| Meta-heuristics provide quite effective strategies to find near optimal solutions for complex problems. With the proliferation of powerful multi-processor workstations, huge computing clusters/clouds and fast communication networks, multi-thread meta-heuristics becomes a natural alternative to speedup the search for approximate solutions. Meanwhile, it should not be neglected that multi-thread meta-heuristics may also bring up improvement of the solution quality due to the intensified effort in both exploring and exploiting the search space. Furthermore, and probably most importantly, multi-thread meta-heuristics could increase its robustness by the use of different combinations of strategies and parameter settings for each thread to offer a consistently high level of performance over a wide variety of problem instances.;In this thesis, we propose a novel bio-inspired multi-agent co-operative searching methodology for global optimization, named Rational Swarm algorithm. It can be used both as a meta-heuristic guiding local search algorithm and as a high-level multi-agent co-operative searching strategy to coordinate multiple agents using meta-heuristics. In this work, the Rational Swarm methodology has been applied to popular meta-heuristics such as Simulated Annealing and a hybrid Genetic Algorithm LKE (the Lin-Kernighan-Helsgaun algorithm) and a pure local search algorithm MSBH (Monotonic Sequential Basin Hopping). Numerical experiments on both continuous and combinatorial optimization problems show Rational Swarm can improve the performance of applied meta-heuristics/heuristics in terms of solution quality and robustness under the same computational budget. Convergence analysis gives the theoretical insights about why the proposed Rational Swarm methodology will work. |
