CN104967353A - Off-grid type wind power generation inverter - Google Patents

Abstract

The invention relates to an off-grid type wind power generation inverter. The off-grid type wind power generation inverter is composed of an MPU controller as well as a rectifier, an energy storage capacitor, a flyback converter, a power frequency inverter bridge and an LCL filter which are connected with one another sequentially between a wind driven generator and a load; and the MPU controller acquires the voltage of the energy storage capacitor and the output voltage of the inverter through a first voltage senor and a second voltage sensor, acquires power frequency inverter bridge-side inductive current of the LCL filter through a current sensor, and controls the flyback converter and the power frequency inverter bridge respectively through two driving modules. According to the off-grid type wind power generation inverter of the invention, the composite controller formed by a neural network inverse model and a PI controller is utilized to control the output voltage of the inverter; at the vicinity of a zero crossing point, neural network inverse model control is mainly adopted, so that the advantage of high response speed of the neural network inverse model control can be fully utilized; at the vicinity of a power grid voltage peak value, PI control is mainly adopted, so that the advantage of high steady-state performance of the PI control can be fully utilized; and therefore, the electric energy quality of an off-grid wind power generation system can be effectively improved.

Description

A kind of from net type wind power generation inverter

Technical field

The present invention relates to a kind of based on complex controll from net type wind power generation inversion device, belong to technical field of power generation.

Background technology

Wind power generation, because having large, the clean characteristic such as renewable of reserves, has become one of Main way of current generation of electricity by new energy.From net type wind generator system, there is cost feature low, flexible for installation, be widely used in the area that the bulk power grid such as remote mountain areas, island cannot provide electric power, and the simple structure adopting minitype permanent magnetism direct driving motor to add inverter more.With the Shunt-connected Wind Power Generation System unlike, from net type wind generator system must implement effective inverter output voltage control, to ensure the stable operation of system.

Existing from the simple PI controller of the many employings of net type blower fan inverter, there is the shortcoming that harmonic wave of output voltage content is high, Ability of Resisting Disturbance is poor.Can harmonic inhabitation to a certain extent by configuration multiple-stage filtering unit, but the method makes system output impedance, and order is too high, easily causes system unstable.Modern society is more and more higher to quality of power supply requirement, must improve constantly the level of control from net type blower fan inverter.

Summary of the invention

The object of the invention is to the drawback for prior art, provide a kind of from net type wind power generation inverter, effectively to improve the quality of power supply from net type wind generator system.

Problem of the present invention realizes with following technical proposals:

A kind of from net type wind power generation inverter, formation comprises MPU controller and is connected to rectifier, storage capacitor, anti exciting converter, power frequency inverter bridge and the LCL filter between wind-driven generator and load successively, described MPU controller gathers storage capacitor voltage and inverter output voltage respectively by the first voltage sensor and the second voltage sensor, the power frequency inverter bridge side inductive current of LCL filter is gathered by current sensor, and control anti exciting converter and power frequency inverter bridge respectively by two driver modules, described MPU controller operates in the following manner:

1. storage capacitor voltage V is gathered by the first voltage sensor _d, gather inverter output voltage V by the second voltage sensor _o, the power frequency inverter bridge side inductive current i of LCL filter is gathered by current sensor _l; Using inverter as object, set up and train corresponding neural network contrary modeling, specific as follows:

A. the inversion model of three layers of BP neural network system is chosen, wherein, input layer nodes is 9, hidden layer neuron nodes is 10, output layer neuron node number is 1, hidden layer neuron transfer function uses hyperbolic tangent function, and output layer neuron transfer function uses S type function;

B. on inverter, service data is gathered;

Under making blower fan be in the condition of rated wind speed, 80% rated wind speed, 60% rated wind speed, 40% rated wind speed and 20% rated wind speed respectively, inverter is made to work in the operating mode of rated power, 80% rated power, 60% rated power, 40% rated power and 20% rated power respectively by changing load under often kind of wind friction velocity, amount to 25 kinds of operating modes, single loop PI controller is all utilized to carry out inverter output voltage control under often kind of operating mode, and continuous acquisition N group storage capacitor voltage V _d, inverter output voltage V _o, anti exciting converter pwm signal duty ratio D, LCL filter power frequency inverter bridge side inductive current i _l, form 25N group service data altogether;

C. the service data structure sample of collection is utilized;

For the k moment, the sample constructed is { V _o(k+1), V _o(k), V _o(k-1), V _o(k-2), i _l(k), i _l(k-1), D (k-1), D (k-2), V _d(k), D (k) }, wherein, V _o(k+1) be the inverter output voltage in k+1 moment, V _ok () is the inverter output voltage in k moment, V _o(k-1) be the inverter output voltage in k-1 moment, V _o(k-2) be the inverter output voltage in k-2 moment, i _lk power frequency inverter bridge side inductive current that () is the LCL filter in k moment, i _l(k-1) be the power frequency inverter bridge side inductive current of the LCL filter in k-1 moment, the anti exciting converter pwm signal duty ratio that D (k-1) is the k-1 moment, the anti exciting converter pwm signal duty ratio that D (k-2) is the k-2 moment, V _dk storage capacitor voltage that () is the k moment, the anti exciting converter pwm signal duty ratio that D (k) is the k moment, 25N group service data forms 25N sample altogether;

D. the 20N randomly drawed in 25N sample data is individual as training sample, as test samples, trains the three layers of BP neural net set up for all the other 5N; In training process, for the sample in k moment, first input getting BP neural net is the inverter output voltage V in k moment _ok (), second input is the inverter output voltage V in k-1 moment _o(k-1), the 3rd input is the inverter output voltage V in k-2 moment _o(k-2), the 4th input is the power frequency inverter bridge side inductive current i of the LCL filter in k moment _lk (), the 5th input is the power frequency inverter bridge side inductive current i of the LCL filter in k-1 moment _l(k-1), the 6th input is the inverter output voltage V in k+1 moment _o(k+1), the 7th input is the storage capacitor voltage V in k moment _d(k), 8th input is the anti exciting converter pwm signal duty ratio D (k-1) in k-1 moment, 9th input is the anti exciting converter pwm signal duty ratio D (k-2) in k-2 moment, and the output of BP neural net is anti exciting converter pwm signal duty ratio D (k) in k moment;

2. utilize the BP neural network contrary modeling that trains and PI controller to control inverter, concrete steps are:

A. utilize the BP neural network contrary modeling trained to carry out the calculating of anti exciting converter pwm signal duty ratio, for the current k moment, get the inverter output voltage V that first input is the k moment _ok (), second input is the inverter output voltage V in k-1 moment _o(k-1), the 3rd input is the inverter output voltage V in k-2 moment _o(k-2), the 4th input is the power frequency inverter bridge side inductive current i of the LCL filter in k moment _lk (), the 5th input is the power frequency inverter bridge side inductive current i of the LCL filter in k-1 moment _l(k-1), the 6th input is the inverter output voltage set point V in k+1 moment ^* _o(k+1), the 7th input is the storage capacitor voltage V in k moment _d(k), 8th input is the anti exciting converter pwm signal duty ratio D (k-1) in k-1 moment, 9th input is the anti exciting converter pwm signal duty ratio D (k-2) in k-2 moment, show that the output of BP neural net is d (k);

B. by subsequent time inverter output voltage set point V ^* _o(k+1) with current k moment inverter output voltage V _ok the deviation of () sends into PI controller, obtain closed-loop control amount d _c(k), namely

$d_C\left(k\right)=K_{p}e\left(k\right)+K_i\underset{j=0}{\overset{k}{\Σ}}e\left(k\right)$

Wherein: e (k)=V ^* _o(k+1)-V _o(k), K _prepresent proportionality coefficient, K _irepresent integral coefficient;

C. neural network contrary modeling and PI controller are formed composite controller, it exports as output d (k) of neural network contrary modeling and the output d of PI controller _ck () be K:(1-K proportionally) superposition, as final anti exciting converter pwm signal duty ratio D (k), that is:

D(k)＝Kd(k)+(1-K)d _C(k)

Wherein, the value of K is:

$K=\left\{\begin{array}{cc}0.9& if\&lsqb;(-\frac{1}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<\frac{1}{12}\mathrm{\&pi;})or(-\mathrm{\&pi;}<\mathrm{\&theta;}<-\frac{11}{12}\mathrm{\&pi;})or(\frac{11}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<\mathrm{\&pi;})\&rsqb;\\ 0.1& if\&lsqb;(-\frac{7}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<-\frac{5}{12}\mathrm{\&pi;})or(\frac{5}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<\frac{7}{12}\mathrm{\&pi;})\&rsqb;\\ 0.5& else\end{array}\right.$

In formula, θ is the phase place of inverter output voltage set point.

Above-mentioned from net type wind power generation inverter, the error back propagation algorithm becoming learning rate is adopted to the training of BP neural net, be specially: first the weight coefficient in each layer neuron and threshold value are initialized as (-1, + 1) random quantity between, again using input layer 9 neuron output values as each neuronic input of hidden layer, 10 neuron output values of hidden layer are as the single neuronic input of output layer, the single neuronic output valve of output layer is network and finally exports, for a kth training sample, calculate the rear actual output of network of the t time training with desired output error:

$\begin{array}{cc}{\mathrm{\&delta;}}_k^t=(y_k^t-c_k^t)& k=1,2,\mathrm{...},20N\end{array}$

And the error of sum square of 20N training sample is:

$E_t=\underset{k=1}{\overset{20N}{\&Sigma;}}{(y_k^t-c_k^t)}^2/=\underset{k=1}{\overset{20N}{\&Sigma;}}{\left({\mathrm{\&delta;}}_k^t\right)}^2/2$

Often calculate one time, compare E _twith anticipation error E _oif, E _t<E _o, then train end condition to meet, training terminates; Otherwise, by E _tcarry out backpropagation along access path, and adjust weight coefficient and the threshold value of each layer one by one, until E _t<E _otill; Then, 5N test samples is utilized to test model, calculate sample standard deviation side error delta MSE, if there is Δ MSE<Tr (Tr is a certain fixed threshold), then think that trained BP neural network model meets required precision, now record each neuronic weight coefficient and threshold value; Otherwise, again BP neural network model is trained, till it meets required precision.

Above-mentioned from net type wind power generation inverter, described anti exciting converter comprises transformer, the first power tube, diode and the second electric capacity, the primary coil of described transformer connects storage capacitor voltage through the first power tube, its secondary coil is powered to power frequency inverter bridge through diode, second electric capacity is connected to the direct-flow input end of power frequency inverter bridge, and the grid of the first power tube connects the output of the first driver module.

Above-mentioned from net type wind power generation inverter, described power frequency inverter bridge is the fully controlled bridge be connected into by four power tubes, and the grid of four power tubes connects the output of the second driver module.

Above-mentioned from net type wind power generation inverter, described LCL filter comprises power frequency inverter bridge side inductance, the 3rd electric capacity and load-side inductance, first ac output end of power frequency inverter bridge side inductance one termination power frequency inverter bridge, the positive pole of an other termination current sensor, the positive pole of the 3rd electric capacity connects the negative pole of current sensor, the negative pole of the 3rd electric capacity connects the second ac output end of power frequency inverter bridge, the measuring-signal output of current sensor is connected with MPU controller, the positive pole of load-side inductance one termination the 3rd electric capacity, the positive pole of an other terminating load.

The composite controller that the present invention utilizes neural network contrary modeling and PI controller to form carrys out the output voltage of control inverter.At near zero-crossing point, the set point change rate of inverter output voltage is maximum, now based on neural network contrary modeling control, to give full play to the advantage of its fast response time; Near line voltage peak value, the set point change rate of inverter output voltage is minimum, now based on PI control, to give full play to the good advantage of its steady-state behaviour, thus effectively improves the quality of power supply from net type wind generator system.

Accompanying drawing explanation

Below in conjunction with accompanying drawing, the invention will be further described.

Fig. 1 is the structured flowchart of wind power generation inverter device;

Fig. 2 is neural network contrary modeling schematic diagram;

Fig. 3 is single neuronic structural representation;

Fig. 4 is wind power generation inverter output voltage curve.

In figure, each list of reference numerals is: GS, blower fan, UT1, the first voltage sensor, UT2, the second voltage sensor, CT, current sensor, D1, diode, Q1 ~ Q5, the first power tube ~ the 5th power tube, C1, storage capacitor, C2, the second electric capacity, C3, the 3rd electric capacity, L1, power frequency inverter bridge side inductance, L2, load-side inductance, TX1, transformer.

In literary composition, each symbol inventory is: V _dfor storage capacitor voltage, V _ofor inverter output voltage, i _lfor LCL filter power frequency inverter bridge side inductive current, D is anti exciting converter pwm signal duty ratio, V _dk storage capacitor voltage that () is the k moment, V _ok () is the inverter output voltage in k moment, V _o(k-1) be the inverter output voltage in k-1 moment, V _o(k-2) be the inverter output voltage in k-2 moment, i _lk LCL filter power frequency inverter bridge side inductive current that () is the k moment, i _l(k-1) be the LCL filter power frequency inverter bridge side inductive current in k-1 moment, V _o(k+1) be the inverter output voltage in k+1 moment, V ^* _o(k+1) be the inverter output voltage set point in k+1 moment, the anti exciting converter pwm signal duty ratio that D (k-1) is the k-1 moment, the anti exciting converter pwm signal duty ratio that D (k-2) is the k-2 moment, the anti exciting converter pwm signal duty ratio that D (k) is the k moment for the actual output of network after t training, for desired output after t training, E _ofor anticipation error, E _tfor the error of sum square of 20N training sample after t training.

Embodiment

Wind power generation inverter of the present invention comprises: rectifier, storage capacitor C1, anti exciting converter is (by transformer TX1, first power tube Q1, diode D1 and the second electric capacity C2 is formed), power frequency inverter bridge (being made up of the second power tube Q2 ~ the 5th power tube Q5), LCL filter is (by the 3rd electric capacity C3, first inductance L 1, second inductance L 2 is formed), first driver module (driver module 1 namely in Fig. 1), second driver module (driver module 2 namely in Fig. 1), MPU controller, first voltage sensor UT1, second voltage sensor UT2 and current sensor CT.

The three-phase input end of described rectifier is connected with the three-phase output end of blower fan, and the single-phase output plus terminal of rectifier is connected with the positive pole of storage capacitor, the single-phase output negativing ending grounding of rectifier; Storage capacitor minus earth; The voltage input anode to be measured of the first voltage sensor is connected with storage capacitor positive pole, the voltage input negativing ending grounding to be measured of the first voltage sensor, and the measuring-signal output of the first voltage sensor is connected with MPU controller; The input anode of anti exciting converter is connected with storage capacitor positive pole, the input negativing ending grounding of anti exciting converter; The output plus terminal of anti exciting converter is connected with power frequency inverter bridge one input, and the output negative terminal of anti exciting converter is connected with another input of power frequency inverter bridge; One output of power frequency inverter bridge is connected with LCL filter one input, and another input of another output and LCL filter of power frequency inverter bridge is connected; One output of LCL filter inputs anode with the voltage to be measured of the second voltage sensor and is connected, and another output of LCL filter inputs negative terminal with the voltage to be measured of the second voltage sensor and is connected; The positive pole of current sensor is connected with other one end of inductance, power frequency inverter bridge side, and the negative pole of current sensor is connected with the positive pole of the 3rd electric capacity, and the measuring-signal output of current sensor is connected with MPU controller; The input of the first driver module is connected with MPU controller, and the output of the first driver module is connected with the first power tube grid in anti exciting converter; The input of the second driver module is connected with MPU controller, one output of the second driver module is connected with the grid of the second power tube in power frequency inverter bridge, the 5th power tube, and another output of the second driver module is connected with the grid of the 3rd power tube in power frequency inverter bridge, the 4th power tube; MPU controller is connected with the measuring-signal output of the first voltage sensor, the second voltage sensor, current sensor, and MPU controller is also connected with the input of the first driver module, the second driver module; Load positive pole connects the other one end of load-side inductance, and load negative pole connects the negative pole of the 3rd electric capacity.

The present invention is further illustrated below in conjunction with the drawings and specific embodiments.

In the present embodiment, blower fan GS is minitype permanent magnetism direct-drive aerogenerator, and rotor diameter is 1.2m, and rated power is 300W, and rated voltage is 24V, rated speed 800r/min, threshold wind velocity 1m/s, rated wind speed 10m/s.

For this wind generator design inverter structure block diagram as shown in Figure 1.This device mainly comprises: rectifier, storage capacitor C1, anti exciting converter, power frequency inverter bridge, LCL filter, the first driver module, the second driver module, MPU controller, the first voltage sensor UT1, the second voltage sensor UT2, current sensor CT; Wherein the high performance float-point digital signal processor TMS320F28335 of TI company selected by MPU controller, and this digital signal processor is integrated with the functional modules such as I/O module, ADC module, PWM module, CAN module, UART and SPI module; Storage capacitor C1 selects 2200uF electrochemical capacitor; In anti exciting converter, transformer TX1 adopts NA5814-AL, and the first power tube Q1 selects TK50X15J1, and diode D1 adopts C2D05120E, and the second electric capacity C2 selects 0.47uF polarity free capacitor; Second power tube Q2, the 3rd power tube Q3, the 4th power tube Q4, the 5th power tube Q5 select IPB60R190C6, and the first driver module and the second driver module are MCP14E4; Power frequency inverter bridge side inductance L 1 in LCL filter selects 0.5mH inductance, and load-side inductance L 2 selects 0.1mH inductance, and the 3rd electric capacity C3 selects 0.47uf polarity free capacitor, and the first voltage sensor UT1 and the second voltage sensor all adopt MCP6022; Current sensor CT adopts Hall current sensor ACS712ELCTR-058-1.

Fig. 2 is set up three layers of BP neural network contrary modeling.Input layer nodes is 9, and hidden layer neuron nodes is 10, and output layer neuron node number is 1.

Fig. 3 is single neuronic structural representation, x ₀, x ₁... x _ifor neuronic input signal, w _ijfor neuronic weight coefficient, θ _jfor neuronic threshold value, then neuronic output is:

$y_j=f(\underset{i}{\&Sigma;}{w}_{ij}*x_i-{\mathrm{\&theta;}}_j)$

Hidden layer neuron transfer function uses hyperbolic tangent function, namely

$f\left(x\right)=\frac{e^{2x}-e^{-2x}}{e^{2x}+e^{-2x}}$

Wherein:

X is transfer function input;

F (x) is transfer function output.

Output layer neuron transfer function uses S type function, namely

$f\left(x\right)=\frac{1}{1+e^{-x}}$

Wherein:

X is transfer function input;

F (x) is transfer function output.

Blower fan is made to be in rated wind speed (10m/s) respectively, 80% rated wind speed (8m/s), 60% rated wind speed (6m/s), under the condition of 40% rated wind speed (4m/s) and 20% rated wind speed (2m/s), inverter is made to work in rated power (300W) respectively by changing load under often kind of wind friction velocity, 80% rated power (240W), 60% rated power (180W), the operating mode of 40% rated power (120W) and 20% rated power (60W), amount to 25 kinds of operating modes, single loop PI controller is all utilized to carry out inverter output voltage control under often kind of operating mode, with the sample frequency continuous acquisition N=4000 group storage capacitor voltage V of 40KHz _d, inverter output voltage V _o, anti exciting converter pwm signal duty ratio D, LCL filter power frequency inverter bridge side inductive current i _l, form 25N=100000 sample altogether, randomly draw wherein 80000 as training sample, all the other 20000 as test samples.

In embodiment, E is set _o=4.0, threshold value Tr=0.02, after repetitive exercise 15422 times, Et=3.92, meets end condition E _t<E _o, training terminates.Now, recycling test samples is tested model, calculates sample standard deviation side error delta MSE=0.014, has Δ MSE<Tr, therefore think and meet inversion model required precision, now record each neuronic weight coefficient and threshold value.

In embodiment, the parameter tuning of PI controller is K _p=0.02, K _i=1200; Inverter output voltage set point V ^* _ochange given by digital signal processor according to discrete sine rule.Write software program according to above-mentioned steps and implant in dsp chip, the blower fan inverter output voltage complex controll based on neural network contrary modeling and PI controller can be realized.

Fig. 4 is the output voltage curve using control method rear fan inverter of the present invention, and visible output voltage shakes without obvious near zero crossing and peak value, and has higher sine degree.

In order to analyze the effect of control method of the present invention further, under different operating mode, itself and conventional PI control method (are got K respectively _p=0.02, K _i=1200) contrast, table 1 is the inverter output voltage harmonic content comparing result of two kinds of methods, known the inventive method has good robustness for the change of operating mode, and the inverter output voltage harmonic content under different operating mode is all significantly less than conventional PI control method.

Table 1: the Contrast on effect of composite control method of the present invention and PI control method

The above; be only the present invention's preferably embodiment; but protection scope of the present invention should be not limited thereto; the core technology that any those skilled in the art of being familiar with disclose according to the present invention, the change that can make or replacement, all should be encompassed within protection scope of the present invention.

Claims (5)

1. one kind from net type wind power generation inverter, it is characterized in that, the rectifier that formation comprises MPU controller and is connected to successively between wind-driven generator and load, storage capacitor (C1), anti exciting converter, power frequency inverter bridge and LCL filter, described MPU controller gathers storage capacitor voltage and inverter output voltage respectively by the first voltage sensor (UT1) and the second voltage sensor (UT2), the power frequency inverter bridge side inductive current of LCL filter is gathered by current sensor (CT), and control anti exciting converter and power frequency inverter bridge respectively by two driver modules, described MPU controller operates in the following manner:

1. storage capacitor voltage V is gathered by the first voltage sensor (UT1) _d, gather inverter output voltage V by the second voltage sensor (UT2) _o, the power frequency inverter bridge side inductive current i of LCL filter is gathered by current sensor (CT) _l; Using inverter as object, set up and train corresponding neural network contrary modeling, specific as follows:

A. the inversion model of three layers of BP neural network system is chosen, wherein, input layer nodes is 9, hidden layer neuron nodes is 10, output layer neuron node number is 1, hidden layer neuron transfer function uses hyperbolic tangent function, and output layer neuron transfer function uses S type function;

B. on inverter, service data is gathered;

Under making blower fan be in the condition of rated wind speed, 80% rated wind speed, 60% rated wind speed, 40% rated wind speed and 20% rated wind speed respectively, inverter is made to work in the operating mode of rated power, 80% rated power, 60% rated power, 40% rated power and 20% rated power respectively by changing load under often kind of wind friction velocity, amount to 25 kinds of operating modes, single loop PI controller is all utilized to carry out inverter output voltage control under often kind of operating mode, and continuous acquisition N group storage capacitor voltage V _d, inverter output voltage V _o, anti exciting converter pwm signal duty ratio D, LCL filter power frequency inverter bridge side inductive current i _l, form 25N group service data altogether;

C. the service data structure sample of collection is utilized;

For the k moment, the sample constructed is { V _o(k+1), V _o(k), V _o(k-1), V _o(k-2), i _l(k), i _l(k-1), D (k-1), D (k-2), V _d(k), D (k) }, wherein, V _o(k+1) be the inverter output voltage in k+1 moment, V _ok () is the inverter output voltage in k moment, V _o(k-1) be the inverter output voltage in k-1 moment, V _o(k-2) be the inverter output voltage in k-2 moment, i _lk power frequency inverter bridge side inductive current that () is the LCL filter in k moment, i _l(k-1) be the power frequency inverter bridge side inductive current of the LCL filter in k-1 moment, the anti exciting converter pwm signal duty ratio that D (k-1) is the k-1 moment, the anti exciting converter pwm signal duty ratio that D (k-2) is the k-2 moment, V _dk storage capacitor voltage that () is the k moment, the anti exciting converter pwm signal duty ratio that D (k) is the k moment, 25N group service data forms 25N sample altogether;

D. the 20N randomly drawed in 25N sample data is individual as training sample, as test samples, trains the three layers of BP neural net set up for all the other 5N; In training process, for the sample in k moment, first input getting BP neural net is the inverter output voltage V in k moment _ok (), second input is the inverter output voltage V in k-1 moment _o(k-1), the 3rd input is the inverter output voltage V in k-2 moment _o(k-2), the 4th input is the power frequency inverter bridge side inductive current i of the LCL filter in k moment _lk (), the 5th input is the power frequency inverter bridge side inductive current i of the LCL filter in k-1 moment _l(k-1), the 6th input is the inverter output voltage V in k+1 moment _o(k+1), the 7th input is the storage capacitor voltage V in k moment _d(k), 8th input is the anti exciting converter pwm signal duty ratio D (k-1) in k-1 moment, 9th input is the anti exciting converter pwm signal duty ratio D (k-2) in k-2 moment, and the output of BP neural net is anti exciting converter pwm signal duty ratio D (k) in k moment;

2. utilize the BP neural network contrary modeling that trains and PI controller to control inverter, concrete steps are:

A. utilize the BP neural network contrary modeling trained to carry out the calculating of anti exciting converter pwm signal duty ratio, for the current k moment, get the inverter output voltage V that first input is the k moment _ok (), second input is the inverter output voltage V in k-1 moment _o(k-1), the 3rd input is the inverter output voltage V in k-2 moment _o(k-2), the 4th input is the power frequency inverter bridge side inductive current i of the LCL filter in k moment _lk (), the 5th input is the power frequency inverter bridge side inductive current i of the LCL filter in k-1 moment _l(k-1), the 6th input is the inverter output voltage set point V in k+1 moment ^* _o(k+1), the 7th input is the storage capacitor voltage V in k moment _d(k), 8th input is the anti exciting converter pwm signal duty ratio D (k-1) in k-1 moment, 9th input is the anti exciting converter pwm signal duty ratio D (k-2) in k-2 moment, show that the output of BP neural net is d (k);

B. by subsequent time inverter output voltage set point V ^* _o(k+1) with current k moment inverter output voltage V _ok the deviation of () sends into PI controller, obtain closed-loop control amount d _c(k), namely

$d_C\left(k\right)=K_{p}e\left(k\right)+K_i\underset{j=0}{\overset{k}{\&Sigma;}}e\left(k\right)$

Wherein: e (k)=V ^* _o(k+1)-V _o(k), K _prepresent proportionality coefficient, K _irepresent integral coefficient;

C. neural network contrary modeling and PI controller are formed composite controller, it exports as output d (k) of neural network contrary modeling and the output d of PI controller _ck () be K:(1-K proportionally) superposition, as final anti exciting converter pwm signal duty ratio D (k), that is:

D(k)＝Kd(k)+(1-K)d _C(k)

Wherein, the value of K is:

$K=\left\{\begin{array}{cc}0.9& if\&lsqb;(-\frac{1}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<\frac{1}{12}\mathrm{\&pi;})or(-\mathrm{\&pi;}<\mathrm{\&theta;}<-\frac{11}{12}\mathrm{\&pi;})or(\frac{11}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<\mathrm{\&pi;})\&rsqb;\\ 0.1& if\&lsqb;(-\frac{7}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<-\frac{5}{12}\mathrm{\&pi;})or(\frac{5}{12}\mathrm{\&pi;}<\mathrm{\&theta;}<\frac{7}{12}\mathrm{\&pi;})\&rsqb;\\ 0.5& else\end{array}\right.$

In formula, θ is the phase place of inverter output voltage set point.

2. according to claim 1 from net type wind power generation inverter, it is characterized in that, the error back propagation algorithm becoming learning rate is adopted to the training of BP neural net, be specially: first the weight coefficient in each layer neuron and threshold value are initialized as (-1, + 1) random quantity between, again using input layer 9 neuron output values as each neuronic input of hidden layer, 10 neuron output values of hidden layer are as the single neuronic input of output layer, the single neuronic output valve of output layer is network and finally exports, for a kth training sample, calculate the rear actual output of network of the t time training with desired output error:

$\begin{array}{cc}{\mathrm{\&delta;}}_k^t=(y_k^r-c_k^t)& k=1,2,...,20N\end{array}$

And the error of sum square of 20N training sample is:

$E_t=\underset{k=1}{\overset{20N}{\&Sigma;}}{(y_k^t-c_k^t)}^2/2=\underset{k=1}{\overset{20N}{\&Sigma;}}{\left({\mathrm{\&delta;}}_k^t\right)}^2/2$

Often calculate one time, compare E _twith anticipation error E _oif, E _t<E _o, then train end condition to meet, training terminates; Otherwise, by E _tcarry out backpropagation along access path, and adjust weight coefficient and the threshold value of each layer one by one, until E _t<E _otill; Then, 5N test samples is utilized to test model, calculate sample standard deviation side error delta MSE, if there is Δ MSE<Tr (Tr is a certain fixed threshold), then think that trained BP neural network model meets required precision, now record each neuronic weight coefficient and threshold value; Otherwise, again BP neural network model is trained, till it meets required precision.

3. according to claim 1 and 2 from net type wind power generation inverter, it is characterized in that, described anti exciting converter comprises transformer (TX1), the first power tube (Q1), diode (D1) and the second electric capacity (C2), the primary coil of described transformer (TX1) connects storage capacitor voltage through the first power tube (Q1), its secondary coil is powered to power frequency inverter bridge through diode (D1), second electric capacity (C2) is connected to the direct-flow input end of power frequency inverter bridge, and the grid of the first power tube (Q1) connects the output of the first driver module.

4. according to claim 3ly it is characterized in that from net type wind power generation inverter, described power frequency inverter bridge is the fully controlled bridge be connected into by four power tubes, and the grid of four power tubes connects the output of the second driver module.

5. according to claim 4 from net type wind power generation inverter, it is characterized in that, described LCL filter comprises power frequency inverter bridge side inductance (L1), 3rd electric capacity (C3) and load-side inductance (L2), first ac output end of power frequency inverter bridge side inductance (L1) termination power frequency inverter bridge, the positive pole of an other termination current sensor (CT), the positive pole of the 3rd electric capacity (C3) connects the negative pole of current sensor (CT), the negative pole of the 3rd electric capacity (C3) connects the second ac output end of power frequency inverter bridge, the measuring-signal output of current sensor (CT) is connected with MPU controller, the positive pole of load-side inductance (L2) termination the 3rd electric capacity (C3), the positive pole of an other terminating load.