Control Strategy of Maximum Wind Energy Capture of Direct-Drive Wind Turbine Generator Based on Neural-Network

Y.F. REN
College of electrical and information engineering Lan Zhou University of Technology LanzhouChina yanfeng505@gmail.com
AbstractThe wind power varies mainly depending on the wind speed. Many methods have been proposed to track the maximum power point (MPPT) of the wind, such as the fuzzy logic (FL), artificial neural network (ANN) and Neuro-Fuzzy. In this paper, a variable speed wind generator MPPT based on artificial neural network (ANN) is presented. It is designed as a combination of the generator speed forecasting model and neural network. The ANN is used to predict the optimal speed rotation using the variation of the wind speed and the generator speed as the inputs. The wind energy control system employs a permanent magnet synchronous generator connected to a DC bus using a power converter is presented. The performance of the control system with the proposed ANN controller is tested for wind speed variation. System simulation results have confirmed the functionality and performance of this method. Keywords-Wind power; The permanent magnet synchronous generator; Speed estimator; Neural network

G.Q.BAO
College of electrical and information engineering LanZhou University of Technology LanzhouChina guangqing.bao@gmail.com (supervised or unsupervised), the neural network tries to correlate the correspondence between the input and target data by adjusting its weights. Following many iterations of incremental training with many different inputoutput pairs, the ANN weights are optimally adjusted by comparing its output and the designated target. This depends on various factors, like, the learning algorithm, network architecture etc. ANN can also be trained off-line to emulate existing controllers [3]. Once well trained, the ANN controller could replace the conventional controller with the advantage of increased speed of execution and fault tolerance. In this paper, ANN principles are applied to implement a novel mechanical sensorless peak power extraction. Simulation results verify the effectiveness of the method. II. MAXIMUM WIND ENERGY TRACKING CONTROL PRINCIPLE

I.

INTRODUCTION

As a type of renewable energy, wind power has received considerable attention for producing electricity due to its cost competitiveness in compare with other types of energy which are conventionally used for power generation. Variable speed operation and direct drive generator have been the recent developments in the technology of wind turbine generation system. The use of the variable speed PMSG in wind turbine application can increase the energy capture from wind, and improve efficiency. On the other hand, if a gearbox is used in a wind turbine system, noise, power losses, additional cost, and potential of mechanical failure can cause many problems. At present, most many wind turbine control systems employ anemometers to measure wind velocity in order to derive the desired shaft speed to vary the generator speed. In most cases, a number of anemometers at some distance away from and surrounding the wind turbine are required to provide adequate information. These mechanical sensors increase the cost and reduce the reliability of the overall system [2]. ANN is particularly useful to implement nonlinear time-varying inputoutput mapping. Depending on the learning method

Based on the Betz theory, the mechanical power extracted from wind power by the wind turbine [4] can be expressed as Pm=0.5ACp () Where, : air density (Kg/m) A: swept area (m) Cp: power coefficient of the wind turbine : wind speed (m) In (1), if the air density, swept area, and wind speed are constant the output power of the turbine will be a function of power coefficient of the turbine. In addition, the wind turbine is normally characterized by its Cp- curve; where, , tip-speed ratio, is given by = R (2) (1)

____________________________________________
This work was supported by The National Nature Science Foundation of China (50877034)

Where R is the radius of blades and is the rotational speed of the wind turbine shaft. Fig. 1 shows the Cp- curve for a wind turbine. As can be seen from Fig. 1, at opt, Cp has its maximum value Cpmax which results in the optimum efficiency; In order to extract the maximum mechanical power, m should vary along with the wind speed. Fig. 2 represents the

978-1-4244-4813-5/10/$25.00 2010 IEEE

relation between generator speed and output power according to wind speed change (v1< v2 <v3 < v4 < v5). By changing the generator speed, in theory, wind turbine can output optimal power when the power of the system is lower to medium power, and maintain the same value after reaching the rated power as shown in Fig. 3 .

Figure 1. Power coefficient versus tip- speed ratio

Fig. 4 shows a schematic of the power circuit topology and control system of a variable speed wind turbine system. It consists of a horizontal axis wind turbine compiled to a PMSG. The system adopts AC-DC-DC-AC conversion, the diode-rectifier is uncontrolled, and the control process can be achieved by the DC-DC circuit and the controllable inverter. When the wind speed is lower than the rated value, to track maximum power by controlling the DC - DC circuit to change generator speed, through controlling the frequency, amplitude and phase of inverter output current to track grid security parallel operation parameters. The process of controlling the frequency, amplitude and phase of inverter output current is that the current sensor is used to test the frequency, amplitude and phase of the network and electric generator, Using closed-loop PI control, the current sampling signal Iref of power grid compares to inverter output voltage and then produces control signal, then compared with the triangle carrier signal, the comparison of the signal to control the each bridge conduction. B. Maximum Wind Energy Tracking Control Scheme When the wind speed is lower than the rated value, the main goal of the control system is to get the maximum input power of the wind power system. The rotor speed of the generator depends on the torque of mechanical power and generator electromagnetic torque, as long as the torque of motive can be controlled, the rotor speed of the generator was controlled. The output current of generator, i.e. generator electromagnetic torque is controlled by changing the output voltage and current of the rectifier. This goal is achieved by the DC - DC circuit as shown in Fig. 4, control structure is shown in the dotted box, and also boost circuit can improve PWM inverter running characteristics when PWM inverter is working at very low input voltage. C. Estimatation of Generator Speed Based on the soft measurement method [6], the generator speed was achieved. Equivalent circuit of the permanent magnet synchronous generators connected to diode rectifier is shown in Fig. 5.

Figure 2. Relationship between the power and the rotational speed

Figure 3.

Ideal power curve

id ean ebn ecn

Id

III.

MAXIMUM POWER EXTRACTION CONTROL SYSTEM FOR DIRECT-DRIVE WIND POWER GENERATOR

Rs
0.5Lms

Lls + Lms

ia
ib ic

R R
R

D1

D3

D5

A. System Description
i rlp
vd

Lls + Lms L
0.5Lms

irlp
vd
D4

0.5 Lms

vdc

Rs

Lls + Lms

D6

D2

iL
d
Qref

T1 T

T3 T4 T5 T6

vw

iL _ ref
iL

I dc_ ref
I dc
Pref

iabc vabc

Figure 5. Equivalent circuit of permanent magnet generator connected to the diode-rectifier

ref

The average output voltage of the rectifier based on Fig. 5 can be formulated as below:
vd = 3 3Em

Figure 4. Power circuit topology and control structure for the wind energy conversion system

3(L M )

id 2RId

(2) (3)

Where

Em=Km

pm = 60

(4)

Substituting from (3) and (4) into (2) and solve it for m results in:
m = d + 2(R + Rs )I d
3 3 p Km 20(Ls + L )I d

vd

il

pe

pm = Jm

dm + pe dt

pm

m
(5)

pm

vw

Where, Km is peak line-to-neutral back emf constant in rpmKm =1.4rpm. Equation (5) can be used to estimate the generator speed just by measuring the average output voltage and current of the diode rectifier. This method can solve the problem that system failure and maintenance costs. IV. THE WIND SPEED PREDICTION BASED ON ARTIFICIAL NEURAL NETWORK

Figure 6. An off-line trained scheme for ANN-based wind velocity estimation

W1

Pm

B1

W2

B2

Vw

The mechanical sensorless peak power tracking control method has been reported in the literature. S. Bhowmik uses a power coefficient polynomial to estimate wind velocity [7]. In this method, an iterative algorithm is employed to determine the polynomial roots of the power coefficient online. Since the polynomial of power coefficient may be seventh order, real-time calculation of the polynomial roots will result in complex and time-consuming calculation, hence, reducing system performance. K. Tan applies a two-dimensional (2-D) lookup table of power coefficient and power-mapping method to estimate the wind velocity directly or indirectly [8]. The power-mapping technique may occupy a lot of memory space. If memory space is saved by reducing the size of lookup table, the control accuracy will be affected. ANN is particularly useful to implement nonlinear time-varying inputoutput mapping. ANN is a newly developed intelligent information processing system based on biological neural network, it is highly parallel, nonlinear global role, good fault-tolerance and associative memory function, but due to the actual situation is great different, there is no neural network model can be applicable to the various conditions. Combined with actual situation and features of network to solve the problem, this paper selects feed forward type BP neural network. This paper used the ANN in the wind control system, and to establish a new intelligent MPPT control algorithm based on the information extraction of the wind power characteristics. The offline training model is shown in Fig. 6. In this scheme, sample data Pm is produced from turbine power equation with pre-selected rotor speed, the output voltage and current of the rectifier. The rotor speed and power samples are then recombined as k data pairs (m(i), Pm(i) |i =1~k) which are employed as input matrix of the neural network. On the other hand, the wind velocity sample are used as target to train a three-layer network as shown in Fig. 7, which is configured with two linear neurons in the input layer, five tan-sigmoid neurons in the hidden layer, and one linear neuron in the output layer.

Figure 7. Wind velocity estimation ANN-based

V.

SIMULATION STUDY

The simulation uses the small power prototype generator which was used in Jiuquan wind power field located in west of China. Jiuquan wind power field is the first construction of the million kilowatt wind farm in China has approved the scale of 500 million kilowatts. A. System parameters 1) wind turbine parameters Using wind machine Aircon10 form Aircon company Power rating:10 kW Rotor diameter:7.4m Rated wind speed:11m/s Cut-in speed:2.5m/s Cut-out speed:37m/s Maximum wind speed:58m/s 2) generator parameters Three-phase 7-kw PMSG Phase resistance: RS=1.764 Phase inductance: LS=4.48mH Flux linkage: =1.05s Number of poles: p=36 Total inertia of turbine and generator: J=2Kgm B. Offline training neural network model In this paper, the sample data was obtained based on the 7KW wind generators simulation system, the average speed range is 8m/s-17m/s, the sampling period is 1 second, average speed increase 1m every 100 seconds, So get 1000 sample data pairs. Sample data including efficient wind velocity, the speed of the turbine, and the generator power samples values of the

wind generator in almost all operating conditions. Fig. 8 shows the curve of ANN training samples.

C. The measurements of output current and voltage Fig. 11 shows the current and the voltage of generator side DC bus across the filter capacitor. The estimated generator speed is calculated based on (5), calculated the wind turbine output power form (6), and sample data will be entered into the trained ANN model to get the predicted wind speed. With estimated wind velocity by the ANN, rotor speed reference will be generated from (1) with a gain K = opt /rm. dr (6) Pm=Pe+Jr dt VI. CONCLUSION In this paper, the ANN is presented for small wind turbine directly driven permanent magnet synchronous generator system. Compared to the traditional control strategies, the new method has the following features: (1) A maximum mechanical power of the wind turbine can be well tracked at both dynamic and steady states; (2) A neural network based wind velocity estimator is developed to provide fast and accurate velocity information to avoid using anemometers; and (3) This method combines the generator speed forecasting algorithm and the wind speed measurement model based on neural network. In the entire system, only the output voltage and current values of the rectifier was tested. This method reduced the system failure rate, and saved the cost of the design. The theoretical concept of the control system has been verified by the simulation study of the PMSG system. REFERENCE
[1] Z. Chen and E. Spooner,Wind turbine power converters: a comparative study, in Proc. 1998 IEE Power Electronics and Variable Speed Drives, Seventh International Conference, pp. 471 476. K. Tan and S. Islam, Optimal control strategies in energy conversion of PMSG wind turbine system without mechanical sensors, IEEE Trans. Energy Convers., vol. 19, no. 2, pp. 392399, Jun. 2004. M. R. Buhl and R. D. Lorenz, Design and implementation of neural networks for digital current regulation of inverter drives, in Conf. Rec.IEEE Industry Applications Society (IAS) Annu. Meeting, Dearborn, MI, 1991, pp. 415421. WANG Cheng-xi, ZHANG Yuan. wind power generation[M]. Beijing:China Electric Power Press[CEPP],2003. WU Di; ZHANG Jian-wen. Control System of Variable-speed Permanent-magnet Generator Directly Driven by Wind Turbine[J]. Large Electric Machine and Hydraulic Turbine, 2006,No6:51-55. Reza Esmaili, Longya Xu,Sensorless Control of Permanent Magnet Generator in Wind Turbine Application IEEE Trans. Ind. Applicat. pp.2070-2075.2006 S. Bhowmik, R. Spee, and J. H. R. Enslin, Performance optimization for doubly fed wind power generation systems, IEEE Trans. Ind. Appl.,vol. 35, no. 4, pp. 949958, Jul./Aug. 1999. K. Tan and S. Islam, Optimal control strategies in energy conversion of PMSG wind turbine system without mechanical sensors, IEEE Trans. Energy Convers., vol. 19, no. 2, pp. 392399, Jun. 2004.

Figure 8. The curve of Neural network training samples

Figure 9. Test samples and estimated samples

[2]

[3] Figure 10. Wind velocity error [4] [5]

[6]

[7]

[8]

Figure 11.

The sampled value of Rectifier output voltage and current

Use the received sample data to train the ANN model as shown in Fig. 5, and then take 200 other samples to test the model, the testing result is shown in Fig. 9. Fig. 10 shows the training error, the average error is 0.35. So the ANN can better tracking the test samples and the ANN model can be used in the wind power generation systems to predict velocity.