1. Introduction
The advancement of photovoltaic (PV) technologies and the drop in manufacturing costs have led to a sudden increase in the number of mega-watt-sized solar projects. Developed countries have planned to wean themselves off their reliance on fossil-fuels, as solar investment becomes an appealing option [
1,
2,
3]. PV energy is superior to other sources because it does not require periodic maintenance, and PV panels have a lifetime of two and a half decades [
4,
5,
6]. PV systems can be either grid-tied systems or off-grid systems [
7]. In the grid-tied systems, the load is powered by the PV system, the main electric grid, or both [
8]. A grid-tied system is less complicated than an off-grid system, which is more commonly seen since it has fewer components and is generally more cost effective; for example, there are usually no batteries in grid-tied systems [
9]. The off-grid system can operate without a battery if the load is set to be powered during peak hours of the day, like a water pump system [
10]. However, if the load is a home appliance, it is better to have enough energy storage devices with enough capacity for a couple of days [
11].
The output power of the PV systems is intermittent and highly depends on weather and sun radiation levels. The sun irradiance affects the PV current directly, and the temperature affects mostly the output voltage [
12,
13,
14]. Therefore, it is essential to have power processing units regulate the output voltage and extract the most out of the PV panels. The load seen by the PV determines the operating point on the power-voltage curve (P-V curve) and often does not match the resistance at the maximum power [
15]. The resistance mismatch results in a loss of power and a reduction of the system’s overall efficiency. The resistance seen by the PV needs to be varied to match the maximum power resistance at various temperatures and solar irradiance levels [
16,
17,
18,
19]. Power processing units enable the operation at the highest power if a maximum power point tracker (MPPT) controls the power processing unit. MPPT is an algorithm or method that tracks the maximum operating point, which is necessary to extract the most out of the PV panel, enhance the operation of the overall system, and extend the lifetime of the PV panels [
20].
Many MPPT methods have been developed and utilized [
21,
22,
23,
24,
25] during the last two decades. Several studies evaluated the performance of such methods [
26,
27,
28,
29,
30] based on both uniform solar irradiance and partially shaded conditions. The essential aspects of the comparisons are the implementation complexity, reaching the real maximum power point (MPP), the required number of sensors, convergence speed, cost, and efficiency [
15,
31]. Several other aspects are considered, such as the dependency on the PV panel or array and whether the control system is analog or digital. The most widely used method is perturbation and observation (P&O). In P&O, the power is calculated from the current and voltage sensors and then compared to the previous power reading. If the current reading is higher than the previous, the reference voltage should be increased until it reaches the maximum power. However, if the current reading is less than the previous reading, the reference voltage is higher, the maximum power voltage (Vmpp). Therefore, the reference voltage needs to be decreased. The P&O conversion speed depends on the perturbation size (
). That is, if
is small, the convergence speed will be slow. On the other hand, when
is large, the convergence speed will be fast. Very large
might cause the control system’s instability, and it might never reach the real maximum power.
Several modifications to P&O have been suggested and experimented with, such as in [
32,
33,
34,
35], where the dynamic perturbation size was introduced. Similar methods to P&O, such as hill-climbing (HC) and incremental conductance (IC), were presented in [
16,
36], respectively, where their speed depends on the size of the step size. Several other methods were introduced, such as fuzzy logic, genetic algorithms, fractional open-circuit voltage (FOCV), fractional short-circuit current, neural networks, ripple correlation control, and many others. The most common methods will be discussed in detail and will be illustrated based on their performance in the next sections.
FOCV is the simplest and easiest to implement tracking approach. However, under the variation of temperature, the tracker cannot reach the true maximum power point. This paper presents an enhancement modification to the FOCV. The feed-forward neural network (FFNN) is used to learn the correlation between the temperature, irradiance, voltage at maximum power, and the open-circuit voltage and enhance the performance by providing the accurate voltage reference for the pulse width modulation controller. That is, the FFNN is used to find the fraction of the voltage at the maximum power and the open-circuit voltage at specific irradiance and temperature instead of guessing or estimating the value of the fraction. Therefore, the advantages of the proposed method are:
It provides accurate results in all weather and irradiance conditions.
It has a stable operation with no oscillation around the MPP.
It is a cost-effective method, which does not require expensive sensing components.
It has a faster transient response and can track the MPP during a sudden decrease in the solar irradiance.
The rest of this paper is organized as follows:
Section 2 discusses several PV models and the models’ performance under weather and temperature variations.
Section 3 illustrates several MPPT methods and compares them to their performance, cost, and implementation.
Section 4 presents the proposed modification to the fractional open-circuit and the implementation details. The proposed method results under several conditions and the performance comparison with other commonly used methods are presented in
Section 5. Finally, the conclusions are presented in
Section 6.
3. Maximum Power Point Tracking Approaches
A single PV cell has a unique mpp, corresponding to the solar irradiance and temperature point. The resistance at that point is represented by
. The
at the STCs can be calculated by:
The variation in solar irradiance and temperature values makes
variable. Therefore, a tracker is needed to allow a PV panel to operate at
and extract the maximum power. Many approaches have been introduced to track the MPP. The commonly used technique in the solar industry is the P&O because of the reasonable trade-off between simplicity and performance. The P&O approach flowchart is seen in
Figure 5. This approach perturbs the duty cycle and observes the power if it increases or decreases. Then, a step-change to the duty cycle takes place. The step-change determines the conversion speed and accuracy. P&O with a large step-change lacks accuracy and might experience instability issues. Making the step-change small reduces the conversion speeds, lacks speed, and might perform poorly at fast transitioning solar irradiance. The adaptive step can be applied, but without the guarantee of stable operation [
36,
38,
39,
40]. Similarly, hill-climbing and incremental conductance have similar performance to P&O, with no significant advantages. The ripple correlation control approach takes advantage of ripples imposed by switching devices on the PV panel [
41,
42]. In RCC, the derivative of the voltage is multiplied by the derivative of the power. If the product is zero, then the PV operates at its maximum power point. If the product is higher or lower than the MPP, then the PV operates on the left or the right of the MPP, respectively. The drawback of using RCC is the required multiplied devices, which are power-consuming devices.
Intelligent tracking approaches, such as fuzzy logic and artificial neural networks, were used to track the MPP. In fuzzy logic, the PV voltage and current are used to calculate the slope of the P-I characteristic and the change in the slope, which will be the input of the fuzzification stage. Fuzzification converts the numeric to linguistic variables to be used in the inference stage to make a decision. After a decision is made by a set of conditionals statements, the defuzzification stage converts the decision into numerical values [
43,
44,
45]. Although the performance of the fuzzy controller is higher than the traditional P&O, the rules of the membership function selection are based on trial and error, which increases the implementation time. The neural network method was used to track the MPP. The technique uses acquired data, usually irradiance and temperature, to train the neurons and provide the optimal weight that makes the correct control decision about the duty [
46,
47]. The NN can track the true MPP and has solid performance, but it uses an expensive pyranometer to read the irradiance values.
The fractional open-circuit voltage (FOCV) method is the simplest MPPT method. In this method, the MPP voltage can be approximated by multiplying the open-circuit voltage with a fraction number
k. The
k is a constant value predetermined by either experimentation or assumption, and it has a typical value ranging between
and
. The open-circuit voltage must be sensed periodically. The challenge is that the open-circuit voltage cannot be measured while the system operates without disconnecting the system. Disconnecting the system momentarily to measure the open-circuit voltage leads to power loss and overall efficiency degradation. In order to find the open-circuit voltage without PV panel disconnection, a pilot cell with the same characteristics as the main PV panel is needed.
Figure 6 shows the implementation of the FOCV with a pilot cell.
5. Simulations
The simulation setup of the proposed method was implemented using MATLAB and PLECS Blockset, as shown in
Figure 12. The model KC200GT was used as an input source to supply a resistive load. The PV system was interfaced using a Ćuk converter, as shown in
Figure 12a. The Ćuk converter was selected because of the non-pulsating input and output currents, which enhance the PV cell’s lifetime and obtain better reading values of the PV current, as well as providing the ability to step-up and step-down the voltage. Then, the resistance of the maximum power point (Rmpp) can be tracked from zero to infinity
∞. The Ćuk converter is designed to operate in continuous conduction mode, where the inductor current at the input level must be above zero in the steady-state. In the continuous conduction mode, the converter goes into two subintervals. In Subinterval 1, the MOSFET is ON, and the diode is in reverse bias mode. The inductor
is charging from the input source, and the capacitor
is discharging to the output. The equivalent circuit of this subinterval is illustrated in
Figure 13a. In Subinterval 2, the MOSFET is OFF, and the diode is in forwarding conduction mode. The capacitor
is charging from the input source and the inductor
. The equivalent circuit of Subinterval 2 is shown in
Figure 13b. The output voltage of the Ćuk converter is inverted and given by:
The parameters used in the simulation are
H,
H,
F,
V,
m
. The MPPT tracker shown in
Figure 12b consists of the pilot PV cell and the neural network. The pilot PV cell is connected directly to a voltage sensor to obtain the open-circuit voltage. The controller uses a pilot cell for open-circuit voltage measurements and a temperature sensor to provide temperature values. Then, the trained neural network takes the
and
values at time
t and provides the corresponding
k. The value of
k will be multiplied by
to provide the reference voltage to the pulse width modulator (PWM). The PWM operates at a 100 kHz switching frequency, and the control bandwidth does not exceed one-tenth of the switching frequency.
5.1. Case1: Variation of Solar Irradiance with a Constant Temperature Value
In this case, the output load and temperature are assumed constant for two reasons. First, the relationship between the output power and solar irradiance is stronger than the relationship between the output load and temperature, which gives an acceptable estimation of the tracking algorithm’s performance if the temperature value is constant. Second, all algorithms need an ideal test, where the load is constant to find their best performance.
Figure 14 shows the performance of the algorithms under the variation of solar irradiance. The proposed method outperforms the P&O algorithm in terms of accuracy and can reach the MPP faster. P&O oscillates around the MPP without reaching the exact MPP. The FOCV method has a similar dynamic performance to the proposed method. However, the accuracy of the FOCV is less than the proposed method.
Figure 15 shows a new way to compare the performance of the two algorithms. Therefore in
Figure 15a, the power difference between the FOCV and the proposed method is shown to indicated positive and negative regions. The positive region is when the difference in power is higher than zero, which indicates that the power produced using the proposed method is higher than the power produced using the FOCV. The negative region is when the difference in power is less than zero and indicates that the FOCV performs better. The comparison with P&O is shown in
Figure 15b.
5.2. Variation of Both Irradiance and Temperature
This case simulates a more realistic situation, where the temperature and irradiance vary. The output power, in this case, is shown in
Figure 16. It can be seen that the proposed method exhibits faster dynamic speed and higher accuracy than the other methods. The varying temperature greatly degrades the performance of the FOCV since the open-circuit voltage depends on it. P&O is also affected by the temperature variation. However, since the P&O algorithm is independent of the panel information, the accuracy is higher than the FOCV method. The proposed method is faster during transitions from one irradiance level to another.
5.3. Energy Analysis
It can be seen from
Figure 14,
Figure 15,
Figure 16 and
Figure 17 that the proposed converter outperforms the other tracking algorithms. The overall comparison can be determined using the average of the difference between the two tracking methods. In the case of fixed temperature, the extracted energy by the proposed method is about 4.27 kW/y higher than the FOCV and 5.1 kWh/y higher than P&O. In the case of variable temperature, the energy extracted by the proposed method is 12.1 kWh/y higher than the extracted energy by the FOCV and 8.9 kWh/y higher than the extracted energy by P&O. Note that the previous analysis is per panel; the difference in energy extraction would be much higher for a large PV system. The performance comparison is summarized in
Table 2. The energy and efficiency of the proposed are higher than the other methods, with better transitioning time and stability.