Experimental validation and enhanced thermal prediction of a shallow solar pond using artificial neural network–based model predictive control for real-time optimization under multiple heat extraction modes

This study presents the experimental validation and enhanced thermal modeling of a Shallow Solar Pond (SSP) operating under three distinct heat extraction modes: direct, open cycle, and closed cycle. A custom-designed SSP, insulated and equipped with double-glazing and a PVC heat exchanger and black-painted bottom, was monitored under clear-sky conditions. 14 Artificial Neural Network (ANN) configurations were trained, with the optimal Bayesian Regularization-based model (two hidden layers, 15 neurons) achieving exceptional predictive accuracy (R 2 = 0.99919, RMSE = 32.7580%, MRE = 0.62%) for temperatures across all system components and operational modes. The ANN model effectively captured complex nonlinear dynamics, including transient heating and cooling phases. The integration of this ANN with a Model Predictive Control (MPC) framework enabled real-time optimization of the fluid flow rate in the closed-cycle mode, which was identified as the most efficient configuration. The MPC-ANN strategy reduced the outlet temperature tracking error by over 50% (MAE = 1.42°C vs. 2.97°C without MPC), demonstrating superior thermal stability and adaptive control under fluctuating solar irradiance. This work confirms the closed-cycle SSP's superiority for stable energy delivery and establishes a robust ANN-MPC framework for real-time operational optimization of solar thermal systems. • The thermal performance of a shallow solar pond has been accurately predicted using artificial neural networks (ANNs). • Various heat extraction modes have been thoroughly studied and analyzed to evaluate their efficiency. • Data collection for the artificial neural network model was conducted through experimental measurements. • The optimization of a shallow solar pond has been extensively examined using artificial neural networks.