Optimization of a Solar Thermal Plant Integrated to an Industrial Process Using Machine Learning and Genetic Algorithms
Abstract
Given Chile’s abundant solar resource, solar thermal energy generation has great potential but must be economically viable. This work aims to develop and apply a Machine Learning–based model to optimize the design of solar thermal plants integrated into industrial processes. The methodology involves hyperparameter tuning, testing preprocessing methods, designing optimization algorithms, and validating results through new simulations. Neural Networks and Decision Trees were trained using two datasets, uniform and random, within defined ranges. Various preprocessing techniques were tested to find the most accurate Machine Learning model and Genetic Algorithms were applied for the optimization model. The results showed Standard Scaler preprocessing produced the lowest prediction errors, especially for Neural Networks trained with the uniform dataset, achieving a 0.12% error, Neural Networks outperformed Decision Trees, and uniform datasets showed stronger interpolation. The study concludes that multi-objective optimization with Neural Networks trained on uniform data yields errors below 2%, confirming their reliability for thermal plant design optimization.
Keywords
Solar Thermal Plant, TRNSYS-18, Optimization, Neural Networks, Genetic Algorithms