501 research outputs found
Short-term forecasting of power production in a large-scale photovoltaic plant
No full textIn this paper, a simple but accurate approach for short-term forecasting of the power produced by a Large-Scale Grid Connected
Photovoltaic Plant (LS-GCPV) is presented. A 1-year database of solar irradiance, cell temperature and power output produced by a
1-MWp photovoltaic plant located in Southern Italy is used for developing three distinct artificial neural network (ANN) models, to
be applied to three typical types of day (sunny, partly cloudy and overcast). The possibility of obtaining accurate results by using solely
the monitored data rather than knowing the actual architecture and details of the plant is a notable advantage; in particular, the method’s
reliability gives to operation and maintenance and to grid operators excellent confidence in the evaluation of the performance of the
plant and in the programming of the dispatching plans, respectively
TinyML for fault diagnosis of Photovoltaic Modules using Edge Impulse Platform
No full textIn this paper a fault diagnosis method for photovoltaic (PV) modules is developed using an open source Machine Learning (ML) platform (Edge impulse). The idea is to develop a TinyML to classify certain defects that can frequently occur on PV modules (e.g. dirty, degradation and dust deposit on PV modules), and then to integrate the impulse into an Edge device for real time application. In this regard a database of infrared thermography image was built and used. The model could be run locally without internet connection. This method could help users to diagnosis their PV modules and make decision about the maintenance schedule (cleaning or replacing of PV modules). Results clearly report the feasibility of the method with a mean accuracy of 93.4 %. The main advantage is that, thanks to this platform, embedded ML model could be developed quickly. Moreover, edge processes are not affected by the latency and bandwidth issues becoming outstanding methods for real-time diagnostics
Adaptive Neural Network-Based Control of a Hybrid AC/DC Microgrid
No full textIn this paper, the behavior of a grid-connected hybrid ac/dc microgrid has been investigated. Different renewable energy sources - photovoltaics modules and a wind turbine generator - have been considered together with a solid oxide fuel cell and a battery energy storage system. The main contribution of this paper is the design and the validation of an innovative online-trained artificial neural network-based control system for a hybrid microgrid. Adaptive neural networks are used to track the maximum power point of renewable energy generators and to control the power exchanged between the front-end converter and the electrical grid. Moreover, a fuzzy logic-based power management system is proposed in order to minimize the energy purchased from the electrical grid. The operation of the hybrid microgrid has been tested in the MATLAB/Simulink environment under different operating conditions. The obtained results demonstrate the effectiveness, the high robustness and the self-adaptation ability of the proposed control system
Hybrid CNN-EML model for fault diagnosis in Electroluminescence images of photovoltaic cells
No full textThe quality inspection of solar module manufacturing is essential to guarantee photovoltaic (PV) power plants' steady. This paper presents the development of an innovative hybrid model that combines convolutional neural networks (CNN) with ensemble machine learning (EML) algorithms. The integration of these approaches was employed in order to develop a ranking weight voting system to the features extracted by the CNN model, by combining three fundamental algorithms: Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF). The advanced CNN-Ensemble Machine Learning (CNN-EML) technique is applied to a dataset of electroluminescence (EL) images featuring the nine most important and frequent defects. The results demonstrated that techniques based on the CNN-EML provide superior classification accuracy, effectively addressing the challenge of diagnosing faults in PV module manufacturing. The CNN-EML model achieved a significant accuracy of 94% in classification of different defects, outperforming CNN algorithm-based methods in the proposed comparative analysis
A Novel Embedded System for Real-Time Fault Diagnosis of Photovoltaic Modules
No full textIn this article, a novel embedded low-cost system for real-time fault diagnosis of photovoltaic (PV) modules is proposed. The idea aims to develop an embedded application to classify certain defects that can frequently occur on PV modules based on infrared (IR) images in different regions (desert and Mediterranean climates). The investigated faults are sand accumulation, dirt on PV modules, degradation, and junction box overheating. After several inspections, these are the most commonly observed defects on PV modules in both regions (south and north of Algeria). A tiny convolutional neural network (TinyCNN) was developed, optimized, and integrated into a low-cost and low-power microcontroller (Arduino Nano 33 BLE sense). In this regard, a database of IR thermography images was built and used. The developed TinyCNN-based model could be run locally, without the need to send the data to the cloud for analysis and processing. Another microcontroller [Arduino Nano 33 Internet of Things (IoT)] was used to remotely monitor the state of the PV modules. Thanks to IoT technology, the results have been visualized and posted online on a dedicated monitoring webpage. The proposed embedded solution could be integrated into an unmanned aerial vehicle for real-time applications. Furthermore, it assists operators in diagnosing their PV modules and making a maintenance schedule. The proposed technique outperforms the existing solutions in terms of cost, consuming power, simplicity, and execution time. Simulation and experimental results clearly report the feasibility of the proposed embedded system, which has an average cost of around 120 US dollars
Performance prediction of 20kWp grid-connected photovoltaicplant at Trieste (Italy) using artificial neural network
No full textThe Photovoltaic Laboratory at the University of Trieste, Italy
No full textThis work deals with the description of the Photovoltaic Laboratory at the Department of Engineering and Architecture of the University of Trieste. The description of the main facilities involved, such as three grid-connected photovoltaic plants and a test facility for the characterization of photovoltaic modules, is being proposed. Moreover, an overview of the recent and future research activities is given
Evaluation of the Multi-Year-on-Year method for degradation analysis in photovoltaic systems
No full textAdvanced Methods for Photovoltaic Output Power Forecasting: A Review
No full textForecasting is a crucial task for successfully integrating photovoltaic (PV) output power into the grid. The design of accurate photovoltaic output forecasters remains a challenging issue, particularly for multistep-ahead prediction. Accurate PV output power forecasting is critical in a number of applications, such as micro-grids (MGs), energy optimization and management, PV integrated in smart buildings, and electrical vehicle chartering. Over the last decade, a vast literature has been produced on this topic, investigating numerical and probabilistic methods, physical models, and artificial intelligence (AI) techniques. This paper aims at providing a complete and critical review on the recent applications of AI techniques; we will focus particularly on machine learning (ML), deep learning (DL), and hybrid methods, as these branches of AI are becoming increasingly attractive. Special attention will be paid to the recent development of the application of DL, as well as to the future trends in this topic
Photovoltaic Plant Output Power Forecast by Means of Hybrid Artificial Neural Networks
No full textThe main goal of this chapter is to show the set up a well-defined method to identify and properly train the hybrid artificial neural network both in terms of number of neurons, hidden layers and training set size in order to perform the day-ahead power production forecast applicable to any photovoltaic (PV) plant, accurately. Therefore, this chapter has been addressed to describe the adopted hybrid method (PHANN—Physic Hybrid Artificial Neural Network) combining both the deterministic clear sky solar radiation algorithm (CSRM) and the stochastic artificial neural network (ANN) method in order to enhance the day-ahead power forecast. In the previous works, this hybrid method had been tested on different PV plants by assessing the role of different training sets varying in the amount of data and number of trials, which should be included in the “ensemble forecast.” In this chapter, the main results obtained by applying the above-mentioned procedure specifically referred to the available data of the PV power production of a single PV module are presented
- …
