Hybrid MPPT Control: P&O and Neural Network for Wind Energy Conversion System

—In the field of wind turbine performance optimization, many techniques are employed to track the maximum power point (MPPT), one of the most commonly used MPPT algorithms is the perturb and observe technique (P&O) because of its ease of implementation. However, the main disadvantage of this method is the lack of accuracy due to fluctuations around the maximum power point. In contrast, MPPT control employing neural networks proved to be an effective solution, in terms of accuracy. The contribution of this work is to propose a hybrid maximum power point tracking control using two types of MPPT control: neural network control (NNC) and the perturbation and observe method (P&O), thus the P&O method can offer better performance. Furthermore, this study aims to provide a comparison of the hybrid method with each algorithm 𝑃 & 𝑂 and NNC. At the resulting duty cycle of the 2 methods, we applied the combination operation. A DC-DC boost converter is subjected to the hybrid MPPT control. This converter is part of a wind energy conversion system employing a permanent magnet synchronous generator (PMSG). The chain is modeled using MATLAB/Simulink software. The effectiveness of the controller is tested at varying wind speeds. In terms of the Integral time absolute error (ITAE), using the P&O technique, the ITAE is 9.72. But, if we apply the suggested technique, it is smaller at 4.55. The corresponding simulation results show that the proposed hybrid method performs best compared to the P&O method. Simulation results ensure the performance of the proposed hybrid MPPT control.


INTRODUCTION
The energy crisis is one of the greatest threats facing our civilization since access to electricity is a mandatory condition for socio-economic development which we cannot do without. The most consuming part of this energy comes from fossil fuels such as petroleum, excessive exploitation of fossil fuels has given rise to problems such as high cost, and negative environmental impacts. Faced with this challenge, we have recourse to renewable energies which have become a competitive alternative given that they are inexhaustible and respectful of the environment to supply energy to the electrical networks [1][2][3][4]. Exploiting energy sources can reduce greenhouse gas emissions and complement and replace the use of fossil fuels [5]. Wind energy is considered a renewable energy source that responds to the increasing demand for energy [6]. Producing electricity with environmental concerns at an affordable price has become a key challenge [7,8]. Because of their many advantages, the technology of variable speed wind power generation has become a research hotspot in this field [9,10]. Compared to fixed-speed wind turbine technology, variable-speed wind turbines have the advantage of increased energy collecting [11][12][13], improved efficiency, and reduced installation and maintenance costs [14,15]. Due to the random and varying conditions of wind speed which changes throughout the day, wind turbines can influence the stability of the electrical systems. For this reason, the energy production control strategy has a great impact on the overall operating performance [16]. Hence, wind power generation systems need to adopt different control methods according to diverse wind speed intervals [17]. In order to extract the maximum energy from the WECS in the MPPT interval where the WECS produces the maximum available wind power, it is necessary to design a maximum power point tracking controller (MPPT) [18][19][20][21][22][23].
Several maximum power point (MPP) extraction techniques have been proposed in the literature [24][25][26][27][28][29][30]. It can be divided into indirect and direct MPPT methods [31,32]. The indirect method is based on knowledge of the wind turbine system properties [33]. Indirect techniques include the optimal torque (OT) method. In this method, control is achieved by monitoring optimum torque to achieve optimum wind energy utilization [34][35][36]. In [37] the authors described the power signal feedback (PSF), an indirect technique for maximizing the output power of a PMSG generator by determining the reference power for a particular wind speed. However, the mentioned indirect approaches have problems with their reliance on the climate and their need for knowledge of the characteristics of the wind turbine [38][39][40][41][42][43]. As opposed to indirect methods, the usage of direct methods no longer requires information about the characteristics of the wind turbine [44]. The example taken in [45], is the perturb & observe (P&O) approach, which tracks the maximum point by constantly varying the maximizing variable and observing the power captured [46]. The last-mentioned approach is one of the conventional MPPT techniques, owning a simple design and ease of implementation, the perturbation and observation technique is characterized by way of its simplicity and independence from the characteristics of the wind turbine [47][48][49][50][51]. The major drawback of this MPPT algorithm is that it cannot respond accurately, and has a poor ability to track the  [52,53]. Larger step sizes lead to larger steady-state oscillations around the MPP, while smaller step sizes slow down the dynamic response. Each of these 2 situations would incur more power loss and slower response [54][55][56]. Therefore, in order to solve this problem, it is necessary to use an adequate method to obtain MPP when the environment changes rapidly. Therefore, to resolve this problem, for obtaining MPP at rapid environmental changes a powerful method must be used. In this context, many researchers have estimated the MPP by the fuzzy logic (FL) method [57][58][59][60], it has a faster response and high accuracy even in the rapid variations of wind speed. FLC-based MPPT control provides high-performance control, but it leads to a high amount of computational burden and its usage is inflexible in complex systems [61,62]. Neural network control (NNC) is also used to detect the maximum power point, this control proposed in [63] is operated for non-linear power electronics circuits, to track the output voltage and improve the performance of power. NNC is essentially a nonlinear control approach and extensively increases system effectivity with greater effectiveness [64][65][66][67][68]. This approach can successfully cope with the nonlinear behavior of the system besides any data about the mathematical model [69][70][71][72][73]. In addition, the dynamic features of the DC-DC converter with the neural network can be realized excellent dynamic characteristics, compared with the conventional methods [74]. Without any prior knowledge, NN controllers can handle the problem, showing good performance under rapidly varying [75,76]. Recently, genetic algorithms [77][78][79][80], pattern swarm optimization (PSO) [81][82][83], ant colony MPPT [84,85], grey wolf MPPT [86], hybrid models such as hybrid particle swarm optimization and salp swarm optimization algorithm [87], and the MPPT control of wind turbines by using ANN-PSO wind speed estimator [88] have been proposed but each of the above-mentioned techniques has complexity levels based on their applications. MPPT controls vary in their effectiveness and complexity. Selecting an MPPT control method that should be able to track MPP properly, at variable wind speeds presents our main aim.
The main contributions of this paper are: the application of hybrid MPPT control using neural networks and P&O control, as well as the proposed hybrid method is compared with each of the conventional P&O and the NNC methods. Considering all the problems summarized above, the ease of implementation of P&O and the accuracy of the neural network are exploited in the suggested novel hybrid method. By this combination, maintaining maximum power and increasing efficiency under varying environmental conditions is guaranteed. Thus, we can make the most of the advantages of the conventional strategy and eliminate its disadvantages. In this research paper, a PMSG-based wind power conversion chain modeling is used in order to apply the hybrid MPPT controls on the DC-DC boost converter which is an element of the chain. The simulation results are confirmed by different cases of wind speed profiles, using the MATLAB/SIMULINK environment.
The present work is composed of 4 parts: the first one is for the introduction, the second part is devoted to presenting modeling of the wind and the components of the aerodynamic system with mathematical equations, the third section presents the MPPT control part giving the research methodology, and then we have the results and discussion part, the last section is consecrated to the conclusion that summarizes the main important points of this paper and gives futures works.

II. WIND MODELING
The wind is a stochastic, intermittent quantity dependent on a set of factors such as location and temperature [89][90][91][92][93]. In order to test the wind power conversion chain in intermittent conditions close to reality, it is necessary to have a wind reconstruction block. To evaluate the control approach proposed in our work, we resort to the application of a variable wind profile modeled in this section. The change in wind speed over time is modeled by a scalar function that changes over time. The wind model is given by a Fourier series representation which presents the wind as a signal consisting of a superposition of several harmonics. It is given by equation (1) [94,95].
The method used for the modeling of the wind consists in generating the temporal characteristic of the wind from a white noise to which is applied a transfer function to be determined which depends on the characteristics of the place and the nature of the wind. The wind speed at a point V 0 (t) can be decomposed into a sum of an average component V avrg which varies slowly and of a variable component representing the fluctuations V t (t) [96].
is the mean wind speeds and ( ) is component representing wind turbulence.
At a given point in space, the wind turbulence is described by the power spectrum, based on the Von Karman filter, in a stochastic manner, and represented by the following transfer function [97].
Where is gain of the filter and is the constant filter time.
This filter is approximated by the transfer function characterized by: 1 = 0.4 and 2 = 0.25.
High-frequency fluctuations are filtered out by the wind turbine. For this purpose, the low pass filter with the transfer function is simplified and given by: Where γ attenuation factor on the rotor (γ = 1.3).
The value of the time constant depends on the diameter of the rotor, the intensity of the wind turbulence, and the average wind speed, (τ1 = 0.11375 ).
This modeling concerning the wind is seen as necessary in order to define the operating conditions of the wind turbine, also describe the stresses that apply to the blades, and refine the modeling of the rotor.

III. AERODYNAMIC SYSTEM MODELING
The wind flow creates a driving torque at the drive shaft. The wind has a certain velocity at a given time and passes through a certain area " ", where " " is the density of the air, The aerodynamic power extracted by the wind turbine is quantified using the following equation [98].
Where is the wind turbine rotor radius, (λ, β), the power coefficient, represents the aerodynamic efficiency of the wind turbine, which describes the capacity of the turbine to transform the wind kinetic power to mechanical power.
is a nonlinear function of the tip-speed ratio and blade pitch angle . The tip-speed ratio is defined as: where is angular shaft speed of the turbine.
The aerodynamic power can be expressed also as follows: Hence, following the previous Equations (8)-(10), the aerodynamic torque Ta on the wind turbine shaft can also be expressed as follows: Where ( , ) is usually modeled by empirical equations it considered approximate expression can be given by [99,100] IV.

MPPT CONTROL
To optimize the energy efficiency of the system by following the optimal operating point, the MPPT technique is used to extract the most power from accessible wind speed by tracking the peak point. In this paper, the MPPT strategy proposed consists of a combination of 2 techniques: neural network and therefore the P&O. Additionally, this paper compares the hybrid technique with each of the standard MPPT techniques perturb and observe technique (P&O), and neural network technique (NNC). The present hybrid approach is applied to the wind energy conversion system. To track the maximum power point and generates the appropriate duty cycle for the switch within the DC-DC boost device, the DC-link voltage is employed as shown in Fig. 1. The proposed topology shown within the Fig. 1 is developed via MATLAB/Simulink software according to the block diagram presented in Fig. 2. As shown within the given system block diagram of the WECS, the chain consists of the wind signal reconstruction block. The variable wind speed generated is transmitted to the turbine, coupled to a synchronous generator with permanent magnets. The PMSG is employed in direct drive. An AC-DC power electronic interface with a diode bridge corrected the output of the PMSG. Then this output is filtered to get rid of large ripple voltage components, and fed to the DC-DC boost converter. Thus, the converter unit is used for the wind turbine's variable speed operation, providing a DC voltage load.
This system is meant to realize maximum power tracking MPPT under wind speed variations, by means of hybrid NNC and P&O control according to the methodology provided through the flowchart in Fig. 3. As shown in the algorithm, the inputs are the current and the voltage pre-requisited on the input side of the converter. Both techniques P&O and NNC work separately depending on the inputs and then the neural network duty cycle output is combined with the P&O duty cycle output. After that, the average of the duty cycle generated is given to the PWM generator for generating pulses to trigger the IGBT switch in the DC-DC converter.
The NNC control employed in this work uses an artificial neural network as a design methodology. NNC is less difficult to design, and performs higher than different controllers. Neurons are linked by using weighted links. The weighted links lift the signal. Each neuron has a single threshold value. The weighted sum of the input is shaped and then subtracted from the threshold value to get the activation sign of the neuron. As illustrated in Fig. 3 in the neural network architecture, it is necessary to define the number of layers to exploit, in our case, there are three layers, that are: the input layer, the hidden layer that functions to companion the input and output layers and the output layer.
The NNC structure is built using gensim instruction in MATLAB. In addition to the NNC method, the algorithm also includes the PO method. By measuring the voltage and current, As shown in Fig. 3, the working precept of the P&O method is to increase the voltage by means of adjusting the duty cycle on the power side DC-DC converter, also the maximum output power of the generator can be optimized. But if delta energy is negative, otherwise, the sign of the perturbation is inverted to tune in the way of growing power. The output power generated is compared to the preceding electrical power. If the generated power increases, then the variable ΔD will be fixed, if it decreases, then the ΔD will change. This is a continuous procedure of observation and perturbation until the operating point converges at the MPP. To evaluate the consistency and effectiveness of the proposed MPP tracking control scheme, the PMSG-based WECS was simulated in the MATLAB/Simulink software environment. In this section, we test the hybrid MPPT technique based on NNC and P&O and compare it with the techniques already in use. The performance of the proposed control is investigated for two different wind speed profiles. The first is the profile where the wind speed changes step by step during the operating time. In addition to the previous wind speed profile, there is a rapidly changing profile that is closer to the real wind speed. For the first scenario, Fig. 4 shows the step change in wind speed, WECS is evaluated for this study with variable wind speed, passes from 8m/s, 12m/s, 9m/s, 10m/s, to 8m/s, then to 12 m/s. Fig. 5, shows the power extracted from the wind conversion chain without application of any MPPT control. It can be observed that it leads to results with the presence of a significant oscillation. It is clear that oscillations occur frequently and that tracking is lost. After applying the MPPT P&O control, the evolution of the extracted power changed as shown in Fig. 6. Compared to the above result, moderate oscillations are present after reaching a maximum limit of power points. The variations of the wind signal show the main drawback of the P&O control, that is, the presence of oscillations around the MPP due to the fixed step size, which leads to a loss of the generated power. Any variation in the wind speed profile introduces a variation in the extracted power represented. The main weakness of this type of MPPT algorithm is the lack of precision. The resulting waveform of the extracted power as a function of wind speed variation with the application of the NNC command is shown in Fig. 7. Result shows that the NNC control give faster responses and less oscillation around the maximum power than the basic P&O control. This present control has advantages such as rapidity and damping of the overshoot when the wind speed changes. The results in Fig. 8 included: mechanical speed of the generator shaft in Fig. 8(a), voltage output in Fig. 8(b), extracted power in Fig. 8(c). As shown in Fig. 8(a), according to the change in wind speed, the mechanical speed is regulated. When the wind speed increases, the mechanical speed increases and the output power increases, in this case, the PMSG delivers the maximum allowable wind energy which is clearly realized in Fig. 8(c). Fig. 8(b) shows the instantaneous output voltage from the WECS, during the step change in wind speed. The output voltage also varies with the wind speed and has the smallest oscillation levels around the MPP, meaning that after reaching the maximum power point, our controlled system becomes comparatively stable. The matching of graphs of the two methods P&O and the proposed method in steady state conditions, indicates the accuracy of the proposed method. The output power in Fig.  8(c) is much smoother and has lesser speed fluctuations, it can be clearly observed that WECS with the hybrid MPPT control has better wind energy tracking capability with fast response time. The simulation of WESC under intermittent wind speed conditions is done in order to evaluate the tracking and to study the robustness of our system adopting hybrid control. Variations in wind speed used in the simulation are shown in Fig. 9, in this figure, the mean wind speed is 10m/s. For mean wind velocity (10 m/s), simulation results have carried out under randomly and fast changeable wind rapidity. The curve presented in Fig. 10 shows the power extracted directly without application of any MPPT control, it is clear that it leads to results with the presence of a significant oscillation. Fig. 11(a)  responses to described wind speed turbulence. As shown, the variation of actual rotor speed is in accordance with wind speed fluctuation. The output voltage is given in Fig. 11(b) varies with the wind speed, and demonstrates the energy management technique and controllers' high stability. From Fig. 11, we can note that the proposed strategy adjusts the generator speed and output power with satisfactory tracking performance. From the overall results, the proposed combined P&O MPPT control with NNC presents outputs with much even, less distortion, and less fluctuation at the rapid various wind speed conditions.  Table 1. Performances of the proposed hybrid are compared with those of the conventional P&O controller, NNC control, conventional SMC, and optimized SMC: second-order sliding mode controller proposed in another similar study. The index , and [101,102] is expressed as:

Simulation result presented in
These indices demonstrate characteristics like, overshoot, settling time and response speed. Also, these are calculated as a function of time and/or error. This is explained by the fact that the hybrid control controller has a smaller peak overshoot than the optimized SMC, but this last controller will tolerate small oscillation in terms of tracking time. To further improve the performance of the system, we can exploit the algorithm optimized SMC.
For Table 2, performance evaluation is based on analysis of the hybrid with other techniques, which are commonly used for the MPPT problem. The hybrid algorithm performed very well, reaching the peak performance point in all cases, including changes in environmental conditions. In comparison, the algorithms like PSF are fast and simple. Nevertheless, for PSF control, sensors are required which is costly. Algorithms like P&O and INC are basic, these are sensor less calculation, which makes it less expensive. However, requires less memory, also the application of these techniques isn't considerable under wind speed variations. with the P&O technique Efficiency is reduced because of the variances around the MPP. The fuzzy control MPPT is great yet Neural Network based MPPT control gives a superior compromise regarding dynamic speed and power reactions of the system, but prior-knowledge of parameters is required, and the complexity is increased. The neural system must be intermittently prepared to ensure the precise MPPT. By using hybrid algorithms, the drawbacks of individual MPPT algorithm can be eliminated for achieving fast tracking speed with higher efficiency. But the system complexity is determined by the participated algorithms on a hybrid algorithm and that is the problem with the Hybrid PSO-FLC. In wind farm installation, it is recommended to apply simple and effective MPPT algorithm, which reduces the cost of sensors. The hybrid P&O, NNC algorithms take into account these features developing new technique that enhance the overall WECS performance with limited shortcomings as deliberated in this section.

VI. CONCLUSION
Wind turbines have grown over the last several years considerably. To ensure high performance, we use the wind turbine with control algorithms, among which is maximum power-point tracking (MPPT). It is used to bring the turbine to the MPP over a full wind speed range and aims to maximize the efficiency of a wind turbine. Within this work, we combine the P&O method and neural network control (NNC). A hybrid algorithm is applied to the variable speed wind power conversion chain including the permanent magnet synchronous generator, by acting directly on the duty cycle of the DC-DC boost converter. Simulation done in MATLAB/SIMULINK software proved the viability of the model. Simulation realized in this work demonstrates that the maximum power point tracking based on the NNC and P&O MPPT control can track and preserve the maximum power carried for every wind speed value. This approach used has been confirmed to be a contributor to successfully extracting the maximum power from the wind energy conversion system (WECS). Hence, the developed system has good prospects in the WECS grid application. As a future work, the proposed MPPT controller method will be experimentally tested to estimate the performance of the control for optimal power generation from realistic wind energy resources, and further evolved as a prototype for validation for application to the real grid-connected wind turbines.