Multi-Objective Evolutionary Algorithms For Solving Groundwater Management Models

Development and application of multi-objective evolutionary algorithms (MOEAs) is a typical and substantial topic in the field of groundwater resources management. The evolutionary algorithms (EAs), possessing the characteristics of self organization, adaption, and self learning, have strong robustness, extensive applicability and overall searching ability. The EAs, regardless of the nature of problems, can handle the complex optimization problems in water science that cannot be eradicated by the traditional methods. Thus the EAs have been widely applied in the field of groundwater resources optimization management. For the real-world applications, however, to balance the tradeoff between the local domination and global diversification of non-inferior solutions and to improve the computational efficiency are crucial for the development of MOEAs. To overcome the mentioned problems, this thesis focuses on the development and applications of MOEAs for optimal design of groundwater systems. Firstly, the reviews of researches and applications of MOEAs in groundwater resources management were carried out, and their shortcomings and several open issues were pointed out. Then, three tabu search (TS) based MOEAs, namely niched Pareto tabu search (NITS), elitist multi-objective tabu search (EMOTS), and niched Pareto tabu search combined with genetic algorithm (NPTSGA), were presented to solve groundwater management models under general hydrogeological conditions. In addition, one more method, probabilistic improved niched Pareto genetic algorithm (PINPGA), was put forward to solve groundwater management models under uncertainty of hydrogeological parameters. (1) The NPTS inherits the structural framework of multiple objective tabu search (MOTS) developed by Baykasoglu and his colleagues, and was made improvements by adding a niche technique, fitness sharing approach, to modify selection strategy for maintaining the diversity on the Pareto-optimal front and conducting the fitness archiving to avoid wasteful repetitive calculations of the objective functions corresponding to the same decision variables during the optimization process; (2) The EMOTS learns the elitist strategy to obtain non-dominated solutions better converging to the true Pareto front from the non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ). Moreover, neighborhood solutions are generated using the Latin hypercube sampling (LHS) for uniform sampling in the neighborhood bound space around each seed solution. Based on elitism strategy and LHS method, the algorithm produces Pareto-optimal solutions converging to the true Pareto front and maintaining uniformly along the tradeoff curves; (3) The NPTSGA aims at balancing the tradeoff between local optimality and global diversity during the search process. It is developed by integrating the global search of NSGA-II with NPTS, in which the global search ability of NPTS is improved by the diversification of the candidate solutions in the evolving NSGA-II population; (4) PINPGA uses two techniques of probabilistic Pareto domination ranking and probabilistic niche technique to find Pareto optimal solutions of groundwater remediation systems under uncertainty. The performance of the four proposed algorithms has been tested through the function optimization problems. Next, the proposed four algorithms are coupled with the commonly used flow and transport code, MODFLOW and MT3DMS/SEAWAT, based on the simulation-optimization framework, to explore several typical issues in groundwater resources management. The optimization results indicate that the proposed MOEAs are the efficient and effective tools for optimizing the multi-objective design of groundwater systems under complicated hydrogeological conditions. Finally, analytic results throughout the thesis were summarized and concluded. The TS-based multi-objective methodologies were explained systematically, and all the new understandings about the groundwater resources management gained in this study were summarized and elaborated. In addition, some suggestions were given to guide the future studies of TS-based MOEA in groundwater management.