Coupling visual and coded programming in computationally expensive optimization problem applied to an energy performance problem

Resumo

The demand for computational optimizations in programs used in the area of architecture has been constant for some time. One of the main software used in this context is the addition of the Grasshopper plugin to Rhinoceros, which, like others of a similar nature, offers a visual programming interface to users. However, significant advances in the area of optimization have been predominantly associated with coded programming, relegating data flow-based programming to a secondary background. A recurring problem in architecture is determining the energy performance of buildings, specifically the one evaluated in this article, as an optimization problem that uses computer simulations to calculate its fitness function. This problem is formulated according to 4 variables: building orientation, transmittance of shading elements, and width of the openings to the east and west. The objective is to find values for these variables that minimize the building's energy consumption. In this work, we evaluate an implementation flow for coding in Pyhton within the Rhinocero+Grasshopper visual programming platform to implement an optimization algorithm based on Differential Evolution. Given the high computational cost associated with executing the algorithm (mainly due to its dependence on simulations), an artificial neural network was implemented to predict the value of the fitness function of some solutions. For this coupling, we used the GHPythonRemote tool, which allows the integration of libraries developed in Python that are not natively available in Grasshopper, such as, for example, scikit-learn. Therefore, the purpose of this article is to demonstrate the optimization process carried out through the coupling between visual and coded programming in a high computational cost optimization problem applied to architecture. The results produced by the algorithm will be compared to the results found by applying a native optimization plugin in Grasshopper based on genetic algorithm. The results demonstrate the potential of coupling by making the implementation of algorithms more flexible, in addition to contributing to improve the optimal solution of the problem and reducing the computational cost of the optimization process by allowing the application of machine learning methods.