Methods to Detect and Address Stall in Particle Swarm Optimization

Particle Swarm Optimization can experience stall in multi-modal search spaces. A stalled swarm is unable to converge and unable to find better solutions because all of its exploratory search solutions are rejected. Remedies to address the stall condition can benefit from knowing the particles that are still performing exploration, so we develop an efficient and accurate real-time, search space independent method to identify these particles. We confirm the benefit of identifying stalled particles through a modification designed for globally convex search spaces. We also discuss opportunities to differentiate search space landscapes and propose future research that can address non-globally convex search spaces.