MACHINE LEARNING AS A POWERFUL TOOL FOR PERFORMANCE PREDICTION AND OPTIMIZATION OF CONCENTRATED PHOTOVOLTAIC-THERMOELECTRIC SYSTEM

Abstract

Because there is a critical necessity to ensure the optimal operation of concentrated photovoltaic-thermoelectric (CPV-TE) systems, various optimization methods such as Paretosearch (PS), Multi-objective genetic algorithm (MOGA), and the hybrid Goal Attainment – Multi-objective genetic algorithm (GOAL-MOGA) are commonly employed. These approaches aim to enhance both the output power and energy efficiency of CPV-TE systems. By combining the Pareto fronts generated by MOGA and GOAL-MOGA, 19 distinct machine learning (ML) algorithms were trained. The findings demonstrate that the Artificial Neural Network (ANN) ML algorithm outperforms others, displaying an average prediction error of 0.0692% on the test dataset. In addition to its prediction capability, the ANN-based ML model can be viewed as an optimization model since it produces optimized outputs similar to those from MOGA and GOAL-MOGA. The ANN-based ML algorithm performs better when trained on a combined dataset from both MOGA and GOAL-MOGA compared to using either MOGA or GOAL-MOGA alone. To enhance the optimization capability of the ANN-based ML algorithm further, more Pareto fronts from other optimization techniques can be added.