Next Article in Journal
Power Loss and Electrothermal Characterization of Hybrid Power Integrated Modules for Industrial Servo Motor Drives
Previous Article in Journal
Performance Assessment of a Municipal Solid Waste Gasification and Power Generation System Integrated with Absorption Heat Pump Drying
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Analytical Computation of the Maximum Power Point of Solar Cells Using Perturbation Theory

by
José G. Tirado-Serrato
1,
Alfredo Sanchez Garcia
2,* and
Serguei Maximov
1
1
Programa de Graduados e Investigación en Ingeniería Eléctrica, Instituto Tecnológico de Morelia, Tecnológico Nacional de Mexico, Campus Morelia, Avenida Tecnológico No. 1500, Lomas de Santiaguito, Morelia 58120, Michoacán, Mexico
2
Sustainable Energy Technology, SINTEF AS, 7465 Trondheim, Norway
*
Author to whom correspondence should be addressed.
Energies 2024, 17(23), 6035; https://doi.org/10.3390/en17236035
Submission received: 19 October 2024 / Revised: 20 November 2024 / Accepted: 28 November 2024 / Published: 30 November 2024
(This article belongs to the Section A2: Solar Energy and Photovoltaic Systems)

Abstract

:
To compute the maximum power point (MPP) from physical parameters of the single-diode model (SDM), it is necessary to solve a transcendental equation using numerical methods. This is computationally expensive and can lead to divergence problems. An alternative is to develop analytical approximations which can be accurate enough for engineering problems and simpler to use. Therefore, this paper presents approximations for computing the MPP of single-junction solar cells. Two special cases are considered: (i) SDM with only series resistance, and (ii) SDM with only shunt resistance. Power series closed-form expressions for the MPP are obtained using perturbation theory and the Lagrange inversion theorem. Validation of the formulas is performed using experimental data from six different technologies obtained from the NREL database and comparing the results with the numerical solution of the SDM and three approximations from the literature. The results show an absolute percentage error (APE) of less than 0.035% with respect to the real MPP measurements. In cases with limited computational resources, this value could be further improved by using a higher- or lower-order power-series approximation.

1. Introduction

Shockley’s diode equation describes the behavior of the electrical variables of an ideal solar cell. In order to include the power losses experienced by an operating solar cell, the resistance of the series R s and the shunt R p need to be included through Kirchhoff’s current law (KCL) and Ohm’s laws [1]. The resulting model is commonly referred to as the five-parameter single-diode model (SDM) [2]. In this case, an implicit equation arises when calculating the current–voltage ( I V ) characteristic, which is necessary to solve numerically. Banwell et al. showed [3] that an explicit expression for the current in terms of the voltage that includes R s can be obtained with the Lambert W -function. A version of this expression that also includes R p was developed in [4]. In [5], formulas were presented for computing the voltage in terms of the current and explicit expressions for short-circuit current I sc and open-circuit voltage V oc .
The maximum power point (MPP) voltage V mpp and current I mpp are the pair of points of an I V curve for which the power defined by P = V I is at its maximum. The MPP is of interest because commercial solar panels are optimized to operate at this point for better efficiency [6]. Furthermore, the MPP used as a simulation tool for predicting generation power [7], estimating the MPP in real time using low-cost hardware [8], or calculating the power density for optimal flow studies [9], to mention only a few applications. The MPP computation problem has been addressed from different approaches. In [10], the Lagrange multiplier method was used, employing an objective function to maximize the power within the constraint of satisfying the I V equation of the photovoltaic (PV) module. A similar methodology was used in [11], with the addition of an objective function to maximize the rectangle inside the I V characteristic curve and study of the double diode model. In [12], the objective was to maximize the power output using the tilt angle of the PV module given an arbitrary location, which was accomplished through the use of neural networks. Using measurements of the slope of the I V curve around points I sc and V oc , ref. [13] constructed a parallelogram by which the MPP was computed using Lagrangian interpolation. Through the use of Thevenin and Norton equivalent circuits, ref. [14] approximated a linearized version of the I V curve of the PV module around the MPP in order to relate the parameters of the equivalent circuits to a variety of weather conditions. On the other hand, a significant problem in PV cell modeling consist of finding an explicit expression for the MPP in terms of the physical parameters of the SDM. Numerical method can be employed to easily determine the MPP with arbitrary accuracy; however, these methods are computationally expensive and may have divergence problems if the initial value is relatively far from the exact value of the solution. Therefore, it is worth finding analytical expressions for I mpp and V mpp with an acceptable level of accuracy. These should allow for quick and easy calculation of the MPP.
Obtaining analytical expressions is based on simplifications that reduce the implicit transcendental equation to an analytically solvable one. The previous literature has addressed this calculation from different approaches. The problem of including only R s was first tackled in [15] provided an explicit expression describing the relationship between voltage and current. From this expression, they proceeded to derive formulas for calculating V mpp and I mpp . Subsequently, expressions for MPP voltage and current were obtained in [8] through analytical methods based on the Mean Value Theorem (MVT) applied to the implicit current equation. Building on this, in [16] the Lambert W-function was employed to refine the PV module’s current equation. It was observed that the arguments of the Lambert W-function take on small values, allowing for an approximation in which these terms are replaced by their arguments to simplify the expressions for V mpp and I mpp . The resulting equations can be analytically solved using the Lambert W-function, providing a more precise approach. The problem of solving the transcendental equation for the MPP becomes more complex when R p is included in the formulation. This problem was initially addressed in [17], where the problem was reduced to a quadratic equation for I mpp by employing some simplifications (such as V mpp / R p 0 and d V d I / R p 0 ) and using a linear approximation to solve the problem analytically. In [7,18], the exact solution of the MPP for the ideal case was presented in terms of the Lambert W -function. The parameters R p and R s were included using the KCL. An alternative approach for analytical computation of the MPP was presented in [19], where an explicit model with parameters associated with the SDM was taken as a reference. Returning to the idea of applying the MVT to the I V curve, ref. [20] applied the MV to the ideal SDM to find an approximation for the MPP, which was dependent on the natural logarithm and included R s and R p by viewing the circuit as a two-port network. Recently, the exact solution to the problem of applying the MVT to MPP computation was finally found in [21].
It is well known that the accuracy of MPP calculation is strongly dependent on the method used to calculate the SDM parameters. In [22], the authors found that with available I V curve measurements, the most accurate method for representing curve is presented in [23], which considers the five-parameter SDM. When data sheet information is available, the most accurate approach is presented in [24,25], which uses the so-called simplified SDM. This model neglects either R p or R s depending on the value of the series-parallel coefficient (SPR) parameter. In the case of SPR > 1 , only R s is considered and R p ; otherwise if SPR < 1 , then R s = 0 , and a finite value of R p is taken. In the current literature there is no approximation for the latter case, only the model where R s or a combination of R s and R p are considered; nevertheless, the expressions obtained in this way lack accuracy due to the multiple simplifications.
In this work, analytical approximations of these two special cases are developed using perturbation theory and the Lagrange inversion theorem (LIT). For the case in which only R s is considered, an implicit equation for I mpp obtained from the SDM is converted into the dimensionless form u ( a , γ ) , where a and γ are dimensionless parameters. From here, we note that for small values of a (which correspond to small values of R s ), it is possible to obtain an analytical closed-form expression for u. Using perturbation theory, it can be argued that this approximation is close to the exact solution, at least at a small distance ξ . Expanding the power series around ξ  = 0, an expression for the exact form is found using the LIT. For the other case, where R p is considered, a similar methodology is followed, using the V mpp estimate of the ideal case as an initial approximation. The expressions are validated using the set of six I V curves measured by the NREL and presented in [26], where they are found to have an error of less than 0.035%.

2. Problem Description

Assuming nondegenerate conditions, the total current I produced by a solar cell follows Shockley’s diode equation [27]. When series and shunt resistance are considered, this equation becomes
I = I L I 0 exp α ( V + I R s ) 1 I R s + V R p ,
where
α = q A N s k B T ,
with N s being the number of cells connected in series, T the module temperature, k B Boltzmann’s constant, q the electron charge, A the ideality factor of the diode, I 0 the recombination current, I L the photogeneration current, R s the series resistance, and  R p the shunt resistance. From Equation (1), the implicit solutions for the current in terms of the voltage (and vice versa) follow [5]
I = I L + I 0 R p V R p + R s W F I α R s ,
V = I L + I 0 R p I ( R s + R p ) 1 α W F V ,
with
F I = α I 0 R p R s R p + R s exp α R p V + I L + I 0 R s R p + R s ,
F V = α I 0 R p exp I L + I 0 I R p α .
At the maximum power point (MPP), it follows that d P d V = d P d I = 0 . The maximum power point voltage V mpp and current I mpp follow the implicit forms [21]
V mpp + R p V mpp R p + R s W F V mpp + 1 = W F V mpp α + R s V mpp I L + I 0 R s R p ( R p + R s ) ,
I mpp R p W F I mpp + 1 + R s = I L + I 0 R p I mpp ( R p + R s ) 1 α W F I mpp ,
where W(x) is Lambert’s W -function, defined by x = W ( x e x ) , and  F V mpp and F I mpp are Equations (4) and (5) evaluated at V = V mpp and I = I mpp , respectively. Equations (6) and (7) are transcendental equations that cannot be solved analytically. Numerical methods can then be employed; however, these may present divergence problems and can be computationally expensive. In the following section, analytical approximations are presented to avoid these problems.

3. Mathematical Background

This section presents two mathematical tools employed in the present work to find the analytical approximations of the MPP: perturbation theory and the Lagrange inversion theorem (LIT).
Let x = x 0 be the exact root of the equation f ( x ) = 0 . Consider the problem of finding the roots of a transcendental equation in the form
h ( x , ε ) = f ( x ) + ε g ( x ) = 0 ,
where ε is a small parameter satisfying | ε | < < 1 . Following perturbation theory, the solution to Equation (8) can be expressed as a power series in ε of the following form:
x = n = 0 x n ε n .
Expanding to the first order in ε gives
x = x 0 + ε x 1 + O ( ε 2 ) ,
where x 1 is the first-order coefficient of the power series. Standard perturbation theory requires substituting Equation (10) in Equation (8) and collecting terms of equal powers of ε . This results in a system of algebraic equations that can be solved analytically. However, in this work we explore a different approach. First, we expand h ( x , ε ) via Taylor series expansion around x 0 to provide
h ( x , ε ) = n = 0 h ( n ) ( x 0 , ε ) n ! ( x x 0 ) n .
Substituting Equation (10) into Equation (11) and simplifying yields
h ( x , ε ) = n = 0 h ( n ) ( x 0 , ε ) n ! ( x 0 + ε x 1 x 0 ) n = n = 0 h ( n ) ( x 0 , ε ) n ! ( ε x 1 ) n = n = 0 h ( n ) ( x 0 , ε ) ε n n ! x 1 n .
Separating the zeroth-order term from the series and denoting ζ n : = h ( n ) ( x 0 ) ε n n ! in Equation (12), it follows that
h ( x , ε ) = h ( x 0 , ε ) n = 1 ζ n x 1 n .
However, according to Equation (8), h ( x , ε ) = 0 ; therefore, it follows that
h ( x 0 , ε ) = n = 1 ζ n x 1 n .
From Equation (14), according to the Lagrange inversion theorem (LIT), if  ζ 1 0 , then there exists a function ψ such that [28]
x 1 = ψ 1 h ( x 0 ) = m = 1 b m ζ 1 , . . , ζ m h ( x 0 ) m ,
where the coefficients b m follow [29]
b m ζ 1 , . . , ζ m = 1 m ζ 1 m k 1 , k 2 , . . . ( 1 ) k 1 + k 2 + + k m · m ( m + 1 ) ( m 1 + k 1 + k 2 + + k m ) k 1 ! k 2 ! k m ! · ζ 2 ζ 1 k 1 · ζ 3 ζ 1 k 2 ζ m ζ 1 k m
with coefficients k i satisfying the constraints k 1 + 2 k 2 + 3 k 3 + = m 1 and k 1 , k 2 , . . . 0 . The first five b m are [30]
b 1 ( ζ 1 ) = 1 ζ 1 , b 2 ( ζ 1 ζ 2 ) = ζ 2 ζ 1 3 , b 3 ( ζ 1 , . . . , ζ 3 ) = 2 ζ 2 2 ζ 1 ζ 3 ζ 1 5 , b 4 ( ζ 1 , . . . , ζ 4 ) = 5 ζ 1 ζ 2 ζ 3 5 ζ 2 3 ζ 1 2 ζ 4 ζ 1 7 , b 5 ( ζ 1 , . . . , ζ 5 ) = 1 ζ 1 9 ( 14 ζ 2 4 21 ζ 1 ζ 3 ζ 2 2 + 6 ζ 1 2 ζ 4 ζ 2 + 3 ζ 1 2 ζ 3 2 ζ 1 3 ζ 5 ) .
The LIT allows us to find x 1 through Equation (15). The root of h ( x , ε ) , x, is then calculated by inserting Equation (15) in Equation (10). The radius of convergence of the reverse series in Equation (15) has the same domain as the original series in Equation (11), according to [31,32].

4. Computation of the MPP Assuming Only Series Resistance

From the general form of the diode equation provided by Equation (1), the case in which only R s is considered can be approximated by taking the limit R p . This yields
I = I L I 0 exp ( V + I R s ) α .
Solving for V gives
V = ln I L I I 0 1 α I R s ,
meaning that the output power of the module P = V · I is provided by
P = I ln I L I I 0 1 α I 2 R s .
At the MPP, the derivative of the power with respect to the current is zero, i.e.,
V + I d V d I | I = I mpp = 0 ,
with
d V d I = 1 α ( I L I ) R s .
Substituting Equations (19) and (22) into Equation (21) and combining and simplifying the like terms yields a transcendental equation for I mpp of the following form:
ln I L I mpp I 0 I mpp 2 R s α + 1 I L I mpp = 0 .
Now, defining the variables a, γ , and u as
a := 2 R s I L α ,
γ := 1 ln I 0 I L ,
u := 1 I mpp I L ,
Equation (23) can be rewritten as
γ = 1 u ln u + a ( 1 u ) ,
which makes the problem dimensionless.
  • Parameter Bounds
Because I mpp > 0 , it follows from Equation (26) that u < 1 . Furthermore, u > 0 must be satisfied to ensure that the solution of Equation (27) lies in the real plane; therefore, the value of u must be bounded by 0 < u < 1 . This is expected, as the value of I mpp that solves Equation (23) is in the range of 0 < I mpp < I L given that I L is the maximum value taken by the current. This value depends on the PV module technology, incoming irradiance, and cell temperature. The analysis in this section shows that the value of u that solves Equation (27) has a well-defined boundary for constant values. Regarding a, because  I L , I 0 , R s , V th , A, and  N s are physical quantities with positive value, it follows that a > 0 for any standard PV module. Finally, because  I 0 I L for any practical scenario, Equation (25) implies that γ > 1 .

4.1. Initial Approximation of the MPP

In the particular case where either I L or R s is small, a in Equation (24) is also small. In this case, γ a and u should be near zero in order to conserve the equality in Equation (27). This occurs in the case of low illumination, resulting in small I L , or when assuming a reasonable quality of the solar cells in the module such that R s is small. Therefore, the significant terms that contribute the most to the solution of Equation (27) are the first two on the right-hand side. In this special case, it is possible to neglect the a u term, causing a transcendentally small error, as follows:
γ a 1 u ln u .
Applying exponentiation to both sides of Equation (28) gives
exp ( γ a ) = exp 1 u ln u = exp 1 u exp ln u = exp 1 u 1 u ,
and using the definition of W finally results in
u 0 = 1 W 0 exp γ a ,
where W 0 ( x ) is the principal branch of Lambert’s W -function. Thus, the MPP is calculated by substituting Equation (29) into Equation (26) for I mpp , then in turn into Equation (19) for V mpp . Returning to the original variables, Equation (29) is represented as follows:
I mpp = 1 1 W 0 I L I 0 exp 1 2 R s I L A N s V t I L .
The expression in Equation (30) was initially presented in [16] for the study of a PV cell. In order for the approximation Equation (29) to satisfy the bounds of u, it must hold that
W 0 exp γ a > 1 .
The main branch of the Lambert W -function has the special value W 0 ( exp 1 ) = 1 ; as it is monotonically increasing, Equation (29) must satisfy exp γ a > exp 1 . Therefore, in order for Equation (31) to be true, it must be the case that
γ a > 1 .
This provides the restriction that the parameters γ and a must be satisfied in order for Equation (29) to hold. Furthermore, Equation (32) provides a limit for the values that R s can take, given by
R s ( lim ) < log ( I L I 0 ) 2 I L α .

4.2. Perturbation Theory

Equation (27) can be rewritten as
γ a 1 u + log ( u ) a u = 0 .
Because a is assumed to be small, note that Equation (34) has the same form as the general expression in Equation (8), with  f ( u ) = γ a 1 u + log ( u ) , ε = a , and  g ( u ) = u . The root of f ( u ) is known to be u 0 , as provided by Equation (29). Then, following perturbation theory, the roots of Equation (34) follow
u = u 0 + a u 1 = u 0 1 + ξ ,
where ξ = a ( u 1 u 0 ) is a new variable introduced for convenience. Substituting Equation (35) in Equation (27) yields
ln u 0 1 + ξ 1 u 0 1 + ξ + a u 0 1 + ξ = a γ .
Taylor series expansion for ξ near zero results in
n = 1 ( 1 ) n + 1 ξ n n 1 u 0 n = 0 1 n ξ n + a u 0 1 + ξ + ln u 0 = a γ ,
which can be represented as
γ 1 u 0 + a ( u 0 1 ) + ln u 0 = n = 1 ( 1 ) n 1 n + 1 u 0 + δ n 1 a u 0 ξ n ,
where δ n m is the Kronecker’s delta. Denoting the left-hand side of Equation (38) as
η = γ 1 u 0 + a ( u 0 1 ) + ln u 0
and the coefficients of the summation of the right-hand side as
σ n = ( 1 ) n 1 n + 1 u 0 + δ n , 1 a u 0 ,
Equation (38) can be written as
η = n = 1 σ n ξ n ,
which has the same form as Equation (14). Then, applying the LIT, it follows that
ξ = m = 1 b m σ 1 , . . , σ m η m .
Notice that η corresponds to Equation (27) evaluated at u = u 0 . Therefore, the closer the initial approximation u 0 is to the exact solution, the smaller η becomes.
The value of ξ is calculated with Equation (42). From here, the value of u can be calculated by substituting u 1 = u 0 ( ξ a ) into Equation (35), with  u 0 provided by Equation (29). The MPP is finally calculated by evaluating Equation (26) with the obtained value of u to estimate I mpp , which is then used in Equation (19) to calculate V mpp .

4.3. Statistical Data of γ and a Parameters

To illustrate the applicability of the approximation provided by Equations (29) and (35), it is important to know the ranges for the values of a and γ for different types of PV technologies and atmospheric conditions. From the I V data published in [26], PV metrics ( V oc , I sc ,…) were calculated using the the analytical method described in [24,25]. This method classifies a module according to the series/parallel ratio (SPR) metric, provided by
SPR = 1 I mpp I sc exp V oc V mpp I sc I mpp I mpp V mpp .
If SPR  > 1 , then only R s is considered; if SPR  < 1 , then only R p is taken into account. The obtained values were used to evaluate Equations (24)–(26).
It was found that 40,891 experimental measurements resulted in SPR  > 1 . These were used to calculate the a and γ values. Table 1 shows a statistical summary of γ a , where it can be observed that Equation (32) is valid for four of the five studied modules. In the case of CIGS technology, Equation (32) is not fulfilled in only five of the 8331 measurements. Analyzing the case where γ a obtains its minimum value, it is found that γ = 7.8155 and a = 7.5364 ; therefore u 0 = 1.484 , indicating that I mpp > I L , which is physically impossible. However, with the use of Equation (35), it is found that ξ = 0.7215 ; using Equation (35) we then have u = 0.4133 and the value provided by the numerical solution of Equation (27) is u num = 0.4076 , with an Absolute Percentage Error of 1.41%, showing that the present methodology performs adequately even in the limiting cases. Furthermore, it is found that 173,822 of the measurements have SPR < 1, emphasizing the importance of finding an approximation for the special case where the SDM model considers only R p .

5. Computation of the MPP Assuming Only Shunt Resistance

In the case where only parallel resistance is considered, the general form of the diode equation in Equation (1) reduces to
I = I L I 0 exp α V V R p .
The electrical power is then provided by
P = V I L V I 0 exp α V V 2 R p
and its derivative with respect to the voltage by
d P d V = I L I 0 exp α V V α + 1 2 V R p .
At the MPP, it follows that d P d V = 0 , which yields
I L I 0 exp α V mpp V mpp α + 1 2 V mpp R p = 0 .
The solution for the ideal case ( R p ) of Equation (47) is provided by [18]
v 0 = 1 α W exp 1 I L I 0 1 .
The method presented in Section 3 can be used to find a solution to Equation (47). Because  R p takes large values for conventional modules, it follows that 1 R p < < 1 . Note that Equation (47) follows the general form of Equation (8) with g ( V mpp ) = 2 V mpp and f ( V mpp ) = I 0 exp α V mpp V mpp α + 1 I L . The root of f ( V mpp ) is v 0 provided by Equation (48). Following perturbation theory, Equation (47) has a solution of the following form:
V mpp = v 0 + v 1 R p = v 0 ( 1 + β )
with β = v 1 R p v 0 . Replacing Equation (49) in Equation (47) yields
I L = Λ exp α v 0 β v 0 ( 1 + β ) α + 1 + 2 R p v 0 ( 1 + β ) ,
where Λ : = I 0 exp α v 0 . Expanding the exponential function via Taylor series expansion for α v 0 β near zero results in
I L = Λ i = 0 α v 0 β i i ! v 0 α + 1 + α v 0 β + 2 R p v 0 ( 1 + β ) .
The summation can be separated as
I L = Λ i = 0 α v 0 β i i ! v 0 α + 1 + i = 0 α v 0 β i i ! α v 0 β + 2 R p v 0 ( 1 + β ) ,
and extracting the first term of the first summation and multiplying the factor α v 0 β in the second provides
I L = Λ ( v 0 α + 1 ) + 2 R p v 0 ( 1 + β ) + Λ i = 1 α v 0 β i i ! v 0 α + 1 + i = 0 α v 0 β i + 1 i ! .
Through an index shift in the second summation and using the factorial property ( i 1 ) ! = i ! i , Equation (53) becomes
φ = Λ i = 1 α v 0 β i i ! v 0 α + 1 + i = 1 i α v 0 β i i ! + 2 v 0 β R p ,
where
φ : = I L 2 R p v 0 Λ v 0 α + 1 .
Grouping the summations and adding the term 2 v 0 β R p with the Kronecker’s delta function results in
φ = Λ i = 1 α v 0 β i i ! v 0 α + 1 + i + δ i , 1 2 R p v 0 .
Now, defining
λ i : = α v 0 i i ! Λ v 0 α + 1 + i + δ i , 1 2 Λ α R p ,
Equation (56) is represented as follows:
φ = i = 1 λ i β i .
Applying the LIT, it can be seen that β follows
β = m = 1 b m ( λ 1 , . . . , λ m ) φ m .
Finally, V mpp is computed by substituting Equation (59) in Equation (49). The latter expression is then used in Equation (44) to calculate I mpp .
Perturbation theory states that in order for Equation (59) to converge, it must be the case that
| φ | < 1 .
Substituting Equation (55) in Equation (60) and splitting the absolute value yields
1 < I L 2 R p v 0 Λ v 0 α + 1 < 1 ,
which with some algebraic manipulations can be rewritten as
1 + I L Λ v 0 α + 1 2 v 0 > 1 R p > I L 1 Λ v 0 α + 1 2 v 0 .
For the ideal case ( R p ), I L can be calculated as follows:
I L Λ ( v 0 α + 1 ) .
Substituting Equation (63) in Equation (62),
1 2 v 0 > 1 R p > 1 2 v 0 .
Because R p is strictly positive, only the left-hand side of Equation (64) is considered, resulting in
R p ( lim ) > 2 v 0 .
This results in a minimum for the value that R p can take, ensuring the convergence of Equation (58).

6. Validation

To validate the performance of the model derived in Equations (26) and (49), I V data from six different photovoltaic technologies were utilized. These data were obtained from the National Renewable Energy Laboratory (NREL) real-time photovoltaic solar resource testing database [33]. A list of the equipment utilized for collecting the I V data is provided in [26], with properties of these experimental measurements such as accuracy, range, and detailed uncertainty calculations provided in [34]. Table 2 summarizes the PV metrics and the simplified SDM parameters for the six studied PV technologies. The characteristic parameters were calculated using the explicit solutions presented in [24,25]. Figure 1 shows the IV curve and MPP for each module and analytical method. Figure 2 displays the base-10 logarithm of the relative error (in %) between the experimental measurements and the simplified SDM model of the current with respect to the normalized voltage. It can be observed that the error is small around the extremes of the curve ( V oc and I sc ) and in the MPP. This is because the methodology seeks to match the experimental curve IV with the theoretical curve at these points. This is sufficient for the methodology presented in this work, which seeks to obtain the highest accuracy in the vicinity of the MPP. Table 3 shows the absolute percentage error (APE) between the calculated and the measured V mpp , I mpp . The APE is provided by
APE = X X exp X exp × 100 ( % ) ,
where X exp represents the experimental measurements of the MPP and X represents calculation of the MPP for each technology using the presented methodology. In addition, three well-established methods from the literature are used to make comparisons, labeled “Batzelis” for the method presented in [35], “Wang” for [20], and “Tirado” for [21]. For the numerical solutions, these are obtained for the implicit Equations (27) and (47) using Matlab’s built-in fzero solver with Equations (35) and (49) as the respective initial values. According to the methodology used to calculate the SDM parameters, the expressions presented in the Section 4 are use for the modules “xSi12922”, “CdTe75669”, and “HIT05662”, while the expressions presented in Section 5 are used for the other three modules.
Of the three methods used for comparison, the one presented in [21] performs best in estimating the MPP of the xSi12922 and HIT05662 technologies, the method presented in [20] performs best for the aSiMicro and CdTe modules, and the method presented in [35] performs best for the CIGS and mSi0188 technologies. However, the smallest mean of APE for the six modules is achieved by the method from [21], with 0.418% and 0.406% for V mpp and I mpp , respectively, followed by [35] with 0.501% and 0.491%. The method with the largest mean APE is [20], with 1.011% for V mpp and 1.096% for I mpp . The best performance is obtained by the numerical solution and the method proposed in this work, both with similar accuracy of 7.077 × 10 3 % for V mpp and 6.195 × 10 3 % for I mpp , showing superior performance compared to the previously published methods. The reason for this is that the previous methods depend on the values of both resistors (series and shunt); thus, if one is neglected, the resulting expression is strongly affected and becomes significantly inaccurate.
Figure 3 and Figure 4 display the curves of V mpp and I mpp with respect to R p and R s (with the scale axis in base-10 logarithm) for the aSiMicro03036 and HIT05662 technologies, respectively. Here, the first five terms of Equations (59) and (42) are compared to the numerical approximation using Equations (47) and (27). The MPP value of the module is marked with a dot, while the limits for R p (from Equation (65)) and R s (from Equation (33)) are indicated with a vertical dashed line. In Figure 3, it can be observed that there is a good fit between the analytical and numerical approximation for values of R p > R p ( lim ) , and V mpp tends to the ideal case as R p continues to increase. On the other hand, the series diverges for values R p < R p ( lim ) ; however, the range covered by r p l is sufficient to satisfy practical cases. From Figure 4, it can be seen that the values calculated with the new analytical model are in good agreement with the numerical reference model for R s < R s ( lim ) . For R s > R s ( lim ) , the value of V mpp calculated with Eqution (42) appears to decrease, while I mpp increases with increasing R s , showing a clear divergence of the power series.
Figure 5 and Figure 6 compare the APE for the numerical and the analytical solutions as a function of R p and R s for the aSiMicro03036 and HIT05662 technologies and for different numbers of terms in the power-series of Equations (59) and (42). In Figure 6, it can be observed that the APE increases as R p decreases; taking one term of Equation (42), the approximation yields an APE of 25.98% for V mpp and 28.51% for I mpp at R p = R p ( lim ) = 298.19 Ω . Increasing R p up to the modulus value R p = 9532 Ω provides an APE of 1.51 × 10 2 % for both V mpp and I mpp , while in the limiting case R p = 1 × 10 5 Ω the APE is equal to 1.36 × 10 6 %. The APE decreases significantly when increasing the number of terms; estimating the performance for R p = R p ( lim ) = 298.19   Ω , an A P E of 0.719% is calculated by taking five terms, significantly improving the accuracy in this extreme case. Figure 5 shows that the APE increases as R s increases; this is because the approximation provided by Equation (42) performs better as R s decreases. It is also observed that in the vicinity of R s = R s ( lim ) , the APE decreases rapidly as R s decreases, from an A P E = 20.127 % for one term down to A P E = 0.295 % using five terms. For the case where R s = 0.2656   Ω , the APE ranges from 1.4148 × 10 4 % for one term to 3.9139 × 10 14 % for five terms, showing significantly improved accuracy of the approximation. Thus, if computational resources are limited, as is the case with real-time MPP estimation using low-cost hardware, it is possible to take the first terms of the series. If requirements are less stringent, it is possible to take as many terms as the level of accuracy requires.
A limitation of the methodology presented here is that it depends on the existence of a small parameter in the equations, introduced in the formulations presented in this work as η (provided by (39)) and φ (provided by (55)). Furthermore, the convergence of the power series depends strongly on how small these parameters are. However, using a large amount of data, we found that the values of these parameters fell within the established limits.

7. Conclusions

In this paper, we have derived formulas for calculating the maximum power point that can explicitly take into account the separate effects of series and shunt resistance. We have found approximations of the transcendental equations for MPP calculation in terms of simplified SDM parameters by employing perturbation theory and the Lagrange inversion theorem. These expressions were validated using the calculated and measured IV curves of six PV modules based on different technologies. The verification results reveal that the APE between the presented approximations and the experimental measurements is less than 0.035%, which is useful in most engineering applications. The formulas can be simplified or made more complex depending on the computational cost and accuracy needs of the specific application. In the future, we intend to extend the presented models to five-parameter SDM using the methods developed in this work.

Author Contributions

Conceptualization, J.G.T.-S. and S.M.; methodology, J.G.T.-S. and A.S.G.; software, J.G.T.-S.; validation, J.G.T.-S.; formal analysis, J.G.T.-S.; writing—original draft preparation, J.G.T.-S. and A.S.G.; writing—review and editing, J.G.T.-S. and A.S.G.; supervision, S.M. and A.S.G. All authors have read and agreed to the published version of the manuscript.

Funding

The authors gratefully acknowledge the financial support of the following Mexican institutions: CONAHCYT, SNI and PRODEP. This research was partially funded by the Norwegian Research Center for Sustainable Solar Cell Technology (FME SUSOLTECH, project number 275639/E20). The center is co-sponsored by the Research Council of Norway and its research and industry partners.

Data Availability Statement

No new data was produced in this work. The data employed for validating the models can be foun in Ref. [33].

Conflicts of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Nomenclature

Nomenclature for the equations describing the solar cell current–voltage characteristics, including symbols, their definitions, and respective units.
SymbolDefinitionUnits
IOutput current of the solar cellA
VOutput voltage of the solar cellV
POutput power of the solar cellW
I L Photogenerated currentA
I 0 Reverse saturation currentA
R s Series resistance Ω
R p Shunt resistance Ω
qElementary chargeeV
ADiode ideality factorDimensionless
N s Number of cells in seriesDimensionless
k B Boltzmann constanteV/K
TTemperatureK
W ( · ) Lambert W-functionDimensionless
I mpp Current at maximum power pointA
I sc Short-circuit currentA
V mpp Voltage at maximum power pointV
V oc Open-circuit voltageV
APEAbsolute percentage error%

References

  1. Femia, N.; Petrone, G.; Spagnuolo, G.; Vitelli, M. Power Electronics and Control Techniques for Maximum Energy Harvesting in Photovoltaic Systems; CRC Press: Boca Raton, FL, USA, 2012. [Google Scholar]
  2. Bana, S.; Saini, R. A mathematical modeling framework to evaluate the performance of single diode and double diode based SPV systems. Energy Rep. 2016, 2, 171–187. [Google Scholar] [CrossRef]
  3. Banwell, T.; Jayakumar, A. Exact analytical solution for current flow through diode with series resistance. Electron. Lett. 2000, 36, 291–292. [Google Scholar] [CrossRef]
  4. Ortiz-Conde, A.; Sanchez, F.J.G.; Muci, J. Exact analytical solutions of the forward non-ideal diode equation with series and shunt parasitic resistances. Solid-State Electron. 2000, 44, 1861–1864. [Google Scholar] [CrossRef]
  5. Jain, A.; Kapoor, A. Exact analytical solutions of the parameters of real solar cells using Lambert W-function. Sol. Energy Mater. Sol. Cells 2004, 81, 269–277. [Google Scholar] [CrossRef]
  6. Qi, B.; Wang, J. Fill factor in organic solar cells. Phys. Chem. Chem. Phys. 2013, 15, 8972–8982. [Google Scholar] [CrossRef] [PubMed]
  7. Farivar, G.; Asaei, B.; Mehrnami, S. An analytical solution for tracking photovoltaic module MPP. IEEE J. Photovoltaics 2013, 3, 1053–1061. [Google Scholar] [CrossRef]
  8. Rodriguez, C.; Amaratunga, G.A. Analytic solution to the photovoltaic maximum power point problem. IEEE Trans. Circuits Syst. I Regul. Pap. 2007, 54, 2054–2060. [Google Scholar] [CrossRef]
  9. Grisales-Noreña, L.F.; Montoya, O.D.; Cortés-Caicedo, B.; Zishan, F.; Rosero-García, J. Optimal power dispatch of pv generators in ac distribution networks by considering solar, environmental, and power demand conditions from Colombia. Mathematics 2023, 11, 484. [Google Scholar] [CrossRef]
  10. Nelatury, S.R. A maximum power point algorithm using the Lagrange method. J. Power Sources 2013, 234, 119–128. [Google Scholar] [CrossRef]
  11. Louzazni, M.; Khouya, A.; Al-Dahidi, S.; Mussetta, M.; Amechnoue, K. Analytical optimization of photovoltaic output with Lagrange Multiplier Method. Optik 2019, 199, 163379. [Google Scholar] [CrossRef]
  12. Chatterjee, A.; Keyhani, A. Neural network estimation of microgrid maximum solar power. IEEE Trans. Smart Grid 2012, 3, 1860–1866. [Google Scholar] [CrossRef]
  13. Kumar, G.; Panchal, A.K. Geometrical prediction of maximum power point for photovoltaics. Appl. Energy 2014, 119, 237–245. [Google Scholar] [CrossRef]
  14. Li, S. Linear equivalent models at the maximum power point based on variable weather parameters for photovoltaic cell. Appl. Energy 2016, 182, 94–104. [Google Scholar] [CrossRef]
  15. Singal, C. Analytical expression for the series-resistance-dependent maximum power point and curve factor for solar cells. Sol. Cells 1981, 3, 163–177. [Google Scholar] [CrossRef]
  16. Garcia, A.S.; Strandberg, R. Analytical modeling of the maximum power point with series resistance. Appl. Sci. 2021, 11, 10952. [Google Scholar] [CrossRef]
  17. Alghuwainem, S. A close-form solution for the maximum-power operating point of a solar cell array. Sol. Energy Mater. Sol. Cells 1997, 46, 249–257. [Google Scholar] [CrossRef]
  18. Farivar, G.; Asaei, B.; Rezaei, M.A. A novel analytical solution for the PV-arrays maximum power point tracking problem. In Proceedings of the 2010 IEEE International Conference on Power and Energy, Kuala Lumpur, Malaysia, 29 November–1 December 2010; pp. 917–922. [Google Scholar] [CrossRef]
  19. Das, A.K. An explicit J–V model of a solar cell using equivalent rational function form for simple estimation of maximum power point voltage. Sol. Energy 2013, 98, 400–403. [Google Scholar] [CrossRef]
  20. Wang, S.N.; Chen, Q.G.; Gao, W.G. An improved analytical solution for MPP parameters of photovoltaic cells. Sol. Energy 2018, 174, 848–854. [Google Scholar] [CrossRef]
  21. Tirado-Serrato, J.G. Exact Solution to the Mean Value Theorem Applied to the Maximum Power Point Estimation. IEEE J. Photovoltaics 2023, 13, 750–755. [Google Scholar] [CrossRef]
  22. Batzelis, E. Non-iterative methods for the extraction of the single-diode model parameters of photovoltaic modules: A review and comparative assessment. Energies 2019, 12, 358. [Google Scholar] [CrossRef]
  23. Toledo, F.; Blanes, J.M. Geometric properties of the single-diode photovoltaic model and a new very simple method for parameters extraction. Renew. Energy 2014, 72, 125–133. [Google Scholar] [CrossRef]
  24. Cannizzaro, S.; Di Piazza, M.; Luna, M.; Vitale, G. Generalized classification of PV modules by simplified single-diode models. In Proceedings of the 2014 IEEE 23rd International Symposium on Industrial Electronics (ISIE), Istanbul, Turkey, 1–4 June 2014; pp. 2266–2273. [Google Scholar] [CrossRef]
  25. Cannizzaro, S.; Di Piazza, M.; Luna, M.; Vitale, G. PVID: An interactive Matlab application for parameter identification of complete and simplified single-diode PV models. In Proceedings of the 2014 IEEE 15th Workshop on Control and Modeling for Power Electronics (COMPEL), Santander, Spain, 22–25 June 2014; pp. 1–7. [Google Scholar] [CrossRef]
  26. Marion, B.; Anderberg, A.; Deline, C.; Del Cueto, J.; Muller, M.; Perrin, G.; Rodriguez, J.; Rummel, S.; Silverman, T.J.; Vignola, F.; et al. New data set for validating PV module performance models. In Proceedings of the 2014 IEEE 40th Photovoltaic Specialist Conference (PVSC), Denver, CO, USA, 8–13 June 2014; pp. 1362–1366. [Google Scholar] [CrossRef]
  27. Shockley, W. The Theory of p-n Junctions in Semiconductors and p-n Junction Transistors. Bell Syst. Tech. J. 1949, 28, 435–489. [Google Scholar] [CrossRef]
  28. Abramowitz, M.; Stegun, I.A. Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables; US Government Printing Office: Washington, DC, USA, 1964; Volume 55. [Google Scholar]
  29. Morse, P.M.; Feshbach, H. Methods of Theoretical Physics; McGraw-Hill Book Company: New York, NY, USA, 1953. [Google Scholar]
  30. Dwight, H.B. Tables of Integrals and Other Mathematical Data; The Macmillan Company: New York, NY, USA, 1934. [Google Scholar]
  31. Harkness, J.; Morley, F. A Treatise on the Theory of Functions; Macmillan and Company: Troon, Scotland, 1893. [Google Scholar]
  32. Van Orstrand, C. XXXVII. Reversion of power series. Lond. Edinb. Dublin Philos. Mag. J. Sci. 1910, 19, 366–376. [Google Scholar] [CrossRef]
  33. Marion, B.; Anderberg, A.; Deline, C.; Muller, M.; Perrin, G.; Rodriguez, J.; Rummel, S.; Silverman, T.; Vignola, F.; Barkaszi, S. Data for Validating Models for PV Module Performance; National Renewable Energy Lab. (NREL): Golden, CO, USA, 2021. [Google Scholar] [CrossRef]
  34. Marion, W.; Anderberg, A.; Deline, C.; Glick, S.; Muller, M.; Perrin, G.; Rodriguez, J.; Rummel, S.; Terwilliger, K.; Silverman, T. User’s Manual for Data for Validating Models for PV Module Performance; Technical Report; National Renewable Energy Lab. (NREL): Golden, CO, USA, 2014. [Google Scholar]
  35. Batzelis, E.I.; Kampitsis, G.E.; Papathanassiou, S.A.; Manias, S.N. Direct MPP calculation in terms of the single-diode PV model parameters. IEEE Trans. Energy Convers. 2014, 30, 226–236. [Google Scholar] [CrossRef]
Figure 1. IV curves and MPP approximations of the six PV technologies.
Figure 1. IV curves and MPP approximations of the six PV technologies.
Energies 17 06035 g001
Figure 2. Relative error for the IV curves of the six PV technologies.
Figure 2. Relative error for the IV curves of the six PV technologies.
Energies 17 06035 g002
Figure 3. Comparison between numerical and analytical approximation of V mpp and I mpp with varying R s and using five terms of the power-series in Equation (59).
Figure 3. Comparison between numerical and analytical approximation of V mpp and I mpp with varying R s and using five terms of the power-series in Equation (59).
Energies 17 06035 g003
Figure 4. Comparison between numerical and analytical approximation of V mpp and I mpp with varying R s and using five terms of the power-series in Equation (42).
Figure 4. Comparison between numerical and analytical approximation of V mpp and I mpp with varying R s and using five terms of the power-series in Equation (42).
Energies 17 06035 g004
Figure 5. APE comparison between numerical and analytical approximation of V mpp and I mpp with varying R s for different numbers of power-series terms in Equation (42).
Figure 5. APE comparison between numerical and analytical approximation of V mpp and I mpp with varying R s for different numbers of power-series terms in Equation (42).
Energies 17 06035 g005
Figure 6. APE comparison between numerical and analytical approximation of V mpp and I mpp with varying R p for different numbers of power-series terms in Equation (59).
Figure 6. APE comparison between numerical and analytical approximation of V mpp and I mpp with varying R p for different numbers of power-series terms in Equation (59).
Energies 17 06035 g006
Table 1. Statistical parameters a and γ of the six PV technologies presented in [26].
Table 1. Statistical parameters a and γ of the six PV technologies presented in [26].
ParametersCIGS8-001CdTe75638HIT05667aSiTriple28324aSiMicro03036mSi0188
γ a γ a γ a γ a γ a γ a
Max8.913912.140942.40927.84507.570224.1751
Min0.27913.17445.92453.82431.77507.5282
Mean3.13978.916719.84686.51697.012014.4199
Std0.54890.47061.48040.51490.77110.5606
SPR >1833110,79921,235471555717
SPR < 134,81531,44922,03642,23443,28837,410
Table 2. PV metrics and SDM parameters.
Table 2. PV metrics and SDM parameters.
ModelDate I sc ( A ) V oc ( V ) I mpp ( A ) V mpp ( V ) I L ( A ) I 0 ( A ) α R s ( Ω ) R p ( Ω )
aSiMicro0303614 Aug 2012—10:30:100.728200.600.599148.020.728 6.486 × 10 4 3.487 × 10 2 -9532
mSi018811 Jan 2013—12:00:022.50723.202.32519.302.507 3.401 × 10 7 6.816 × 10 1 -3441
CIGS1-00114 Aug 2012—10:30:382.20939.591.94831.012.209 8.679 × 10 5 2.560 × 10 1 -1790
xSi1292211 Jan 2013—12:00:094.69023.154.34018.854.690 7.856 × 10 7 6.738 × 10 1 0.104-
CdTe7566914 Aug 2012—10:30:531.03981.380.90060.341.039 9.052 × 10 5 1.149 × 10 1 3.821-
HIT0566214 Aug 2012—10:30:054.89047.264.53038.454.890 3.756 × 10 7 3.466 × 10 1 0.266-
Table 3. APE (%) between the analytical approximations and experimental measurements of V mpp and I mpp .
Table 3. APE (%) between the analytical approximations and experimental measurements of V mpp and I mpp .
ModelProposed ModelTiradoBatzelisWangNumerical
V mpp I mpp V mpp I mpp V mpp I mpp V mpp I mpp V mpp I mpp
aSiMicro03036 2.03 × 10 7 2.02 × 10 7 8.65 × 10 1 7.76 × 10 1 7.27 × 10 1 7.40 × 10 1 6.01 × 10 1 7.01 × 10 1 1.88 × 10 7 2.17 × 10 7
mSi0188 7.31 × 10 6 4.74 × 10 6 1.23 × 10 1 1.23 × 10 1 3.18 × 10 2 3.19 × 10 2 1.40 × 10 0 1.53 × 10 0 7.31 × 10 6 4.74 × 10 6
CIGS1-001 3.36 × 10 6 2.92 × 10 6 3.74 × 10 1 3.75 × 10 1 1.79 × 10 1 1.80 × 10 1 1.96 × 10 0 2.12 × 10 0 3.36 × 10 6 2.92 × 10 6
xSi12922 2.63 × 10 3 1.71 × 10 3 1.71 × 10 1 1.73 × 10 1 3.30 × 10 1 3.23 × 10 1 1.11 × 10 0 1.19 × 10 0 2.63 × 10 3 1.71 × 10 3
CdTe75669 3.47 × 10 2 3.21 × 10 2 8.12 × 10 1 8.22 × 10 1 1.32 × 10 0 1.27 × 10 0 1.20 × 10 1 1.10 × 10 1 3.47 × 10 2 3.21 × 10 2
HIT05662 5.11 × 10 3 3.36 × 10 3 1.67 × 10 1 1.68 × 10 1 4.17 × 10 1 4.06 × 10 1 8.76 × 10 1 9.25 × 10 1 5.11 × 10 3 3.36 × 10 3
MAPE 7.08 × 10 3 6.20 × 10 3 4.18 × 10 1 4.06 × 10 1 5.01 × 10 1 4.91 × 10 1 1.01 × 10 0 1.10 × 10 0 7.08 × 10 3 6.20 × 10 3
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Tirado-Serrato, J.G.; Garcia, A.S.; Maximov, S. Analytical Computation of the Maximum Power Point of Solar Cells Using Perturbation Theory. Energies 2024, 17, 6035. https://doi.org/10.3390/en17236035

AMA Style

Tirado-Serrato JG, Garcia AS, Maximov S. Analytical Computation of the Maximum Power Point of Solar Cells Using Perturbation Theory. Energies. 2024; 17(23):6035. https://doi.org/10.3390/en17236035

Chicago/Turabian Style

Tirado-Serrato, José G., Alfredo Sanchez Garcia, and Serguei Maximov. 2024. "Analytical Computation of the Maximum Power Point of Solar Cells Using Perturbation Theory" Energies 17, no. 23: 6035. https://doi.org/10.3390/en17236035

APA Style

Tirado-Serrato, J. G., Garcia, A. S., & Maximov, S. (2024). Analytical Computation of the Maximum Power Point of Solar Cells Using Perturbation Theory. Energies, 17(23), 6035. https://doi.org/10.3390/en17236035

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop