Conference Paper
Proceedings of BSA Conference 2024: Sixth Conference of IBPSA-Italy
Computational Cost Reduction of a Simulation-Based Optimization Process Through Machine Learning Methods: Neural Networks vs. Random Forest
Computational Cost Reduction of a Simulation-Based Optimization Process Through Machine Learning Methods: Neural Networks vs. Random Forest
Iuri Praça Verginio, Rafael de Paula Garcia, Mario Alves da Silva, Joyce Correna CarloDOI: https://doi.org/10.13124/9788860462022_04Abstract: Simulation-based optimization (SBO) processes are computationally expensive and the combination with machine learning (ML) methods appears as an alternative capable of reducing computational time consumption without losing the robustness of the solutions. This study compares neural network and random forest algorithms as approaches to replace simulations during the SBO processes. The main objective is to define the best machine learning algorithm and the most reliable ratio between simulations and predictions. The problem was implemented in the Grasshopper + Rhino platform and aimed to minimize the annual energy consumption with artificial conditioning in an office building. Comparing the convergence and reliability of the hybrid processes, the results show that the neural network achieved the best results. The results also show that for this particular/specific problem, the ideal budget comprises 80% of simulations and 20% of predictions, maintaining the results' reliability and reducing the computational cost. Pages: 25 - 30 Paper: