Process Plan Optimization using a Genetic Algorithm

In this paper we present our ongoing research project GAPO (Genetic Algorithm Process Optimization) that focuses on the development of an optimization module for process plans. GAPO is part of a 4D-Toolbox that conflates different modules for the 4D planning of construction projects. Among other modules, the 4D-Toolbox consists of a DES (Discrete Event Simulator) that automatically sequences activities into a network plan 3 taking structural and process constraints into account. Thereafter, GAPO is used to optimize the generated network plan in terms of time, cost and resource management forming a multiple criteria optimized process plan 4 This process plan can then be joined with the 3D-Model of the construction project forming a 4D Model and visualized through the 4D-Player, another module of our 4D-Toolbox. GAPO is based on a Genetic Algorithm (GA) approach to perform its optimization. GA’s are a class of heuristic search methods based on the Darwinian principle of evolution. They mimic and exploit the genetic dynamics underlying natural evolution to search for optimal solutions of general combinatorial optimization problems [1]. Our Evolution Model starts with an initial population of randomly generated process plans. A subsequent population will then be assembled using five strategies which can be weighted by the user. A fraction q of the best individuals will be directly passed to the next population. This guarantees that the quality of the most suited candidates will monotonically increase from generation to generation. A second fraction r of individuals will be passed to the next population after a mutation. On one side, this process opposes early convergence in a local optimum and thereby helps to open new search regions. On the other side, it also allows a fine tuning of suitable solutions by applying small chances on them. A third fraction s of the new population is created by recombining individuals from the old generation. This process forces convergence into an optimum. A fourth fraction t is created by recombining individuals but instead of passing them di-