A Review on Machine Learning Applications for Solar Plants

A solar plant system has complex nonlinear dynamics with uncertainties due to variations in system parameters and insolation. Thereby, it is difficult to approximate these complex dynamics with conventional algorithms whereas Machine Learning (ML) methods yield the essential performance required. ML models are key units in recent sensor systems for solar plant design, forecasting, maintenance, and control to provide the best safety, reliability, robustness, and performance as compared to classical methods which are usually employed in the hardware and software of solar plants. Considering this, the goal of our paper is to explore and analyze ML technologies and their advantages and shortcomings as compared to classical methods for the design, forecasting, maintenance, and control of solar plants. In contrast with other review articles, our research briefly summarizes our intelligent, self-adaptive models for sizing, forecasting, maintenance, and control of a solar plant; sets benchmarks for performance comparison of the reviewed ML models for a solar plant’s system; proposes a simple but effective integration scheme of an ML sensor solar plant system’s implementation and outlines its future digital transformation into a smart solar plant based on the integrated cutting-edge technologies; and estimates the impact of ML technologies based on the proposed scheme on a solar plant value chain.


Introduction
Solar plant systems have complex nonlinear dynamics with uncertainties since the system's parameters and insolation fluctuate [1]. Thereby, it is complicated to approximate these complex dynamics with classical methods, while ML methods provide the required performance [2]. In modern sensor systems, ML methods are crucial units to increase the quality of big dataset processing for solar plant design, forecasting, maintenance, and control [1,2]. Within the EU COVID-19 strategic reply, the smart energy standards define a cloud platform specification for a distributed solar big data ecosystem that will provide the creation of effective ML technologies for smart solar energy [3]. The long-term contribution of solar energy is dependent on overcoming the remaining issues of grid integration, high costs, and low efficiency, mainly through the research and development of a smart solar plant system based on ML methods on account of traditional methods' ineffectiveness. Within breakthrough studies, ML technologies collected, analyzed, and converted a huge number of sensory datasets into ML knowledge. These big data sets are collected by supervisory control and data acquisition (SCADA) systems [4]. The SCADA system is able to integrate the sensor system and ML technologies into an ML sensor system based on software that implements ML sensor models and integrates with SCADA through API. Further, the application of ML technologies for the digital transformation of solar plant systems has a massive potential to increase their stability, reliability, dynamic response, cost-effectiveness, and other essential advancements, easing their integration into electric grids.
The contribution of this article is threefold. First, we reviewed more than 100 research papers devoted to state-of-the-art ML technologies of solar plant systems, most of the articles were published in the last five years. Second, we reviewed resources where researchers can find open datasets, source code, and ML framework and simulation environments to create ML technologies for a solar plant system. Third, in contrast with other review articles, our review proposes a simple but effective pipeline scheme for an ML sensor solar plant system's implementation and outlines its future digital transformation into a smart solar plant based on integrated, cutting-edge technologies; estimates the impact of the ML technologies based on the proposed scheme on a solar plant value chain; sets benchmarks for performance comparison of the reviewed ML models for a solar plant's system based on the comparative studies' results summaries; and briefly summarizes our self-adaptive models for sizing, forecasting, maintenance, and control of a solar plant based on a modified fuzzy neural net (MFNN) that is automatically created with regard to tasks' complexities and overfitting problems [5][6][7][8]. Our research provides a mapping of the recently reported ML methods and quantification of their advantage and shortcomings as compared to classical methods, which are usually employed in the sensor system, hardware, and software of solar plants; an effective integration scheme of ML technologies into the sensor systems and software of solar plants; a future prospect of the integrated cutting-edge technologies, including ML for digital transformation of solar energy into smart solar energy. In addition, we provide some directions and insights for the future development of a smart solar plant system.
The rest of the article proceeds as follows. Section 2 reflects an analysis of ML technologies for a solar plant system. In Section 2.1, we describe an ML sensor system of a solar plant based on an ML sensor model and its life cycle. Sections 2.2 and 2.3 provide brief introductions of the main ML methods and frameworks which are being applied in solar plant systems, correspondingly. In Section 2.4, we briefly describe the open datasets and source code to create ML technologies for a solar plant system. In Section 3, we analyze, discuss, and summarize recently reported research into ML applications for solar plant systems, their advantages, and shortcomings as compared to classical methods. In addition, in Section 3, we briefly describe an ML sensor system based on a developed software that integrates with SCADA through API. The subsections of Section 3 reflect an analytical review of ML technologies for the design, forecasting, maintenance, and control of solar plants. Section 4 presents the future prospect of integrated, cutting-edge technologies, such as ML, cloud, edge computing (EC), internet of things (IoT), etc., to create a smart solar plant system that provides the digital transformation of solar energy into smart solar energy. Finally, in Section 5, we conclude the article with a brief summary of this review and a discussion about the current locus and opportunities for future development in the field of ML to create a smart solar plant system.

Machine Learning Technologies for a Solar Plant's System
Real-life solar plant systems have complex, nonlinear dynamics due to variations in system parameters and insolation. Thus, ML methods have been proposed to approximate this complex dynamic. The recent studies [1,2,[5][6][7][8][9][10][11][12][13] prove that ML technologies for a solar plant's design, forecasting, maintenance, and control increase the effectiveness and reliability of the solar plant as compared to conventional methods. In smart sensor systems of solar plants, ML methods are crucial units to increase the quality of datasets processing the solar plant's design, forecasting, maintenance, and control. SCADA is a control system architecture that uses sensors, programmable logic, and discrete PID controllers to control the processes of a solar plant system. The solar plant's system includes advanced sensors. Big data from SCADA are collected 24/7. Combined with weather big data, this enables the creation of ML technologies to solve complex tasks of a solar plant's design, forecasting, maintenance, and control.

ML Sensor System of a Solar Plant
Smart models based on ML technologies have the advantage of parallel computation through modern graphical processing units, which significantly decreases the time cost in SCADA datasets processing for solar plant design, forecasting, maintenance, and control [12].
The reliability, accuracy, and other demanded quality parameters must be composed as the performance of an ML model. This model must be created effectively with high-quality datasets to have optimal performance [14]. Figure 1 shows the basic life cycle of an ML sensor model. Smart model creation has two phases: data preparation (DP phase) and model creation (MC phase). They should be elaborated by the Cross-Industry Standard Process for Data Mining cycle (CRISP-DM) [15] and Open Neural Network Exchange (ONNX) format [16]. The CRISP-DM cycle [15] provides a pipeline for the implementation of smart models in real-time scenarios.
The sensor data of a solar plant are compiled into raw SCADA datasets. Then, these datasets are preprocessed ( Figure 1) in a simple way (standardization or encoding). Data preparation methods include dimensionality reduction (principal component analysis (PCA)), sampling (subsampling, oversampling), transformation, encoding, feature extraction, and selection [14]. Feature extraction is a crucial step in a smart sensor system's creation because it provides knowledge for ML model creation [14]. The DM methods generate features. The most relevant data are further separated into train, validating, and test datasets ( Figure 1). An ML model to solve either classification or regression tasks is trained based on a train dataset. When a smart model provides the demanded performance, its weights are frozen. The ML frameworks, which we review in Section 2.3, provide an automatic MC phase, including validating ( Figure 1). The trained ML model is deployed. If a monitored ML model does not provide optimal performance, then it is retrained based on updated datasets.

ML Methods for Smart Sensor Creation
An ML sensor model can be developed based on neural network (NN) or non-NN algorithms [14]. The last ones include PCA, Random Forest (RF), support vector machine (SVM), and Decision Tree (DT). In contrast with non-NN methods, NN architectures can include various neurons which are specified by ONNX [17], highly effective learning, and extracting features. A deep neural learning/network (DL/DNN), such as a recurrent neural network (RNN), convolutional neural network (CNN), and transformers, is part of the ML methods with feature learning that use multiple layers, complex connectivity architectures, and different transfer operators to automatically mine meta features from the input. NNs, such as artificial neural networks (ANNs), radial basis function neural networks (RBF-NNs), generative adversarial networks (GANs), RNNs, and CNNs have recently made major progress in practical applications of solar energy [1]. Figure 2 shows two NN methods' classes and the ML method groups according to the task they solved for a solar plant system [2]. The ensemble's types are bagging, boosting, and stacking/blending [18,19]. Table 1 presents the comparison of ensemble techniques [18]. There are constant and dynamic weighting ensemble approaches. In recent studies, the most used ensemble methods are RF, Extreme Gradient Boosting (XGBoost), Extreme Learning Machine (ELM), etc.
Model training methods that optimize performance include quasi-Newton, stochastic gradient descent (SGD), evolutionary computation, genetic programming, etc. [15]. The creation of the ML model is the most complex and important task which includes the creation of an optimal ML model's architecture and requires a multidimensional global optimization (GO).
The bias and variance estimate the effectiveness of a model. The improvement of a model's bias always makes gains at the expense of variance and vice versa. The performance of ML models highly correlates with the representativeness of a dataset. A lot of techniques provide a model's evaluation, including cross-validation, kfold, holdout with a different performance including accuracy (ACC), mean squared error (MSE), precision, receiver operating characteristics (ROC), recall, Matthew's correlation coefficient (MCC), F1, area under the curve (AUC), mean absolute error (MAE), and root-MSE (RMSE). The relative errors, such as normalized RMSE (nRMSE), normalized MAE (nMAE), etc., facilitate the comparison between models that are tuned based on datasets with different scales.
With the goal to develop intelligent models for sizing, forecasting, and control of a solar plant system and to make an RNN more adaptive with regard to a task's complexity and overfitting problem, we developed an MFNN [5][6][7][8]. The MFNN includes RNNs with fuzzy units and/or a convolutional block to process images. An RNN approximates a membership function in contrast to an Adaptive Network-Based Fuzzy Inference System (ANFIS). We combined the modified multidimensional quantum-behaved particle swarm optimization (PSO) with the Levenberg-Marquardt algorithm (MD QPSO) and developed a hierarchical encoder of the particle's dimension component [5][6][7][8] to automatically create an optimal architecture of an MFNN and improve the convergence. Provides the optimal combination of base learners, reduces variance, and bias [18] In the case of huge datasets, the computational time increases sufficiently as each classifier is working independently on the huge dataset.
We implemented an MFNN and its life cycle, which includes automatic creation and self-adaptation as an intelligent framework based on the authors' software [20]. This intelligent framework provides the automatic creation of the optimum architecture of an MFNN with regard to a task's complexity.
All the above-mentioned ML methods and algorithms were implemented as software by an ML Framework, which represents a tool to create a smart sensor system.

ML Frameworks
ML frameworks implement many ML methods [15]. Table 2 shows the comparison of popular ML frameworks.
Big data ecosystems, namely Apache Flink, Apache Spark, and Cloudera Oryx 2, include built-in ML libraries for large-scale DM. These ML libraries evolve presently, but the potency of the entire ecosystem is significant.
Chainer, Theano, Deeplearning4, and H2O are also appropriate DL libraries and frameworks for smart sensor system creation.
The high-level DL wrapper libraries such as Keras, TensorLayer, and Gluon are developed on top of the DL frameworks. They provide a simpler but more computationally expensive way for smart sensor system creation.
The ML frameworks provide an automatic MC phase of an ML model, including validating ( Figure 1). An ML sensor system can be implemented as software based on an ML framework that supports ONNX. Such implementation will provide flexibility and all an ML framework's advantages for a developed ML sensor system.

Open Resources for ML Research in a Solar Plant System
The open solar energy data sources, including big data, provide the development of cutting-edge ML technologies in solar energy.

Machine Learning Applications for a Solar Plant System
This section presents a review of research studies that have been published mostly in the last five years on the topic of ML applications for a solar plant. The literature review process elaborates on the articles' search queries in Scopus/ScienceDirect, IEEEXplore, ResearchGate, and Google Scholar with the following keywords: machine learning, neural networks, DL, PV, and solar plant. We focused on four important tasks' categories in the solar plant systems, as shown in Figure 3: design, forecasting, maintenance, and control. We are persuaded that tasks of these categories are most in demand in solar plant systems where ML can be applied with high efficiency. Figure 3 identifies the number of publications devoted to ML for a solar plant's design, forecasting, maintenance, and control that have been published mostly during the last five years. We prepared the data based on the considerable contributions from the most cited journals. We have not covered cybersecurity in a solar plant system since it was covered in-depth in study [1].   Figure 4 also presents the various types of feature spaces-to create a smart sensor system based on an ML method. It specifies the essential preprocessing and ML models to create a smart sensor [14].
The researchers in [10] noted that a pipeline implementation of an ML system is demanded. Therefore, we proposed a simple but effective pipeline scheme of an implementation (implementation step in Figure 1) of an ML sensor system for a solar plant. Figure 5 shows this simple scheme of a solar plant system based on ML technologies for a solar plant's design, forecasting, maintenance, and control. The center of a solar plant controller controls all devices and data of the solar plant and congregates datasets from the sensors, meteorological stations, and inverters [3]. The SCADA system is able to integrate sensor systems and ML technologies into an ML sensor system based on software that implements ML sensor models and integrates with SCADA through API. This software through API can transmit a control signal which is generated by an ML sensor model to a solar plant controller [14]. These ML sensor models for a solar plant's design, forecasting, maintenance, and control are implementations of a basic ML model class which is represented in Figure 5 as a UML class diagram. A method "Train" of a basic ML model class implements the MC phase, including validating. Thus, the impact of the ML technologies based on the proposed scheme ( Figure 5) on a solar plant value chain will mostly be associated with the cost of software development (including API development and the developed software's integration with SCADA) and maintenance. This developed software implements an ML sensor system based on an ML framework that supports ONNX. Most ML systems, which we review in subsections of Section 3, can be implemented on a solar plant based on the proposed scheme. Such implementation will provide flexibility and all ML framework's advantages for the developed ML sensor system and its digital transformation into a smart sensor system which we outlined in Section 4.

ML Technologies for Design of the Solar Plants
The optimal design of a solar plant is a very complex task that requires the fulfillment of models for a solar plant's components as well as the usage of global optimizers.

Parameter Identification in a Solar Plant System
The parameter extraction models for the single (SDM), double (DDM), or triple diode solar cell model (TDM) with RMSE as the performance metric are highly demanded for simulation and fault detection of a solar plant system.

SDM Low
Chaos Game Optimization (CGO) [43] W, TG, MD, applied chaotic reproduction optimization (CARO) [41], modified simplified swarm optimization algorithm (MSSO) [48], Cuckoo search algorithm (CSA) [49], Biogeography optimization algorithm-based heterogeneous cuckoo search (BBO-HCS) algorithm [50] SDM Low In [45], the parameter identification models for 17 different industrial solar cells/modules are reported. The hybrid bee pollinator flower pollination algorithm (BPFPA) [46] has the lowest RMSE and highest convergence as compared to all 21 reviewed parameter identification metaheuristic algorithms. Table 5 summarizes the comparative results of papers [42,43,[45][46][47] to set benchmarks for the performance comparison of the parameter identification models based on different metaheuristic algorithms for the 57 mm dia RTC France solar cell.  Summarizing, we highlight a need to assess more benchmarks for a performance comparison of the parameter identification models including ML methods.

Sizing of a Solar Plant
Within the research literature, a whole array of differing sizing methods for a solar plant has been proposed. These sizing methods of a solar plant are classified as intuitive, numerical, and analytical algorithms. The intuitive algorithms do not provide effectiveness and reliability. The numerical algorithms require a long time series of insolation. Many of the analytical algorithms use a concept of the system's reliability or loss of load probability. ML technologies provide an estimation of the optimal number of panels, storage capacity of batteries, tilt, and azimuth angles for a solar plant. Moreover, several ML technologies have been developed to size a solar plant. Table 6 shows a brief comparison of ML sizing methods of a solar plant [5,[51][52][53][54][55]. Two-year dataset of total insolation, meteorological parameters which was collected at the site of Abakan MAE% is 0.6% which is superior to PSO Automatic creation, self-adaptation MFNN based on the authors' software [20] ML optimization method based on ANN and heuristic optimizers [55] One-month datasets of meteorological parameters which were collected at the different climatic China regions The annual equivalent overall output energy increased by 4.48% as compared to a Taguchi standard orthogonal array Within the application of smart cities researchers design a renewable system that includes solar-to-electricity conversion.
Summarizing, we highlight a need to assess more benchmarks for a performance comparison of the PV sizing ML models. In addition, DL methods, including RNN, that extract knowledge from time series and effectively approximate insolation and load under small disturbances of a PV system dynamic, including degradation, are promising alternatives.

ML Technologies for Insolation and Power Forecasting of Solar Plants
Energy production of a solar plant is highly dependent on weather conditions such as insolation and temperature. Thus, it is difficult to balance the production and consumption of the electric grid with integrated solar plants where production levels fluctuate. In case of a deviation from an hourly plan schedule of solar plant power, the energy market charges penalties. Hence, many ML methods have been implemented to forecast insolation and the output power from a solar plant. Figure 6 presents specifics of the energy market to forecasting and classification of ML forecasting models based on a forecasting horizon [1,56]. The surveys of insolation and power forecasting of a solar plant in [57][58][59][60][61][62][63] appraise various approaches and methods to increase the performance of forecasting models under uncertainties. According to the reviews, ANNs are the most popular method for forecasting, as they are easy to implement and quite effective as compared to classical methods, such as conventional autoregressive integrated moving average (ARIMA), etc.

ML Technologies for Power Forecasting of Solar Plants
The power forecasting of a solar plant provides safety and effectiveness of grid control. There are mainly three ways to power forecast for a solar plant: • only historical output power recorded is used, • forecasted meteorological parameters are used as input, • combination of the historical power data with forecasted meteorological parameters is used.
Recent studies present the ML methods which effectively forecast a solar plant's power.
The simple (in [61], preprocessing generated normalized insolation; in [73], preprocessing elaborated k-means) and complex data preprocessing algorithms (in [71], four CNNs with different filters mine simple features from a sequence of time series; a single-kernel CNN mines the meta features from the simple features) provide for the ML model better performance (Table 7).
Due to forecast power, in [69,70], researchers integrated a PV-performance model into ML methods such as RF, SVR, CNN, LSTM, and hybrid CNN-LSTM. The results indicated that the proposed ML models provide the best performance regardless of the model's type and forecasting horizon (Table 7). Table 7 shows that indirect, very short-term forecasting ML models [61,67] provide higher accuracy as compared to direct ones. Table 7 shows that the dataset's length has a positive correlation with forecast performance (an average correlation coefficient of normalized corresponding columns is 0.34). Table 7 displays that the forecast horizon has a negative correlation with forecast performance (an average correlation coefficient of normalized corresponding columns is −0.31).

ML Technologies for Insolation Forecasting of the Solar Plants
ML technologies for insolation forecasting provide great benefits to smart grid integration and solar plant management. ML insolation forecasting is a necessary step for indirect power forecasting that provides higher accuracy as compared to a direct one. Thus, output of an insolation forecasting ML model can be used as an additional input signal for an indirect power forecasting ML model.
Summarizing, we highlight a need to assess more datasets and benchmarks for the performance comparison of ML technologies for insolation and solar plant power forecasting. The number of data preprocessing algorithms has a negative correlation with a forecast's performance. The dataset's length and forecast horizon have positive and negative correlation with a forecast's performance, correspondingly. A one-year test dataset is enough to create and validate a robust ML model. Indirect power forecasting provides higher accuracy as compared to a direct one. In addition, DL methods including transformers based on an attention mechanism that hierarchically preprocess and mine knowledge from datasets are promising alternatives.

ML Technologies for Maintenance of Solar Plants
ML methods solve the most complex tasks, which include failure classification, detection, localization, and automated solar panel diagnostics, based on solar plant sensor data ( Figure 4). Thus, grid operators can greatly increase the effectiveness and reliability of their solar plants based on ML technologies. ANN, FL, DT, RNN, RF, and different ensembles automatically detected basic solar plant faults based on data from ordinary sensors (Figure 4). DL and various types of CNN automatically perform analysis of infrared (IFR) images that are tracked by Unmanned Aerial Vehicles (UAVs). In this field of research, usually a dataset is highly unbalanced, i.e., it has unlabeled data and/or has rare failures. For this reason, the Balanced Accuracy, F1 score, Cohen's Kappa, or MCC better reflect the model's performance as compared to traditional accuracy metric.
Most of the ML models were created based on the dataset which was generated from simulation. A limited number of failure classes were considered, with the exception of a number of works in [91,92] in which 10 or more faults were considered (Table 9).

ML Technologies for Failure Diagnosis of the Solar Plants
According to study [93], there are six different categories of solar plant systems failures: shading, open-circuit, degradation, line-to-line, bypass diode, and bridging.
Frequent faults are failure in a component, system isolation, inverter shutdown, shading, and inverter MPP. In recent years, ML techniques that process data from ordinary sensors ( Figure 4) have been highly applied for fault classification and, in some cases, to identify the location of a failure.
In studies [99][100][101][102][103][104][105], researchers detect, classify, and localize [100] different failures of a solar plant system based on CNNs. For this purpose, researchers tuned CNNs based on the created dataset which sample represented a two-dimensional or three-dimensional transformation of data from ordinary sensors (Figure 4) namely, a scalogram [101], a two-dimensional time series graph [99], a three-dimensional image [103] and a polarcoordinate image [105]. This transformation can be simple (in [99], only PV current and voltage were composed into a two-dimensional time series graph) or complex (in [103], the direct current and alternating current values of a PV system were composed into a three-dimensional image based on a Gramian Angular Field; in [105], the time domain waveform signals were composed into a polar-coordinate image based on a symmetrized dot pattern (SDP)).
We proposed a failure forecasting system of a wiring losses' failure free operating period of a PV box based on an MFNN that has two RNNs with fuzzy units [5]. We created the MFNN based on a two-year historical dataset which included 20 kW PV array's signals. The developed fault forecasting system of the solar plant based on the tuned MFNN effectively forecasted a wiring losses' failure free operating period of a PV box. The relative error of the tuned MFNN was 0.0006.
In Table 9, we summarize the ML models for PV failure diagnosis from studies [91,92,[94][95][96][97][98][99][100][101][102][103][104][105].  Summarizing, we highlight a need for open datasets to assess experimental results on real testbeds and an open tool to generate and process scalograms based on transformers with an attention mechanism which feasibly outperforms other ML methods, such as CNNs. For failure detection and classification, there is a need to study the MPPT algorithms based on Reinforcement Learning (RL) and a spiking neural network under failure conditions.

ML Technologies for Solar Panel Diagnostics
The drop in solar plant productivity due to deviant maintenance modes caused by nonclean module surfaces, cell damage, delamination, or hot spots, demands a solar panel diagnostic based on the ML image sensors that process the panels' images ( Figure 4).
In studies [106][107][108][109], researchers localized and identified different failures of a solar plant system based on CNNs that process the solar panels' images, including thermographic images [106][107][108]. In Table 10, we summarize the ML technologies for PV diagnostics from studies [106][107][108][109][110][111][112].  DIP filters and SVM classifier [111] Classification into 10 different classes (1 healthy and 9 failure modes including warm module/substrings/cells, hot spot, etc.) The average accuracy on test dataset is 94.4% The thermographic images dataset that includes 16,000 samples (1600 for each class) VGG16 [112] Localization and classification into 6 different classes (1 healthy and 5 failure modes including overheated module/substrings hot spot, etc.) The mean F1-score is 94.52% Dataset of thermal infrared images was collected from 28 solar plants, which have 93220 solar panels Summarizing, we identify an opportunity to collect and make datasets available in which new ML models for solar plant diagnostics can be tested. In the reviewed studies, a considerable number of smart sensors process images almost perfectly. In the reviewed studies, the smart sensors where signals of image sensor and the CNN blocks strongly correlate provide high performance. There is an argumentative direction to substitute non-NN smart models with a DNN-based model for the solar plant's maintenance because DNN provides better information processing quality and performance as compared to non-NN smart models. In addition, ML methods such as GANs can be applied to generate artificial thermal images and create knowledge of the failure. Moreover, future research can comprise the elaboration of a pipeline for implementing a real time solar plant diagnostic system based on DNN or spiking neural network.

ML Technologies for Control of Solar Plants
The application of ML methods for the MPPT of solar plant systems has massive potential to increase their stability, reliability, dynamic response, and other essential advancements and easing their integration to electric grids.

ML MPPT Technologies of Solar Plants
The insolation and cell temperature of solar panels primarily define the total generated power by a solar plant. In the research reviews, a whole array of differing MPPT algorithms has been revealed [1,2,4]. Among them, the perturbation and observation (P and O) and incremental conductance (INC) algorithms are the most popular due to their easy and simple implementation. However, controllers which were created on the basis of these algorithms for solar plant systems have very bad speed of the response times, a long time to settle down from oscillating around the reference state. Furthermore, under PS, the MPPT task demands GO. Thus, traditional methods for MPPT do not provide global MPPT (GMPPT) and decrease efficiency in solar power production.
There are a lot of GO algorithms to create a GMPPT model [1,2,4], but all these models have the following disadvantages: power oscillations in the calm mode; the initialization is a critical issue that decrease power; very slow convergence to a GMPP under insolation's variation, etc. Due to all the above-mentioned disadvantages, GO-based, real-time GMPPT of a solar plant are ineffective while ML technologies provide the required performance.
In [114], researchers integrated the trained RL control agent into a fuzzy-logic-sliding mode control and incremental conductance-sliding mode control (RL FL INC) and gained better performance as compared to a classical RL agent (Table 11).
In [120], researchers created an MPPT controller based on a fuzzy logic search of variable voltage step size and fuzzy adaptive RBF-NN. The simulation results reflect the superiority of the developed MPPT controller as compared to the conventional P and O and RBF-NN.
In [121], we introduced the GMPPT system based on an MFNN that has five convolutional blocks to process the PV array's images, RNNs, and fuzzy units. Figure 7 shows the proposed GMPPT system based on an MFNN, where Im i is image of solar plant's modules; x i = V i , P i−1 , dI/ dV i and u i -input and output signal of MFNN, correspondingly; µ j -membership function of the fuzzy sets A j (A 1 is the rapidly increased uniform insolation, A 2 is nonuniform insolation); .z = indmax j (µ j ) = {j |∀k = j µ j ≥ µ k triggers the rule, which corresponds the z fuzzy set and RNN F z . The performance and control speed in GMPPT under PS of the created MFNN were superior as compared to the PSO and RNNs.  Summarizing, we highlight a demand for implementing more benchmarks for performance comparison of the real-time MPPT ML models based on ML frameworks, which we presented in Section 2.3. In addition, a real-time MPPT model based on a spiking neural network is a promising alternative.

ML Technologies for Control of Reconfigurable Solar Plants
The technology of reconfigurable PV arrays (rPV) by switching the electrical interconnection maximizes the generated PV array power in case of PS [127,128]. There are two classes of rPV: static and dynamic. Researchers proposed a lot of rPV's structures, including Honey Comb, Series Parallel, Total Cross Tied (TCT), etc. [127][128][129][130][131][132][133][134][135][136][137][138][139]. According to the articles [140][141][142], the last one generates more power in case of PS as compared to other structures. The GMMPT of an rPV array in case of PS represents a GO task.
The comparative analysis of recent rPV methods in [130] revealed that a TCT rPV based on a Static Shade Dispersion Physical Array Relocation (SD-PAR) algorithm and Modified Harris Hawks Optimizer (MHHO) algorithm that generated a switching matrix generates more power under PS as compared to other methods. Although, all metaheuristic optimizers do not provide a GMMP in real time mode because of a slow convergence.
The goal of study [136] is a GMPPT of an rPV array based on the MFNN in a case of PS. We created an optimal MFNN based on the dataset that contains the 20 kW PV array's signals under PS including PV array images that were congregated at the town Abakan from 31 January 2018 through 31 December 2018. Figures 8 and 9 display the insolation of the four solar panels' groups for the time period 9:20 am 3 December 2018-9:21 am 3 December 2018. Figure 10 shows that the rPV system based on the MFNN outperforms an rPV system based on GA because last one does not provide GMPP in this case. Similarly, we evaluated the performances of the rPV system based on the MFNN and rPV system based on GA on 100 test samples from the time period 1 December 2018-31 December 2018. The comparative simulation results show the superiority in terms of robustness and control speed of the created intelligent rPV system under PS that provides on average 30% more energy as compared to a TCT rPV system based on GA.

TCT rPV based on Static Shade Dispersion Physical Array Relocation (SD-PAR) algorithm and Modified Harris
Hawks Optimizer (MHHO) [130] Disadvantage: GO-based, real-time GMPPT of a solar plant are ineffective because of the slow convergence Technology generates more power under PS as compared to other methods.
Reconfiguration methods based on a GA [129,131] Disadvantage: GO-based, real-time GMPPT of a solar plant are ineffective because of the slow convergence The simulation results in Simulink for TCT rPV revealed that the developed method increased power: by 16.68% and 6.8% in three PS scenarios as compared to the TCT and the Su Do Ku scheme [129]; in four PS scenarios as compared to TCT.
ANFIS and an OCS [132,133] -Created method provided faster GMMPT and an average of 21% more generated power as compared to the P and O algorithm Fuzzy controller [134] Disadvantage: the proposed scheme does not provide MPPT under dynamic PS due to constant threshold-based switching of a fuzzy controller.
MFNN that contains: a convolutional block, RNNs and fuzzy units [136] Advantage: MFNN is implemented by authors software [20]. The trained MFNN by processing of the signals from ordinary sensors and PV array's image creates the GMMP interconnection matrix and GMMP voltage in case of PS.
The results show the superiority of the created intelligent rPV system under PS in terms of robustness, control speed that provides on average 30% more energy, as compared to a TCT rPV system based on GA  Summarizing, we identify an opportunity to use RNN for rPV that provides a GMMP interconnection matrix and GMMP voltage under dynamic PS. Nevertheless, an rPV's payback period is about 20 years [127] solely in places where PS happens daily, or over the full year leastwise in the seasons where solar production is great.

Future Technologies for Smart Solar Energy
The long-term contribution, including increased capacity of solar energy, depends on solving the remaining tasks of grids integration, high costs, and low efficiency, mainly through the research and development of a smart solar plant system based on integration of cutting-edge technologies, including DNN [137][138][139][140][141][142][143][144][145][146][147]. To attain the smart optimization and high efficiency of solar energy, the cloud, big data, ML, EC, IoT, quantum, and sensor technologies need to be adaptively combined and implemented as smart grid, home, and city applications. Figure 11 reflects the overlapping integration of these technologies into a smart solar plant system. The integration of the above-mentioned cutting-edge technologies provides high efficiency of ML technologies for the solar plant's design, forecasting, maintenance, and control. Implementation of such cutting-edge ML technologies for the solar plant's design, forecasting, maintenance, and control provides digital transformation of solar energy into smart solar energy. These ML technologies are implementations of a basic ML model class which is represented on Figure 11 as an UML class diagram. Figure 11 shows a method "Add" of a basic ML model class. This method adds a quantum layer into a classical ML model to create a quantum ML model. This method can be implemented by an integrating framework (Pennylane) for quantum computer simulators [140]. A quantum-based solar plant failure detection model was developed in [141]. IoT ( Figure 11) provides an optimal solution to collect solar energy big data wirelessly ( Figure 11). In [137], the solution researchers integrated a solar plant failure detection ML model. Future research can comprise the elaboration of a pipeline for implementing a real time solar plant diagnostic system based on IoT, EC, and/or TinyML technologies [138]. In [139], researchers developed based on EC a lightweight ML real-time solar plant failure detection model. Recent cloud-based monitoring solutions were developed in [141,142]. Forthcoming ML technologies for solar energy will integrate cloud-based solutions in which these technologies take full benefits of ML parallelism, data parallelism, practically limitless big data and ML knowledge storage, and almost boundless parallel computational resources.
The most complex issue of a smart sensor system is the self-learning of a sensor system. The potential methods for smart sensor's adaptive learning are memristors and a spiking neural network [143]. In the future, a smart solar plant system will integrate a self-supervised learning ML model zoo [144] that provides optimum ML technologies for the solar plant's design, forecasting, maintenance, and control.
Within the EU COVID-19 strategic reply, the smart energy standards define a cloud platform specification for distributed solar big data ecosystem that will provide creation of effective ML technologies for smart solar energy. The open solar energy data sources, including big data, provide the development of cutting-edge ML technologies in solar energy. Therefore, more open datasets with real data from solar sensor systems should be shared with the research community.
The integration of the cloud, big data, ML, EC, IoT, quantum, and sensor technologies will provide high efficiency of ML technologies for the solar plant's design, forecasting, maintenance, and control. Implementation of these technologies for the solar plant's design, forecasting, maintenance, and control provides digital transformation of solar energy into smart solar energy. The integrated electric grids are becoming increasingly reliable and overall solar production costs are minimized.

Conclusions
We presented a structured (mostly in benchmark tables) review of the advances in ML technologies for the solar plant's design, forecasting, maintenance, and control where most of the reviewed articles were published within the last five years.
ML methods are key elements of smart sensor systems of solar plants because they automatically create smart models for the solar plant's design, forecasting, maintenance, and control and more effectively analyze exponentially growing big data as compared to traditional methods. In this review, we briefly summarized our self-adaptive models for sizing, forecasting, maintenance, and control of a solar plant based on an MFNN that were automatically created with regard to a task's complexity and overfitting problem.
In the reviewed studies, the smart sensors where signals of image sensor and the CNN blocks strongly correlate provide high performance. There is an argumentative direction to substitute non-NN smart models with a DNN-based model for the solar plant's design, forecasting, maintenance, and control because DNN provides better information processing quality and performance as compared to non-NN smart models. The impact of the ML technologies based on the proposed implementation scheme on a solar plant value chain will mostly be associated with the cost of software development which implements a ML sensor system based on ONNX, a developed software's integration with SCADA, and maintenance.
The most complex issue of a smart sensor system is the self-learning of a sensor system. The potential methods for adaptive sensor learning are memristors and a spiking neural network.
In addition, we have outlined several problems that can be considered for future research in field of smart solar energy:

•
In forecasting and failure detection, the usage of the DNNs such as transformers based on an attention mechanism is a promising alternative.

•
For failure detection and classification, there is a need to study the MPPT algorithms based on RL and a spiking neural network under failure conditions. • For diagnosis of a solar plant system based on thermal images, the usage of GANs is a promising alternative.

•
There is a need to propose a pipeline for implementing a real-time solar plant diagnostic system based on IoT, EC, and/or TinyML technologies.

•
The development of ML algorithms for real-time processing and decision making are most in demand in solar plant systems.
The long-term contribution, including increased capacity of solar energy, depends on solving the remaining tasks of coupling to electric grids, high costs, and low efficiency, mainly through the research and development of a smart solar plant system based on the integration of cutting-edge technologies, including DNN. Within the EU COVID-19 strategic reply, the smart energy standards define a cloud platform specification for a distributed smart solar big data ecosystem that will provide the creation of effective ML technologies for smart solar energy. The open solar energy data sources, including big data, provide the development of cutting-edge ML technologies in solar energy. Therefore, more open datasets with real data from solar plant sensor systems should be shared with the research community. In order to achieve the smart optimization and high efficiency of solar energy, the cloud, big data, ML, EC, IoT, quantum, and sensor technologies need to be adaptively combined and implemented as smart grid, home, and city applications. The integration of the above-mentioned cutting-edge technologies will provide high efficiency of ML technologies for the solar plant's design, forecasting, maintenance, and control. Implementation of these technologies for the solar plant's design, forecasting, maintenance, and control will provide digital transformation of solar energy into smart solar energy. The integrated electric grids are becoming increasingly reliable, and overall solar production costs are minimized. Forthcoming ML technologies for solar energy will integrate cloudbased solutions, in which these technologies take full benefits of the ML parallelism, data parallelism, practically limitless big data and ML knowledge storage, and almost boundless parallel computational resources.  Acknowledgments: The reported study was fulfilled according to the research project "Development and implementation as software of an intelligent system for indirect forecasting power generation from solar plant based on the modified fuzzy neural network" during the activity "Development of intelligent systems for forecasting and maximizing power generation based on the original modified fuzzy neural network, their implementation as software and the implementation at a renewable power plant" within the program of the Worldclass Scientific Educational Center "Yenisei Siberia".

Conflicts of Interest:
The authors declare no conflict of interest.