How Can a Neural Network Predict the Temperature of a Solar Collector?

A research team employed an artificial neural network (ANN) to estimate the outlet air temperature and thermal efficiency of an air solar collector. The collector featured porous absorber surfaces made from metallic scourers. Three different absorber configurations were tested, varying inlet air temperature, solar irradiance, airflow rate, and surface temperatures measured at four points. Experimental data fed the ANN, which learned to link operating conditions with thermal outcomes. After training, the network's predictions were compared to measured values: the correlation coefficient R reached 0.99987, nearly perfect, and the mean squared error (MSE) was 0.0901, very low. These results demonstrate that neural networks can become reliable tools for optimizing solar collectors, aiding the design of more efficient systems and better renewable energy management.

Artificial neural networks (ANNs) are computational models inspired by biological neurons. In this work, a feed‑forward multilayer perceptron (MLP) with sigmoid activation functions was used, trained via back‑propagation gradient descent. The dataset contained 1,200 samples, split 80% for training and 20% for validation. Data were normalized using min‑max scaling to ensure uniform variable ranges and reduce overfitting risk. Hyperparameters (hidden layer neurons, learning rate) were tuned through grid search, evaluated with 5‑fold cross‑validation. Results indicate that the ANN outperforms traditional multiple linear regression models, thanks to its ability to capture complex nonlinear relationships among irradiance, airflow, and thermal properties.