Document not found! Please try again

Robust Control of Doubly Fed Induction Generator for Wind ... - Core

3 downloads 0 Views 497KB Size Report
In addition, the power converter is usually rated at 25-30% of the generator power rating ... wind into electrical energy through a generator calibrated according to .... wind power control “MPP” (Maximum Power Point) from the wind for a wide.
Available online at www.sciencedirect.com

ScienceDirect Energy Procedia 74 (2015) 886 – 899

Robust Control of Doubly Fed Induction Generator for Wind Turbine Under Sub-Synchronous Operation Mode Khouloud Bedouda,b *, Mahieddine Ali-rachedic, Tahar Bahid, Rabah Lakelb, Azzeddine Grida a Welding and NDT Research Centre (CSC). BP 64 Cheraga, Algeria. Automatic Laboratory and Signals, Badji Mokhtar University, Annaba, Algeria. c Preparatory School in Sciences and Technics, Annaba, Algeria. d Electrical department, Badji Mokhtar University, Annaba,Box 12, 2300 Algeria. b

Abstract This paper presents a modeling and a robust control of doubly fed induction generator for wind generation system. The whole system is presented in d-q-synchronous reference frame. The regulation of the electromagnetic torque , stator reactive power control and neuronal controller are applied in order to control the rotor currents of the DFIG. For to improve the controller robustness, the study is validated through simulation using software Matlab/Simulink, studies on a 1.5 MW DFIG wind generation system compared with conventional proportional integral controller. Performance and robustness results obtained will be presented and analyzed. © by Elsevier Ltd. This is an openLtd. access article under the CC BY-NC-ND license © 2015 2015Published The Authors. Published by Elsevier (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Euro-Mediterranean Institute for Sustainable Development (EUMISD). Peer-review under responsibility of the Euro-Mediterranean Institute for Sustainable Development (EUMISD)

Keywords: Wind power generation, Modeling, Control, Doubly fed induction generator, Neuronal controller, Performances;

* * Corresponding author. Tel.: +213-388-759-82; fax: +213-388-760-05 E-mail address:[email protected]

1876-6102 © 2015 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4 .0/). Peer-review under responsibility of the Euro-Mediterranean Institute for Sustainable Development (EUMISD) doi:10.1016/j.egypro.2015.07.824

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

887

1. Introduction Faced to the worldwide demand for the conservation of nature and maintaining the biodiversity of natural environments, the world is heading towards renewable energy to produce electricity. Wind power is one of the cleanest sources of renewable energy that allow producing the green energy. However, wind energy is a natural resource that features many advantages since while producing electricity they do not propagate any gas greenhouse effect, do not degrade the quality of the air and do not pollute nor the soils or waters. Furthermore, it do not produce toxic or radioactive waste [1, 4]. Nowadays, wind generation system based on a doubly fed induction generator (DFIG) are employed widely in large wind farms fat has its many advantages: a very good energy efficiency, robust sensorless operation [5,6] as well as ease exploitation and control, fault-tolerant, auto-fault detection power converters [7]. In addition, the power converter is usually rated at 25-30% of the generator power rating [8-11] For such several advantages, this machine has generated a lot of curiosity on the part of researchers have tried to develop strategies to best exploit its strong points [12]. The conventional "PI" controller is extensively used in the control of doubly fed induction generator which his parameters are initially calculated according to the DFIG parameters. But in practice, the machine parameters change inevitably during the time. The problematic studied in this work is to find a compromise between the change of the parameters of the DFIG and the efficiency of the control. To solve this problem, we used the neuronal controller (NNT). The paper is organized as follows. Section 2, describes the modeling studied system. The various control algorithms for optimal turbine operation and control of active and reactive power will be presented in section 3. The proposed NNT controller performance is compared with conventional proportional integral controller (PI), the results of simulations obtained are presented and discussed for validating the proposed controller in Section 4. Finally, the conclusion is drawn in section 5. Nomenclature ܲ௦ , ܳ௦ ܲ௥ , ܳ௥ ܶ௘௠ ݀, ‫ݍ‬ ܸ௦ௗ,௤ ܸ௥ௗ,௤ ݅௦ௗ,௤ ݅௥ௗ,௤ ߮ୱୢ,୯ ߮୰ୢ,୯ ܴ௦ , ܴ௥ ‫ܮ‬௦ , ‫ܮ‬௥ ‫ܮ‬୫

stator active and reactive power rotor active and reactive power DFIG electromagnetic torque (N m) synchronous reference frame index stator d–q frame voltage rotor d–q frame voltage stator d–q frame current rotor d–q frame current stator d–q frame flux rotor d–q frame flux stator and rotor Resistances stator and rotor self Inductances mutual inductance

߱s, ߱r ߩ ܸ R Ȝ π௧௨௥ π௠ G ‫ܬ‬ ‫ܬ‬௧௨௥ , ‫ܬ‬௚௘௡ ݂ ı CNV1/CNV2

synchronous and rotor angular frequency air density wind speed rotor radius tip-speed ratio aeroturbine rotor speed generator speed gearbox ratio turbine total inertia rotor and DFIG inertia turbine total external damping Coefficient of dispersion. First converter/Second converter

2. Wind Turbine Conversion System Modeling A wind turbine conversion system is a system that converts the wind turbines mechanical energy obtained from wind into electrical energy through a generator calibrated according to nominal turbine speed, number of generator pole-pairs and network frequency [11, 13, 14]. Fig. 1 shows the synoptic scheme of the studied system.

888

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

3ph, 50

Ps

Gearbox

Wind

Ÿtur

Pr

Ÿmec DFIG

ࢂࢊࢉ CNV1

CNV2

Rotor side converter

Grid side converter

Filter

࢏࢘ ࢊ,ࢗ_࢓ࢋ࢙

Aeroturbine rotor

Fig. 1. Schematic diagram of DFIG wind turbine

The main parts of this scheme are the wind turbine, the gearbox and generator. The rotor-side converter (RSC) and the grid-side converter (GSC) connected back-to-back by a dc-link capacitor. The two converters, network side CNV2 and rotor side CNV1 are ordered in Pulse Width Modulation [15-16]. In typical mode of operation, the rotor delivers active power to the grid when the DFIG operates in super-synchronous speed and absorbs active power from the grid (or from the generator stator) when the DFIG operates in sub-synchronous speed. The DFIG stator usually delivers active power to the grid in both sub-synchronous and super-synchronous speeds [10]. The models of the different components of the wind-turbine generation system are described below. 2.1. Turbine Modeling The theoretical power produced by the wind is given by [17- 19]: ܲ௧௨௥ = ‫ܥ‬௣ .

ఘ.ௌ.௏ య

(1)



Where C୮ denotes power coefficient of wind turbine, its evolution depends on the blade pitch angle ( ) and the tip-VSHHGUDWLR Ȝ ZKLFKLVGHILQHGas [17- 19]:

ɉ=

ୖ.Ÿ౪౫౨

(2)



From summaries achieved on a wind of 1.5 MW, the expression of the power coefficient for this type of turbine can be approximated by the following equation [20]-[21]: C୮ = ቀ0.35 െ ൫0.0167(Ⱦ െ 2)൯ቁ ൭sin ቆ

஠(஛ା଴.ଵ)

ቇ൱ െ ൫0.00184(ɉ െ 3)(Ⱦ െ 2)൯

ቀଵହ.ହି൫଴.ଷ(ஒିଶ)൯ቁ

(3)

The aerodynamic torque expression is given by [22]: ௉

ܶ௧௨௥ = Ÿ೟ೠೝ = ‫ܥ‬௣ . ೟ೠೝ

ఘ.ௌ.௏ య ଶ





(4)

೟ೠೝ

The gearbox is installed between the turbine and generator to adapt the speed of the turbine to that of the generator [22]: Ÿ௠௘௖ = ‫ܩ‬. Ÿ௧௨௥

(5)

The friction, elasticity and energy losses in the gearbox are neglected. ்

‫் = ܩ‬೟ೠೝ

೘೐೎

(6)

889

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

The mechanical equations of the system can be characterized by [22]:

‫ܬ‬

ௗπ೘೐೎ ௗ௧

With: ‫= ܬ‬

+ ݂π௠௘௖ = ܶ௠௘௖ െ ܶ௘௠ ௃೟ೠೝ ீమ

(7)

+ ‫ܬ‬௚௘௡

2.2. Modeling the DFIG with stator field orientation The Park model of DFIG is given by the equations below [21, 23-24]: ௗఝ

ܸ௦ௗ = ܴ௦ ݅௦ௗ + ௗ௧ೞ೏ െ ߱௦ ߮௦௤ ቐ ௗఝೞ೜ ܸ௦௤ = ܴ௦ ݅௦௤ + ௗ௧ + ߱௦ ߮௦ௗ

(8)

ௗఝ

ܸ௥ௗ = ܴ௥ ݅௥ௗ + ௗ௧ೝ೏ െ ߱௥ ߮௥௤ ቐ ௗఝೝ೜ ܸ௥௤ = ܴ௥ ݅௥௤ + ௗ௧ + ߱௥ ߮௥ௗ

(9)

As the d and q axis are magnetically decoupled, the stator and rotor flux are given as: ߮௦ௗ = ‫ܮ‬௦ ݅௦ௗ + ‫ܮ‬௠ ݅௥ௗ ߮௦௤ = ‫ܮ‬௦ ݅௦௤ + ‫ܮ‬௠ ݅௥௤ ߮ =‫ ݅ ܮ‬+‫݅ ܮ‬ ൜ ߮௥ௗ = ‫ܮ‬௥ ݅௥ௗ + ‫ ܮ‬௠݅ ௦ௗ ௥௤ ௥ ௥௤ ௠ ௦௤

(10)



(11)

With: ‫ܮ‬௦ =‫ܮ‬௙௦ +‫ܮ‬௠ ‫ܮ‬௥ =‫ܮ‬௙௥ + ‫ ܯ‬ଶ ‫ܮ‬௠ The active and reactive powers are defined as: ܲ௦ = ܸ௦ௗ ݅௦ௗ + ܸ௦௤ ݅௦௤ ܳ௦ = ܸ௦௤ ݅௦ௗ െ ܸ௦ௗ ݅௦௤ ܲ௥ = ܸ௥ௗ ݅௥ௗ + ܸ௥௤ ݅௥௤ ൜ ܳ௥ = ܸ௥௤ ݅௥ௗ െ ܸ௥ௗ ݅௥௤

(12)



(13)

Consequently, the d-q orientation has to be synchronized with the stator flux (see Fig.2). ࡿ࢈

࣐ࢊ࢙

ࡾ࢈ ࢂ࢙ࢗ

d-q frame

d q

șV șU

ࡾࢇ Rotor axis

șH 0

Stator axis

ࡿࢇ

ࡿࢉ ࡾࢉ Fig. 2. Orientation of the d-q frame

The DFIG model is presented in synchronous dq reference frame where the d-axis is aligned with the stator flux linkage vector ߮௦ , and then, (߮௦௤ = 0, ߮௦ௗ = ߮௦ ) [19, 24]. In addition, considering that the resistance of the stator

890

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

winding (ܴ௦ ) is neglected and the grid is supposed stable with voltage ‫ݒ‬௦ and synchronous angular frequency ( ߱௦ ) constant what implies ߮௦ௗ = ܿ‫ݐݏ‬, the voltage and the flux equations of the stator windings can be simplified in steady state as [23-26]: ௗఝ

ܸ௦ௗ = ೞ೏ = 0 ௗ௧ ܸ௦௤ = ߱௦ . ߮௦ௗ = ܸ௦ The stator and rotor voltages are given by [21]: ቊ





ோ ߮ െ ೞ ‫ܮ‬௠ ݅௥ௗ ௅ೞ ௦ௗ ௅ೞ ோ െ ௅ೞ ‫ܮ‬௠ ݅௥௤ + ߱௦ ߮௦ௗ

ܸ௦ௗ = ܸ௦௤ =

(14)

ோೞ

(15)



ܸ௥ௗ = ܴ௥ ݅௥ௗ + ߪ. ‫ܮ‬௥ ܸ௥௤ = ܴ௥ ݅௥௤ + ߪ. ‫ܮ‬௥

ௗ௜ೝ೏

ௗ௜ೝ೜ ௗ௧

ௗ௧

+ ݁௥ௗ

+ ݁௥௤ + ݁ఝ

(16)

Where: ݁௥ௗ = െߪ. ‫ܮ‬௥ . ߱௥ . ݅௥௤ ‫ۓ‬ ۖ ݁௥௤ = ߪ. ‫ܮ‬௥ . ߱௥ . ݅௥ௗ ெ

݁ఝ = ߱௥ . ௅ . ߮௦ௗ ೞ ‫۔‬ ଶ ெ ۖ ߪ = 1െ൬ ௅ ൰ ‫ە‬ ඥ௅ೞ ೝ

The stator and the rotor active and reactive power can be rewritten as follows [21]: ୚౩.୑ ‫ ۓ‬Pୱ = െ ୐౩ . i୰୯ మ ۖ ۖQ = ୚౩ െ ୑.୚౩ . i ୱ ୰ୢ ୐౩ ன౩

‫۔‬ ۖ ۖ ‫ە‬

P୰ = g. Q୰ = g.

୐౩ ୚౩ .୑ ୐౩ ୚౩.୑ ୐౩

. i୰୯

(17)

(18)

. i୰ୢ

The electromagnetic torque is as follows [21]: ெ

ܶ௘௠ = െܲ. ௅ ߮௦ௗ . ݅௥௤ ೞ

(19)

3. Wind Turbine Control System 3.1. Control structure The block diagram of the whole system control in Park reference frame is illustrated in Fig.3. In this work, the two controls studied are: - extraction of the maximum wind power control “MPP” (Maximum Power Point) from the wind for a wide range of wind speeds, - control of CNV1. These controls will be considered separately in the following sections.

891

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

Grid

Ps Ÿtur Pr

Ÿmec DFIG

Filter

࢏࢘ ࢊ,ࢗ_࢓ࢋ࢙

ࢀ‫࢓ࢋכ‬

dq

dq PWM

abc ࢂࢗࢌ_࢘ࢋࢌ

ࢂࢊ࢘_࢘ࢋࢌ abc

࢏ࢊ,ࢗ ࢘_࢓ࢋ࢙

ࣂ࢘

ࢂࢊࢌ_࢘ࢋࢌ

Filter currents control

6 ࢂࢗ࢘_࢘ࢋࢌ dq

Decoupling/compensation

6

PWM

࢏ࢊ࢘_࢘ࢋࢌ

Decoupling/compensation

࢏ࢗ࢘_࢘ࢋࢌ

Rotor currents control

Rotor currents calculation

ࡼࡿ_࢘ࢋࢌ

CNV2

CNV1

MPP

ࡽࡿ_࢘ࢋࢌ

abc ࢏ࢊ,ࢗ ࢌ_࢓ࢋ࢙

ࢂࢊࢉ

Ÿ࢓ࢋࢉ

2

࢏ࢊࢌ_࢘ࢋࢌ ÷ DC bus control

࢏ࢗࢌ_࢘ࢋࢌ

ࢂࢊࢉ_࢘ࢋࢌ

ࢂ࢙ࢗ

ࣂ࢙

ࢂࢊࢉ

ࡽࢌ_࢘ࢋࢌ

Fig. 3. Block diagram of the wind control system [27]

3.2. Control objectives The operation of a wind turbine at variable speed is generally more beneficial over constant speed operation. Fig.4 shows the crucial advantage of variable speed wind turbines compared to fixed speed. If the wind speed varies ܸଵ and ܸଶ to the speed ɘଵ of the DFIG is unchanged, the power varies from ܲଵ to ܲଶ . In addition, the maximum power is equal to ܲଷ . In case we want to extract the maximum power should be changed by ɘଵ and ɘଶ thus make the variable speed depending on the wind speed. The extraction of maximum power control is to adjust the torque of the DFIG to extract maximum power. The red dotted line indicates the optimal power points for respectively ܸ = 10m/s and 12m/s, where the ‫ܥ‬௣ coefficient is kept at its maximum value. 4

12

x 10

Turbine Power (W)

P3 P2 10

w1

w2

MPP

8 V2=12 P1 6

4 V1=10

2 0 0

100 200 300 Rotor Speed (rad/sec) Fig. 4. Turbine power versus rotor speeds of generator

400

892

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

The power extracted from the wind is maximized when the rotor speed is such that the power coefficient is optimal ‫ܥ‬௣௢௣௧ . Therefore, we must set the tip speed ratio on its optimal value ɉ୭୮୲ . Fig. 5 illustrates the principle of MPP control without speed control of the rotation speed [28-29]: Aeroturbine rotor ࣅ



(3)

(2)

Gearbox π࢚࢛࢘

π࢓ࢋࢉ

(5)

࡯࢖ ࢂ

(4)

ࢀ࢚࢛࢘

(6)

ࢀ࢓ࢋࢉ ࢀ‫࢓ࢋכ‬

Model

π࢓ࢋࢉ

(7)

+

-

Control ࡯࢖࢕࢖࢚ . ࣋. ࣊. ࡾ૞ ૛ . π࢓ ૛. ࡳ૜ . ࣅ૜࢕࢖࢚ MPP control without speed control Fig. 5. Block diagram without speed control

Fig. 6 show the variation of the power coefficient (‫ܥ‬௣ ) versus the tip-speed ratio ( ) for the values of the pitch angle ߚ=2°, 3° and 4°. This figure indicates that there is one specific point (ߣ௢௣௧ , ‫ܥ‬௣௢௣௧ ) at which the turbine is most efficient for ߚ=2°. 0.4

Power coefficient Cp

0.35

X: 7.98 Y: 0.35

0.3

ࢼ = ૛° ࢼ = ૜° ࢼ = ૝°

0.25 0.2 0.15 0.1 0.05 0 0

2

4

6

8

10

12

Tip-speed ratio lumbda Fig. 6. Power coefficient versus tip speed ratio

3.3. Rotor side Converter Control CNV1 As seen in equation (19) and (20) that the electromagnetic torque and the stator reactive power can be controlled by means of the DFIG current i୰୯ and i୰ୢ respectively. The model of DFIG in d-q reference frame with stator field orientation shows that the rotor currents can be controlled independently. Fig. 7 shows the bloc rotor currents control diagram with ݅௥ௗ_௥௘௙ , ݅௥௤_௥௘௙ are given by:

893

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

݅௥ௗ_௥௘௙ =



ఝೞ೏ ெ



െ ெ.௏ೞ . ܳ௦_௥௘௙

݅௥௤_௥௘௙ = െ

ೞ೜

௅ೞ ெ.௉.ఝೞ೏

(20)

‫כ‬ . ܶ௘௠

ࢋ࢘ࢊ

࢏࢘ࢊ_࢓ࢋ࢙ ࢏࢘ࢊ_࢘ࢋࢌ

+



+

PI

ࢂ࢘ࢊ_࢘ࢋࢌ

+

ࢋ࢘ࢗ ࢏࢘ࢗ_࢘ࢋࢌ

+

PI



+

࢏࢘ࢗ_࢓ࢋ࢙

+

ࢂ࢘ࢗ_࢘ࢋࢌ

+

ࢋ࣐ Fig. 7. Bloc diagram of rotor currents control

3.4. Design of the Neuronal Network Controller The objective of this work is to design a new controller in order to assure a robust control of the DFIG facing the parametric changes. In this section, we presented the design of the proposed neuronal network controller. A network can have several layers. Each layer has a weight matrix ܹ, a bias vector ܾ, and an output vector ܽ. The number of hidden layers and the number of neurons in each layer are not definitive. The determination of the combinations of hidden layers and the number of neurons in each layer which can give the best performance for a given problem has no general guidelines. In our case, the number of hidden layers and the number of neurons in each hidden layer were chosen heuristically on a trial and error basis [30]. After several attempts, we opted for two hidden layers containing 06 neurons where the first hidden layer has 04 neurons and the second has 02 neurons, Fig.8 show the structure of the neuronal controller. The type of numerical optimization techniques for neural network training is LevenbergMarquardt (LM) which is a combination of gradient descent and Newton’s Method with the mean square error performance function. It was designed to approach second-order training speed without having to compute the Hessian matrix [31-33]. This algorithm appears to be the fastest method for training moderate-sized feedforward neural networks (up to several hundred weights). It also has an efficient implementation in MATLAB software, because the solution of the matrix equation is a built-in function, so its attributes become even more pronounced in a MATLAB environment [34]. The linear transfer function “purelin” are used in the final layer of multilayer networks that are used as function approximators. The sigmoid activation function “tansig“ is commonly used in the hidden layers of multilayer Networks [35].

Input

Hidden layers

f ࢈૚

࢝૚૚

ࢀ‫࢓ࢋכ‬ + -

ࢀࢋ࢓_࢓ࢋ࢙

࢝૚૛ ࢝૚૜ ࢝૚૝

Output

࢝૚૞ ࢝૚૟

f

࢝૛૞

f

࢈૛

࢝૛૟

࢈૞

f

࢝૜૞ ࢝૜૟ ࢝૝૞

f

࢈૜

f

࢈૟

࢝૝૟

࢈૝ Fig. 8. Structure of the neuronal controller

࢝૞ૠ

f ࢝૟ૠ

࢈ૠ

࢏࢘ࢗ_࢘ࢋࢌ

894

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

4. Simulation and interpretation In this section, the behavior of the doubly fed induction generator was carried out using the MATLAB/Simulink environment under the variation of machine parameters, notably: rotor self inductances, stator self inductances and rotor resistance are studied. Indeed, the problem with the PI controller is that its parameters (K and I) are initially adjusted according to the nominal parameters of the machine, knowing that the machine parameters change inevitably during the time. Therefore, our interest consists to the use of the neuronal controller in order to have a robustness control vis-à-vis parametric changes. The system performance with the proposed controller is compared with a conventional PI controller. The system parameters, MPP control and controller gains are given in Appendix A and B respectively. In the case where the machine parameters are nominal, Fig.9 shows that the d and q components of rotor currents, electromagnetic torque and reactive power follow their references according to table 2 (see appendix A). All these grandeurs evolve without notable overshoot (D) and static error, and that their response time is acceptable. In the aim to test the robustness of the proposed controller, the parameters value has been changed as follows: variation of the stator and the rotor self inductances (+20%). However, the effects of the considered variations have been studied separately (Fig.10 and Fig.11). The variations of ‫ܮ‬௠ , ‫ܮ‬௦ , ‫ܮ‬௥ and ܴ௥ have been tested and simultaneously represented (Fig.12). Concerning the variation of ‫ܮ‬௦ (+20%), the simulation results are shown in Fig.10, where it is found that the system response with neuronal controller remains remarkably insensible to the variation of the stator inductance. Contrary to the system response with PI controller, it witness an impotently overshoot and response time (tr) of the electromagnetic torque equal to 0.775 × 10ସ and 0.162s respectively. Similarly, for the change of ‫ܮ‬௥ (+20%), the overshoot (0.85 × 10ସ ) and response time (tr=0.171s) are very important as in the previous case. Finally, Fig.12 shows the case where all the parameters ‫ܮ‬௠ (+20%), ‫ܮ‬௦ (+25%), ‫ܮ‬௥ (+25%) and ܴ௥ (+50%) are changed simultaneously. In this case, the system performances with PI controller are highly degraded and the system becomes even unstable. Sum up, the simulation results of NNT controller show good performances in term of: response time and overshoot (see Appendix A table 3). 2000

2500 I

1500

rd Ref

Ird NNT

Ird PI

Irq Ref

2000

Irq NNT

Irq PI

1500 1000

1000

500

500

0

0 -500

-500

-1000 -1000

-1500

-1500 0

0.5

1 Time (sec)

1.5

2

1000

0.5

1 Time (sec)

1000

500

1.5

Irq

NNT

Irq

PI

0

0

I rd PI I

-1000

rd NNT

-500

-2000 0

1

1.05

1.1

1

1.05

1.1

2

895

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

4

2

6

x 10

2 T

1.5

T

em Ref

T

em NNT

em PI

x 10

Qs NNT

Q

1.5

s Ref

Qs PI

1

1

0.5

0.5

0 0

-0.5

-0.5

-1

-1

-1.5

-1.5 0

0.5

1 Time(sec)

1.5

2

-2 0

0.5

1 Time(sec)

1.5

2

Fig. 9. System response without machine parameters changes

Ird Ref

2000

Ird NNT

4000

Ird PI

Irq Ref

Irq PI

Irq NNT I

rq PI

2000

1000

Irq NNT

Ird NNT

0 -1000

0

I

-2000 0

0.5 4

6

x 10 2

0.5

0

2

1.5

1 Time(sec)

-2000

rd PI

2

1 Time (sec)

1.5

Qs NNT

Q

x 10 T

em Ref

T

T

em NNT

2

em PI

Q

s Ref

s PI

Qs PI

0.775*104

1

1 T

em NNT

Q

s NNT

0

0

T

em PI

-1 -1 0

tr =0.162

0.5

1 Time(sec)

1.5

2

-2 0

0.5

1 Time(sec)

Fig. 10. System response with stator self inductance change Ls (+20%)

1.5

2

896

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899

Ird Ref

2000

Ird NNT

Ird PI

Irq Ref

3000

Irq NNT

Irq PI

2000 1000

1000 Ird NNT

0

0

-2000

4

2

1.5

1 Time(sec)

0.5

-3000 0

Tem

T

Ref

T

em NNT

em PI

2 Tem

2

1.5

-1

em PI

0

1.5

s PI

Qs NNT

-1

tr =0.171

1 Time(sec)

Q

Qs PI

1 T

0.5

Qs NNT

Qs Ref

4

NNT

0

0

1 Time(sec)

x 10

0.85*10

1

0.5 6

x 10 2

Irq NNT

-1000

Ird PI

-1000

Irq PI

0

2

-2 0

0.5

1 Time(sec)

1.5

2

Fig. 11. System response with rotor self inductance change Lr (+20%)

I

3000

I

rd Ref

rd NNT

I

6000

rd PI

2000

I

rq Ref

I

rq NNT

rq PI

4000

1000

2000

0

I

-1000

I 0

rd NNT

I

-2000 -3000 0

I

0.5

I

rq PI

-2000

rd PI

1 Time(sec)

rq NNT

1.5

2

-4000 0

0.5

1 Time(sec)

1.5

2

897

Khouloud Bedoud et al. / Energy Procedia 74 (2015) 886 – 899 6

4

x 10 3

Tem Ref

Tem NNT T

2

3

Tem PI

x 10

Qs Ref

2

Q

Qs PI

s NNT

Q

s PI

em PI

1

1 T 0

Qs NNT 0

em NNT

-1

-1

-2

-2 0

0.5

2

1.5

1 Time(sec)

0

0.5

1 Time(sec)

1.5

2

Fig. 12. System response with machine parameters changes Lm (+20%), Ls (+25%), Lr (+25%) and Rr (+50%)

5. Conclusion In this paper, a modeling of DFIG based wind turbine under sub-synchronous operation mode and NNT controller is proposed. To achieve, these works are validated through simulation studies on a 1.5 MW DFIG wind generation system compared with conventional proportional integral current control design. The simulation results using software Matlab/Simulink show that the NNT controller gives good performance, efficient and more robust under parameters variations of the DFIG. The calculation of weight matrix (W) and bias vector (b) of NNT controller in function of the input/output data without knowledge of the DFIG parameters, explain his robustness to the parameter variations compared with PI controller which is initially adjusted according to the nominal parameters of the DFIG. Appendix A. Table 1. Font Simulated DFIG Wind Turbine Parameters Rated power Rotor diameter Gearbox ratio Friction coefficient : ࢌ Moment of inertia ‫ࡶ ׷‬ Stator voltage/Frequency ࡾ࢙ / ܴ‫) ( ݎ‬ ࡸ࢓/ࡸࢌ࢙/ࡸࢌ࢘ (H) Number of pole pairs:࢖

1.5MW 45m 100 0.0024 1000 690V/50Hz 0.00297/0.00382 0.01212/0.000121/0.0000573 2

M

1

Table 2. Operation Statues of the Simulated DFIG Status

Time (sec)

1 2 3 4

0 < t ൑ 0.5 0.5 < t ൑1 1 < t ൑ 1.7 1.7 < t ൑ 2

Reactive power (MVar) 0 -1 0.8 -0.7

Time (sec) 0 < t ൑ 0.3 0.3 < t ൑ 1 1