Development of a computational tool based on Artificial Intelligence for the performance prediction of a solar energy system

Accurate prediction of photovoltaic module performance under real operating conditions benefits from frameworks that connect field degradation evidence, internal electrical behavior, operating-temperature information, and long-term uncertainty under site-specific conditions. This doctoral thesis develops a site-specific and uncertainty-aware predictive framework by combining field current--voltage measurements, double-diode model parameter extraction, photovoltaic cell-temperature estimation, and field-informed degradation projection.

The first contribution quantifies the degradation of polycrystalline silicon photovoltaic modules after 10 years of field exposure in the Atacama Desert. STC-corrected current--voltage measurements showed maximum-power degradation rates between 0.87\% and 1.58\%/year, with projected useful lifetimes of approximately 13 to 23 years at the 80\% power-retention threshold. The second contribution implements PVMSim, a MATLAB-based workflow for traceable and reproducible double-diode parameter extraction from measured current--voltage curves. The third contribution develops a site-specific thermal model for photovoltaic cell-temperature estimation under tropical operating conditions, supporting the thermal-modeling layer of the thesis. The final contribution proposes a hybrid circuit-based and data-driven degradation framework that integrates field-derived aging descriptors, double-diode modeling, machine-learning emulation, Monte Carlo uncertainty propagation, measured irradiance and module-temperature records, and annual energy estimation.

Under the correlated equivalent module-level degradation scenario, maximum power decreased from 279.5~W at year 0 to 204.6~W at year 20 and 183.8~W at year 25. The first-order probability of falling below the 80\% power-retention threshold reached 99.6\% at year 20 and 100\% at year 25 under the adopted uncertainty model. The median annual energy projection decreased from 747.4~kWh for the non-aged reference condition to 540.1~kWh at year 20 and 483.9~kWh at year 25. These projections are scenario-based estimates conditioned on the available field data, the equivalent-circuit representation, and the adopted uncertainty model. Overall, the thesis supports the hypothesis that uncertainty-aware PV performance prediction can be strengthened by integrating operating-temperature estimation, equivalent-circuit parameter identification, and time-dependent degradation modeling.