TEG Maximum Power Point Tracking Using an ... - Science Direct

Available online at www.sciencedirect.com

ScienceDirect Energy Procedia 105 (2017) 14 – 27

The 8th International Conference on Applied Energy – ICAE2016

TEG Maximum Power Point Tracking using an Adaptive Duty Cycle Scaling Algorithm Trevor Hocksun Kwan, Xiaofeng Wu* School of AMME, University of Sydney, Darlington, 2006, Sydney, Australia

Abstract The thermoelectric generator (TEG) is a clean and noiseless renewable electrical power source that requires no moving parts. Unfortunately, the practicality of TEGs is currently limited by its typical low conversion efficiencies. Subsequently, researchers have taken many approaches to improve the efficiency of the TEG. One of such approaches is the utilization of maximum power point tracking (MPPT) techniques. MPPT techniques are popularly used in literature for maximizing the power that is extracted from solar panels. Such techniques can be reused for the TEG scenario because TEGs also have I-V and P-V characteristics that follow the same principles as that of solar panels. This paper presents a “Lock-On Mechanism” MPPT algorithm and applies it specifically to the TEG application. In comparison to conventional fixed step based MPPT algorithms, the proposed algorithm improves the MPP tracking performance by adaptively scaling the DC-DC converter duty cycle whenever the MPP is located. In doing so, the steady state oscillations become negligibly small thus be considered eliminated and a smooth steady state MPP response is achieved. Simulation and experimental results prove that the proposed algorithm is fast and stable in comparison to the conventional fixed step hill climbing algorithm. © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license © 2016 The Authors. Published by Elsevier Ltd. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Selection and/or peer-review under responsibility of ICAE Peer-review under responsibility of the scientific committee of the 8th International Conference on Applied Energy.

Keywords: PWM, MPPT, SEPIC Converter, TEG

1. Introduction The thermoelectric generator (TEG) is a semiconductor device which generates electric power via a temperature differential between the ends of the device’s thermocouples. The TEG not only has no moving parts, but is also a clean and noiseless renewable energy source. Due to these beneficial merits, the TEG finds many applications such as extracting energy from an automotive heat waste system [1] or the combustion chamber [2]. Unfortunately, the practicality of TEGs is currently limited mainly because of its relatively low conversion efficiency from heat to electricity. As such, researchers have taken various different

* Corresponding author. Tel.: +0-000-000-0000 ; fax: +0-000-000-0000 . E-mail address: [email protected] .

1876-6102 © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 8th International Conference on Applied Energy. doi:10.1016/j.egypro.2017.03.274

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

approaches, all of which have the common goal of improving the conversion efficiency of the TEG. For instance, one approach involves the search for better materials to be used as the thermocouple of the ஑஢ device. Such an approach is based on the characterization of the TEG by its figure of merit ZT = T ஛ where a higher Z means a higher TEG performance. State of the art materials used for the TEG thermocouples include bismuth telluride [3, 4], a material with optimized performance at room temperature, lead telluride [5] and skutterudites [6]. Recently, researchers have even found a new class of materials known as nanostructured materials which is found to achieve an even better ZT performance [7, 8]. Alternatively, another approach to achieve better TEG conversion efficiencies is by improving its geometric design. For instance, references [9-11] provided comprehensive mathematical models of the TEG in operation and based on these models, the TEG geometric design can be optimized. The two stage TEG was also proposed in [12] where the TEG mathematical model was modified to include the extra stage. Optimization of the TEG utilizing the genetic algorithm (GA) has also been conducted in [13, 14] although these analyses are focused on thermoelectric cooling systems as opposed to electricity generation. On the other hand, the authors of this paper have previously proposed a GA optimization of the TEG as utilized in a hybrid solar panel/TEG application when used in an outer space environment [15]. The third approach, which is the focus of this paper, is to use the principles of maximum power point tracking (MPPT). Similarly to solar panels, when electrically connected to an external load, TEGs have power characteristic curves where if the impedance of the external load has a particular unique size, maximum power is retrieved. This unique impedance varies considerably with both the design of the TEG and its operating conditions. Subsequently, a maximum power point tracking is utilized to actively emulate the impedance such that the TEG is always operating at the maximum power point (MPP). As the I-V and P-V characteristics of the TEG follow the same principles as that of the solar panel, MPPT techniques that are commonly used for solar panels can be reused for the TEG. For instance, in [1, 16], the perturb and observe (P&O) algorithm was used which simply involves perturbing the input voltage of the DC-DC converter until the MPP is located. Perturbation of the duty cycle directly (which is often known as hill climbing) was also used in [17] for the TEG application. The popularly known incremental conductance technique was also utilized in [18, 19] although in [19], the focus is on a seamless transfer between MPPT and the power matching mode. Nevertheless, the aforementioned MPPT techniques suffer from steady state oscillations around the MPP which results in significant power losses. By recognizing the nature of the TEG equivalent circuit model, a recent study [20] used the fractional short circuit current method to evaluate the MPP. While this method is simple in nature, it requires a periodic disconnection of the load from the TEG in order to measure the short circuit current. Moreover, this method assumes 100% accuracy on the TEG circuit model which, in reality, is often not the case. This paper proposes an algorithm that adaptively scales the DC-DC converter duty cycle such that steady state oscillations around the MPP are so small that they can be considered eliminated. The proposed algorithm will be demonstrated through simulations and a hardware experiment of the TEG with a large temperature differential between its two end surfaces. The resulting performance will be compared to that of the fixed step counterpart. Despite the simplicity and the minimal computational intensity, the proposed algorithm achieves a very fast and stable tracking response regardless of the applied conditions on the TEG. The remainder of this paper is organized as follows. Section 2 describes the equivalent circuit model of the TEG which is a prerequisite for the subsequent sections. Section 3 provides the full details of the proposed algorithm. Section 4 describes the simulation parameters and Section 5 presents the simulation

15

16

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

itself as well as its results. Section 6 provides an experimental verification of the proposed algorithm and Section 7 concludes this paper. 2. TEG Equivalent Circuit Model The equivalent circuit model of the TEG is shown in Figure 1. The model consists of the internal TEG resistance (்ܴாீ ) connected in series with a voltage source. The voltage of the voltage source equals the TEG’s open circuit voltage which is a function of the Seebeck Co-efficient and the number of thermocouples as shown in Eq. (1). In Eq. (1) and (2), m is the number stages of the TEG which obviously equals one for a single stage TEG. The short circuit current, Iୗେ(୘୉ୋ) , which is defined as the flowing current when the load resistance ܴ௅ equals zero, is easily obtained through Ohm’s law (Eq. (3)).

Figure 1: The equivalent circuit model of the TEG.

௠

ܸை஼(்ாீ) = ෍ ݊௜ ߙ௜ (ܶ௜ െ ܶ௜ାଵ )

(1)

௜ୀଵ ௠

்ܴாீ = ෍ ݊௜ ܴ௜

(2)

௜ୀଵ

‫ܫ‬ௌ஼(்ாீ) =

ܸை஼(்ாீ) ்ܴாீ

(3)

V୓େ(୘୉ୋ) , R ୘୉ୋ and hence Iୗେ(୘୉ୋ) are all dependent on both the TEG design and the temperature differential that is applied to it. In literature, various mathematical models have already been proposed and can be used to accurately determine these parameters [9, 15]. For the sake of brevity, the presentation of such models will not be included in this paper and the reader is referred to the aforementioned references. It is noted however that the mathematical model used in this research is based on that presented in the authors’ previous publication [15] although the solar panel given in that model is removed. 3. The Proposed MPPT Algorithm 3.1. Direction of Perturbation

17

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

Define a function u which defines the direction of perturbation and takes the discrete values of 1, 0 or -1. The appropriate direction of the duty cycle perturbation can be found by using conventional techniques such as hill climbing or incremental conductance. Eq. (4) represents the direction as derived from the hill climbing technique and Eq. (5) represents the direction as derived from the incremental conductance technique. In both of these equations, the sign function (Eq. (6)) returns the sign of the input parameter and can be easily implemented in software. Theoretically, for a given operating condition, both Eq. (4) and Eq. (5) will give exactly the same results. However, a practical implementation will often involve measurements that are noisy in nature. Hence, Eq. (4) will potentially give an inaccurate interpretation of u[n] when very small duty cycle perturbations (Eg. < 1%) are applied. This is because such perturbations will result in small values of dP[n] and dV[n] which will be dominated by the presence of noise. On the other hand, the magnitude of the function in Eq. (5) is not dependent on the duty cycle perturbation size and depends rather on the relative distance to the MPP. Subsequently, Eq. (5) has been adopted in the proposed algorithm of this paper. ‫ = ]݊[ݑ‬െ‫)]݊[ܸ݀(݊݃݅ݏ )]݊[ܲ݀(݊݃݅ݏ‬

(4)

ο‫]݊[ܫ‬ ) οܸ[݊]

(5)

‫ = ]݊[ݑ‬െ‫ ]݊[ܫ(݊݃݅ݏ‬+ ܸ[݊] 1 ‫ = )ݔ(݊݃݅ݏ‬൝ 0 െ1

‫>ݔ‬0 ‫=ݔ‬0 ‫ 0.075 |I[n]| k=k–1 Delay = 1 end elseif Eq. (7) is Satisfied AND Delay < (Delay_Max – Delay_Min) Delay = Delay_Max k=k+1 if k > k_max k = k_max end elseif Delay > 0 Delay = Delay – 1 else k=k–1 Delay = 1 end if k < 0 k=0 end ୅ οD[n] = ౡ // Eq. (8) ୆ ‫ ݊[ܦ = ]݊[ܦ‬െ 1] + ‫]݊[ݑ‬οD[n] Update Historic Variables (ie. [n-1], [n-2] etc. variables) Return

Figure 2: The overall pseudocode of the proposed MPPT algorithm. 4. Simulation Parameters 4.1. Chosen TEG Module The commercial off the shelf Marlow Industries Inc. TG12-6L TEG was selected as the test subject of this paper. It is a 4x4cm with a total of 127 thermocouples. The operating electrical characteristics vary according to the applied temperature differential and thus will be discussed in Section 4.4. 4.2. DC-DC Converter The SEPIC converter was chosen over other DC-DC converter topologies mainly because of its proven stability and also because it does not invert the output. The circuit diagram for the SEPIC converter is shown in Figure 3.

19

20

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

Figure 3: The SEPIC Converter Circuit.

The output voltage is controlled by pulse width modulation (PWM) switching of the switch S. In this paper, the switching frequency is chosen to be 100 kHz where this frequency is high enough to obtain accurate control and is also low enough for operation in a typical microcontroller. In the steady state condition, the output voltage is given as: ܸ௢௨௧ =

‫ܦ‬ ܸ 1 െ ‫ ܦ‬ூ௡

(11)

Where VIn and D are respectively the input voltage and the PWM duty cycle. The components for the SEPIC converter used in this paper are selected as follows: 1.

‫ܮ‬ଵ = ‫ܮ‬ଶ = 5 mH

2.

‫ܥ‬ଶ = 300 uF

3.

‫ܥ‬ଵ = ‫ܥ‬ଷ = 470 uF

4.

FQP30N06L MOSFET for switch S

5.

1N5819 Schottky Diode is used as diode D

The inductances and capacitances were chosen based on the state space averaging method [23] under the criteria that the transient response is stable and reaches the steady state within the selected sampling time of 0.02s. The FQP30N06L MOSFET was selected because of its low gate threshold voltage which allows for better compatibility with microcontrollers as well as lower power requirements on the MOSFET drivers. 4.3. Experimental Setup In the experiment, the TEG module is heated with a hot air soldering gun on the hot side. A passive heat sink is then used to cool the TEG cold side. This design is typical of a scaled down model of a real application scenario and thus is sufficient for testing the proposed MPPT algorithm. For the sake of simplicity, the passive heat sink is modelled in simulation as a large TEG outlet area which is 15 times the inlet area (4x4cm). The convection coefficient of the soldering air gun to the inlet TEG is tailored to match experimental observations of its I-V and P-V characteristics. 4.4. Test Temperature Profile In the simulation, the hot side temperature profile shown in Figure 4 is used as a test subject. Initially, the air gun was set to apply air with a temperature of 600K for 2.5s. This was then followed by a sudden drop to 500K. After another 2.5s, the air temperature is then suddenly returned to the original 600K and the simulation terminates at 7.5s. The sudden jumps and drops between each temperature levels are useful in testing the adaptability of the proposed algorithms to sudden changes in the applied temperature. The cold side temperature is set to be constant at 300K throughout the entire simulation.

21

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

Figure 4: The applied hot side temperature profile with respect to time for the simulation.

By simulating the TEG using the conditions specified in Section 4.3 and solving the parameters of the thermodynamic model, the following characteristics are obtained for the two applied hot side temperatures. The Seebeck coefficient (ߙ) and the thermal conductance (K) are estimated based on the temperature dependant curves provided in [15]. Table 1: Electrical Characteristics of the TEG for the given applied temperature differentials

Temperature (K) 500 600

V୓େ(୘୉ୋ) (V) 1.073 1.800

R ୘୉ୋ Ÿ 2.372 2.622

ߙ(ܸ/‫)ܭ‬ 3.596 × 10ିସ 3.589 × 10ିସ

‫ܹ( ܭ‬/‫)ܭ‬ 0.0046 0.0046

5. The Simulation The proposed MPPT algorithm was simulated in MATLAB/Simulink and the overall diagram is shown in Figure 5. The model is composed of the TEG power source, the SEPIC converter, the MPPT algorithm and of course, the external load. For the purposes of flexibility and legibility, the aforementioned components were defined in subsystem blocks. By using such a configuration, the MPPT algorithm block can be easily interchanged between the proposed and previous algorithms. It is worth noting that during operation, the TEG’s voltage and current both contain ripples which is simply a result of the PWM switching in the SEPIC converter. Subsequently, the TEG’s operating voltage and current are each passed through a low pass filter with a cut off frequency of 100 Hz. This cut off frequency was chosen on the requirement that it is to both pass the MPPT algorithm’s sampling frequency and to reject the PWM switching frequency of S.

22

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

Figure 5: The overall Simulink model that implements the proposed MPPT system.

5.1. Evaluation and Benchmark The performance of the proposed algorithm shall be compared to that of the conventional fixed step hill climbing algorithm that uses a constant duty cycle perturbation of 0.01. Moreover, it shall also be compared to that of using Eq. (12) to evaluate ȟ‫ ܦ‬where M is set as ‫ = ܯ‬െ0.06. For the sake of brevity, the algorithm that uses Eq. (12) will be called the previous algorithm. The proposed algorithm is given the following parameters: N = 3, A = 0.08, B = 2 and k_max = 4.

ο‫ܯ = ]݊[ܦ‬

οܲ[݊] οܸ[݊]

(12)

A sampling time of 0.02 seconds is utilized for both algorithms and the externally connected load has a resistance of 22Ÿ7KHLQGXFWRUVRIWKH6(3,&FRQYHUWHUDUHDVVXPHGWRKDYHDUHVLVWDQFHRIŸDQG this effect is modelled by adding a series resistor of the corresponding size. For a conservative analysis, the VZLWFK 6 LV DVVXPHG WR KDYH DQ 21 UHVLVWDQFH RI Ÿ ZKLFK LV KLJKHU WKDQ WKDW VSHFLILHG LQ WKH datasheet of the selected MOSFET. Snubbers for all switching components are ignored. On the other hand, a forward voltage of (datasheet specified) 0.6 Volts is imposed on the diode D. 5.2. Simulation Results Figure 6 shows the electrical power that was drawn from the TEG and Figure 7 shows the electrical power that was delivered to the load resistor. According to these results, all 3 algorithms achieve a high tracking performance in terms of speed. They are also able to quickly track the new MPP point when the applied temperature differential is changed. Based on these results, it is clear that the proposed lock-on mechanism is properly tracking the MPP with a high degree of accuracy. In terms of comparing the tracking response of the 3 algorithms, it is clear that the proposed lock on mechanism algorithm is achieving a better tracking performance than that of the other two algorithms. For instance, as shown in the beginning of Figure 6, the proposed algorithm achieves a much faster initial response than that of the previous algorithm. Moreover, unlike the fixed step hill climbing algorithm, the proposed algorithm does not achieve any significant steady state oscillations about the MPP.

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

Subsequently, it is expected that in a scaled up TEG application which involves high power and where the output power is very sensitive to small changes in duty cycle perturbations, in comparison to the other two algorithms, the proposed algorithm will improve the output power efficiency of the TEG.

Figure 6: Power drawn from the TEG as simulated on Simulink

Figure 7: Power delivered to the resistor as simulated on Simulink

6. Experimental Verification The proposed algorithm shall now be further verified by implementing it into a real hardware system. The components of the SEPIC converter are the same as that used in the simulation and the controller is implemented using the MSP430F5529 microcontroller. The built-in 12 bit ADC converter is used to convert both the voltage and the current analogue readings into the digital format for use in the MPPT algorithm. In the meantime, the current is sensed by using the 5A ACS712 current sensor where the output is an analogue voltage that varies proportionally to the flowing current. The inherent noise in the raw current signal is reduced by amplifying the signal with an active 1st order low pass filter circuit with a cut off frequency of around 130 Hz. The overall experimental apparatus that was used to test the proposed MPPT algorithm is shown in Figure 8.

23

24

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

Figure 8: The experimental apparatus for testing the proposed MPPT algorithm.

Figure 9 shows the I-V and P-V curves of the TEG as obtained from experimental observations. This data was obtained by exploiting the linear operating region of a MOSFET where a varying gate-source voltage is applied. While a small difference in the open circuit voltage is noticed for both temperatures, the output power is roughly the same as that in simulation.

Figure 9: The I-V Curve of the TEG when 600K hot air from the soldering air gun is applied.

In the experiment, a temperature profile similar to that used in the simulation (Figure 4) is readopted in the experiment. However, the jumps in the temperature (From 600K to 500K and vice versa) were set to be 40s and 80s instead respectively. This was done to observe MPP tracking over a longer period of time. Moreover, the hot solder gun requires time to adjust the applied air temperature whenever the command temperature is changed. The proposed algorithm shall be validated by comparing it to that of the conventional algorithm with fixed duty cycle perturbations of 0.01. Both algorithms shall have their duty cycles be initiated from 0.5. 6.1. Experimental Results Figure 10 shows the power drawn from the TEG when both the proposed and fixed step algorithms are utilized. Figure 11 then shows the corresponding power delivered to the load resistor as measured on the

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

resistor itself. As shown here, the tracked input power for both algorithms is consistent with the peak power shown in Figure 9 which indicates that the MPP is indeed, being tracked. Overall, both algorithms exhibit very similar high performance MPP tracking in terms of speed of response and low steady state oscillations. This indicates that the proposed algorithm is working and is operating as intended.

Figure 10: Power drawn from the TEG.

Figure 11: Power delivered to the load resistor.

7. Conclusion In this paper, an adaptive duty cycle scaling algorithm, known as the lock on mechanism, is proposed to achieve MPPT in the TEG application. By utilizing such an algorithm, steady state oscillations that would otherwise exist in conventional fixed step based MPPT algorithms are removed. The proposed algorithm achieves a high performance tracking response both in terms of speed and low steady state oscillations and these characteristics are verified through simulation and a hardware experiment involving a single

25

26

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

TEG module. Future work on this research includes repeating the experiment using a practical sized TEG which can be typically used in applications such as automotive heat waste power recovery.

References 1. 2.

3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16.

17. 18.

19.

Yu, C. and K.T. Chau, Thermoelectric automotive waste heat energy recovery using maximum power point tracking. Energy Conversion and Management, 2009. 50(6): p. 1506-1512. Aranguren, P., et al., Experimental investigation of the applicability of a thermoelectric generator to recover waste heat from a combustion chamber. Applied Energy, 2015. 152(0): p. 121-130. Puneet, P., et al., Preferential Scattering by Interfacial Charged Defects for Enhanced Thermoelectric Performance in Few-layered n-type Bi2Te3. Sci. Rep., 2013. 3. Ivanova, L. and Y.V. Granatkina, Thermoelectric properties of Bi2Te3-Sb2Te3 single crystals in the range 100–700 K. Inorganic materials, 2000. 36(7): p. 672-677. Dughaish, Z.H., Lead telluride as a thermoelectric material for thermoelectric power generation. Physica B: Condensed Matter, 2002. 322(1–2): p. 205-223. Scherrer, H., et al., Solar thermolectric generator based on skutterudites. Journal of Power Sources, 2003. 115(1): p. 141-148. Kanatzidis, M.G., Nanostructured Thermoelectrics: The New Paradigm?†. Chemistry of Materials, 2009. 22(3): p. 648-659. Pichanusakorn, P. and P. Bandaru, Nanostructured thermoelectrics. Materials Science and Engineering: R: Reports, 2010. 67(2–4): p. 19-63. Gou, X., H. Xiao, and S. Yang, Modeling, experimental study and optimization on lowtemperature waste heat thermoelectric generator system. Applied Energy, 2010. 87(10): p. 3131-3136. Kim, S., Analysis and modeling of effective temperature differences and electrical parameters of thermoelectric generators. Applied Energy, 2013. 102: p. 1458-1463. Wang, Y., C. Dai, and S. Wang, Theoretical analysis of a thermoelectric generator using exhaust gas of vehicles as heat source. Applied Energy, 2013. 112: p. 1171-1180. Chen, L., et al., Performance optimization of a two-stage semiconductor thermoelectricgenerator. Applied Energy, 2005. 82(4): p. 300-312. Najafi, H. and K.A. Woodbury, Optimization of a cooling system based on Peltier effect for photovoltaic cells. Solar Energy, 2013. 91: p. 152-160. Cheng, Y.-H. and W.-K. Lin, Geometric optimization of thermoelectric coolers in a confined volume using genetic algorithms. Applied Thermal Engineering, 2005. 25(17–18): p. 2983-2997. Kwan, T.H. and X. Wu, Power and mass optimization of the hybrid solar panel and thermoelectric generators. Applied Energy, 2016. 165: p. 297-307. Zhang, X. and K.T. Chau, An automotive thermoelectric–photovoltaic hybrid energy system using maximum power point tracking. Energy Conversion and Management, 2011. 52(1): p. 641647. Eakburanawat, J. and I. Boonyaroonate, Development of a thermoelectric battery-charger with microcontroller-based maximum power point tracking technique. Applied Energy, 2006. 83(7): p. 687-704. Kim, R.-Y., et al., Analysis and design of maximum power point tracking scheme for thermoelectric battery energy storage system. Industrial Electronics, IEEE Transactions on, 2009. 56(9): p. 3709-3716. Kim, R.-Y. and J.-S. Lai, A seamless mode transfer maximum power point tracking controller for thermoelectric generator applications. Power Electronics, IEEE Transactions on, 2008. 23(5): p. 2310-2318.

Trevor Hocksun Kwan and Xiaofeng Wu / Energy Procedia 105 (2017) 14 – 27

20. 21.

22.

23.

Laird, I. and D.D.-C. Lu, High step-up DC/DC topology and MPPT algorithm for use with a thermoelectric generator. Power Electronics, IEEE Transactions on, 2013. 28(7): p. 3147-3157. Hsiao, Y.-T. and C.-H. Chen. Maximum power tracking for photovoltaic power system. in Industry Applications Conference, 2002. 37th IAS Annual Meeting. Conference Record of the. 2002. IEEE. Tey, K.S. and S. Mekhilef, Modified incremental conductance MPPT algorithm to mitigate inaccurate responses under fast-changing solar irradiation level. Solar Energy, 2014. 101: p. 333-342. Robert, W.E. and M. Dragan, Fundamentals of power electronics. 2001, New York, USA: MA: Kluwer Academic Publishers.

27