Abstract

In this study, we present a new sensorless control for Wind Energy Conversion System (WECS) based on Cascaded Doubly Fed Induction Generator (CDFIG). The aim of this control is to extract the maximum wind power and to control the power generation only by measuring phase voltages and currents. The proposed control scheme is ensured without the need of using any rotor speed or wind sensors from the aspect of reliability and increase in cost. For that a neuronal observer is introduced to estimate the rotor speed. To impose the Maximum Power Point (MPPT) an extended luenberger observer was introduced in cascade with the neuronal observer to estimate the optimal aerodynamic torque. After an approximation with an another artificial neuronal network, observer makes it possible to estimate the wind velocity. This later will be injected in the algorithm control to impose the MPPT. Simulation results demonstrate the sensorless control device excellent performances.

---

INTRODUCTION

Various types of aero-generators are spread out in the world. The Cascade Doubly Fed Induction Generator (CDFIG) is introduced because it arises as an alternative to the Doubly Fed induction Generator (DFIG) and the Permanent Magnet Synchronous Generator (PMSG). The CDFIG is composed of two doubly fed induction machines called Power Machine (PM) and Control Machine (CM). They are used with the rotors mechanically and electrically coupled.

The great advantages of the CDFIG are the low maintenance requirement avoiding the use of slip ring and brushes and the possibility of reducing or even eliminating the gearbox (Patin et al., 2009; Adamowicz et al., 2009; Adamowicz and Strzelecki, 2008; Basic et al., 2002, 2003; Boardman et al., 2002). In addition, the power flow in the power machine can be controlled via a fractionally power converter connected to the control machine stator (Boardman et al., 2002; Basic et al., 2002).

For that CDFIG have been proposed to apply in new renewable energy generation systems, i.e., small hydropower plants, variable speed windmills, shaft generators of vessels and embedded applications (Patin et al., 2009; Adamowicz et al., 2009; Adamowicz and Strzelecki, 2008; Basic et al., 2002). As all variable-speed constant-frequency generators, we can applied one of the sensorless vector controls to the CDFIG to maximise the system efficiency and to eliminate the rotor speed sensor. In literature, many techniques have been developed for sensorless control for induction machine, PMSG and DFIG applications. The most effective sensorless control techniques are MRAS (Schauder, 1992; Brahmi et al., 2009; Cardenas et al., 2004), the luenberger observer (Senjyua et al., 2006), neuronal network observer (Rahman and Hoque, 1998; Hui et al., 2005; Batzel and Lee, 2003; Rajaji and Kumar, 2009; Qiao et al, 2009; Barambones et al., 2010) and a sliding mode observer (Paponpen and Konghirun, 2007). In this study, we propose a new sensorless control stategy of CDFIG wind turbines based on ANN.

The rotor speed and wind velocity are estimated from the measurement of the power machine currents and grid voltages. The estimated quantities are used then to determine the optimal output power reference by using an MPPT block, based on knowledge of the WECS dynamics and power characteristics for controlling the CDFIG wind turbine.

Fig. 1:	WECS composition

Consequently, the WECS is optimally controlled to extract the maximum wind power without the need of using any rotor speed or wind sensors. The WECS under study is shown in Fig. 1. It is based on a CDFIG where the power machine stator is directly connected to the power system grid and the control machine stator is connected to a PWM rectifier; allowing an optimal power extraction by the use of an MPPT algorithm. A PWM inverter ensures the injection of the produced power to the AC grid. Between the two converters, a capacitor is used as a voltage DC bus. This system is connected to the grid via a filter to reduce the higher-frequency components.

MATERIALS AND METHODS

Wind turbine modelling: The wind power acting on the swept area of the blade S is a function of the air density ρ and the wind velocity w. The transmitted power P is generally deduced from the wind power using the power coefficient C_p:

(1)

The power coefficient is a non-linear function of the tip speed-ratio λ which depends on the wind velocity and the rotation speed of the shaft Ω_t:

(2)

where, R is the blade radius (m). There is a value of λ = λ_opt for which C_p is maximized where the wind turbine captures the maximum wind power. The aerodynamic torque is given by:

(3)

where, η is the gear ratio. For each wind speed, there exists a specific point in the wind generator power characteristic called Maximum Power Point (MPPT) where the output power is maximized.

CDFIG modelling: The model of the CDFIG can be derived from the models of two DFIG connected as shown in Fig. 1. To obtain the state model of the CDFIG, we adopt the hypotheses of the oriented constant power machine stator flux and we neglect the power machine stator resistor then we have:

And:

Where:

Where:

vgd and vgq	=	d and q components of the grid voltage, respectively
and	=	d and q components of the power machine stator flux, respectively
l and Vg	=	Grid inductance and voltage magnitude, respectively

The CDFIG model placed in (d, q) reference frame which is synchronized with the power machine stator flux rotates at the angular speed ω_p = ω_g can be written as:

(4)

(5)

Where:

vscd and vscq	=	d and q components of the control machine stator voltage, respectively
iscd and iscq	=	d and q components of the control machine stator current, respectively
ωg	=	Synchronous pulsation
Ωmec	=	Aero generator speed
pp and pc	=	Power and control machine pole pairs, respectively
rsc	=	Control machine stator resistor

Let:

The control machine stator fluxes are given by the following equations:

(6)

(7)

where, l_sc, l_mc and l_rc are the control machine stator, mutual and rotor inductance, respectively; l_sp, l_mp and l_rp are the power machine stator, mutual and rotor inductance, respectively with arrangement of Eq. 4-7 and the CDFIG model can be written under the following form:

(8)

(9)

Where:

eφ = k1(ωg-(pp+pc)Ωmec)φspd

The mechanical equation is given by the following equation:

(10)

where, T_em is the electromechanical torque given by:

(11)

and j is the total inertia which appears on the shaft of the generator (kg m²).

Converters modelling: The converters are modelled in the d-q reference frame according to the switching function concept (Brahmi et al., 2009). The rectifier provides the voltages V_scn = [v_scd v_scq]^T from the capacitor voltage U_c and the modulated current r_ec:

(12)

(13)

where, u_recd, u_recq are the rectifier modulation functions. The inverter yields, the inverter voltages:

Vinvn = [vinvd vinvq]T

from the capacitor voltage and the inverter modulated current i_inv from the line currents I_fn = [i_fd i_fq]^T:

(14)

(15)

where, u_invd, u_invq are the inverter modulation functions. The DC bus voltage is given by the following differential equation:

(16)

where, C is the capacitance of the DC bus. It is assumed that the filters reduce the higher-frequency components. In the d-q reference frame, the voltage balance across the filter inductors and resistors is:

(17)

(18)

Sensorless control strategy: The WECS include the wind turbine; the CDFIG and the rectifier (Fig. 1). The latter is used to control the control machine stator currents. For that a vector control is necessary. Relating to the CDFIG model given by Eq. 8 and 9, the vector control allows fixing the (d - q) control machine stator voltage as follows:

(19)

Where:

ed = -1'sc (ωg - (pp+pc) Ωmec) iscq_ref eq = 1'sc(ωg - (pp+pc) Ωmec) iscd_ref

v_d and v_q are the de-coupled voltages given by:

where, k_i, k_p, k_i’, k_p’ are the regulators parameters and i_{scd_ref} and i_{scq_ref} are the control machine stator currents reference values. This later are provided by an MPPT block , based on knowledge of the WECS dynamics and power characteristics to extract the maximum wind power for each wind speed. Consequently, the control scheme requires wind velocity and mechanical speed measures. To increase the installation cost and limit the measurements only on the power machine stator size coupled directly to the grid, a sensorless vector control will be considered. Thus, the control machine stator currents and the mechanical speed in the control laws given by Eq. 19 will be substituted by the estimated one provided from a neuronal network observer.

Artificial neuronal network controller: The ability of the Artificial Neuronal Network (ANN) to approximate nonlinear functions is the most significant. With the potential to shorten and generalize the control scheme configuration and model identification task, neural networks are usually used in many control applications (Brunskill, 2010). In literature, many rotor speed estimation methods using artificial neuronal network has been developed for sensorless control for PMSG (Rahman and Hoque, 1998; Batzel and Lee, 2003; Brahmi et al., 2009) and for DFIG (Rajaji and Kumar, 2009; Qiao et al., 2009; Barambones et al., 2010) wind turbine applications. In the proposed algorithm, the rotor speed is estimated by using the grid voltages and currents of the power machine which is connected directly to the grid. The proposed algorithm use three layers Radial Basis Function Network (RBFN). The three layers were the input, output and hidden layers. The proposed ANN is constituted of 4 inputs and a single output. The overall input-output mapping for the RBFN is given by:

(20)

where, Ω_mec is the output of the RBNF that represents the rotor speed estimated value; x = [i_spd, i_spq, v_gd, v_gq] is the input vector; σ_jmR and C_jmRⁿ are the width and center of the jth radial basis function in the hidden layer, respectively; w_j and b are the weight and a bias term between the hidden layer and output one. The parameters of the RBFN including the numbers of Radial Basis Function (RBF) units, the RBF width and centers and the output bias term and weights are determined by using a training data set generated from simulation of the complete drive system.

Once trained, the parameters of the RBFN are then fixed for online estimation of the rotor speed. The proposed RBFN used for rotor speed estimation is shown in Fig. 2. In the network, the circles represent the neurons. The input, output and hidden layers in proposed network are composed, respectively by 4, 5 and 1 neurons.

Fig. 2:	RBFN-based rotor speed estimation algorithm

Wind velocity and output torque estimation: The maximum point of power MPPT depends on the aerodynamic torque and the wind velocity. From the aspect of reliability and increase in cost, wind velocity sensor is not preferred. To eliminate the latter, wind velocity and output torque estimation techniques may be used.

In some researches, the torque and wind velocity are estimated by an analytical expression deduced from the characteristic curve of the wind turbine aerofoil (Morimoto et al., 2005; Senjyua et al., 2006). In this study, an extended order luenberger observer was introduced in cascade with the neuronal observer to estimate the optimal aerodynamic torque.

After an approximation with a neuronal network radial basis function makes it possible to estimate the wind velocity. The latter will be injected in the algorithm control to impose the MPPT.

Output torque estimation: By using an extended order luenberger observer, we can extend the estimation to the optimal aerodynamic torque and we can suppose that:

(21)

The luenberger observer is constructed using the CDFIG mechanical Eq. 10 and the optimal aerodynamic torque variation Eq. 21. The observer model is represented by:

(22)

Where:

and G is the matrix gain observer which can written as:

To ensure the observer stability, we determine the matrix G coefficients so that (A_u - GC_u) is stable. In Eq. 22, T_em and Ω_mec which are not measurable in the control algorithm are substituted by the estimated values obtained from the neuronal observer.

Fig. 3:	RBFN-based wind speed estimation algorithm

Wind speed estimation: In the proposed algorithm, the wind speed is estimated by using the estimated aerodynamic torque and rotor speed. We use three layers Radial Basis Function Network (RBFN). As shown in Fig. 3, the network proposed is constituted of 2 inputs corresponding to the estimated aerodynamic torque and rotor speed and a single output corresponding to the estimated wind speed. The latter is used after to impose the optimal output power reference by using an MPPT block, based on knowledge of the WECS dynamics and power characteristics. Figure 4 shows the global sensorless control for wind generation system based on cascade doubly fed induction generator using Artificial Neuronal Network (ANN).

RESULTS AND DISCUSSION

We present the simulation results of the sensorless optimal vector control proposed. The parameters describing the WECS simulated are shown in Table 1. We have considered a real wind speed ranging between 6 and 12 m sec^-1 with an average value of 9 m sec^-1 and we chose the unity power factor strategy, i.e., the reactive power injected to the grid via the power machine stator and through the rectifier is set to zero. Figure 5 shows the real and estimated wind speed which is ranged between 6 and 12 m sec^-1. Figure 6 shows the estimated rotor speed using the proposed method it was consistent with real method it was consistent with the real shaft speed.

Fig. 4:	Block diagram of global sensorlesss control for wind generation system based on CDFIG

 

Table 1:	Characteristics of the WECS simulated

 

 

Fig. 5:	Wind speed

 

 

Fig. 6:	Rotor speed

 

The output torque and the aerodynamic power shown in Fig. 7 and 8, respectively, are also well estimated.

Fig. 7:	Output torque of the wind turbine

 

Fig. 8:	Aerodynamic output power

 

 

Fig. 9:	D-component of the control machine stator current

 

The d and q control machine stator currents components are given, respectively in Fig. 9 and 10. We note that the currents are correctly estimated during the time.

Fig. 10:	Q-component of the control machine stator current

CONCLUSION

From the aspect of reliability and increase in cost, a sensorless vector control of a WESC based on ANN was proposed. In the proposed method, a three layers radial basis function network is used for estimating the shaft speed. In addition, an extended luenberger and another neuronal observer were used to estimate the wind velocity to impose the MPPT. Simulation result was illustrated to improve the sensorless control strategy.

How to cite this article:

A. Ouali and S. Nouali. Sensorless Control Strategy for Wind Energy Conversion System Based on Cascaded Doubly Fed Induction Generator Using Artificial Neuronal Network.
DOI: https://doi.org/10.36478/ijepe.2011.42.48
URL: https://www.makhillpublications.co/view-article/1990-7958/ijepe.2011.42.48