CN103835878B - Maximum power point tracing control method based on neural network optimization starting rotating speed - Google Patents

Abstract

The invention provides a maximum power point tracing control method based on the neural network optimization starting rotating speed. According to the method, on the basis of an existing MPPT control method, a neural network is adopted to obtain the best starting generator rotating speed according to wind speed condition dynamic optimization compensation factors, and then wind energy capture efficiency is further improved. The adopted neural network takes the average wind speed and turbulence intensity as input, and takes the best compensation factor as output. Mass draining data obtained through the traversal algorithm are used for training the neural network, the trained neural network is adopted for obtaining the corresponding best compensation factor through calculation according to the changed wind speed conditions, and then the best compensation factor is used for optimizing the starting generator rotating sped, so that the best MPPT tracking interval is obtained, and the wind energy capture efficiency is further improved. Compared with various traditional MPPT control methods, the effectiveness and superiority of the algorithm are verified.

Description

Maximum power point-tracing control method based on Neural Network Optimization starting rotating speed

Technical field

The invention belongs to wind power generation field, particularly a kind of maximum power point based on Neural Network Optimization starting rotating speed Tracking and controlling method.

Background technology

In order to improve the Wind energy extraction efficiency interval less than rated wind speed, speed-variable frequency-constant wind-driven generator group is typically using most High-power point tracking（Maximum Power Point Tracking, MPPT）Control strategy.Power curve method（Also referred to as power Signal feedback transmitter or torque curve method）It is one of most widely used MPPT control method.

Traditional MPPT controls, particularly power curve method, more based on system stabilization design, and have ignored blower fan system and exist The dynamic process tracked between different steady operation points.But, cause wind wheel constantly to increase in the face of the continuous single-machine capacity for being lifted Rotary inertia and its more slow dynamic response performance, and wind speed is frequently in the wave process and to be difficult short-term pre- Survey, the blower fan system overwhelming majority time under traditional MPPT controls is in dynamic process, rather than operates on steady operation point. Therefore, the actual wind speed tracking effect of blower fan is still to be improved.

For this purpose, the L.J.Fingersh and P.W.Carlin in National Renewable Energy laboratory propose first utilization Generator electromagnetic torque helps the improved though of blower fan acceleration or deceleration；On this basis, Johnson K.E. et al. are proposed Reduce gain of torque（Decreased Torque Gain, DTG）Control.The control method is not only carried by reducing electromagnetic torque High acceleration of the blower fan when crescendo fitful wind is tracked, more first Application is imitated with the rotating-speed tracking for abandoning the low wind speed section in part Fruit exchanges the control thought of the high Wind energy extraction efficiency of high wind speed section for；Further, it is contemplated that DTG controls adopt constant gain Coefficient, Johnson K.E. et al. design self adaptation direct torque again, using adaptive algorithm and the system of history run operating mode Count, iterative search and on-line amending optimum gain coefficient, to respond the change of the wind friction velocity in iteration cycle time scale. But, the crucial problem that the thinking faces is that the optimum state of torque adjustment is closely related with wind friction velocity, but is difficult to find Direct quantitative relationship between them.In consideration of it, Yin Minghui et al. is proposed based on the improvement maximum power point for shrinking trace interval Tracing control, is shortened by tracking distance and reaches the purpose for improving MPPT tracking effects.

The studies above work can be summarized as two Research Thinkings, i.e., by adjusting electromagnetic braking torque and shortening tracking road Two aspects of journey breach conventional power curve method and ignore the dynamic office of tracking improving the dynamic property and tracking effect of blower fan It is sex-limited.However, based on the trace interval that MPPT maximum power point tracking control does not provide optimum that improves for shrinking trace interval（I.e. most Good penalty coefficient）, because the setting of trace interval will significantly affect the tracking effect of MPPT, thus for the optimization of trace interval Seem particularly significant.But associated description is there is no in prior art.

The content of the invention

Technical problem solved by the invention is to provide a kind of peak power based on Neural Network Optimization starting rotating speed Point-tracing control method, obtains optimum rotating-speed tracking interval further to improve Wind energy extraction by the initial generating rotating speed of optimization Efficiency.

The technical solution for realizing the object of the invention is：A kind of peak power based on Neural Network Optimization starting rotating speed Point-tracing control method, based on the improvement power curve method based on initial adjustment of rotational speed, using neutral net adjustment starting Generating rotating speed is realizing MPPT maximum power point tracking control, formula used by the improvement power curve method based on initial adjustment of rotational speed For：

In above formula, M is rotary inertia, T_mFor the Mechanical Driven torque of wind wheel, T_eFor electromagnetic braking torque, v is wind speed, ω For the angular speed of wind wheel,For wind wheel angular acceleration, ρ is atmospheric density, and R is wind wheel radius, C_PFor power coefficient, λ= ω R/v are tip speed ratios, ω_bgnFor the i.e. initial rotating speed of initial generating rotating speed, λ_optFor optimum tip-speed ratio,For in a cycle Wind-speed sample value sequence mean value, α is penalty coefficient, for periodically adjusting initial rotating speed, T_optFor the optimum of blower fan Torque curve, specially：

T_opt(ω)=K_mω²

In above formulaFor maximal wind-energy usage factor；

Wherein initial generating rotational speed omega_bgnThe step of being adjusted using neutral net is as follows：

Step 1, initialized, i.e., to wind-speed sample frequency and initial rotating speed update cycle T_rIt is configured, its apoplexy Fast sample frequency is 1~4Hz；Empty T_rCorresponding wind-speed sample value sequence, by ω_bgnIt is initialized as blower fan maximum power point The minimum speed in tracing control stage；Initial rotating speed update cycle T_rPreferably 20 minutes.

Step 2, training neutral net, i.e., obtain various mean wind speeds using ergodic algorithmWith correspondence under turbulence intensity TI The optimal compensation factor alpha_opt, as the training sample of neutral net；During training, with the mean value of wind speedIt is strong with turbulent flow Degree TI is the input variable of neutral net, with the optimal compensation coefficient as output variable；

Step 3, into new initial rotating speed update cycle T_r, with wind-speed sample cycle T_wIn the cycle T_rIn to air speed value Being sampled measuring wind speed value of read, and preserved to wind-speed sample value sequence；

Step 4, judgement current starting rotating speed update cycle T_rWhether complete, if completing, execution step 5；Otherwise, continue With wind-speed sample cycle T_wIn the cycle T_rIn air speed value sampled and preserved to wind-speed sample value sequence；

Step 5, the mean value for asking for wind-speed sample value sequenceWith turbulence intensity TI；

The mean value of step 6, the wind-speed sample value sequence asked for step 5It is input with turbulence intensity TI, calls god Jing networks obtain corresponding the optimal compensation factor alpha_opt；

Step 7, the optimal compensation factor alpha obtained according to step 6_optTo initial rotational speed omega_bgnIt is adjusted, specially：

By ω_bgnIt is adjusted toAfterwards with the ω after renewal_bgnInto new update cycle T_r, Electromagnetic braking torque is still adjusted as follows

Wind-speed sample value sequence is emptied afterwards, and skips to step 3.

Compared with prior art, its remarkable advantage is the present invention：1）The present invention is using the control of Neural Network Optimization MPPT Trace interval, further increases Wind energy extraction efficiency；2）The present invention adjusts trace interval according to specific wind friction velocity so as to Wind speed can preferably be tracked；3）The present invention directly estimates optimum starting rotating speed with recent mean wind speed, is not easily susceptible to wind speed bar The impact of part change；4）The present invention does not need complicated iterative search procedures, and algorithm is very simple；5）Requirement of the present invention The information of survey is few, and computation burden is light.

The present invention is described in further detail below in conjunction with the accompanying drawings.

Description of the drawings

Fig. 1 is the flow process of the maximum power point-tracing control method based on Neural Network Optimization starting rotating speed of the present invention Figure.

Fig. 2 is the construction wind series of high degree of fluctuation.

Fig. 3 is the maximum power point-tracing control method based on Neural Network Optimization starting rotating speed and its other party of the present invention The comparison diagram of method.

Specific embodiment

The present invention will propose to be compensated according to wind friction velocity dynamic optimization using neutral net based on existing MPPT control method Coefficient is obtaining optimal starting generating rotating speed（I.e. optimal tracking is interval）, and then further improve Wind energy extraction efficiency.

With reference to Fig. 1, a kind of maximum power point-tracing control method based on Neural Network Optimization starting rotating speed of the present invention, Based on the improvement power curve method based on initial adjustment of rotational speed, adjust initial generating rotating speed to realize most using neutral net High-power tracing control, formula used by the improvement power curve method based on initial adjustment of rotational speed is：

In above formula, M is rotary inertia, T_mFor the Mechanical Driven torque of wind wheel, T_eFor electromagnetic braking torque, v is wind speed, ω For the angular speed of wind wheel,For wind wheel angular acceleration, ρ is atmospheric density, and R is wind wheel radius, C_PFor power coefficient, λ= ω R/v are tip speed ratios, ω_bgnFor the i.e. initial rotating speed of initial generating rotating speed, λ_optFor optimum tip-speed ratio,For in a cycle Wind-speed sample value sequence mean value, α is penalty coefficient, for periodically adjusting initial rotating speed, T_optFor the optimum of blower fan Torque curve, specially：

T_opt(ω)=K_mω²

In above formulaFor maximal wind-energy usage factor；

Wherein initial generating rotational speed omega_bgnThe step of being adjusted using neutral net is as follows：

Step 1, initialized, i.e., to wind-speed sample frequency and initial rotating speed update cycle T_rIt is configured, its apoplexy Fast sample frequency is 1~4Hz；Empty T_rCorresponding wind-speed sample value sequence, by ω_bgnIt is initialized as blower fan maximum power point The minimum speed in tracing control stage；Initial rotating speed update cycle T_rPreferably 20 minutes.

Step 2, training neutral net, i.e., obtain various mean wind speeds using ergodic algorithmWith correspondence under turbulence intensity TI The optimal compensation factor alpha_opt, as the training sample of neutral net；During training, with the mean value of wind speedIt is strong with turbulent flow Degree TI is the input variable of neutral net, with the optimal compensation coefficient as output variable；

Step 3, into new initial rotating speed update cycle T_r, with wind-speed sample cycle T_wIn the cycle T_rIn to air speed value Being sampled measuring wind speed value of read, and preserved to wind-speed sample value sequence；

Step 4, judgement current starting rotating speed update cycle T_rWhether complete, if completing, execution step 5；Otherwise, continue With wind-speed sample cycle T_wIn the cycle T_rIn air speed value sampled and preserved to wind-speed sample value sequence；

Step 5, the mean value for asking for wind-speed sample value sequenceWith turbulence intensity TI；

The mean value of step 6, the wind-speed sample value sequence asked for step 5It is input with turbulence intensity TI, calls god Jing networks obtain corresponding the optimal compensation factor alpha_opt；

Step 7, the optimal compensation factor alpha obtained according to step 6_optTo initial rotational speed omega_bgnIt is adjusted, specially：

By ω_bgnIt is adjusted toAfterwards with the ω after renewal_bgnInto new update cycle T_r, Electromagnetic braking torque is still adjusted as follows

Wind-speed sample value sequence is emptied afterwards, and skips to step 3.

Further detailed description is done to the present invention with reference to embodiment：

By simulation calculation and statistical analysis to simulating wind series, to proposed by the present invention based on Neural Network Optimization The maximum power point-tracing control method and conventional power curve method of initial rotating speed, self adaptation direct torque and based on shrink with The interval improvement MPPT controls of track（the improved MPPT control based on reduction of tracking Range, referred to as RTR-MPPT are controlled）It is compared, to verify effectiveness of the invention and superiority.

1st, the simulation model of embodiment

1）Simplify the parameter of blower fan model

The major parameter of blower fan model is set to：Fan capacity 1.0MW, rotor diameter 52.67m, rotary inertia 1.1204 ×10⁶Kgm².The maximum C of blower fan_PValueFor 0.4109, optimum tip-speed ratio λ_optFor 8.0.

2）The construction of wind series

The building method of simulation wind series is illustrated with reference to Fig. 2.According to autoregressive moving average（ARMA）Method, construction 80 The individual duration is the wind speed period of 20 minutes.And permutation and combination is carried out to the above-mentioned wind speed period by wind speed mean value, construct The acclivity of mean wind speed or rise/fall replace slope（As shown in Figure 2）, to construct the wind friction velocity of high degree of fluctuation. The turbulence intensity of each wind speed period is set to the A classes defined in IEC-614000-1 standards（It is high）Turbulent flow rank.

3）The setting of neutral net

The present embodiment is using based on RBF（Radical basis function, RBF）Neutral net, its mistake Difference performance indications are set to 0.1.Error performance index determines fitting degree of the neutral net to training data, will finally determine god Approximation accuracy of the Jing networks to the optimal compensation coefficient.

4）The Realization of Simulation based on the maximum power point-tracing control method of Neural Network Optimization starting rotating speed

ω is periodically optimized using neutral net according to step described in the content of the invention_bgnIt is capable of achieving to be based on neutral net The maximum power point-tracing control method of the initial rotating speed of optimization, it is specific as follows：

Generated electricity based on the improvement power curve method based on initial adjustment of rotational speed and using neutral net adjustment starting and turned Speed periodically optimizes ω realizing MPPT maximum power point tracking control using neutral net_bgnThe step of it is as follows：

Step 1, initialized, i.e., to wind-speed sample frequency and initial rotating speed update cycle T_rIt is configured, its apoplexy Fast sample frequency is 4Hz；T_rFor 20 minutes, T is emptied_rCorresponding wind-speed sample value sequence, by ω_bgnIt is initialized as blower fan maximum The minimum speed in power points tracing control stage；

Step 2, training neutral net, i.e., obtain various mean wind speeds using ergodic algorithmWith correspondence under turbulence intensity TI The optimal compensation factor alpha_opt, as the training sample of neutral net；During training, with the mean value of wind speedIt is strong with turbulent flow Degree TI is the input variable of neutral net, with the optimal compensation coefficient as output variable.

Step 3, into new initial rotating speed update cycle T_r, with wind-speed sample cycle T_wIn the cycle T_rIn to air speed value Being sampled measuring wind speed value of read, and preserved to wind-speed sample value sequence；The wind-speed sample cycle T_wFor 0.25s；

Step 4, judgement current starting rotating speed update cycle T_rWhether complete, if completing, execution step 5；Otherwise, continue With wind-speed sample cycle T_wIn the cycle T_rIn air speed value sampled and preserved to wind-speed sample value sequence；

Step 5, the mean value for asking for wind-speed sample value sequenceWith turbulence intensity TI；

Step 6, asked for step 5Neutral net is called to obtain corresponding the optimal compensation factor alpha for input with TI_opt；

Step 7, pressTo initial rotational speed omega_bgnIt is adjusted, afterwards with the ω after renewal_bgn Into new update cycle T_r.Wind-speed sample value sequence is emptied, and skips to step 3.

2nd, the comparative analysis of Wind energy extraction efficiency

The present invention constructs 75 groups of emulation experiment examples according to above-mentioned wind speed building method.For every group of example, respectively should With being turned based on Neural Network Optimization starting for conventional power curve method, self adaptation direct torque, RTR-MPPT controls and the present invention The maximum power point-tracing control method of speed, and the average wind energy utilization P for passing through the statistical analysis wind speed period_favgCome in comparison State method.P_favgIt is defined as follows：

P_wy=0.5 ρ π R²v³cos³ψ

Wherein, n is a statistical time range（That is iteration cycle）Interior sampling number；ψ be yaw error angle, ignore herein for 0 degree.

The Wind energy extraction efficiency of above-mentioned 4 kinds of methods is contrasted with reference to Fig. 3.Above-mentioned 4 kinds of methods are applied to into the bag of above-mentioned construction Wind series containing 80 wind speed periods, can respectively calculate the P corresponding to every kind of method and the wind series_favgIt is average Value, is designated asSpecially：

Further, calculate what 75 groups of simulation examples were obtainedMean value, be designated asAs shown in Figure 3.By scheming 3 is visible, the maximum power point-tracing control method based on Neural Network Optimization starting rotating speed proposed by the present invention（4th Individual block diagram）2.03% is improve than conventional power curve method, than self adaptation direct torque 0.84% is improve, than RTR-MPPT controls System improves 0.14%.The present invention is demonstrated based on the maximum power point-tracing control method of Neural Network Optimization starting rotating speed Superiority.

Claims (3)

1. a kind of maximum power point-tracing control method based on Neural Network Optimization starting rotating speed, with based on initial adjustment of rotational speed Improvement power curve method based on, adjust initial generating rotating speed to realize MPPT maximum power point tracking control using neutral net, Formula used by the improvement power curve method based on initial adjustment of rotational speed is：

$M\stackrel{\·}{\mathrm{\ω}}=T_m(v,\mathrm{\ω})-T_e\left(\mathrm{\ω}\right)$

$T_m(v,\mathrm{\ω})=\frac{0.5{\mathrm{\ρ\πR}}^5{C}_p\left(\mathrm{\λ}\right)}{{\mathrm{\λ}}^3}{\mathrm{\ω}}^2$

$T_e\left(\mathrm{\&omega;}\right)=\left\{\begin{array}{cc}0& \mathrm{\&omega;}<{\mathrm{\&omega;}}_{bgn}\\ T_{opt}& \mathrm{\&omega;}>{\mathrm{\&omega;}}_{bgn}\end{array}\right.$

${\mathrm{\&omega;}}_{bgn}={\mathrm{\&lambda;}}_{opt}({\stackrel{\&OverBar;}{v}}_T-\mathrm{\&alpha;})/R$

In above formula, M is rotary inertia, T_mFor the Mechanical Driven torque of wind wheel, T_eFor electromagnetic braking torque, v is wind speed, and ω is wind The angular speed of wheel,For wind wheel angular acceleration, ρ is atmospheric density, and R is wind wheel radius, C_PFor power coefficient, λ=ω R/v It is tip speed ratio, ω_bgnFor the i.e. initial rotating speed of initial generating rotating speed, λ_optFor optimum tip-speed ratio,For the wind in a cycle The mean value of fast sampled value sequence, α is penalty coefficient, for periodically adjusting initial rotating speed, T_optFor the optimum torque of blower fan Curve, specially：

T_opt(ω)=K_mω²

$K_m=\frac{0.5{\mathrm{\&rho;\&pi;R}}^5{C}_p^{\max }}{{\mathrm{\&lambda;}}_{opt}^3}$

In above formulaFor maximal wind-energy usage factor；

Characterized in that, initial generating rotational speed omega_bgnThe step of being adjusted using neutral net is as follows：

Step 1, initialized, i.e., to wind-speed sample frequency and initial rotating speed update cycle T_rIt is configured, wherein wind-speed sample Frequency is 1~4Hz；Empty T_rCorresponding wind-speed sample value sequence, by ω_bgnIt is initialized as blower fan MPPT maximum power point tracking control The minimum speed in stage processed；

Step 2, training neutral net, i.e., obtain various mean wind speeds using ergodic algorithmWith under turbulence intensity TI it is corresponding most Good penalty coefficient α_opt, as the training sample of neutral net；During training, with the mean value of wind speedWith turbulence intensity TI For the input variable of neutral net, with the optimal compensation coefficient as output variable；

Step 3, into new initial rotating speed update cycle T_r, with wind-speed sample cycle T_wIn the cycle T_rIn air speed value is carried out Measuring wind speed value is read in sampling, and preserves to wind-speed sample value sequence；

Step 4, judgement current starting rotating speed update cycle T_rWhether complete, if completing, execution step 5；Otherwise, continue with wind Fast sampling period T_wIn the cycle T_rIn air speed value sampled and preserved to wind-speed sample value sequence；

Step 5, the mean value for asking for wind-speed sample value sequenceWith turbulence intensity TI；

The mean value of step 6, the wind-speed sample value sequence asked for step 5It is input with turbulence intensity TI, calls neutral net Obtain corresponding the optimal compensation factor alpha_opt；

Step 7, the optimal compensation factor alpha obtained according to step 6_optTo initial rotational speed omega_bgnBe adjusted, afterwards with renewal after ω_bgnInto new update cycle T_r, electromagnetic braking torque still adjusts as follows

$T_e\left(\mathrm{\&omega;}\right)=\left\{\begin{array}{cc}0& \mathrm{\&omega;}<{\mathrm{\&omega;}}_{bgn}\\ T_{opt}& \mathrm{\&omega;}>{\mathrm{\&omega;}}_{bgn}\end{array}\right.$

Wind-speed sample value sequence is emptied afterwards, and skips to step 3.

2. it is according to claim 1 based on Neural Network Optimization starting rotating speed maximum power point-tracing control method, its It is characterised by, initial rotating speed update cycle T in step 1_rFor 20 minutes.

3. it is according to claim 1 based on Neural Network Optimization starting rotating speed maximum power point-tracing control method, its It is characterised by, to initial generating rotational speed omega in step 7_bgnIt is adjusted specially：

By ω_bgnIt is adjusted to