A team of researchers from the University of Lisbon in Portugal has developed a groundbreaking machine learning algorithm that can accurately classify and predict inverter failures in utility-scale PV plants. By monitoring the inverter subsystems and analyzing data, the algorithm is able to send alarms when maximum and minimum values are reached, providing early warnings of potential failures.
Unlike traditional methods, which rely on manual inspection and analysis, this new algorithm utilizes machine learning techniques to categorize variables based on historic values. By identifying the types of failures through inverter errors and occurrences, the algorithm can determine a range of issues including grid faults, overvoltage, undervoltage, low voltage, output overload, and more.
To test the effectiveness of the approach, the researchers conducted experiments on two ground-mounted PV systems, both equipped with inverters from renowned German manufacturer SMA. By analyzing the variables of each inverter, the algorithm successfully identified different types of failures.
The data collected from the experiments were then processed using fine tree, medium tree, and coarse tree prediction models. These tree-based models employ a set of splitting rules to divide the feature space into smaller regions with similar response values, allowing for accurate predictions.
One of the key findings of the study was the algorithm’s ability to detect seasonal variations in inverter failures, enabling better reliability analysis. The researchers emphasized the importance of data-driven evaluation in classifying failure modes within inverter subsystems.
Furthermore, the researchers proposed a solution to protect inverters from inrush and overcurrent by incorporating clamp circuits to the resonant capacitance in parallel. This approach not only ensures high power-conversion efficiency but also regenerates the clamp current to the input voltage source.
The results of this study, presented in the publication “Machine Learning for Monitoring and Classification in Inverters from Solar Photovoltaic Energy Plants,” provide valuable insights into the application of machine learning in the field of renewable energy. The algorithm offers a cost-effective and efficient solution for identifying potential inverter failures, ultimately improving the overall performance and reliability of solar PV plants.
The source of the article is from the blog newyorkpostgazette.com