Presentation Details
| Forecasting solar panel soiling recovery using environmental data during rain events in Africa (yes) Grace S.W.Liu, Brendan Wright, Ali Shakiba, Abhnil Prasad, Ziv Hameiri. University of New South Wales, Kensington, Australia |
Abstract
Generation loss from soiling has been estimated to cost the photovoltaic industry billions of dollars globally. Therefore, it is essential to account for soiling in financial analyses. Since directly measuring soiling is expensive and often unreliable, several soiling models have been developed to estimate the soiling losses based on particulate matter concentrations and precipitation levels. However, most of these models are highly simplified and assume complete cleaning when precipitation exceeds a certain rainfall threshold. This study addresses this gap by training and testing models that consider a wide range of environmental parameters—such as precipitation, humidity, wind, and temperature—to quantify panel recovery from soiling during rainfall events. The statistical and machine learning models that were assessed and compared included logistic regression, multiple linear regression, and random forest. While all three methods outperformed the baseline model, random forest achieved the best performance. Random forest was further analysed using explainable artificial intelligence to dissect and gain deeper insight into its decision-making process.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.