3 A Threshold Approach for Multi-Agent SI Systems to Improve Optimization Performance

Optimization problems may require a non-polynomial time to find a solution, so heuristic approaches are widely adopted. Swarm intelligence (SI) is one such approach proven to be efficient. An SI system has simple agents working together to quickly accomplish a complex task. However, the efficiency does not guarantee a global optimum, because agents can be trapped at local optimums without progression. For this reason, we compute a metric to compare with the assigned threshold to determine when to escape from traps in order to improve performance. The metric will quickly approach the threshold when agents are hard to find a better solution. Once reached, the ongoing ran of the system terminates and a brand new ran begins to counter the trap. We study the traveling salesman problem (TSP) and show that under the same conditions the threshold approach often finds a better solution more promptly.