International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 08 | Aug 2024
p-ISSN: 2395-0072
www.irjet.net
Predictive Maintenance of Electrical Grid Assets Using Machine Learning Sri Sai Tarun Kurapati1, Dr Chow Siing Sia2 1MSc. Advanced Computer Science, Cardiff Metropolitan University, Cardiff, United Kingdom 2 Professor, Dept. of Computer Science, Cardiff Metropolitan University, Cardiff, United Kingdom
---------------------------------------------------------------------***--------------------------------------------------------------------plays a crucial role in maintaining the overall health and Abstract - Predictive maintenance (PdM) is a technology efficiency of the electrical grid.
that can make electric grid assets more reliable and efficient. It uses machine learning to find possible problems before they happen, so repairs can be done on time and preventive actions can be taken. This helps reduce power outages, makes things safer, and saves money. In a study, we used a dataset from the UCI Machine Learning Repository to check how well six different machine learning models work for electric grid assets. We trained these models using 12 features like voltage, current, and temperature measurements. The results showed that the tuned gradient boosting model performed the best, with a very high accuracy of 99.2%. This thesis suggests that PdM with machine learning is a promising way to improve the reliability and efficiency of electric grid assets. However, there are still some challenges to overcome, like getting large and accurate datasets. As technology advances, we can expect even better models to be developed in the future...
Traditionally, electrical grid assets have been subject to maintenance based on either a time-based or conditionbased approach. However, these methods are often inefficient, leading to avoidable periods of inoperability and increased financial outlays [Yang & Li, 2019]. The application of predictive maintenance, which employs data-driven techniques including Measurement-based Methods (MMs) and machine learning (ML)-based approaches, offers the potential for significant reductions in maintenance costs and downtime, along with notable improvements in the dependability and operational efficiency of the electrical grid. 1.1 Research Background: The integration of machine learning methods in predictive maintenance has demonstrated promising results across diverse industrial sectors. However, the application of these methodologies to the electrical grid presents unique challenges due to its complex structure and the massive scale of data that requires processing. A prominent hurdle in predictive maintenance for electrical grid assets is the seamless integration of heterogeneous data from multiple origins. Furthermore, the creation of accurate and reliable predictive models constitutes an additional obstacle.
Key Words: Predictive Maintenance, Machine Learning, Electrical Grid, Gradient Boosting, Reliability, Efficiency
1.INTRODUCTION The electric power grid, often described as the most gigantic engineering feat ever built, is facing a quantum leap to an even more complicated structure. This transformation is driven by the increased integration of heterogeneous Renewable Energy Sources (RES) and everincreasing load demand [Zhang et al., 2022]. Consequently, the Electrical Power System (EPS) is beginning to operate quite close to its stability boundary. The reason for this is that RES and load consumption behavior are characterized by high intermittency, which may compromise power systems' stability. The uncertainty and uncontrollability of RES make maintaining Power Grid Stability (PGS) a challenging issue, presenting one of the fundamental concerns for futuristic grid systems.
1.2 Problem Statement: The electrical assets that constitute the power grid require periodic maintenance to ensure their peak operational efficiency and mitigate the occurrence of failures that can result in significant downtime and pose potential safety hazards. Conventional maintenance methodologies are often inefficient and can lead to high maintenance costs. The design of effective Measurement-based Methods (MMs) is a complex and challenging task [Wang et al., 2021]. The increased unpredictability of states of instability due to complex operational conditions represents a significant limitation for point forecasting [Liu & Wang, 2020].
PGS can be classified into several categories, including voltage stability, transient stability, frequency stability, rotor angle stability, resonance stability, and converterdriven stability [Chen & Wang, 2021; Liu & Wang, 2020; Yang & Li, 2019; Wang et al., 2021]. Each of these aspects
© 2024, IRJET
|
Impact Factor value: 8.226
1.3 Research Aim & Objectives: The main aim of this research is to create a predictive algorithm that incorporates machine learning methodologies to cater to the needs of electrical grid assets. The objectives of this study are:
|
ISO 9001:2008 Certified Journal
|
Page 99