This paper addresses the problem of system optimization using meta-heuristical algorithms inspired by biological processes. In particular, we focus on Ant Colony Optimization (ACO), Wasp Swarm Optimization (WSO) and Genetic Algorithms (GA), where GA is used alone and for post-optimization.
The meta-heuristics inspired by social insects share a common background, and it is interesting to see how they compare, how they can be applied to a problem, and to which problem each algorithm is more adequate. The table shows the main structure of both ACO and WSO algorithms. As it can be seen, the processes behind them have many similarities, with the pheromones of the ants being the equivalent to the force based hierarchy of the wasps. x is the system variable, computed in both algorithms from a probability matrix. ACO and WSO basic algorithm Initialization of the pheromone matrix t Computation of the forces f(x) Computation of the probability matrix p(t ) Computation of the probability matrix p(f) Computation of the solution x(p) Computation of the solution x(p) Update of the pheromone matrix t Update of the forces f(x)
ACO is based on how unsupervised colony agents cooperate to achieve a common goal, and as such it is a natural way of optimizing systems where cooperation is advantageous. With some bigger or smaller modifications it can be applied to several different problems. WSO is based on how wasps compete between themselves, and as the optimization of logistic systems or network routing usually imply some kind of competition, for resources, power, etc, WSO can be applied successfully in many cases.
The paper has four main sections. First, we do a brief introduction. In the second part we present a point by point comparison between ACO, WSO and GA, analyze their potential in optimizing different kinds of example systems and how to determine which one is the best option for a given situation. We follow this study with a comparison of the five variants ACO, WSO, GA, ACO-GA and WSOGA when applied to two systems, a theoretical benchmark problem and a real world logistic system at Fujitsu-Siemens Computers. We end the paper with a global conclusion of the matters discussed, and a summary of the future work. We conclude that WSO produces better results than ACO and GA for the considered logistic problem. We also conclude that the hybrid WSO/ACO-GA performs better than their stand-alone counterparts, and that WSO/GA performs better than ACO/GA due to the GA’s part of the algorithm having a more diversified population.