Scientists Use Machine Learning to Schedule Off-Grid Solar Module Cleaning

The results are relevant for non-governmental organizations, governments, and energy service companies to improve the maintenance level of their solar facilities

October 3, 2020


A group of scientists have developed a new low-cost machine learning technology to run scheduled cleaning of solar photovoltaic (PV) modules installed in remote locations that are off the grid.

The research team from France’s Sorbonne Université, the École Normale Supérieure de Rennes (ENS Rennes), and the University of Paris-Saclay, applied four machine learning algorithms on data collated from a remote PV system.

The algorithm helped attain a reliability of 95% with 3.5 MHz voltage, current, and temperature signals. These results would be useful for non-governmental organizations, governments, and energy service companies to improve the maintenance level of their PV facilities.

Losses incurred due to soiling remain a significant concern for remote power systems that rely on solar energy. Efficient power loss analysis is available for monitoring sizeable solar PV projects, which helps develop an optimal cleaning schedule. However, things are different for remote monitoring of standalone solar systems employed in rural and remote regions.

The process for large-scale solar PV projects relies on a costly and dirt-sensitive irradiance sensor. However, the team investigated the possibility of low-cost monitoring of cleaning interventions on solar modules during the daytime.

According to the researchers, it is helpful to know whether the soiling is regular and to assess whether it is necessary to carry out additional cleaning operations.

A classification task was formulated using a time window of temperature, voltage, and current measurements of the PV array, to identify the occurrence of a cleaning intervention automatically.

To achieve such a classification task, the team examined the machine learning tools based on logistic regression, support vector machines, and artificial neural networks. They also studied the influence of the temporal resolution of the signals and the feature extraction on the classification performance.

These experiments were conducted on a real dataset and showed promising results with a classification accuracy of up to 95%.

The cleaning of solar panels is essential for efficient power generation from a project. Dust particles, bird droppings, and other particulate materials lead to a decrease in energy generation, highlighting the importance of solar module cleaning.

Robotic cleaning of solar modules is gaining momentum in India amid rising water scarcity.

Recently, India’s Central Electronics Limited (CEL), a public sector enterprise under the Department of Scientific and Industrial Research, had issued an expression of interest to tie-up with start-up companies as a technology partner for waterless solar photovoltaic module cleaning systems.

Similarly, BHEL had also issued a tender for the supply of module cleaning systems for 50 MW of solar projects in Maharashtra.

Image credit: Blokleen Solar Private Limited

Rahul is a staff reporter at Mercom India. Before entering the world of renewables, Rahul was head of the Gujarat bureau for The Quint. He has also worked for DNA Ahmedabad and Ahmedabad Mirror. Hailing from a banking and finance background, Rahul has also worked for JP Morgan Chase and State Bank of India. More articles from Rahul Nair.