array}{c}{\mathrm{\&sigma;}}_{L\_j}^2=-{\mathrm{\&eta;}}_4\frac{\&part;E}{\&part;{\mathrm{\&sigma;}}_{L\_j}^2}=-{\mathrm{\&eta;}}_4\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;{\mathrm{\&sigma;}}_{L\_j}^2\left(N\right)}\\ ={\mathrm{\&eta;}}_4{\mathrm{\&delta;}}_j^2\frac{2{(y_i^1-m_j^2)}^2}{{\left({\mathrm{\&sigma;}}_{L\_j}^2\right)}^3},j=1,2,...,6\end{array}---\left(35\right)$

$\begin{array}{c}{\mathrm{\&sigma;}}_{R\_j}^2=-{\mathrm{\&eta;}}_5\frac{\&part;E}{\&part;{\mathrm{\&sigma;}}_{R\_j}^2}=-{\mathrm{\&eta;}}_5\frac{\&part;E}{\&part;{\mathrm{net}}_j^2\left(N\right)}\frac{\&part;{\mathrm{net}}_j^2\left(N\right)}{\&part;{\mathrm{\&sigma;}}_{R\_j}^2\left(N\right)}\\ ={\mathrm{\&eta;}}_5{\mathrm{\&delta;}}_j^2\frac{2{(y_i^1-m_j^2)}^2}{{\left({\mathrm{\&sigma;}}_{R\_j}^2\right)}^3},j=1,2,...,6\end{array}---\left(36\right)$

Wherein, η ₄, η ₅represent the learning rate of algorithm;

The left standard of asymmetric Gaussian function is poor and right standard is poor with rule of iteration as follows:

${\mathrm{\&sigma;}}_{L\_j}^2(N+1)={\mathrm{\&sigma;}}_{L\_j}^2\left(N\right)+{\mathrm{\&Delta;\&sigma;}}_{L\_j}^2---\left(37\right)$

${\mathrm{\&sigma;}}_{R\_j}^2(N+1)={\mathrm{\&sigma;}}_{R\_j}^2\left(N\right)+{\mathrm{\&Delta;\&sigma;}}_{R\_j}^2---\left(38\right)$

According to above-mentioned steps, by parameter c _ik, with repetitive exercise is carried out, parameter respectively according to formula (25), (30), (34), (37), (38) c _ik, with initial value is set to 0 respectively, and 1,1,1; Parameter initial value is set to-1,0 respectively, and 1 ,-1,0,1; Learning rate η ₁~ η ₅initial value is zero.

6. the photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net according to claim 5, it is characterized in that, ring controller adoption rate integral controller in Boost boost chopper in described step S5, the input signal of pi controller selects input signal by switch according to the output mode signal of electric network fault controller, system works when pattern I, by the output signal under MPPT maximum power point tracking pattern (MPPT) with the real output signal V of photovoltaic array _pVask deviation, using the input signal that deviation signal controls as Boost boost chopper inner ring; When system works is at pattern II, by three-phase inverter DC bus-bar voltage reference value with actual value V _dcthe input signal controlled as Boost boost chopper inner ring after asking for deviation; In Boost boost chopper, the output signal of ring controller compares with triangular carrier and forms the pwm pulse control signal of Boost boost chopper switching device.

7. the photovoltaic generating system powerless control method based on Probabilistic Fuzzy neural net according to claim 5, is characterized in that, the three-phase photovoltaic inverter inner ring current controller model in described step S5, adoption rate integral control, realizes output current i _a, i _b, i _cto reference signal tracing control, the output signal of three-phase photovoltaic inverter inner ring current controller is compared with triangular carrier and forms the pwm pulse control signal of each switching device of three-phase inverting circuit, thus realize photovoltaic generating system cutting-in control and under grid fault conditions to the target that electrical network dynamic reactive power supports.