An artificial intelligence approach to predict energy parameters in a photovoltaic-thermal system within a greenhouse

The ever-increasing energy demands in various agricultural sectors, especially in greenhouse facilities, require exploring feasible solutions. Utilizing renewable energy sources, along with implementing artificial intelligence (AI) to predict and analyze energy consumption data, offers a promising approach to tackle this challenge. In this research, various machine learning models are used to predict energy parameters (such as output power, electrical efficiency, thermal efficiency, and total efficiency) of a photovoltaic-thermal system based on nanofluids (Al₂O₃, SiO₂, Al₂O₃-SiO₂) both inside and outside a greenhouse environment. The modeling is carried out using time-delay neural networks (TDNN), multilayer perceptron (MLP), and nonlinear autoregression (NARX) methods, incorporating a logarithmic activation function. The results of the modeling for predicting different energy parameters indicate that the NARX network achieves the highest accuracy, with average statistical indicators of R² = 0.9979 and RMSE = 0.1062. In contrast, the MLP network shows the lowest accuracy, with average statistical indicators of R² = −0.1657 and RMSE = 3.4482. Furthermore, a comparison of the energy parameter modeling results shows that simulations conducted outside the greenhouse have better statistical indicators, with an average R² = 0.7038 and RMSE = 0.9358, compared to simulations conducted inside the greenhouse, which yielded an average R² = 0.5162 and RMSE = 1.5267. Additionally, an analysis of the convergence times for the different networks reveals that the MLP, TDNN, and NARX networks require average times of 0.4057 h, 37.3864 h, and 103.5006 h, respectively.