Gafai (2022). Comparative Performance Analysis of Machine Learning Algorithms for Predicting Output Power of Soiled PV Modules. Nigeria Journal of Engineering Science Research (NIJESR), 5(3): 39-50 Comparative Performance Analysis of Machine Learning Algorithms for Predicting Output Power of Soiled PV Modules Gafai, N.B Department of Electrical and Computer Engineering, Faculty of Engineering, Baze University, Abuja, Nigeria, najashi.gafai@bazeuniversity.edu.ng (09055020008) Keywords: Photovoltaic, Hammerstein-Weiner, ANFIS, Prediction, Soiling INTRODUCTION Due to its economic merit (quick price reduction) and environmental merit (emission), solar energy conversion devices have been urged to replace conventional non-sustainable energy gathering systems(Faskari et al., 2022). Solar photovoltaic devices are considered to be the most promising clean energy source among renewable energy technologies, owing to the continual growth in fossil fuel prices and their environmental impact (Badamasi et al., 2021; Faskari et al., 2022). Areas inside the worldwide Sun Belt with abundant solar irradiation appear to have a high concentration of air dust which has been recognized as the primary degrading element that has a negative impact on PV performance, particularly in the desert, arid, and semi-arid regions(Kazem and Chaichan, 2019). Nigerian Journal of Engineering Science Research (NIJESR). Vol. 5, Issue 3, pp.39-50, September, 2022 Copyright@ Department of Mechanical Engineering, Gen. Abdusalami Abubakar College of Engineering, Igbinedion University, Okada, Edo State, Nigeria. ISSN: 2636-7114 Journal Homepage: https://www.iuokada.edu.ng/journals/nijesr/ Abstract: Photovoltaic (PV) system soiling losses have been proven to vary between areas and time. These losses can add a lot of unpredictability to the output of a PV system. With a greater understanding and predictability of these losses, the variability of PV system output, operation, and maintenance costs associated with cleaning systems might be better understood. Due to the nonavailability of solar radiation measuring equipment at the meteorological stations, especially in developing countries, it is essential to deploy the usage of machine learning algorithms to predict the performance of PV systems under soiling conditions. In this work, a comparative performance analysis of two promising Techniques: Hammerstein-Weiner (H-W) and Adaptive Neuro-Fuzzy Inference System (ANFIS) is presented. Data used in this work was gathered from the University of Abuja's Faculty of Engineering, Nigeria. These two unique models are simple to use and are an upgrade to the existing models in the literature. The models' prediction accuracy were assessed using metrics such as the R and R 2 while the error analysis was conducted using MSE and Root Mean Square Error (RMSE). The results revealed the models developed in this work performed well in estimating the predicted power of a soiled PV module with minimal error. H-W model has a better performance in the training phase with R and R 2 having 0.9904 and 0.9809 respectively in training and testing respectively. However, the ANFIS model recorded a better performance of 0.9735 and 0.9477 for R and R 2 respectively in the testing phase. Manuscript History Received: 14/06/2022 Revised: 07/08/2022 Accepted: 15/09/2022 Published: 30/09/2022 39