Degree

Doctor of Philosophy (PhD)

Department

Systems Engineering

Document Type

Dissertation

Abstract

Floating photovoltaic (FPV) systems have become a transformative renewable energy technology because of their cooling effects on PV performance and ability to prevent water evaporation in land-constrained areas. Although FPV systems have the potential to become a commercially viable technology, their large-scale deployment remains constrained by uncertainties in thermal behavior, sustainability, and grid-operational variability. This dissertation identifies and characterizes these three key issues and presents an integrated, measurement-based evaluation of a 130 kW FPV installation located at the Passaúna reservoir in Brazil. In the first contribution, four temperature models, including physical and empirical models, were developed and comparatively evaluated to predict the FPV module temperature. All models performed substantially better than existing models, with Root Mean Squared Error (RMSE) ranging from 2.73°C to 3.21°C, outperforming the previously proposed empirical temperature model (RMSE 4.49°C). Importantly, it is shown that there exists a water-temperature driven seasonal component to the temperature of FPV modules in the standard approaches. The second contribution translated these thermal advantages into sustainability values by developing a ‘7E’ framework and a novel Floating Equivalence Ratio (FER). In the Passaúna system, an FER of 1.20 and 4.42% increase in energy production was achieved compared to traditional ground-mounted photovoltaic (PV) systems. Also, 81,459 m³ of water and 257.7 tons of CO2 could be saved in the process over 30 years, and no negative ecological impacts were observed. Finally, the operational challenge of the grid integration of FPV was addressed through a ramp-aware ultra-short-term FPV power forecasting framework. A ramp-aware forecasting framework, KAN-HiTS (DILATE), was developed by combining the Kolmogorov-Arnold Network (KAN) spline layer within the hierarchical temporal interpolation block and trained on an alignment-aware loss function. Improvements in ramp detection were observed, with the lowest RMSE of 12.52 kW and highest recall of 0.372. Additionally, Mondrian conformal prediction integrated with Adaptive Conformal Inference (ACI) has been proposed to generate calibrated conditional uncertainty intervals for probabilistic power forecasting and uncertainty quantification. Together, these three contributions present a comprehensive approach to enhancing the performance, assessment, and integration of large-scale FPV systems, thereby facilitating the wider deployment of FPV systems in the water-energy nexus.

Date

2-6-2026

DOI

https://proquest.com/docview/3347951617

First Committee Chair

Terrence Chambers

Second Committee Chair

Afef Fekih

First Committee Member

Farzad Ferdowsi

Second Committee Member

Giovana Wiecheteck

Third Committee Member

Raju Gottumukkala

Fourth Committee Member

Zhongqi Pan

Download

Included in

Engineering Commons

Share

COinS