Date of Award
2026-05-01
Degree Name
Master of Science
Department
Computer Science
Advisor(s)
Martine Ceberio
Abstract
The growing demand for resilient and sustainable energy generation has driven interest in Hybrid Floating Photovoltaic-Hydropower (HFPVH) systems. Operating these systems effectively requires making water release decisions that satisfy physical and regulatory constraints while maximizing energy production. Prior work by Vega (2024) developed a Dynamic Outlier Filter Long Short-Term Memory (DOF-LSTM) architecture for forecasting reservoir release patterns. That predictive work is valid and addresses an important component of the HFPVH decision pipeline. However, the constraint system in that work operates externally to the learning process, and the behavior of penalty-based constraint enforcement had not been studied independently in this context. This thesis contributes a systematic separation of the constraint handling problem from the predictive modeling problem. Rather than immediately integrating penalty methods into the LSTM training loop, we isolate and study how penalty-based constrained optimization behaves across a range of problem structures. This disassociation clarifies what AI and machine learning can contribute (prediction and pattern recognition) versus what penalty methods can add (constraint enforcement and feasibility guarantees), establishing a foundation applicable to future integration with the DOF-LSTM or other learning systems. Four penalty update strategies are designed and evaluated on benchmark problems drawn from state-of-the-art constrained optimization repositories. The validated penalty methods are then applied to a real hydropower reservoir optimization problem using operational data. A proof-of-concept meta-loop architecture is demonstrated in which the interval constraint solver RealPaver provides guaranteed feasible solutions that serve as a safety net for the penalty optimizer. Finally, the penalty strategies are integrated into the DOF-LSTM training process, where the progressive strategy simultaneously improves prediction accuracy and reduces constraint violation severity, demonstrating that constraint enforcement can act as a beneficial regularizer in neural network training.
Language
en
Provenance
Received from ProQuest
Copyright Date
2026-05
File Size
99 p.
File Format
application/pdf
Rights Holder
Edwin Horacio Trejo
Recommended Citation
Trejo, Edwin Horacio, "Penalty Approach To Constrained Optimization Problems In Water Reservoir And Energy Generation Management" (2026). Open Access Theses & Dissertations. 4813.
https://scholarworks.utep.edu/open_etd/4813