Many-Objective Evolutionary Algorithms and Applications to Water Resources Engineering


Over the last ten years, Multi-Objective Evolutionary Algorithms (MOEAs) have undergone tremendous development and have been successfully applied in various fields to solve complex problems that were formulated typically with two or three objectives. However, an increasingly expanding basis of our knowledge of natural processes and engineering requirements, and the need of taking into account sustainability indicators in order to compensate for the socio-environmental damage caused by the human activity, has increasingly led to formulation of the real world optimisation problems requiring a large number of management criteria. This has posed a serious challenge for traditional MOEAs, which are known to progressively lose their effectiveness as the number of performance criteria of a problem increases, and has led researchers to begin investigating the issues related to the evolutionary optimisation of problems with more than three objectives, usually referred to as Many-Objective Problems (MOP), and devising remedial measures. Motivated by the need to provide researchers and practitioners interested in water resources engineering with an optimisation method to effectively solve problems formulated with many objectives, this thesis is an effort to advance in the field of Evolutionary Many-objective Optimisation . In order to meet this ambitious goal, following an attentive review and analysis of the issues arising in this setting, a new methodology that can be easily incorporated into existing MOEAs is presented. Extensive experiments suggest that the methodology proposed improves significantly the performance of a state-of-the-art algorithm and suggest that other MOEAs could benefit to a larger extent from it. In addition, a novel methodological framework is developed providing formal tools to analyse the theoretical properties of the methodology introduced. The relevance to the water resources engineering field of the optimisation methodology developed was illustrated through a case study dealing with the automatic calibration of an urban drainage model. A novel calibration framework, enabling the concurrent consideration of spatially distributed temporal time series, was developed leveraging the optimisation method proposed, and successfully applied to the calibration of the model BEMUS, for a 25ha experimental urban catchment.