Publication Details
Abstract
One of the major obstacles in solar thermal system design is the temporal mismatch between solar energy availability and thermal demand, especially at high-irradiance places in arid regions. Phase Change Materials (PCMs) provide high energy-density storage at near-isothermal conditions; however, the inherent non-linear enthalpy dynamics in PCMs and low thermal conductivity makes using traditional controllers based on Proportional-Integral-Derivative (PID) principles reactive and thermodynamically inefficient. We propose and empirically validate a novel integrated MPC-ML framework by fusing a one-step-ahead Random Forest (RF) machine learning model for solar irradiance prediction with an online receding-horizon Model Predictive Control (MPC) strategy applied to a lumped-parameter phase-change material (PCM) storage system. The RF model uses a single year (2020) of PVGIS hourly data at the typical high-irradiance coordinate point (Lat: 33.3152°N, Lon: 44.3661°E). A 30% chunk of data sits aside to check how well the model performs, judged by RMSE, MAE, and R². Through a three-day stretch in summer, sunlight conditions run in three versions: normal cloud-free skies, one with light cut down to 40%, while another plays out random clouds drifting through. That model gave an error score of 47.3 W/m² along with 31.6 W/m² average deviation, hitting a match level near 0.964 when checked on unseen data. When skies stayed clear, the new method cut heat loss signals down - dropping from 2.125 to 1.479 kWh - while keeping stored energy somewhat below normal levels. Cloudy shifts brought harder tests; even then, losses fell by 9.2 percent under weak light, nearly a quarter during flickering sun patterns. Demand gaps barely showed up at all, peaking at just over 0.059 kWh when weather turned messiest. The Choir and Orchestra of them proposed MPC-ML framework can statistically consistently achieve efficiency benefits over standard rule-based control for different irradiance conditions, indicating the approach is a viable, scalable solution to enhanced solar-PCM thermal storage performance. This fully reproducible open-data methodology serves as an actionable template for implementation in high-solar-resource environments.