Robust Solutions in Resource Leveling via Genetic Algorithm Populations

Abstract

Algorithms for solving the resource leveling problem (RLP) in construction projects are proven to increase efficiency, create predictability, and balance resource demand across adjacent time periods of the project's duration while observing time and resource constraints. Leveling resources reduces the amount of change between one time period and the next in the project's time-use resource histogram. Conventional optimization methods of the RLP can become difficult as the number of decision variables grow, because the solution space increases exponentially as variables are added. Genetic algorithms are very capable when applied to large-scale instances of the RLP, and here the author applies a genetic algorithm testing multiple objective functions from literature with different performance measures, including kurtosis, the Resource Improvement Coefficient, and non-parametric statistical analysis. Results show that given a large problem based on practical data, genetic algorithms produce an unexploited large range of robust options via their embedded search population for decision-makers and planners. These alternatives can even be selected mid-project up to a certain point, while maintaining solution strength. There is a demonstrated tradeoff between number of time-feasible alternate solutions and variance of the alternate decision variables.