Research On Nature-Inspired Optimization Algorithms And Their Applications

Nature-inspired algorithm is a new method of optimization based on co-evolution of a group of individuals in recent years and provides a new tool for complex optimization problems. Due to its advantages of intelligence, wide applicability, global search ability and parallelism, nature-inspired optimization algorithm has become a hot research area and has been widely used in many fields such as computer science, telecommunication network, knowledge discovery, robots and so on. In this dissertation, studies are focused on two typical models of nature-inspired computation: electromagnetism-like mechanism (EM) algorithms and genetic algorithms (GAs). Several novel nature-inspired optimization algorithms are proposed for different optimization problems, respectively. The main contributions of this thesis can be summarized as follows:1. In order to avoid the charge overflow and parameter sensitivity in standard electromagnetism-like mechanism algorithm, new formulas of particle charge and force between any two particles are presented, in which the magnitude of the force is related to the difference of their objective function values and the distance between two particle. The particle moves towards a local optimal solution in its neighborhood along the direction of total force. Based on these, an improved EM algorithm is proposed for unconstrained optimization problems. Simulation results on benchmark problems demonstrate that the proposed algorithm is effective and can find the global optimal solution quickly.2. An electromagnetism-like mechanism method is proposed for solving constrained optimizations on the basis of pattern search. First, the violation degree function is introduced for the constrained optimization problems. Second, inspired from the idea that feasible particles are always better than infeasible ones, the new computational equation of the particle charge is designed. Third, a formula for computing the force between two particles is presented, which can guide infeasible particles to move towards feasible ones. Furthermore, the convergence of the new algorithm is proved and the computer simulations are made. Compared with simulation results of the existing algorithms, the proposed algorithm has the advantages of quick convergence and good performance and is very potential for constrained optimization problems. 3. For unconstrained multi-objective optimization problems (UMOP), a new algorithm is proposed. In order to guide the search along potential direction, an effective crossover operator and a simple mutation operator are presented. The convergence of the proposed algorithm is proved using the theory of probability. The simulation results show that the proposed algorithm can more quickly track the Pareto optimal solutions in comparison with other existing algorithms.4. A novel method for graph coloring problem (GCP) is presented. A new encoding scheme is presented to avoid the redundancy of the integer representation. An effective crossover is designed according to the characteristic of the GCP. In order to enhance its ability of exploration, a greedy local search is integrated into the crossover operator. Based on these, a novel genetic algorithm for GCP is presented and its convergence to global optimal solution with probability one is proved. The proposed algorithm was tested on 10 standard test problems with the number of vertices from 11 to 1000. Experimental results indicate that the proposed algorithm performs well and is competitive with other existing algorithms.