**1. Introduction**

Energy is essential in our society, being the motor of almost every sector. Fossil-fuels are historically the most important source of energy, representing 80.2% in 2019 [1] These kinds of energies have different problems; one of them is their scarcity, since they are limited resources that have been exploited for a long time. Another critical problem is the pollution caused by the burning and extraction of these fuels, which is hazardous for people [2] and the environment [3]. To solve these problems, other energy sources can be used. These alternative energies, renewable energies, have two main benefits. First of all, they are based on unlimited resources that will not run out, even with extensive exploitation. Their exploitation is also nonpolluting. Investment in these energies has been rising in the last years, even with a crisis such as the COVID-19 pandemic [1,4].

One of the most important green energies is solar energy. This energy is composed of solar, thermal and photovoltaic (PV). The latter has been found to be more useful and profitable for industry production [5,6] and has been growing steadily in recent years. As we can see in Figure 1, the share of PV systems is increasing, and it is expected to be one of the prime energy sources in the next years [7].

PV energy is produced by photovoltaic modules. Each module is composed of different sub-units, called solar cells, which absorb the energy emitted by the sun [8]. PV panels

**Citation:** Mateo Romero, H.F.; González Rebollo, M.A.; Cardeñoso-Payo, V.; Alonso Gómez V.; Redondo Plaza A.; Moyo, R.T.; Hernandez Callejo, L. Applications of Artificial Intelligence to Photovoltaic Systems: A Review. *Appl. Sci.* **2022**, *12*, 10056. https://doi.org/ 10.3390/app121910056

Academic Editor: Fabrice Goubard

Received: 7 September 2022 Accepted: 3 October 2022 Published: 6 October 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

are usually connected in series to each other, this is known as a PV array. Each PV array is connected to a power inverter to control the production and check the performance of the array [9]. PV farms (also known as PV plants) are usually composed of many PV arrays. The maintenance of these factories is extremely complex. The production of the modules depends on different conditions, this makes mechanisms of control to optimize the production necessary. Solar modules are also vulnerable to physical defects, which can reduce or even nullify the production of one cell, or even the whole modules in the worst cases. This is usually dealt with by human labor, checking each module in a certain period of time.

Solutions to maintenance problems in PV systems have been traditionally circumscribed either to simplistic automatic supervision approaches [10] or costly direct human supervision.

In the past recent years, however, artificial intelligence (AI)-based approaches have emerged. AI techniques are being applied in almost every research field or industry to improve services or solve problems which are impossible for traditional methods [11].

These techniques can also be applied to solve the problems of PV systems. In this review, we analyze how AI is being applied to the PV sector. PV systems face different problems during installation but also during exploitation, since PV modules are vulnerable to the climate conditions' unpredictable events. An analysis of the problems found in PV plants can be found in [12,13]; also, the maintenance of the modules is key in order to secure maximum production and to improve the security of the installations [14]. In order to reduce the scope of this review, only the four most critical problems related to energy optimization and maintenance are considered:


The most important problems and the technologies that have been used for dealing with each one of these problems are addressed. The final objective of this review is to analyze the most important techniques used and how they have improved the solutions of the problems in order to have a clear understanding of the state of the art in the area.

The paper is structured as follows: First, an explanation of the problem is described in Section 2, in order to provide more information to readers who are not familiarized with this topic. After that, the artificial intelligence techniques used of each of the problems are explained. The Maximum Power Point Tracking is discussed in Section 3.2; Section 3.3 is about the forecasting; Section 3.4 presents the parameter estimation, and defect detection problems are discussed in Section 3.5. Finally, an analysis of the different problems is performed in Section 4, identifying the tendencies and flaws in the research of each one of them.

## **2. Relevant PV Problems**

As we explained before, PV installations have to face a large amount of problems. The most important ones are related to optimizing energy production, since it is the final objective of an energy installation. They are also related to the maintenance of equipment, mostly the solar modules. In this section, four problems related to this issue are explained in order to give the reader a basic context of the different problems.

### *2.1. Maximum Power Point Tracking (MPPT)*

PV cells have a complex relationship between their environment and the power they can produce. Along the IV curve (Figure 2) of solar cells there is a point where the power will be maximized, this is called the Maximum Power Point (MPP). This point usually changes depending on conditions such as irradiation, temperature or the state of the PV cell. These conditions can change the shape of the curve, making the problem nonlinear and time-varying due to the changes produced by the atmospheric and load conditions.

Another problem is that it is not possible to directly obtain the IV curve of a single PV cell. The IV curves are usually taken from one single module or even from a PV array. The measured curves are more complex than the IV curve of a cell. The more complex a curve is, the harder it is to track the MPP.

**Figure 2.** Different IV curves. Green circle: Max Power Point; Red Star: Open-Circuit Voltage (*VOC*); Yellow Square: Short-Circuit Current (*ISC*).

The algorithms to solve this problem can be classified according to different criteria; one of the most important ones is according to the number of variables used for measuring the tension. Another interesting approach is to classify the method according to the control strategy used. A brief explanation of the most important traditional methods can be found in Table 1, and more information can be found in works such as [15,16].


**Table 1.** Traditional Methods for MPPT.
