Urban form optimization for air quality applications using simulated annealing and genetic algorithms

In the present work, the application of two optimization algorithms in the solution of an urban form optimization problem (UFOP) is demonstrated for idealized urban geometries. Given ‘p' buildings, ‘q' emitters, and ‘r' decision variables (where p, q, and r are positive integers), the UFOP comprises identifying optimum decision variables that result in a certain minimum objective. Simply stated, the main objective is to find building layouts that result in low pollutant concentrations. The Quick Urban and Industrial Complex (QUIC) dispersion modeling system implemented on graphics processing units (GPU) was adopted as the forward model because of its ability to produce reasonably accurate simulations extremely rapidly. At present, the decision variables that are optimized include spatial locations and physical dimensions of buildings, as well as wind speed and direction over the domain of interest. One of the important initial tasks of this project has been the identification of suitable objective functions that when minimized produce desired building configurations. In this project we present two different objective functions and illustrate their effect on the final solution. The first objective function is based on minimizing the maximum average concentration at the street level, and the second is based on minimizing the total number of exceedances (concentration values above a certain threshold) over the region of interest. Simulated annealing (SA) and genetic algorithms (GA) are used to identify the optimum decision variables, and their performance is compared against each other, and pure random search. Simple urban geometry test problems are used to enable validation of the optimization algorithms through comparisons with exhaustive search. Even so, many thousands of simulations must be run. At the conference, results for simple test cases along with challenges encountered during implementation will be presented.