Prescreening and Repairing in a Genetic Algorithm for Highway Alignment Optimization

Abstract

Abstract: The method of handling infeasible solutions in an evolutionary search algorithm [e.g., genetic algorithms (GAs)] is crucial to the effectiveness of the solution search process. This problem arises because solution search steps, techniques, and operators used in GAs (such as reproduction, mutation, and recombination) are normally “blind”to the constraints, and thus GAs can generate solutions that do not satisfy the requirements of the problems. In GA‐based highway alignment optimization (HAO), many infeasible solutions, which violate model constraints, are also possibly generated, and evaluation of such solutions is wasteful. This study focuses on ways to avoid wasting computation time on evaluating infeasible solutions generated from the GA‐based HAO, and develops a prescreening and repairing (P&R) method for an efficient search of highway alignments. The key idea of the P&R method is to repair (before the very detailed alignment evaluation) any candidate alignments whose violations of design constraints can be fixed with reasonable modifications. However, infeasible alignments whose violations of constraints are too severe to repair are discarded (prescreened) before any detailed evaluation procedure is applied. The proposed P&R method is simple, but significantly improves computation time and solution quality in the GA‐based HAO process. Such improvements are demonstrated with a test example for a real road project. Through the example study, it is shown that the model incorporating the P&R method can find a good solution much faster (by approximately 23%) than the model with the conventional penalty method. In addition, the P&R method allows the model to evaluate about 70% more solutions than that it can evaluate with the penalty method for the same number of generations.