Paper Conference

Proceedings of Building Simulation 2017: 15th Conference of IBPSA

     

A Modified Genetic Optimization Algorithm Using Ancestor Path Extrapolation

Aaron Powers
Johnson Controls, Inc., Memphis, TN

DOI: https://doi.org/10.26868/25222708.2017.598
Abstract: Optimization techniques have become increasingly popular in building performance simulation and design. The relatively high computational expense of simulating a building for a full year has traditionally been a major obstacle in the widespread use of these techniques. Consequently, most practical implementations employ some variety of meta-heuristics which do not necessarily guarantee an optimal solution but also do not require an impractical number of simulations to run. Still, further gains in convergence speed are needed for automated optimization to become truly practical in building design. This paper introduces a modified genetic optimization algorithm which implements a new creation operator to improve the rate of convergence in building performance optimization. The performance of the new algorithm was compared to that of a traditional genetic algorithm using a test case building model with 26 real valued design variables over several optimization runs. The new algorithm displayed a faster rate of convergence in all tests. The improved convergence rate was independent of the initial population fitness and building location. These results show that genetic algorithms with ancestor path extrapolations can greatly reduce the computational time required in building performance optimization.
Pages: 2193 - 2198
Paper:
BS2017_598