Improvement of Rooftop Solar Panels Efficiency using Maximum Power Point Tracking Based on an Adaptive Neural Network Fuzzy Inference System

: Rooftop solar panels are a strategy for achieving Indonesia's renewable energy goals, but their non-linear characteristics make them difficult to control, especially in the face of extreme weather changes. An effective controller is needed to optimize the power output of solar panels. This study proposes a Maximum Power Point Tracking (MPPT) controller based on an Adaptive Neural network Fuzzy Inference System (ANFIS) to address this control problem. The capacity of the rooftop solar panels is 3,430-Watt peak (Wp) and they are connected to a 220-Volt (V) grid system. The system is designed, simulated


Introduction
Indonesia has great potential for solar energy because of its location at the equator. All Indonesian regions are exposed to sunlight throughout the year and sunlight energy is straightly converted into electricity. Solar energy potential in Indonesia is 207.8 GWp, as reported by the Indonesian National Energy Council (DEN, 2019). Based on this report, the Ministry of Energy and Mineral Resources Republic of Indonesia released a regulation on how to build rooftop solar panels. This regulation ensures that the quality and optimal conversion of sunlight into electrical energy work effectively. Every electricity customer can utilize solar-based electricity for personal needs and sell it to the electricity grid. Electricity customers are allowed to install a rooftop solar panel capacity at the same as their customer capacity contract (ESDM, 2021). Zero-emission energy in Indonesia's target by 2050 (IESR, 2021) will be supported by this regulation. This regulation is also supported by the United State Institute for international development on technical assistance (USAID, 2020).
Erratic and extreme weather occur due to climate change phenomena, so this situation makes solar panels' performance decrease. Roof solar panels are affected directly by these weather changes because the solar panels are placed on the roofs of houses or buildings. Based on global solar atlas data, the West Nusa Tenggara Republic of Indonesia, the air temperature, average direct irradiation, and specific photovoltaic power output are as follows: 11.3-27.8°C; 1.53-5.25 kWh/m 2 and 2.82-4.62 kWh/kWp, respectively (Group, 2022). These data show that solar radiation and temperature fluctuate and affect rooftop solar panel performances. One of the scenarios to increase panel efficiency is to stabilize the panel temperature by the circulating water method. This method can increase panel efficiency from 7.57 to 8.11% (Pido et al., 2018). Another option to increase the efficiency of rooftop solar panels is to use a tracking system technique.
There are two tracking system models so far, namely mechanical and electrical. However, the mechanical tracking system has been abandoned because it requires much energy to control the solar panels to follow the sun's movement direction. At the 2 same time, the electric tracking system is being developed actively to produce optimal output from the photovoltaic system (Abdel-Salam et al., 2019). The electrical tracking system tracks the maximum power of the solar modules electronically. Solar modules have nonlinear characteristics because they are made from semiconductor materials. The Maximum Power Point Tracking (MPPT) electric tracking system uses offline, online, and artificial intelligence techniques. The offline technique has the disadvantage of tracking results due to partial shade conditions, sudden irradiation changes and ramps, and other non-ideal conditions. Online technique such as Perturb and Observe (P&O) is proposed by Abdel-Salam et al. (2018) to overcome the disadvantages of offline techniques. The P&O technique provides stable conditions with efficiency and dynamic efficiency up to 96.98 and 91.9%, respectively. On the other hand, soft computing techniques (artificial intelligence techniques) are developed rapidly and are used to solve complex problems in MPPT field research. These techniques include Fuzzy Logic Control (FLC), Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference Systems (ANFIS), difference evolution and genetic algorithms, particle swarms, artificial bee colonies, grey wolf optimizers, and many others. The MPPT fuzzy logic control is applied to control the DC-DC converter through the duty cycle to get the maximum output power from the PV system (Eltamaly and Farh, 2020). It is reported that the FLC MPPT tracks environmental condition changes rapidly (Loukriz et al., 2013), and makes power and efficiency of PV systems increased for this condition (Chekired et al., 2014;Hasan et al., 2013;Kamarzaman and Tan, 2014;Messai et al., 2011). Another soft computing technique implemented for MPPT calculation is ANN because the ANN is very effective in solving non-linear problems. It is more accurate and efficient in tracking uniform conditions than other conventional MPPT techniques (Rai et al., 2011;Messalti et al., 2017;Laudani et al., 2014;Mancilla-David et al., 2014).
ANFIS is defined as a soft computing technique that combines FLC and ANN algorithms. The FLC-type with the harmonic search metaheuristic method can provide the best approach for generating optimal vector values in optimizing the fuzzy type-1 and interval type-2 controller membership function in non-linear problems (Patel, 2022). For the system under uncertainty conditions, the fuzzy type-1 and type-2 intervals are implemented to improve various algorithms in control fields (Patel and Shah, 2022a-f). Artificial Conditional Network (ACN) is a popular research topic due to their better ability to handle modeling and parameter uncertainties (Raval et al., 2023). A combination of the FLC and ACN methods is enhanced on each method's deficiencies. The Fuzzy Logic System membership function is generated by training based on the uncertainty of system parameter data using ANN. The ANFIS has been proven to increase the stability of a single-machine electric power system (Ginarsa and Zebua, 2014). The ANFIS control is applied to more than just a single machine. Moreover, this control maintains a large power system's stability (Muljono et al., 2018). A Power System Stabilizer (PSS) based on ANFIS control is implemented on integrated systems with renewable energy generation (Ginarsa et al., 2020). Transient condition in a High Voltage Direct Current (HVDC) system is corrected by the additional controller based on ANFIS (Ginarsa et al., 2022a-b). In the MPPT research field, the ANFIS is applied to adjust the boost converter duty cycle to increase the PV system efficiency (Alfian et al., 2022). Moreover, ANFISbased MPPT design is also used to track change properly in solar irradiation and temperature in rooftop solar panel systems connected to single-phase grid systems . Artificial intelligence applications for solar panel control have been developed intensively by previous researchers. However, research topic change in radiation followed by an extreme change in temperature has not been investigated so far.
This study reports on the implementation of ANFIS as an MPPT to increase the efficiency of rooftop solar panels connected to a single-phase power grid system with varying irradiation dan followed by extreme changes in temperature. Materials and research methods are explained after the introduction, followed by a discussion of the research results. In the final section, a conclusion is given on the efficiency of the ANFIS MPPT technique compared to the no MPPT and MPPT Perturb and Observe (P&O) methods.

Materials and Methods
The Simulink tool 8.5 on MATLAB version R2015a is employed for the simulation. The solar module is Trina Solar TSM-250PA05.08. The rooftop solar panel consisted of 14 modules arranged in series to produce a voltage at a maximum power point of 434 V for a 3,498 Wp. The weight of each module is 8.6 kg and the total weight of the roof solar panel system is 260.4 kg. The area required for installation on the roof was 22.92 m 2 . The module specifications are Table 1.
The rooftop solar panel is modeled at Simulink as a PV array with a configuration of 14 Trina Solar TSM-250PA05.08 modules connected in series. The system model was analyzed for the range of solar irradiance of 2-1000 W/m 2 and temperature 15-55°C. On the output side, the solar panels were loaded with varying resistive loads to obtain the characteristics/behavior for each combination of irradiation and temperature . The simulation results (irradiation, temperature, and voltage at maximum power point) were stored and used for ANFIS training data. The amount of data to train the ANFIS MPPT was 1000 pairs of inputs (irradiation and temperature) and output (voltage at maximum power point).
3 The fuzzy inference system on ANFIS MPPT uses sub-clustering with a range of influence of 0.5; squash factor of 1.25; accept a ratio of 0.5 and reject ratio of 0.15. The number of FIS membership functions for irradiation and temperature each has six membership functions. The input membership function is of type gaussmf, while the output membership function is linear. Then ANFIS was trained using the hybrid optimization method with 1000 epochs. Figure 1(a-b) show irradiation and temperature membership functions. Figure 2(a-b) show the control surfaces and the structure of the ANFIS MPPT.
The next step is to validate the ANFIS training results with test data. The test data are 14 pairs of training data in the range 50-1000 W/m 2 and 25-55C for irradiation and temperature. The test aims to ensure that the rules generated from the ANFIS MPPT learning outcomes produce the same results between the ANFIS MPPT and the test data. Comparison results are significant to ensure the quality control of ANFIS MPPT is effective.
Rooftop solar panels were connected to a 220 Volt electric power network model on Matlab/Simulink. An inverter device was installed on the rooftop solar panels output to convert the DC voltage/current to AC voltage/current in a single-phase grid system. The single-phase inverter was equipped with a controller. The Sinus PWM method was applied to control the inverter. The inverter specifications were the universal bridge with two arms, where the Rs, Cs, and Ron were 106 Ohms, in farads, and 10 -3 Ohms, respectively. An Insulate Gate Bipolar Transistor was included as a switching component. The inverter was also equipped by PLL as a controller, current regulator, four pulses bipolar modulation for the PWM generator, and carrier frequency. The carrier frequency was operated on 63 × 50 or 3150 Hz (network frequency: 50 Hz). The load on the grid side was set at S = 5+ j2 kVA. Grid power is defined as the difference between load power and power from rooftop solar panels. Equations 1 and 2 are the energy equations for the DC and AC sides. Equation 3 is a formula to calculate efficiency: where, PDC and t are the DC output power of rooftop solar panels and operating time, respectively: where, cos() is the load power factor:

Results and Discussion
The test results of the ANFIS MPPT model after training with input/output data are compared with the test data to determine its accuracy, Table 2. The maximum error of the ANFIS MPPT model is 2.02% at 50 W/m 2 and 55°C and the minimum error is 0.00% at 500 W/m 2 and 25°C; 900 W/m 2 and 25°C; 900 W/m 2 and 55°C; and 1000 W/m 2 and 55°C. It is founded that highnonlinearity performance is shown by the rooftop solar panels, especially at low irradiation and temperature. More diverse training data is needed to find more detailed results.
The grid-connected solar panel model in simulink is Fig. 4. Three simulation scenarios were studied: Gridconnected solar panels with no MPPT, P&O MPPT, and ANFIS MPPT. The input values in the inverter control block are replaced with 1, 2, or 3 for systems with no MPPT, P&O MPPT, and ANFIS MPPT to change the scenario. Figure 5 shows varying irradiation and constant temperature. Irradiation varied from 250, 750; 1000; 800; 400; and 200 W/m 2 and temperature at 25°C. The change of irradiation from one to the next is with a slope of +/-8000. Figure 6 shows the operating results of gridconnected rooftop solar panels for variable irradiation at a constant temperature. Figures 6(a, d) show that the three methods have almost the same output power and the current flowing from the rooftop solar panel to the load for each change in irradiation. The comparison results of energy and efficiency are Table 3. Energy yields on the DC and AC sides of the roof solar panels during operation for each MPPT method are calculated using Eq. 1 and 2. The efficiency values are obtained from Eq. 3. Table 3 shows that the most energy is provided by the DC side of the rooftop solar panel with the grid-connected P&O MPPT. But on the AC side, the biggest energy from rooftop solar panels occurred on ANFIS MPPT. The efficiency was achieved at 36.562, 36.569, and 36.586% for systems with no MPPT, P&O MPPT and ANFIS MPPT, respectively. The efficiency of the ANFIS MPPT was at 0.0664% compared to the no MPPT and 0.0474% compared to the P&O MPPT. The ANFIS MPPT results are better than the other competing methods. Figure 6(b) shows the reference voltage (Vref) output of each MPPT. The Vref fluctuation of the MPPT P&O control is larger than the others. So, the voltage in the AC network is more fluctuate for MPPT P&O, Fig. 6(c).

The result in
The solar panel's temperature increases with increasing solar irradiation and vice versa. The following simulation is a change in irradiation followed by a change in solar panel temperature. The irradiation changes were the same as in the previous simulation. Irradiation changes were followed by the temperature of rooftop solar panels at 25, 35, 50, 40, 35, and 27°C. The varying irradiation and extreme temperature changes are shown in Fig. 7. The irradiation gradient is the same as the previous simulation, while the temperature gradient is +/-80.
The output results of rooftop solar panels for variations in irradiation and temperature. Figure 8(a) shows the DC output power of ANFIS MPPT is better than the method with no MPPT and P&O MPPT. Similarly, Fig. 8(d) shows that the current supplied by the rooftop solar panels to the electrical load for the ANFIS MMPT method is more significant than the other methods comparison results are Table 4.  6 The result in Table 4 shows that the ANFIS MPPT method for rooftop solar panels produces greater energy on the DC and AC sides than other methods. In terms of efficiency, the ANFIS MPPT method is the most. The efficiency of the no MPPT, P&O MPPT,and ANFIS MPPT methods was 36.505,36.629,and 36.670%. So ANFIS MPPT is 0.452% better than the method with no MPPT and 0.112% than the P&O MPPT method.
Variations of irradiation and temperature, as Fig. 7, the method with no MPPT produces a constant Vref. However, in the P&O MPPT ANFIS MPPT method, Vref changes depending on the panel temperature. The MPPT ANFIS method provides a more stable and faster Vref change than the MPPT P&O method, Fig. 8(b). Figure 8(c) is the voltage change on the side of the AC network due to a change in Vref. Changes in Vref as a control voltage for voltage detection at maximum power (such as the system modeling) were investigated earlier (Eltamaly and Farh, 2020). The maximum voltage (Vm) at maximum power was found not constant with changes in irradiation and Vm decreased at a higher panel temperature.

Conclusion
Applying the ANFIS MPPT method to rooftop solar panels connected to a single-phase network can increase the efficiency of rooftop solar panels. At fluctuating irradiation and constant temperature, the increase in efficiency of the ANFIS MPPT method reached 0.0664 and 0.0474% against the method with no MPPT and P&O MPPT. At fluctuating irradiation and extreme temperature, the efficiency increases of the ANFIS MPPT method reached 0.4523 and 0.1115% compared to the methods with no MPPT and P&O MPPT. The efficiency of ANFIS MPPT is better in all operating conditions compared to the no MPPT and P&O MPPT methods.