MPPT Control of Wind Turbines with Wind Speed Estimation Using Direct Adaptive Fuzzy-PI Controller Sanaz Sabzevari, Ali Karimpor, Mohammad bagher Naghibi sistani, Mohammad Monfared Electrical Engineering Department Ferdowsi University of Mashhad Mashhad, Iran Sanaz.sabzevari@kiaeee.org, karimpor@um.ac.ir, mb-naghibi@um.ac.ir, m.monfared@um.ac.ir Abstract—This paper presents a maximum power point tracking (MPPT) technique based on the tip speed ratio control for small scale wind turbines (WTs). In this paper, artificial neural network (ANN) with particle swarm optimization (PSO) (ANNPSO) has been trained offline to learn the turbine-characteristic surface power versus wind and machine speeds and has been implemented online to estimate the varying wind speed. It is essential to include a controller that can track the maximum peak of energy regardless of wind speed changes. Therefore, this work provides a novel robust direct adaptive fuzzy-PI controller during MPPT process for permanent magnet synchronous generator (PMSG) driven by a WT. The proposed method has successfully reduced the ripples of coefficient of power (Cp) which is the index of MPPT mode, under wind speed variations in comparison with conventional controller. Finally, a systematic analysis is presented as well as simulation results proving the effectiveness of proposed strategy. Keywords: Maximum power point tracking (MPPT), artificial neural network, particle swarm optimization, direct adaptive fuzzy-PI control, PMSG wind turbine. I. INTRODUCTION Over the past decade, wind energy systems have gained tremendous attention as one of the most promising renewable energy sources due to the probable depletion, high costs, and negative environment impacts of conventional energy sources [1, 2]. Among renewable energy generation systems, like those based on wind, photovoltaic, fuel cell, or geothermal power, wind generators are, in general, more cost effective than other renewables of the same power rating [3]. Wind turbines are controlled to operate only in a range of wind speed from cut-in (V cut-in) to cut-out (Vcut-out) speeds. This paper focuses on the moderate wind speed region from cut-in speed at which the turbine starts working, to the rated speed (Vrated), at which the turbine produces its nominal power. In this region the maximum power point tracking (MPPT) algorithms are usually employed to maximize the turbine energy conversion efficiency through regulating the rotational speed of variable speed wind turbines. Recently, lots of MPPT algorithms have been proposed by researchers, which can be categorized into four main groups based on the specified performance and measurement requirements: 1) tip speed ratio (TSR) control [4], 2) optimal torque (OT) control [5], 3) power signal feedback (PSF) control [6], and 4) perturbation and observation (P&O) control [7-9]. A detailed performance comparison carried out in [10] shows that the TSR control presents more advantages among others. It can be noted that for each wind turbine, there is a specific ratio called the optimum TSR for which the extracted power is maximum [1]. In TSR control method as the wind speed changes instantaneously, it is necessary for the rotational speed to change accordingly to always maintain the optimal TSR at all times. Despite its promising features like fast convergence speed, robustness, simplicity, and no need for prior training and memory, it needs some kind of sensors to measure the wind speed. In general, the efficiency of methods that does not directly measure the wind speed is lower than compared to other TSR control methods which directly measure the wind speed. This is due to the fact that wind changes are not reflected instantaneously and significantly on the reference signal [5, 10]. In recent years, wind speed sensor-less MPPT controls have been reported in [11-15], in which the wind speed is estimated online from the instantaneous inputs [11, 14-15]. In detail, Bhowmik et al. use a look-up table to approximate the wind speed; it is estimated online by calculating the roots of a polynomial of wind turbine power coefficient using an iterative algorithm [12]. However, real-time calculation of the polynomial roots may require external memory for highly accurate estimation, therefore, degrading system performance. In [13], Tan and Islam employ a power equation of an autoregressive statistical model to predict the wind speed. This method is a time-consuming task and also the predicted wind speed is not precise for online MPPT control. Artificial neural networks (ANNs) are beneficial tool to implement nonlinear complex time-varying input-output mappings [14]. This paper presents a novel TSR-based MPPT control algorithm which does not need any information about the wind speed. Indeed, in the proposed method, the wind speed sensor is replaced by a very short term wind speed estimator, which let achieve superior performance compared to other sensor-less techniques. In this work, in order to have more precision and simplicity with less training data, an ANN input-output mapping with two hidden layers is utilized for wind speed prediction. The layer’s weights are updated through a particle swarm optimization (PSO). Other contribution of this paper is that instead of conventional PI controllers, a direct adaptive fuzzy-PI controller is employed which significantly reduces the dependence on system parameters. Fuzzy controllers are supposed to work in situations where there is a large uncertainty or unknown variation in plant parameters and structures [16, 17]. The main advantages of adaptive fuzzy control over non-adaptive fuzzy control are: (i) better performance is usually achieved because the adaptive fuzzy controller can adjust itself to the changing environment, and (ii) less information about the plant is required because the adaptation law can help to learn the dynamics of the plant during the real-time operation [18]. Figure 1. A Simplified block diagram of the proposed PMSG wind-energy system [1] Figure 2. The characteristic of the power coefficient as a function of the tip speed ratio [10] This paper is organized as follows. Section 2 explains the system description of PMSG wind turbine. Section 3 describes ANN-PSO wind speed estimator. Section 4 explains the MPPT control by direct adaptive fuzzy-PI controller. Section 5 provides the simulation results. Finally, section 6 concludes the paper. II. SYSTEM DESCRIPTION Fig. 1 shows the scheme of small wind turbine system based on a permanent magnet synchronous generator (PMSG). The PMSG is connected to the resistive load basically through the rectifier and a boost converter. Wind turbine converts the wind energy into mechanical energy, which then runs a generator to create electrical energy. The mechanical power generated by a wind turbine can be expressed as [19–21]: 1 Pm = 2 ρπR2 Vw3 Cp (λ,β) Figure 3. Characteristics of the turbine power as a function of the rotor speed for a different wind speeds [10] (1) The turbine power coefficient (Cp) describes the power extraction efficiency of the wind turbine. It is a nonlinear function of both the tip speed ratio (λ) and the blade pitch angle (β). Although its maximum theoretical value is approximately 0.59, practically it lays between 0.4 and 0.48 [22]. The tip speed ratio is a variable expressing the ratio of the linear speed of the blade tips to the rotational speed of the wind turbine [19–21], and can be written as: λ= ωm R Vw (2) Figure 4. Block diagram of ANN-PSO based wind speed estimator In this paper, Cp is defined as [23]: Cp (λ,β)=0.5176 1 λi 116 λi 1 - 21 -0.4β-5 e λi +0.0068λ = λ+0.08β - 0.035 β3+1 (3) (4) In the present work, due to the premise of a fixed pitch rotor, the angle (β) is set to zero. Thus, the characteristics of Cp mainly depend on λ. Fig. 2 shows Cp as a function of λ. Based on this figure, there is only one optimal point, denoted by λopt, where Cp is maximum. Continuous operation of the wind turbine at this point by the tip speed ratio control, guarantees that it will extract the maximum available power from the wind at any speed, as shown in Fig. 3. III. ANN-PSO WIND SPEED ESTIMATOR A. Estimation of the Wind Speed by ANN-PSO The basic idea of the proposed wind speed estimation is to retrieve more precision on MPPT mode for any instantaneous value of the wind speed. It can be noted that the artificial neural network based particle swarm optimization (ANNPSO) basically behaves as a virtual anemometer, permitting the free wind speed to be estimated by the inversion of the wind turbine model. The block diagram of the algorithm is shown in Fig. 4. The direct model of the wind turbine, given by (1) – (4), gives the output power of the turbine by means of an involved nonlinear function of the free wind speed Vw and the rotor speed ωm. The approximation of the direct model is not a challenging problem and can be easily carried out by any neural network (NN) for function approximation (multilayer perceptron or RBF). However, this is not the case for the computation of the inverse model of the turbine, which gives the free wind speed as a function of the machine speed and power. This is because it is not known a priori if there exist discontinuities or multiple branches, which makes conventional NNs unsuitable [3]. According to the direct turbine model, a complete training set of data has been created, as well as a test set, and then used to train an ANN with two hidden layers which their weights are updated through a particle swarm optimization (PSO). B. Particle Swarm Optimization (PSO) PSO is a population-based searching algorithm. PSO randomly produces npopu particles in searching space, and each particle includes position X i and velocity Vi, where Xi is the position of i-th particle in the searching space, Xi=( Xi1,…, Xij,….,Xik), and Vi is the velocity of i-th particle in the searching space, Vi=( V i1,…, V ij,….,Vik) [24, 25]. The position Xi of i-th particle represents a solution of the problem and the velocity Vi of i-th particle represents its displacement in the searching space. Pbesti is the optimal position that the i-th particle has experienced, and pbesti is the optimal fitness that the i-th particle has experienced. Gbest is the optimal position that all particles have experienced and gbest is the optimal fitness that all particles have experienced. When Fit (.) is the fitness function for solving the maximum value, the optimal position of each particle is shown in (5). Pbesti (t+1)= Pbesti(t)for Fit(Xi (t+1))≤Fit(Pbesti (t)) (5) Xi (t+1)for Fit(Xi (t+1))>Fit(Pbesti (t)) To improve the convergence, gbest and Gbest are selected by comparing the experiences of others. Therefore, the i-th particle is guided to three vectors (Vi, Pbesti, and Gbest). The inertia weight method, shown as in (6) and (7), is applied to update velocity and position of the particles [24]. Vnew ij =w.V ij+c1 .rand1. Pbestij -Xij +c2 .rand2. Gbestij -Xij (6) Xnew ij =Xij +V ij (7) Where Pbest i=( Pbesti1,…,Pbestij,….,Pbest ik); Gbest=(Gbest 1,…,Gbestj,….,Gbest k); w=wmax -iter.(wmax -wmin )/itermax ; c1 , c2 are acceleration coefficients; w is the coefficient of the inertia weight; iter is the current iteration number. Given the PSO method described above, the process of the PSO is shown as the following steps: Step 1) Generate equivalent npopu quantity of position and velocity randomly, and record Pbesti, pbesti, gbest, and Gbest. Step 2) Calculate each fitness value of particles. Step 3) If stopping criterion is satisfied (e.g., maximum iteration number), the procedure would go to the end; otherwise, proceed to step (4). Step 4) Update the Pbesti and pbesti. Step 5) Update the gbest and Gbest. Step 6) Update particles position and velocity by applying steps (6) and (7), and then go back to step (2). IV. MPPT CONTROL BY DIRECT ADAPTIVE F UZZY-PI CONTROLLER A direct adaptive fuzzy-PI controller is designed in order to track λopt on MPPT mode for a wide range of wind speeds in the presence of noises and disturbances. Suppose that D (duty cycle of the boost converter) is the output of an adaptive fuzzy system in the normalized form with the inputs and . Three fuzzy membership functions are considered for each fuzzy input, hence the whole control space will be covered by nine fuzzy rules. The linguistic fuzzy rules are proposed in the Mamdani type of the form FRl: ℎ (8) where FRl denotes the l-th fuzzy rule for l=1,…, 9. In the l-th rule, , , and are fuzzy membership functions belonging to the fuzzy variables , and D, respectively. Three ( ), named as Positive Gaussian membership functions, (P), Zero (Z), and Negative (N) are defined for input in the operating range of manipulator as shown in Fig. 5. Three Gaussian membership functions, ( ), named as P, Z, and N in the same shape as illustrated in Fig. 5, are used for input . Nine symmetric Gaussian membership functions, ( ), named as Very Positive High (VPH), Positive High (PH), Positive Medium (PM), Positive Small (PS), Zero (Z), Negative Small (NS), Negative Medium (NM), Negative High (NH), and Very Negative High (VNH) are defined for D ( ) = exp(−(( − )/ ) ) for l= 1,…,9 in the form of in the operating range of output. The symmetric Gaussian function depends on two parameters and . In this work, is adjusted by an adaptive law, whereas is considered as a fixed scalar. The fuzzy rules should be defined such that the tracking control system goes to λopt. An expert’s knowledge, the trial and error method, or an optimization algorithm such as PSO may be used to design the fuzzy controller. The obtained fuzzy rules are given in TABLE I. In this paper, the centers of Gaussian membership functions for D are adapted to optimize the performance of the control system. If the product inference engine, singleton fuzzifier, center average defuzzifier, and Gaussian membership functions are used, the fuzzy system is as the following: ( , ∑ )= ( ∑ ) ( ( ) ) ( ) (9) ( ) ∈ [0,1] and ( ) ∈ [0,1] are the where membership functions for the fuzzy sets and , respectively, and is the center of fuzzy set . An important contribution of fuzzy systems theory is to provide a systematic procedure for transforming a set of linguistic rules into a nonlinear mapping as stated in (9). One can easily manipulate this transformation by regulating from (9) such that ( | )=∑ where = = [ … ] and ( =∑ = 0.5( ) ( ) … ) ( + ∫ ) + ( ∗ − ∗ )( where is a positive scalar. We differentiate to time to get ̇ = + ∫ ̇+ − ( − ) (17) with respect ∗ − ) ̇ (18) From (16) we have = − ∗ ∫ + − (19) Substituting (19) into (18) yields ̇ = ∗ − + − 1 ̇ (10) + is a positive value expressed as ( =[ For regulating , we use (16) to suggest a positive definite function of the form in (17), , (11) ) ] (12) + ̇− ( ∫ ∫ ) (20) ̇ in (20) includes two terms. Therefore, we can control ̇ by regulating which is in the first term of ̇ . This leads to propose an adaptive law which should be evaluated afterward as follows: ∗ − + − ∫ ̇ =0 (21) Therefore, the law of adaption is obtained by ̇= which + ∫ is given by = Figure 5. Membership functions of the input (22) (23) In order to evaluate the adaptive law (22), we substitute (21) into (20) to obtain TABLE I. FUZZY RULES ̇ = + ̇− ( ∫ ∫ ) (24) P Z N N Z P Taking the derivative of gives in (19) with respect to time Z NS VNH PM Z NM VPH PS Z ̇ = − Using the input scaling factors and , we have and Where to scale = , (13) = ̇ (14) is + ∗ ∫ = ( − ) (15) + ∫ = ( ∗ − ) (16) ) (26) Substituting (25) into (24), to satisfy ̇ < 0 in (24), it is required that + ∫ < + ∫ (27) According to the Cauchy-Schwartz inequality, it is clear that: By substituting (10) into (15), ( ∗ = ∗ Let in (15) be the ideal control. Hence, the closedloop dynamics of the fuzzy control system can be written as: (25) + Suppose that ∫ ≤ + ∫ . | |. (28) | |< TABLE II. WIND-TURBINE PARAMETERS (29) where is a positive scalar. Thus, to satisfy inequality (27), it is sufficient that R [m] 2.5 3 ρ [kg/m ] 1.225 λopt 8.1 cannot reduce further to be ∫ . > + ∫ . Considering (30), it verifies that the minimum value of error will be small if we use a large gain value . The tracking error is affected by the upper bound of Cpmax 0.48 Gear box ratio 5 Pole-pairs of PMSG 4 Generator rated power [w] 2827.43 as stated by (30).In summary, the simplified direct adaptive fuzzy control system is shown in Fig. 6. Generator rated speed [rpm] 4500 Rated wind speed [m/s] 7.56 . Note that < + ∫ (30) + V. SIMULATION RESULTS The simulations have been carried out using MATLAB/Simulink software. The parameters of the wind generator system are summarized in TABLE II. Fig. 10 shows Cp during the transient time, which confirms the results of Fig. 9. Indeed, PI controller cannot follow Cpmax with zero steady state error. Wind Speed (m/s) 10 5 Real Estimated 0 1 2 3 4 5 6 7 time (s) Figure 7. Real and estimated wind speeds 7.8 7.6 7.4 Wind Speed (m/s) Fig. 7 shows the real (measured) wind-speed profile as well as estimation by the ANN-PSO. It can be observed that the ANN-PSO properly estimates the wind speed both in steady state and in transients, with negligible estimation errors. Moreover, for evaluating the robustness of the adaptive fuzzy-PI controller compared with the conventional PI controller which is tuned for nonlinear plant at the nominal operating, a sudden wind change is being applied according to Fig. 8. Fig. 9 depicts the error of tracking λopt for each controller. Although PI controller has fewer oscillations for the rated wind speed before step change at 3 s in comparison with the proposed method, but it can be observed that a step change in the wind speed causes a steady state error for classic PI controller. In contrast the proposed method can track λopt in about 0.05 s during the MPPT process. 15 7.2 7 6.8 6.6 6.4 0 1 2 3 4 5 time (s) Figure 8. A step change in wind speed 4 3 Tracking Error 2 1 0 -1 -2 Proposed controller PI controller -3 -4 1 1.5 2 2.5 3 3.5 4 4.5 5 time (s) Figure 6. The direct adaptive fuzzy-PI control system Figure 9. Error tracking of proposed method and PI controller 5.5 0.48 Coefficient of Power (Cp) 0.46 [12] 0.44 0.42 Proposed controller 0.4 [13] PI controller 0.38 0.36 [14] 0.34 0.32 0.3 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 time (s) [15] Figure 10. The Power coefficient VI. CONCLUSION This paper proposes a robust direct adaptive fuzzy-PI control for MPPT control. To track the maximum power point over a wide speed range based on tip speed ratio control, the information of the wind speed is required. Hence, we have used an ANN–PSO to estimate the wind tangential speed. In addition, the authors analysed a simulation on a PMSG wind turbine using rectifier and a boost converter. By controlling the duty cycle of the converter, the apparent load developed by the generator can be adjusted, and thus, its shaft speed can be tuned. The proposed method obtains the maximum average value of Cp and maintains it at its maximum even with changes in wind speed. The simulation results show the effectiveness of the proposed method. REFERENCES [1] Abdullah MA, Yatim AHM, Tan CW, “A study of maximum power point tracking algorithms for wind energy system, ” In: 2011 IEEE Clean Energy and Technology Conf., pp. 321–6. [2] Saidur R, Islam MR, Rahim NA, Solangi KH, “A review on global wind energy policy, ” Renewable and Sustainable Energy Reviews, vol. 14, pp.1744–62, 2010. [3] Marcello Pucci, Maurizio Cirrincione, “Neural MPPT control of wind generators with induction machines without speed sensors,” IEEE Trans. Industrial Electronics, vol. 58, pp. 37-47, 2011. [4] Thongam JS, Ouhrouche M. (2011). MPPT control methods in wind energy conversion systems, Carriveau R (Ed)., Quebec, CA. [online]. Available: http://www.intechopen.com/books/fundamental-andadvancedtopics-in-wind-power/mppt-control-methods-in-windenergy-conversion-systems [5] Reza Kazemi SM, Goto H, Hai-Jiao G, Ichinokura O, “Review and critical analysis of the research papers published till date on maximum power point tracking in wind energy conversion system,” In: 2010 IEEE Energy Conversion Congress and Exposition Conf., pp. 4075-82. [6] Barakati SM, Kazerani M, Aplevich JD, “Maximum power tracking control for a wind turbine system including a matrix converter, ” IEEE Trans. Energy Conversion, vol. 24, pp. 705–13, 2009. [7] Ching-Tsai P, Yu-Ling J, “A novel sensorless MPPT controller for a high-efficiency micro scale wind power generation system,” IEEE Trans. Energy Conversion, vol. 25, pp. 207–16, 2010. [8] Hong M-K, Lee H-H, “Adaptive maximum power point tracking algorithm for variable speed wind power systems, ” In Proc. 2010 Sustainable Energy and Environment Intelligent Computing Conf., pp. 380–8. [9] Raza K, Goto H, Hai-Jiao G, Ichinokura O, “A novel algorithm for fast and efficient Maximum power point tracking of wind energy conversion systems, ” In: 2008 ICEM Electrical Machines Conf., pp. 1– 6. [10] Abdullah MA, Yatim AHM, Tan CW, “A review of maximum power point tracking algorithm for wind energy systems,” Renewable and Sustainable Energy reviews, vol. 16, pp. 3220–3227, 2012. [11] Ahmed G. Abo-Khalil and Dong-Choon Lee, “MPPT control of wind generation systems based on estimated wind speed using SVR,” IEEE [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] Trans. Industrial Electronics, vol. 55, no. 3, pp. 1489–1490, March. 2008. 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. 949–958, Jul./Aug. 1999. K. Tan and S. Islam, “Optimum control strategies in energy conversion of PMSG wind turbine system without mechanical sensors,” IEEE Trans. Energy Conversion, vol. 19, no. 2, pp. 392–399, Jun. 2004. Wei Qiao, Wei Zhou, José M. Aller, and Ronald G. Harley, “Wind speed estimation based sensorless output maximization control for a wind turbine driving a DFIG,” IEEE Trans. power electronics, vol. 23, no. 3, May. 2008. H. Li, K. L. Shi, and P. G. McLaren, “Neural-network-based sensorless maximum wind energy capture with compensated power coefficient,” IEEE Tran. Ind. Appl., vol. 41, no. 6, pp. 1548–1556, Nov. /Dec. 2005. M. Karbakhsh, H. Abutorabi, A. Khazaee, “An enhanced MPPT fuzzy control of a wind turbine equipped with permanent magnet synchronous generator,” presented at the 4th Iranian. Conf. Electrical and Electronics Engineering, 2012. Marwan Rosyadi, S.M. Muyeen, Rion Takahashi, and Junji Tamura, “Transient stability enhancement of variable speed permanent magnet wind generator using adaptive PI-fuzzy controller, ” In: 2011 IEEE Trondheim PowerTech Conf., pp. 1-6. L. X. Wang, A course in fuzzy systems and control. New Jersey: Prentice Hall, 1997, pp. 290-307. Freris LL, Wind energy conversion system. London, UK: Prentice Hall, 1990. Abdelkafi A, Krichen L, “New strategy of pitch angle control for energy management of a wind farm,” Energy, vol. 36, pp. 1470–9, 2011. Nema P, Nemab RK, Rangnekar S, “A current and future state of art development of hybrid energy system using wind and PV-solar: a review,” Renewable and Sustainable Energy Reviews, vol. 13, pp. 2096-103, 2009. Zhe C, Guerrero JM, Blaabjerg F, “A review of the state of the art of power electronics for wind turbines,” IEEE Trans. Power Electronics, vol. 24, pp. 1859–75, 2009. T. Senjyu, Y. Ochi, E. Muhando, N. Urasaki, and H. Sekine, “Speed and position sensor-less maximum power point tracking control for wind generation system with squirrel cage induction generator,” in Proc. 2006 PSCE Conf., pp. 2038–2043. Clerc, Maurice, Particle swarm optimization. Paul & Co. Pub Consortium, 2006. J. Kennedy, R. Eberhart , “Particle swarm optimization,” in Proc. 1995 IEEE Neural Network Conf., vol. IV, Perth, Australia, pp. 1942-1948.