General Correlation of Building Energy Use Via Hybrid Genetic Programming/Genetic Algorithm

Abstract

A hybrid algorithm that combines genetic programming (GP) and genetic algorithms (GAs) that deduce a closed-form correlation of building energy use is presented. Throughout the evolution, the terms, functions, and form of the correlation are evolved via the genetic program. Whenever the fitness of the best correlation stagnates for a specific number of GP generations, the GA optimizes the real-valued coefficients of each correlation in the population. When the GA, in turn, stagnates, correlations with optimized coefficients and powers are passed back to the GP for further search. The hybrid algorithm is applied to the problem of predicting energy use of a U-shape building. More than 800 buildings with various foot-print areas, relative compactness (RC), window-to-wall ratio (WWR), and projection factor (PF) values were simulated using the VisualDOETM energy simulation engine. The algorithm tries to minimize the difference between simulated and predicted values by maximizing the R2 value. The algorithm was able to arrive at a closed-form correlation that combines the four building parameters, accurate to within 4%. The methodology can be easily used to model any type of data behavior in any engineering or nonengineering application.