You are on page 1of 11

Design of an intelligent ANFIS Controller for a

Photovoltaic Irrigation System


Siwar Bellahirich* Dhafer Mezghani Abdelkader Mami
UR-LAPER, Faculty of Sciences of UR-LAPER, Faculty of Sciences of UR-LAPER, Faculty of Sciences of
Tunis, University of Tunis El Manar, Tunis, University of Tunis El Manar, Tunis, University of Tunis El Manar,
Tunis, Tunisia Tunis, Tunisia Tunis, Tunisia
siwar.bellahirich@etudiant-fst.utm.tn dhafer.mezghani@fst.utm.tn abdelkader.mami@gmail.com

Abstract— Taking into account its strategic Climatic diversity: Mediterranean climate in the
geographical location, Tunisia benefits of a panoramic north, semi-arid in the center and hot desert in the
climate favoring the diversity of agricultural activities south; plant production is distinguished by great
and thus making it one of the most important pillars diversity.
of the country’s economy.
Water is essential for agricultural activity and for
In addition, Tunisia has a very high level of the life of certain crops.
sunshine and a maximum temperature that often rises The classic irrigation methods used manifest
to around 50 ° C; then, the wise exploitation of solar themselves in dramatic water losses and unwanted and
energy in Tunisia represents a very promising unconscious overconsumption.
investment area which will be able to compensate the
massive use of polluting and exhaustible non- Or, global climate change threatens the existence of
renewable energies. several water resources in several countries, of which
Tunisia is no longer an exception.
On the other hand, the growing use of IT tools and It is therefore necessary to provide for methods
the development of regulation techniques in the likely to save water use, while ensuring the best
industrial field has become a reality which is continuity of agricultural activities and good
manifested through the reliability and power of the performance in the sector.
digital computer which offers the possibility of
implementing control laws with a great flexibility. The increase in temperature presents a double edge
for the country, because Tunisia has a sunshine rate
In this context, we propose to design an intelligent that exceeds 3000 hours / year, in addition to a
control strategy based on a neuron-fuzzy controller, temperature that is close to 50 ° C.
particularly the ANFIS architecture, combined with a
scalar control in order to control a photovoltaic These two factors represent a great asset for
irrigation system. investment in the field of renewable energies, in
particular solar energy, which exploitation in this
This process is coupled with a meteorological country remains limited.
(temperature Ta), pedological (sunshine Ec) and On another side, neuron-fuzzy logic is an important
technical (water flow rate delivered by pump Q) research axis on which many researchers are focusing.
database and will be applied to the PWM block inlet
of the pumping system. Technological applications are now available, both
in the general public (cameras, washing machines,
The simulation results show an immediate microwave ovens, etc.), and in the industrial field
optimization of water pumping, hence the robustness (regulation and control of complex energy-related
of the intelligent neuron-fuzzy control used; in processes, transport, material transformation, robotics,
addition to the good use of the country's natural etc.).
resources (water and solar energy), thus making a Fuzzy logic represents a logic which substitutes for
revolutionary economic and energy ratio compared to binary logic, based on variables which can take (true 1)
other conventional irrigation methods. or (false 0), it takes the intermediate values "true" or
"false" with a certain probability (a certain degree of
Keywords—– Irrigation, solar energy, intelligent belonging, or, real numbers between 0 and 1).
control, fuzzy-neuron, ANFIS architecture, scalar
control, photovoltaic, PWM Artificial neural networks are inspired from human
brain function and learning capacities, hence, they
I. ITRODUCTION represent a mathematical modeling of the human brain.
Tunisian agriculture represents a vast and This structures are organized around a set of cells
important field for the economy of the country and the (neurons) interconnected according to a certain
individual’s activity. architecture by weighted links that can be modified in a
procedure called ‘learning’.
Until the 1990s, these two techniques, neural and II. PHOTOVOLTAIC IRRIGATION SYSTEM
fuzzy, appeared to be quite distinct approaches, each
with their advantages, disadvantages and their specific A. Photovoltaic Generator
areas. However, they also have some common points: The photovoltaic generator converts instantly the
- The two methods applied to the control have good light energy of solar radiation and delivers an
qualities of robustness and are capable of extrapolation electrical energy [3].
and generalization. This model is called mono-diode coupled to a
- Neither method requires a mathematical model of start up capacity illustrated in the following figure
the system to be controlled; then, they can therefore [2], [9], [10], [13]:
both be applied to nonlinear systems without particular
complications.
The idea is to create an optimized fuzzy interference
system, seeking to use supervised learning methods to
automatically optimize certain parameters of a fuzzy
interference system.
For that, there are three fuzzy-neuron architectures:
the Nomura architecture, the LSC architecture and the
ANFIS architecture.
In the following, we will present an interesting Fig. 2: Electrical diagram of photovoltaic generator with startup
application of the neuron-fuzzy control, precisely the capacity
ANFIS architecture, for a photovoltaic pumping system
consisted of a photovoltaic generator, a three-phase The current voltage relationship is given by [4]: 
inverter, an asynchronous motor-pump set plus pipes
and valves as its shown in the following figure:

The photo-current is given by:

The junction temperature is given by:

B. Three phase inverter


It is a static voltage converter controlled using
PMW pulse width modulation; it generates an AC
Fig. 1: Synoptic diagram of the photovoltaic installation voltage from a DC voltage source, then, connection
of the solar panel to the power grid.
This application presents the fruit of the
combination of renewable resources and new control The electrical diagram of the three phase inverter
technologies by way of switching to a modern, more is shown in figure 3 [10]:
efficient and economical irrigation technique, able to
conserve national water reserves, spare harmful non-
renewable resources and maintain, or even reinforce,
local production with the lesser quantities of water used
compared to other conventional unfavorable, inefficient
and expensive irrigation methods (surface irrigation,
sprinkler irrigation, etc.).
Thus, the proposed design shows a crucial economic
contribution, not only for the governance of water use,
but also for a structural point of view and minimal
components used in relation to other research work on
photovoltaic pumping, while maintaining better
performance.

Fig.3: Electrical diagram of the three phase inverter mounted to


three phases motor
We consider that the inverter is ideal; the D. Centrifuge pump
switches are perfect and switch instantly. In the case of our installation, the tanks are
We take the case of a balanced three phase load always connected to each other and on the same
verifying: plane:

The matrix writing is given by [6]:


Therefore, the flow equation becomes [11]:

=

We note by: E. Pipes and valves models


Ciref, where i= {1, 2, 3, 4, 5, 6}, control signals The following table summarizes the necessary
where Ciref = 1 for Ti passing and Ciref = 0 for Ti hydraulic parameters to use in our system [1], [5],
blocked. [12], [19]:
Ti, where i= {1, 2, 3, 4, 5, 6}, transistor (bipolar, TABLE.1: SUMMARY TABLE OF HYDRAULIC MODELING
MOSFET…) considered as perfect switches. PARAMETERS
Element Parameter formula
C. Asynchronous motor
The equations of the induction machine in the
Dissipative effect
two-phase reference (d,q) related to the three-phase
reference through the Park transformation are [10],
[15]:
Valve

Park’s matrixis equivalent to:

With :
III. ANFIS CONTROLLER

The Adaptative-Network-based Fuzzy Inference


System (ANFIS) is a multi-layered network whose
And connections are not weighted, or all its weights are
equal to 1.
There are two different types of nodes depending
on their functionality: square (adaptive) nodes
contain parameters, and circular (fixed) nodes have
no parameters.
forms a constant and invertible linear
transformation. However each node (square or circular) applies a
function to its input signals.
The electrical diagram of the three phase
induction machine is shown in figure 4 [10]: The output of node i of layer k (called node
(i, k)) depends on the signals coming from layer k-1
and the parameters of node (i, k), i.e:

Where is the number of nodes in layer k-1,


and a, b, c, are the parameters of node (i, k). For a
circular node, these parameters do not exist.
In the network of the following figure, the nodes
of the same layer have functions derived from the
same family that we explain above:

Fig.4: Electrical diagram of the three phase inverter mounted to


three phases motor
rate of water Q mesured at the pump’s output. The
frequency fs represents the ANFIS controller’s output.

Fig.5: ANFIS network equivalent to fuzzy reasoning

Layer1: each node in this layer is a square node:

Fig.7: ANFIS editor window


Where x is the input of node’s i, and the linguistic
term associated with its function. To train the ANFIS network, 30 data sets are used for
temperature Ta, irradiance Ec, and flow rate Q as it is
Layer2: each node i of this layer is a circular node Π shown in the next figure:
which generates the product of its inputs.
This product represents the degree of activation:

Layer3: each node i of this layer is a circular node N;


they estimate the normalized weight of each rule.

Layer4: each node of this layer is a square node with:

Where is the output of layer 3, and { } are


the set of output parameters of rule i.
Layer5: this layer has a single circular node which
affects the sum of the signals coming from layer 4, i.e:
Fig.8: Training data sets

The training error with 1000 epochs tends towards zero


as it is drawn in figure 9:
The number of Layer 1 nodes is always equal to the
total number of defined linguistic terms.
The number of nodes in layers 2, 3 and 4 is always
equal to the number of fuzzy rules.
Thefollowing figure shows the design of the intelligent
controller:

Fig.6: Block diagram control with ANFIS

Sunshine Ec and temperature Ta of the photovoltaic


panel are used as inputs of the ANFIS controller, plus, the
Fig.9: Training error with 1000 epochs
The following figure represents the ANFIS structure:

Fig.13: Membership function of input3: Q

The next figures represent the surface between two


inputs and the output (Ta and Ec), (Ta and Q) and (Ec and
Q):

Fig.10: ANFIS structure

The input membership functions of temperature,


irradince and masured water flow rate are represented
respectively in figure 11, 12 and 13:
Fig.14: Surface viewer window between (Ta,Ec) and fs

Fig.11: Membership function of input1: Ta Fig.15: Surface viewer window between (Ta,Q) and fs

Fig.12: Membership function of input2: Ec Fig.16: Surface viewer window between (Ec,Q) and fs
Rules are presented by: IV. SCALAR CONTROLLER

The scalar command is based on the equivalent phase


diagram of the asynchronous motor as it is presented:

Fig.19: Equivalent diagram per phase of the asynchronous motor

The magnetizing current is calculated by []:

Scalar control allows the:


- Maintain constant magnetizing current

- Flow’s motor reduction


- Supply voltage decrease (for fs < fnom )

In this intelligent design, the output of the ANFIS


Fig.17: Rule viewer controller fs is used as the input of the scalar controller
block.
The final step is to export all those data to workspace
Using this frequency, we deduce the Vs voltage and
and build the Fuzzy Inference System (FIS) in order to use
produce the PWM signals:
it in MATLAB Simulink environment:

Fig.18: Simulink design of ANFIS controller

Fig.20: Simulink design of the Scalar controller


V. SIMULATION RESULTS

The results are obtained by simulating the Simulink


models over a period of 10 seconds.

The Simulink model of the photovoltaic irrigation


system is
The Simulink model of the Vs/fs controller is giving
by:

VI. CONCLUSION
VII. DISCUSSION

Fig.21: Simulink model of the intelligent ANFIS controller design for In order to ensure the proper functioning of the
the photovoltaic puping system system;We kept a single panel in parallel and we increased
the number of panels in series until the adequate voltage is
The PV voltage Vp and current Ip are represented in the reached at the output which is almost equal to 544V for
following figure: nominal conditions: Ta=25°C and Ec=1000W/m².
The frequency fs generated by the ANFIS controller
designed at the same conditions is equal to 50Hz:

Fig.22: Vp characteristic for Ta=25°C and Ec=1000W/m²

Fig.24: Characteristic of ANFIS controller output

The Voltage Vs generated by the scalar controller at the


same conditions is equal to 220V; its characteristic is
shown in figure 25:

Fig.23: Ip characteristic for Ta=25°C and Ec=1000W/m²


PWM signals will be used to control the three-phase
voltage inverter:

Fig.24: Characteristic of the scalar controller output

The PWM Simulink model is:


Fig.27: PWM output signals

As its name indicates, the controlled inverter generates


three sinusoidal voltages with 120° shift phase of 120
each; and of amplitude near to 400V:

Fig.25: PWM Simulink model

The PWM input voltages are represented as follow:

Fig.28: Inverter output voltages

Thus, the voltage at the output of the inverter will


represent the source of the three-phase asynchronous
machine.
The Simulink model of the induction machine is shown
in the following figure:

Fig.26: PWM input voltages


Fig.29: Induction machine Simulink model

The torque and the rotation speed of the three-phase


asynchronous motor will be illustrated in the following
figures:

Fig.30: Induction machine torque (N.m) Fig.31: Induction machine speed (tr/min)

The torque undergoes a slight positive vibration which In these nominal conditions, the characteristic of the
does not exceed 2.6 for a period of 2s then tends to 0.6 rotation speed of the induction machine, shows two
N.m. operating modes, a transient regime for 2s and an
established regime where the speed stabilizes and reaches
its maximum value close to 2800tr/min.
After incorporating the valve and pipe parameters artificial neural networks and even the ANFIS architecture
described in table 1; the characteristics of the flow rates employed to control the MPPT [].
delivered by the pump Q and that passing through the one
pipe are as follows: The perspective that it would be interesting to give to
this research work for the purpose of its innovation is to
implement this intelligent ANFIS design control laws on
an embedded target as Raspberry Pi.

DISCUSSION

On a national scale, this new design of intelligent


ANFIS controller combined with the classic scalar
controller and based on the simulation results presented
here, this design is very promising for the development of
the agricultural sector.
This work illustrates a very beneficial economic energy
relationship in the short and long term.
With the help of this application, Tunisia will conserve
a lot of poorly exploited hydraulic resources and will
minimize the use of polluting fossil resources through the
adoption of sustainable and affordable renewable energies,
in particular solar energy.
Thus the country will be able to achieve its energy,
water and food self-sufficiency with the least expenditure.
Funding: This work was not had any grant a funding
from any Organization or Institution.
Conflict of Interest: The authors declare that they
have no conflict of interest.
Fig.32: Flow rates characteristics (l/min) APPENDIX
Both cureves show two operating modes, a transient
regime for 2s and an established regime where the flow TABLE A1: IRRIGATION SYSTEM’S NOMENCLATURE
rates stabilize and reache their maximum values close to Variable Value
34 l/min and 11 l/min. : Photovoltaic current of the
photovoltaic generator (A)
The rotation speed of the induction machine and the : Reverse saturation current or
flow delivered by the pump are proportional. diode leakage current (A)
: Photovoltaic voltage of the
Similarly, the flow delivered by the pump and the flow photovoltaic generator (V)
of water flowing in the branches are proportional. : Junction temperature
: Photo-courrent (A)
The synoptic diagram opted, allows to divide the flow : Short circuit current (1.19A)
delivered by the pump in an equal way (same flow for : Temperature of photo-corrent
every irrigation branche). coefficient of the photovoltaic
module (0.075%/°C)
: Sunshine received by the cell
(W/m²)
CONCLUSION : Reference sunshine 1000
W.m-2
: Reference temperature 25°C
The work presented in this paper concerns the design : Reference temperature 298 K
of an intelligent ANFIS controller for a photovoltaic : Ambient temperature (°C)
irrigation system. NOCT : Rated operating temperature
(45°C)
Thanks to the intelligent design chosen, the system
performances get improved in a unique way.
TABLE A2: KANEKA 60 CHARACTERISTICS
The structure is stable, precise and much faster Electric quantities Values
compared to its response when controlled by the scalar Power max Pmpp (W) 60
control only []. Voltage max Umpp (V) 67
Current max Impp (A) 0.9
The obtained response is very competitive compared to Vacuum voltage Uoc (V) 92
other smart control designs too, including fuzzy logic, Short circuit current Isc (A) 1.19
Time coeff Isc 0.075
[ %/ °C ]
Time coeff Uoc [ mV / °C ] -280
Time coeff -0.23
[ % / °C ]

TABLE A3: MOTOR PUMP PARAMETERS


Electric quantities Values
Useful power (W) 370
Electric power (W) 550
Nominal flow (l/min) 35.6
Height max (m) 7
Statoric resistance ( ) 24.6
Rotoric resistance ( ) 16.1
Mutual inductance (H) 1.46
Rotoric inductance (H) 1.48
Statoric Inductance (H) 1.49
Pump’s coefficients
C1 (Kg.m-2) 1.75 10-3
C2 (Kg.m-4.s-2) 7.5 10-6
J (Kg.m-1) 6.5 10-3
b0 (min².m.tr-2) 4.52 10-4
b1 (min².m.tr-1.L-1) -1.966 10-3
b2 (min².m.L-2) -0.012
(min².m.L-2) 4.0816 10-3

TABLE A4: HYDRAULIC NETWORK PARAMETERS


Physique quantities Values
Length and diameter Cp (m) 0.1 ;; 0.04
Length and diameter Cs (m) 0.25 ;; 0.03
Length and diameter Ct (m) 5.0 ;; 0.03
Length and diameter Valve (m) 0.05 ;; 0.03
Density 1000
(Kg / m3)
Kinematic viscosity (m²/s) 1.005 10-6
Darcy-Weisbach coeff 0.02475
Weight constant g (m/s²) ou 9.80665
(N/Kg)

REFERENCES

You might also like