Abstract
Large optimization problems that involve either a large number of decision variables or many objectives pose great challenges to nature inspired optimization algorithms. On the one hand, nature inspired optimization algorithms suffer from the curse of dimensionality in the decision space. With an exponentially increased volume of the decision space as well as the complexity of the search landscape, it is challenging for nature inspired optimization algorithms to perform efficient search within limited execution time, in particular when the decision variables are non-separable due to the interactions between them. On the other hand, nature inspired optimization algorithms suffer from the curse of dimensionality in the objective space. In high-dimensional objective space, due to the sparse distribution of the candidate solutions, efficient population diversity management and convergence pressure preservation become particularly important. In this thesis, we present several pieces of work to address the above challenges. Firstly, we present two variants of the particle swarm optimization algorithm to tackle single-objective large problems. By enhancing the swarm diversity, the two algorithms are capable of handling problems with as large as 5000 decision variables. Secondly, we present an inverse model based evolutionary algorithm to tackle multi-objective large problems. By building inverse models that map candidate solutions from the objective space to the decision space, the algorithm is able to enhance the computational efficiency in the optimization of bi- or three-objective problems with a large number of decision variables. Thirdly, we present a reference vector guided evolutionary algorithm to tackle optimization problems with many objectives. By decomposing the high-dimensional objective space into subspaces using a set of reference vectors, the algorithm is able to handle problems with as many as 10 objectives. Finally, we present a benchmark test suite which can be used to assess the performance of nature inspired optimization algorithms on problems with both a large number of decision variables and many objectives. The performance of all proposed algorithms have been verified on widely used benchmark problems in comparison with some other state-of-the-art algorithms. Moreover, the proposed reference vector guided evolutionary algorithm has been successfully applied to the optimization of a seven-objective hybrid electric vehicle controller model designed at the Honda Research Institute Europe.