Optimization of a hybrid photovoltaic/thermal-ground source heat pump system using artificial neural networks and multi-objective genetic algorithms

The current study focuses on determining the optimal operating conditions of a cycle using multiobjective genetic algorithms; this cycle harnesses energy from the sun and the ground to supply a multi-zone and large sports and administrative complex. The cycle is designed to incorporate various control scenarios; numerous decision variables, some of which are related to these control scenarios, are utilized to optimize the performance of the cycle. COP, ground temperature variation over a year, and reduced CO2 emissions are the objective functions of the optimization problem, which is divided into two distinct problems. In both problems, there are 36 decision variables that inherently originate from four independent variables, which are modified over the months. So, the optimization process is time-consuming, and computations are reduced by developing surrogate models via artificial neural networks. Also, a total of 1,080 different artificial neural network architectures and training data sets, each with 36 inputs (corresponding to the decision variables of the optimization problems) and one output (representing one of the objective functions) are analyzed to identify the best-performing architectures using hyperparameter tuning. Then, the Pareto fronts of the mentioned optimization problems are extracted, proposing a number of optimal conditions. The highest and the lowest amounts of the coefficient of performance on the Pareto fronts of the problems are 2.28 and 3.02, respectively. The highest possible amount is 41.15%, 7.35 %, and 7.27% more than three baseline scenarios. Among all the suggested conditions on the Pareto fronts, ten conditions (including the Knee points) are extensively studied.