A Machine Learning Approach for Fault Detection and Diagnosis in Four-legged Inverters

Peykarporsan, Rasool; Heidary, Jalal; Oshnoei, Soroush; Lie, Tek Tjing

doi:10.1109/kpec58008.2023.10215460

Cited by 3 publications

(1 citation statement)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The resulting data is processed by a Random Forest (RF) classifier to detect if there is a fault or not. In the case of inverters, statistical models are not only used to detect faults once they are operative, as previously commented in [9] or in [10], these techniques can be used also for conducting reliability analysis of power converters during the manufacturing process [11]. In [12] an Artificial Neural Network (ANN) model, applied to photovoltaic fault detection, is used to assess the power production and compare it with the power production trend of the plant.…”

Section: Introductionmentioning

confidence: 99%

Detection of Abnormal Operation of PV Inverters Based on Regressive Prediction Models With Recursive Least Squares Training

Moreno,

Laguna,

Cipriano

et al. 2024

IEEE J. Emerg. Sel. Top. Ind. Electron.

View full text Add to dashboard Cite

The large scale integration of photovoltaic (PV) power plants has launched the massive deployment of PV inverters. In fact, just a single multi-MW PV plant may have thousands of them, which can be also found isolated in low scale distributed generation applications. The number of grid-connected assets can be no longer managed and maintained effectively without using AI tools, able to analyze their operation, detect faults and support decision-making maintenance tools. In this paper, a Linear Regression method able to detect abnormal operation in PV systems, based on a Recursive Least Squares (RLS) training algorithm, which requires a low amount of data, mainly energy generation measurements and meteorological data, is proposed. In addition, two different applications of this methodology will be presented, one based on issuing a simplified model for realtime analysis, and another one consisting of a complex model for long-term diagnosis. The first one is focused on detecting faults and abnormal operation profiles in real-time, while the second one permits assessing the historical efficiency decay of PV plants in longer periods. Both will be used for detecting abnormal operation of PV inverters and panels. The performance and behavior of these algorithms will be tested using the data of 22 PV plants, placed at different climatic areas and with different peak powers. The results will show the good performance of the proposed fault detection method in both applications.

show abstract