Comparative analysis of artificial neural network‐based predictive models for hybrid compound parabolic concentrator ‐ Photovoltaic thermal—Thermoelectric generator collector across multiple performance metrics

Abstract This study presents the predictive performance evaluation of a hybrid CPC‐PVT‐TEG (Compound Parabolic Concentrator – Photovoltaic Thermal – Thermoelectric Generator) collector under Delhi's climatic conditions using 3 artificial neural network (ANN) models: Feedforward Neural Network (FFN), Multilayer Perceptron (MLP), and Radial Basis Function (RBF). The system consists of a CPC (concentration ratio = 3), a c‐Si based PV module thermally coupled with a thermoelectric generator (TEG) comprising 255 thermocouples ( N = 255), and fluid ducts carrying water under constant mass flow condition (0.003 kg/s) at the TEG's cold end. Input parameters include useful solar radiation (I U ) and ambient temperature (T a ), while 6 key output parameters—electrical efficiency (η HC ), electrical energy (E HC ), thermal efficiency (η TH ), thermal exergy (E XC ), overall exergy (E EX ), and exergy efficiency (η EX ) are predicted. A dataset of 480 hourly samples covering 4‐day types (Type A, B, C, and D) was generated using thermal modeling with 360 samples used for training and 120 for testing. Model evaluation revealed that MLP provided the highest short‐term accuracy (overall MAPE = 0.3177%, η TH MAPE = 0.0231%, accuracy = 0.9997), while RBF performed best in long‐term predictions (APE: E HC,annual = 1.53%, E EX,annual = 1.63%). FFN showed strong intermediate performance (RMSE = 0.0115, R = 0.9999).