Artificial Neural Network power manager for hybrid PV-wind desalination system

Artificial Neural Network power manager for hybrid PV-wind desalination system

Available online at www.sciencedirect.com ScienceDirect Mathematics and Computers in Simulation xxx (xxxx) xxx www.elsevier.com/locate/matcom Origin...

2MB Sizes 0 Downloads 22 Views

Available online at www.sciencedirect.com

ScienceDirect Mathematics and Computers in Simulation xxx (xxxx) xxx www.elsevier.com/locate/matcom

Original articles

Artificial Neural Network power manager for hybrid PV-wind desalination system O. Charroufa ,∗, A. Betkaa , S. Abdeddaima , A. Ghamrib a

LGEB Laboratory, Electrical Engineering Department, University of Biskra, Algeria MSE Laboratory, Electrical Engineering Department, University of Biskra, Algeria

b

Received 7 February 2019; received in revised form 18 August 2019; accepted 13 September 2019 Available online xxxx

Abstract In this paper, Artificial Neural Network (ANN) power management for a reverse osmosis desalination unit fed by hybrid renewable energy sources solar PV and wind turbine associated to battery bank as storage element is studied. The ANN power management system has as main objective to ensure the smooth transfer of the generated power by these sources under the variability and intermittency of the wind speed and the irradiation during 24 h of operation considering the limitation constraints of the RO unit and the need water profile. The design, the modeling and the control strategies of all the components are made in this study using Matlab/Simulink. The results show the ability of the ANN power manager to define the operating modes based on the proposed flow chart. c 2019 International Association for Mathematics and Computers in Simulation (IMACS). Published by Elsevier B.V. All rights ⃝ reserved. Keywords: Desalination; PV; Wind; Artificial neural network; Power management

1. Introduction Water and energy are two inseparable elements that govern human life and promote civilizations [20,37]. Obviously, the social and economic health of the modern world depends on the sustainable supply of energy and water. However, nowadays about three billion people do not have access to a safe source of fresh water and about 1.76 billion people live in areas already facing a high degree of freshwater scarcity according to the 2015 United Nations World Report indicating that 75% of the Arab population lives below the level of water scarcity [1]. To deal with this announced water scarcity, emerging techniques as well as desalination have been widely deployed throughout the world however almost of the desalination plants have been installed near the sea. Consequently, the remote areas which are generally not covered by the electrical grid and in possession of large quantities of saline water did not benefit of these processes. Using the renewable energy sources, these regions will attain the two primordial conditions of the modern life: water and energy. The combination of desalination technologies with renewable energy sources is nowadays subject of several research studies [5,26,27,36,43]. The large number of possible combinations between these two technologies ∗ Corresponding author.

E-mail addresses: [email protected] (O. Charrouf), [email protected] (A. Betka), [email protected] (S. Abdeddaim), [email protected] (A. Ghamri). https://doi.org/10.1016/j.matcom.2019.09.005 c 2019 International Association for Mathematics and Computers in Simulation (IMACS). Published by Elsevier B.V. All rights 0378-4754/⃝ reserved.

Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

2

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Nomenclature λ R V Ωmec Ωt Paer S ρ Cp G SOC Np Ns Ipv Vpv VB rb ibat DOD Vd , Vq Rd I d , Iq Ld , Lq w p ϕf Tem Pres Qf Pm-p A Pf PF Qp Qc Cf cc πp B T* Vdc AD PMSM HPF LPF GPV

Tip speed ratio Blade turbine radius (m) Wind speed (m/s) PMSM rotational speed (rd/s) Turbine rotational speed (rd/s) Turbine aerodynamic power (W) Useful surface traversed by the wind (m2 ) Air density (kg/m3 ) Wind turbine aerodynamic efficiency Gear box coefficient State of charge Pv number in parallel Pv number in series Pv current (A) Pv voltage (V) Battery voltage (V) Battery internal resistor () Battery current (A) Depth of discharge d, q Axis voltages of PMSG (V) Resistor of the PMSG () d, q Axis currents of PMSG (A) d, q Axis inductances of PMSG (H) PMSG rotor pulsation (rd/s) Number of pole pairs PMSG flux (Wb) PMSM electromechanical torque (N m) The head applied by the pump in meter The flow rate at the input of the RO module in m3 /h. The power transmitted to the pump by the PMSM (kW) Membrane solvent permeability constant (kg m−2 S−1 Pa−1 ) Feed stream pressure (Pa) Concentration Polarization factor Permeate flow (kg/m3 ) Concentrate flow (kg/m3 ) Feed water concentration kg/m3 Concentrate water concentration (kg/m3 ) Osmotic pressure (Pa) Membrane solute permeability constant (kg m−2 S−1 Pa−1 ) PMSM reference torque DC-link voltage Algerian Dinar (0.0083 $) Permanent synchronous machine High pass filter Low pass filter Photovoltaic generator

Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Pnet Ppv Pwind Pload Pn Pref e c k Isource Iconv Batt WT FOC ESC KI a WH WL

3

Difference between Source power and load power GPV power WInd energy conversion system power Load power Nominal power of the PMSM Reference power Error Capacitor (F) Constant Source current (A) Converter current (A) Battery Wind turbine Field-oriented control Extremum seeking control. Integrator gain Sinusoidal perturbation signal magnitude High pass filter pulsation Low pass filter pulsation

offers to researchers several scenarios that can be considered in terms of design, control, power management and techno-economic analysis [4,24,28]. The feasibility of RO desalination systems combined to RE sources has been demonstrated in earlier papers [29,32,38]. Solar and wind energy sources are the most practiced in this field (19% wind-RO, 32% PV-RO). France and Spain was among the first countries in Europe to promote the Wind-RO desalination plants [40]. In these first plants, the wind source power was associated to electrical grid or batteries to power the RO desalination plants. However it was Carta et al. who demonstrated, among first, experimentally the Wind/RO plant operating under variable wind power source. An automatic operational control strategy has been used based on connect/disconnect load [10]. Also, Miranda and Infield developed and tested an RO unit driven by a 2.5 kW wind generator without batteries with the capacity of 500 l/h of produced water [33]. The effect of the variable power source on the RO membranes has been studied precisely by some authors [15,34,39]. However, it was Feron [41], in earlier paper, the first to report the concept of safe operating window (SOW) for the RO membranes where the operating limits of the device are considered. In recent research papers, Park et al. studied the effect of wind speed fluctuations on the performances of the Wind-RO plants which was unknown before [16,42]. He confirmed the non-effect of the RO permeate flow rate and Nacl concentration during the variable power operation (0 to 0.6 oscillation range was tested). The design of PV-RO desalination systems consists in a combination of reverse osmosis membranes and photovoltaic (PV) modules. The wide use of this combination is probably due to the fact that photovoltaic energy technology is the first to have conquered the markets, it constitutes the most dynamic market [4,10,11,15– 18,21,24,27–29,32–34,38–42,44]. The PV-RO desalination systems can be designed with or without batteries [11,17,21,44]. The PVs are used to supply the pumps that generate the pressure required to supply the reverse osmosis membranes with generally brackish water with low salinity. Over the last decade, reverse osmosis systems powered by photovoltaic (PV) panels have been implemented in several remote areas throughout the world [9,14,23]. The intermittency of the renewable energy sources and their unpredictable character leads to the hybridization of the RES-Desalination systems. The interest of hybridization of the RE driven desalination systems has the main goal to satisfy the load demand in terms of water production [7,28,47]. Indeed, photovoltaic energy production always follows a parabola during the day and vanishes during the night. Moreover, its amplitude varies according to the weather conditions during the seasons. Also, it is extremely variable in the short term during cloudy days under the effect of partial or total shading. The wind source is also characterized by variable energy production during the Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

4

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Fig. 1. Hybrid desalination system.

year. Its production is strongly related to the weather conditions and wind speeds available during the year without neglecting moments of calm that may occur unpredictably. It is therefore necessary to integrate storage units with the power systems to act as a buffer between production and demand. Therefore, the use of a Power Management System (PMS) become an important key for optimal operation. Many alternative power management strategies can be used to manage a standalone hybrid energy system: Conventional control strategies as well as stereotype techniques depending on linear programming and PI controllers or advanced control strategies like genetic algorithm (GA), differential evolution (DE), neural network, fuzzy system, and neuro-fuzzy. In this study, an Artificial Neural Network (ANN) algorithm has been performed for the power management of a small scale Reverse osmosis (RO) desalination system driven by hybrid wind–solar conversion system with battery bank as storage element. Among the several ANN learning paradigms, feed-forward network with back propagation algorithm using training data of specific region in the Algerian Sahara namely Illizi has been used. Considering the variability conditions of the renewable energy sources and the water demand, The NN power manager suggests the power references of the wind turbine and the batteries to overcome the operating limitation constraints of the whole system based on developed flowchart. The PMS has been developed using Matlab library. 2. Description and mathematic model of the hybrid system The mathematic models of the various components of a typical reverse osmosis plant driven by solar–wind conversion system with batteries are described. The typical reverse osmosis plant is shown in Fig. 1. To ensure the objectives of energy autonomy, the final design of the studied system is as following: • A wind generator (2.2 kW) composed of wind turbine driving a permanent magnet synchronous generator connected to the DC bus via a DC–DC converter. • A photovoltaic generator (2.5 kW) associated with DC–DC converter connected to the same DC bus. • Batteries (100 AH, 48 V) associated with reversible DC–DC converter, connected to the same node. • A permanent magnet synchronous motor (2.2 kW) powered by DC–AC converter used to drive a high-pressure pump (HP). • A high pressure pump (2.2 kW). • A module of reverse osmosis (RO).

Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

5

Table 1 Technical data of the wind energy conversion system. Component

Parameters

Symbols

Values

Wind turbine

Blade radius Power coefficient max Optimal TSR Voltage L-L Flux Moment of Inertia Friction coefficient stator Resistor Inductance Poles Gear box coefficient

R Cp-max λop

1.6 m 0.4061 6.8

Un ϕf J F Rs Ld = Lq 2P G

220 V 0.175 Wb 0.089 0.005 0.2  0.0085 H 8 3/7

PMSG

Fig. 2. Equivalent circuit of the solar cell.

2.1. Model of wind turbine The power received by the wind turbine can be written as follows [6,12,45]: Paer =

1 ρ SV 3 C p 2

(1)

where ρ is the density of air, V (m/s) the wind speed, S (m2 ) the useful surface traversed by the wind and Cp is the aerodynamic efficiency of the wind turbine presented as a function of the ratio of speed λ given by the following relationship: Ωt R . V In Eq (2), R and Ω t are respectively the radius of the blade of the wind turbine and rotation speed. Technical parameters of the wind energy conversion system are listed in Table 1. λ=

(2)

2.2. Model of the photovoltaic generator The photovoltaic generator comprising Ns panels in series and Np branches in parallel, thus forming a matrix of (Ns × Np) modules has at its terminals a voltage Vpv = NsVp and delivers a current Ipv = NpIp. The electrical behavior of a photovoltaic cell of a module can be described with good precision by the equivalent circuit [13,46], whose schematic diagram is detailed in Fig. 2. It consists of a photovoltaic current source IPH in parallel with a diode, shunt resistor Rp and second resistor in series Rs. The relationship that determine the well known I–V characteristic of any Photovoltaic generator is given by the following equation: [ V +R I ] S V + RS I V T I = I P H − I0 e −1 − (3) Rsh Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

6

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

2.3. Model of the battery A simple expression defined by a close model consisting of an equivalent voltage source VB in series with a resistor Rbat can be used. The charging voltage is deduced from the no-load voltage by the following equation: Vbat = VB − rbats i bat

(4)

When operating the batteries, the Power manager developed must at all times identify the state of charge of the battery derived from the depth of discharge (DOD) of the batteries by the following equation: SOC = 1 − DOD.

(5)

The state of charge at any operating time must be constrained as following: 20% .S OC < S OC < 80% .S OC

(6)

2.4. Model of the PMGS The model chosen for this application is a permanent magnets synchronous generator. The high efficiency and low inertia of these machines make them privileged in these applications. The model of the permanent magnet synchronous generator in a motor mode is described by the following equations [19,48]: d Id − ωL q Iq dt d Iq Vq = R Iq + L q + ωL d Id + ωϕ f dt The electromagnetic torque linked to the rotor frame is written as: ) ] 3 [( Tem = p L d − L q Id Iq − ωϕ f Iq 2 Vd = R Id + L d

(7)

(8)

2.5. Model of the pump In this study, the well known pump GRUNDFOS CRN2-23 (2.2 kW) associated with the synchronous motor is used. The pressure-flow and power-flow characteristics provided by the manufacturer for a frequency of 50 Hz have been interpolated using Matlab fitting Tool to obtain the dynamic behavior of the pump. The results are shown in Fig. 3. The relation between the pressure at the flow rate and the power transmitted to the pump based on this interpolation gives the following cubic polynomial equations: { Pr es = 0.75Q f 3 − 7.66Q f 2 + 3.35Q f + 1.64 (9) Pm− p = −0.0141Q f 3 + 0.0123Q f 2 + 0.466Q f + 0.99 Such as: – Pres is the head applied by the pump in meter – Qf is the flow rate at the input of the RO module in m3 /h. – Pm-p is the power transmitted to the pump by the PMSM (kW) The torque of the centrifugal pump which represents the PMSM-resistant torque is given by: Tr p = a Q 2f + bQ f Ω

(10)

In Eq. (10), a and b are pump parameters while Q and Ω are respectively the pump flow rate and the mechanical speed. Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

7

Fig. 3. Head-flow and power-flow characteristics of the high pressure pump.

Fig. 4. Block diagram of RO module.

To define the changes in the capacity of the pump: pressure, power and flow when a change of speed is made, similarity laws can be adopted as following [35]: ⎧ ( ) ⎪ ⎪ Q = Q Ω2 ⎪ 2 1 ⎪ ⎪ Ω( ⎪ 1 ⎪ ) ⎨ Ω2 2 (11) Pr es2 = Pr es1 ⎪ Ω(1 ) ⎪ ⎪ ⎪ ⎪ Ω2 3 ⎪ ⎪ ⎩ Pm− p2 = Pm− p1 Ω1

2.6. Model of the reverse osmosis membrane The RO membrane is a semi-permeable membrane that separates the liquid under the pressure gradient. The flow is carried out continuously tangentially to the membrane through the semi-permeable film which allows only the water molecules to pass through. Part of the solution to be treated (flow Qf) divides at the level of the membrane into two parts of different concentrations: • Part 1 (flow Qp) passes through the membrane (permeate). • Part 2 that does not pass through the membrane (concentrate or retentate) and contains the molecules or particles retained by the membrane. In Fig. 4. is represented the block diagram of the reverse osmosis (RO) module Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

8

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Fig. 5. Neural network architecture.

The diffusive water flux is a function of the surface area of the membrane, the difference in pressure and the osmotic pressure defined by [2,22]: Q p = ASr o (P f − 0.5 × 7.857.10−4 (c p + cc )PF + π p )

(12)

The osmotic pressure is depending of the permeate water concentration and can be expressed as following: π p = 7.857.10−4 C p The concentration polarization factor (PF ) is given by the following equation: ( ) Qp PF = exp K Qf

(13)

(14)

The permeate salt concentration at the outlet of RO module is given by: Cp =

B.Sr o (c f + cc )PF 2Q p

(15)

3. Proposed Artificial Neural Network (ANN) power manager The neural network approach as illustrated in Fig. 5 commonly known as the Artificial Neural Network (ANN) has been used in this research as algorithm to govern the multi-source system feeding the RO desalination unit It can be described as a mathematical model that reproduces the structure and function of the system of human biological neurons [25,31]. Most mathematical models of Neural Networks (NN) use learning algorithms. The most popular and the most effective is the back propagation learning algorithm [30]. Learning mechanisms are the most interesting properties of neural networks because some of them try to copy the process of memorizing knowledge of the human brain. There are two families of learning: supervised learning and unsupervised learning. The algorithms with supervised learning determines synaptic weights from labeled examples of forms in which a teacher associates desired responses or targets with similar labeling and a specific strategy. Unsupervised learning copies the functioning of the human brain that retrieves information by association. At the input of the network are presented known examples and the network is organized itself around attractors which correspond to stable configurations of the nonlinear dynamic model associated with the network. Learning is accomplished using rules that change or adjust the weight of synaptic coefficients according to the examples presented at the entrance and in some cases depending on the desired outputs. The proposed power manager based on supervised learning for the system will be primarily responsible for: (1) Flexible permutation between the operating modes, serving for a reliable flow of the energy. (2) Smart use of information from subsystems to manage their operating states. (3) Control the optimum routing of subsystem powers to ensure the production–demand balance of the entire system while respecting the operating constraints of each subsystem. Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

9

3.1. Power management strategy The power management for 24 h is illustrated in the flowchart of Fig. 6. The strategy is based on the availability of the power of solar and wind sources and to the daily demand of fresh water. Therefore, the difference between the total power and the load power Pnet must be determined given by the following equation: Pnet = Ppv + Pwind − Pload

(16)

If the difference is greater than zero, the excess will be stored by the batteries. Otherwise, the power stored by batteries will be depleted as an additional source to cover the needs of the load. This management is based on the determination of the power Pnet which represents the difference of power between that produced by the two sources GPV and wind generator on one hand and the power demanded by the load on the other hand. Then, according to the power generated by the wind source and the state of charge of the batteries, the power references will be imposed for the wind turbine and the batteries as indicated in the flowchart of the energy management algorithm. Throughout the day, the photovoltaic generator is controlled in MPPT by an (Extremum Seeking Control) which allows GPV to generate all its maximum power when it is in operation even under the conditions of partial shading. The state of charge of the batteries and the operating zones of the power curve of the wind turbine will determine the operating modes of the energy manager which are as follows: Mode1: This mode of operation occurs during the following cases: (a) If the power generated by the renewable sources is less than the power demanded by the load, the battery power stock must be used while the wind turbine will operate in MPPT mode. This case may occur during the night and at the beginning and end of the day. Since the sizing of the system is optimal the reserves of the batteries will cover the power deficit until the intervention of the photovoltaic generator. (b) If the power generated by the renewable sources is greater than the power required by the load and the state of charge SOC of the batteries <80% while the power generated by the wind turbine remains below its nominal power (Operating zone I), the excess power must be stored by the batteries while the wind turbine will operate in MPPT mode. Mode 2: This mode of operation occurs during the following cases: (a) If the net power Pnet is positive while a surplus of power is noticed, it is necessary to estimate the state of charge SOC of the batteries. Realizing that SOC > 0.8, we must estimate the difference between the power generated by the wind turbine and its nominal power. Thus, two cases will occur: If the difference in power is positive, this indicates that the wind turbine is in operating zone II which corresponds to its nominal power, we will proceed to the charging of the batteries at a reference permitting to store the excess power recorded while the wind turbine must switch to a power limitation reference equal to its nominal power. (b) If the net power Pnet is negative then a power deficit is detected, it is then necessary to estimate the difference between the power generated by the wind turbine and its nominal power. Thus, if the difference in power is positive, the wind turbine will operate at its nominal power and we must proceed to unload the batteries at a power reference that will cover the power deficit. Mode 3: This mode occurs when the batteries are fully charged (SOC 80%) with a surplus of recorded power (Pnet > 0), while the wind turbine enters in operating zone II which corresponds to its nominal power. The reference power of the wind turbine switches to a reference power corresponding to the power difference between the photovoltaic source and the load. This will avoid charging the batteries. The different torque references will be imposed by the Power manager according to the operating conditions of the proposed power management algorithm as illustrated by the following equations: ⎧ ⎪ ⎪ ⎪T ∗ = K op Ωmec 2 → ModeI ⎪ ⎪ ⎪ ⎨ Pn T∗ = → ModeI I (17) ⎪ Ω mec ⎪ ⎪ ⎪ ⎪ ⎪ ⎩T ∗ = Pr e f → ModeI I I Ωmec where: C p−max ρπ Rt 5 1 K op = (18) op 3 2 G3 Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

10

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Fig. 6. Flowchart of the energy management algorithm.

4. Control of the hybrid desalination system 4.1. PV side control approach The PVG is considered as the main source of the system. is controlled by an ESC (Extremum seeking control) strategy. The PVG operates during the day by giving its maximum power. The advantage of such controllers is the ability to overcome the partial shading conditions that occur during the day. Indeed, the effect of the partial shading gives rise to global and local MPPs on the P-V characteristic. The Conventional MMPT techniques can fail in the local MPP while using the ESC controller, it converges to the global MPP due to its ability to resolve the problem without knowing the internal system parameters. The architecture of the ESC controller is presented in Fig. 7. It consists of sinusoidal wave signal generator, High pass filter, Low pass filter and integrator. At the output of the HPF, the measured PV power is multiplied by a sinusoidal perturbation signal to determine the gradient of the derivative. The signal obtained pass through the LPF which filter the dither signal effect then is added to a small sinusoidal perturbation; the signal obtained at the output of the integrator gives the voltage reference at the MPP. The tuning of the ESC parameters which are the integrator gain KI and the perturbation signal magnitude do not require an analytical method but just an adjusting to control the speed of convergence. 4.2. PMSM side control strategy The control of the PMSM that drives the RO module pump is subject to the demand for freshwater represented by the equivalent freshwater power that is required as the operating reference for the PMSM. It is therefore imperative to ensure the routing of the power imposed by the NN manager to the MSAP. For this purpose, we propose a torque control based on a vector control with two external and internal cascade loops as shown in Fig. 8. The first loop is dedicated to the regulation of the DC-link around a fixed value by a robust regulator based on the theory of Lyapunov. As for the second loop, its role is through the control of d-axis and q-axis currents to ensure the transfer of all the power generated by the hybrid system to the PMSM. Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

11

Fig. 7. ESC control diagram.

Fig. 8. Schematic of the PMSM vector control.

4.2.1. DC -link regulation The establishment of the DC-link is necessary to ensure the flow of the source power to the PMSM driving the RO unit. For that, the Lyapunov-based controller is used to stabilize the voltage at fixed value. The equation of the DC-link currents is given by the following relation: d Vdc c = Isour ce − Iconv (19) dt For the control of the DC-link, the error between the reference voltage and the DC-link voltage is chosen such as: e = Vdc − Vdcr e f

(20)

To ensure that the tracking error goes to zero, quadratic Lyapunov function is chosen, which takes the following expression [8]: 1 Vdc = e2 (21) 2 From Eq. (21), The derivative takes the following expression: V˙dc = ee ˙ (22) It suffices as a sufficient condition of stability to ensure the negative definite derivative of the error such as: e˙ = −ke

(23)

Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

12

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Fig. 9. DC-link voltage control bloc.

Fig. 10. Solar input profiles.

From Eqs. (19) and (23), one can obtain: Isour ce − Iconv = −cke

(24)

The q-axis current reference coming from the regulation loop and which ensures the transmission of the power generated by the hybrid system at any time is given by: Pcr e f − Psour ce I qs − r e f = (25) V qs In Fig. 9 is shown the diagram of the Lyapunov-based control of the DC-link voltage. 5. Simulation results and discussion 5.1. Input data of Illizi In this study, powers corresponding to real data of solar irradiation and wind speeds related to the city of Illizi in the Algerian Sahara are chosen. These powers have been used as learning quantities for the ANN algorithm. The state of charge of the batteries is an input variable varying between 0.2 to 0.8 of the battery bank capacity. Two wind and solar profiles corresponding to two different dates and seasons shown in Figs. 10 and 11 are taken as inputs. The profile 1 concerns winter season and Profile 2 summer season in the region of Illizi in Algeria. The load capacity is calculated from the water requirements profile of a residential agglomeration of 60 inhabitants in the city of Illizi. This load profile, which varies during the year according to the seasons, is used to calculate the power equivalent profile that can be assuming a typical recovery rate of 30%. Fig. 12 corresponds to the load power demand at the specified dates of the year in Illizi. 5.2. Training results The shown regression curves in Fig. 13 show the results of the network outputs of training, validation, and target testing. For a perfect fit, the data must fall on a 45-degree line, where the outputs of the network are equal to the Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

13

Fig. 11. Wind input profiles.

Fig. 12. Load power demand for two seasons: summer (–) and winter (-).

targets. The fit is good enough for all data sets, with the values of R (the percentage of data falling on a 45-degree line) in each case worth 0.99. 5.3. Power delivered by the hybrid system Operating modes I and II imposed by the energy manager appears in Fig. 14 for the profile 1. While at the beginning of the day the photovoltaic generator is at a standstill, only the wind turbine and the batteries generate the power required by the load. During the day, the photovoltaic generator intervenes and the surplus energy is stored while the wind turbine is in MPPT mode. At the end of the day, due to the lower illumination of the GPV and the presence of a weak wind power, the batteries are solicited to fill the indicated power deficit and the manager stays in Mode I as long as the batteries are not fully charged or unloaded. Fig. 15 shows the progress of the power generated by the power generators for the second profile. We can see in the figure the effect of partial shading and the moment of calm of the wind. The GPV under the effect of shading loses power during peak hours. We can also see that the power generated by the wind turbine perfectly follows the shape of the wind profile. It is interesting to see that Mode II is reached at the moment when the turbine enters Zone III of its power curve and the power is limited to its nominal power of 2000 W. Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

14

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Fig. 13. Regression curves.

Fig. 14. Power variation during 24 h of operation for profile 1.

In Fig. 16 are depicted the electrical and mechanical performances of the PMSM driving the desalination unit. The FOC method principle adopted can appear on the shape of the d-axis and the q-axis currents. The direct Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

15

Fig. 15. Power variation during 24 h of operation for profile 2.

Fig. 16. Electrical and mechanical performances of the PMSM.

current is maintained at zero during the operating cycle with no disturbance indicating that the PI controllers are very efficient. The q-axis current which is the image of the electromagnetic torque of the PMSM takes the same shape. We can see on Fig. 16 that the rotation speed of the machine after the transient regime approximate closely its nominal rotation speed in the middle of the day; accentuated by the strong load request during this time of the day. The electromagnetic torque of the machine, which is the controlled quantity, has a variable appearance according to the reference which takes the shape of the speed and the load power. The voltage of the dc-link as it is visible in the same figure is kept constant during the day at the value of. 300 V. It is clear that the controller of the dc-link is very robust due to the absence of disturbances on the voltage. Despite the moment of calm and the effect of shading, the RN manager manages to ensure the energy balance between the power sources and the load demand. As can be seen in Fig. 17, the equivalent load capacity in fresh water and the power transmitted by the MSAP to the pump are very close in steady state at the same pace. Table 2 gives representative numerical values of the performance of the hybrid desalination system implemented in Illizi in southern Algeria. The daily quantity produced for two periods of the year varies between 3.10 M3 /day and 4.17 M3 /day. These results give the system a specific energy consumption (SEC) varying between 9.16 kWh/m3 Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

16

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

Fig. 17. Load reference power and power transmitted to the pump (profile 1). Table 2 Performances of the hybrid desalination system for the two profiles. Profile

PV energy (kWh)

Wind energy (kWh)

Batteries energy (kWh)

Equivalent energy for the water demand (kWh)

Fresh water quantity M3 /day

Profile 1 Profile 2

17.4 14.86

12.19 12.77

4.1 −0.529

38.2 26.7

4.17 3.10

and 8.61 kWh/m3 . From these results, we can estimate the price of the water production which is around 37 AD/m3 considering the price of the Algerian kilowatt hour. In order to calculate the cost of the produced desalinated water of the hybrid renewable energy system, one can firstly calculate the initial cost based on the prices of the components of the system according to the following costs: – GPV (25 × 100 W pv panels: 25 × 97$ = 2425$) – Wind turbine (1900$) – DC–AC converter (400$) – Battery 100 AH(170$) – RO unit (1500 $) – other costs: 150 $ Therefore the initial cost is 6445 $. The online cost analysis tool provided by NREL (National Renewable Energy Laboratory) [3] can be used to determine the Simple Levelized Cost of Renewable Energy which gives 0.03 $/kWh. According to the previous results in Table 2, the produced water cost is varying between 30.5 AD/m3 and 28.67 AD/m3 . 6. Conclusion and perspectives The main target of this research is the study of a reverse osmosis desalination system powered by a hybrid energy source: solar–wind. We presented the results obtained from the simulators and programs developed using mathematical models and control strategies adopted. The results of energy management by the neural network algorithm allowed the system to intelligently manage the energy-output-demand balance of the variable exogenous inputs as well as the variable load profile while exploiting the information obtained from different inputs of the sub-systems forming the conversion chain avoiding operating conflicts and limitations of different power sources. MPPT controller of the GPV ensures the maximization of the generated power even under the shading conditions that occur during certain periods while the control of the batteries by the use of the RN manager has ensured Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

17

the continuity of service of the desalination system during the shutdown moments of one of the renewable energy sources. As perspectives to this work, we intend to investigate the experimental implementation of such controllers besides the technical–economic evaluation of other topologies based on the specific energy consumption. References [1] World Water Assessment Programme (Nations Unies). The United Nations World Water Development Report 2015. United Nations Educational, Scientific and Cultural Organization, 2015. [2] Dow, Dow’s water and process solution. URL http://msdssearch.dow.com/PublishedLiteratureDOWCOM/dh_0885/0901b8038088586b. pdf?filepath=/609-00071.pdf1995(fromPage=GetDoc). [3] Levelized Cost of Energy Calculator tool available on: https://www.nrel.gov/analysis/tech-lcoe.html access on august 2019. [4] Muhammad Tauha Ali, Hassan E.S. Fath, Peter R. Armstrong, A comprehensive techno- economical review of indirect solar desalination, Renew. Sustain. Energy Rev. 15 (8) (2011) 4187–4199. [5] Aamer Ali, et al., Membrane technology in renewable-energy-driven desalination, Renew. Sustain. Energy Rev. 81 (2018) 1–21. [6] Karima Berkoune, et al., Robust control of hybrid excitation synchronous generator for wind applications, Math. Comput. Simulation 131 (2017) 55–75. [7] R. Bourbon, et al., Energy management optimization of a smart wind power plant comparing heuristic and linear programming methods, Math. Comput. Simulation (2018). [8] Y. Boussairi, A. Abouloifa, I. Lachkar, C. Aouadi, A. Hamdoun, State feedback nonlinear control strategy for wind turbine system driven by permanent magnet synchronous generator for maximum power extraction and power factor correction, Vib. Anal. Control Mech. Struct. Wind Energy Convers. Syst. (2018) 75. [9] Upeksha Caldera, et al., Role of seawater desalination in the management of an integrated water and 100% renewable energy based power sector in Saudi Arabia, Water 10 (1) (2018) 3. [10] J.A. Carta, J. Gonzalez, V. Subiela, Operational analysis of an innovative wind powered reverse osmosis system installed in the Canary Islands, Sol. Energy 75 (2) (2003) 153–168. [11] P.C. Carvalho, L.A. Carvalho, J.J. Hiluy Filho, R.S. Oliveira, Feasibility study of photovoltaic powered reverse osmosis and pumping plant configurations, IET Renew. Power Gener. 7 (2) (2013) 134–143. [12] Alexandros C. Charalampidis, Antonios E. Chaniotis, Antonios G. Kladas, Current waveform optimization techniques for synchronous machines and numerical evaluation in the case of a PMSM wind turbine generator, Electr. Eng. 99 (2017) 525–533. [13] O. Charrouf, A. Betka, H. hadef, M. Djebabra, M. Tiar, Degradation evaluation of PV modules operating under Northern Saharan environment in Algeria, in: AIP Conference Proceedings (Vol. 1814, (1) 020030), AIP Publishing, 2017. [14] Ali M. El-Nashar, M. Samad, The solar desalination plant in Abu Dhabi: 13 years of performance and operation history, Renew. Energy 14 (1–4) (1998) 263–274. [15] P. Feron, The use of wind power in autonomous reverse osmosis sea water desalination, Wind Eng. 9 (1985) 180–199. [16] M. Forstmeier, F. Mannerheim, F. D’Amato, M. Shah, Y. Liu, M. Baldea, A. Stella, Feasibility study on wind-powered desalination, Desalination 203 (1) (2007) 463–470. [17] Vasilis Fthenakis, et al., New prospects for PV powered water desalination plants: case studies in Saudi Arabia, Prog. Photovolt., Res. Appl. 24 (4) (2016) 543–550. [18] Akhtar Hussain, Syed Muhammad Arif, Muhammad Aslam, Emerging renewable and sustainable energy technologies: State of the art, Renew. Sustain. Energy Rev. 71 (2017) 12–28. [19] Dariusz Janiszewski, Extended Kalman Filter Based Speed Sensorless PMSM Control with Load Reconstruction, Intech, Kalman Filter, 2010. [20] Meer A.M. Khan, S. Rehman, Fahad A. Al-Sulaiman, A hybrid renewable energy system as a potential energy source for water desalination using reverse osmosis: A review, Renew. Sustain. Energy Rev. 97 (2018) 456–477. [21] Senthil Kumarasamy, Shankar Narasimhan, Sridharakumar Narasimhan, Optimal operation of battery-less solar powered reverse osmosis plant for desalination, Desalination 375 (2015) 89–99. [22] Senthil Kumarasamy, Shankar Narasimhan, Sridharakumar Narasimhan, Optimal operation of battery-less solar powered reverse osmosis plant for desalination, Desalination 375 (2015) 89–99. [23] A. Lamei, P.V.D. Zaag, E.V. Munch, Impact of solar energy cost on water production cost of seawater desalination plants in Egypt, Energy Policy 36 (2008) 1748–1756. [24] Sangkeum Lee, Hong-Yeon Cho, Dongsoo Har, Operation optimization with jointly controlled modules powered by hybrid energy source: A case study of desalination, Renew. Sustain. Energy Rev. 81 (2018) 3070–3080. [25] Sonia Leva, et al., Analysis and validation of 24 h ahead neural network forecasting of photovoltaic output power, Math. Comput. Simul. 131 (2017) 88–100. [26] Zhenyu Li, et al., Towards sustainability in water-energy nexus: Ocean energy for seawater desalination, Renew. Sustain. Energy Rev. (2017). [27] C. Lupangu, R.C. Bansal, A review of technical issues on the development of solar photovoltaic systems, Renew. Sustain. Energy Rev. 73 (2017) 950–965. [28] Akbar Maleki, Design and optimization of autonomous solar-wind-reverse osmosis desalination systems coupling battery and hydrogen energy storage by an improved bee algorithm, Desalination 435 (2018) 221–234.

Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.

18

O. Charrouf, A. Betka, S. Abdeddaim et al. / Mathematics and Computers in Simulation xxx (xxxx) xxx

[29] J.F. Manwell, J.G. McGowan, Recent renewable energy driven desalination system research and development in North America, Desalination 94 (3) (1994) 229–241. [30] Phil Mars, Learning Algorithms: Theory and Applications in Signal Processing, Control and Communications, CRC press, 2018. [31] Stephen Marsland, Machine Learning: An Algorithmic Perspective, Chapman and Hall/CRC, 2011. [32] R. McBride, R. Morris, W. Hanbury, Wind power a reliable source for desalination, Desalination 67 (1987) 559–564. [33] Marcos S. Miranda, David Infield, A wind-powered seawater reverse-osmosis system without batteries, Desalination 153 (1) (2003) 9–16. [34] F. Moreno, A. Pinilla, Preliminary experimental study of a small reverse osmosiswind- powered desalination plant, Desalination 171 (3) (2005) 257–265. [35] Bruce Roy Munson, et al., Fluid Mechanics, Wiley, Singapore, 2013. [36] Kim Choon Ng, Muhammad Wakil Shahzad, Sustainable desalination using ocean thermocline energy, Renew. Sustain. Energy Rev. 82 (2018) 240–246. [37] Isidro Padrón, et al., Assessment of hybrid renewable energy systems to supplied energy to autonomous desalination systems in two islands of the canary archipelago, Renew. Sustain. Energy Rev. 101 (2019) 221–230. [38] G. Petersen, S. Fries, J. Mohn, A. Muller, Wind and solar powered reverse osmosis desalination units: Description of two demonstration projects, Desalination 31 (1979) 501–509. [39] Robert Pohl, Martin Kaltschmitt, Robert Holländer, Investigation of different operational strategies for the variable operation of a simple reverse osmosis unit, Desalination 249 (3) (2009) 1280–1287. [40] Ma a Qingfen, Hui Lu b, Wind energy technologies integrated with desalination systems: Review and state-of-the-art, Desalination 277 (2011) 274–280. [41] B.S. Richards, G.L. Park, T. Pietzsch, A.I. Schäfer, Renewable energy powered membrane technology: Brackish water desalination system operated using real wind fluctuations and energy buffering, J. Membr. Sci. 468 (2014) 224–232. [42] B.S. Richards, G.L. Park, T. Pietzsch, A.I. Schäfer, Renewable energy powered membrane technology: Safe operating window of a brackish water desalination system, J. Membr. Sci. 468 (2014) 400–409. [43] S.M. Shalaby, Reverse osmosis desalination powered by photovoltaic and solar Rankine cycle power systems: a review, Renew. Sustain. Energy Rev. 73 (2017) 789–797. [44] M. Thomson, D. Infield, A photovoltaic-powered seawater reverse-osmosis system without batteries, Desalination 153 (2002) 1–8. [45] Nanfang Yang, et al., Techno-economic design methodology of hybrid energy systems connected to electrical grid: An application of hybrid railway power substation, Math. Comput. Simulation (2018). [46] Mahdi Jedari Zare Zadeh, Seyed Hamid Fathi, A new approach for photovoltaic arrays modeling and maximum power point estimation in real operating conditions, IEEE Trans. Ind. Electron. 64 (2017) 9334–9343. [47] Malek Zaibi, et al., Smart power management of a hybrid photovoltaic/wind stand-alone system coupling battery storage and hydraulic network, Math. Comput. Simulation 146 (2018) 210–228. [48] M. Zhou, Y. Wu, L. Zhong, Q. Chen, J. Guo, Adaptive robust control for high- speed permanent magnet synchronous motor systems, in: Control Science and Systems Engineering (ICCSSE), 2017 3rd IEEE International Conference on (Pp. 346-350), IEEE, 2017.

Please cite this article as: O. Charrouf, A. Betka, S. Abdeddaim et al., Artificial Neural Network power manager for hybrid PV-wind desalination system, Mathematics and Computers in Simulation (2019), https://doi.org/10.1016/j.matcom.2019.09.005.