Improved Particle Swarm Optimization Using Two Novel Parallel Inertia Weights

To solve the premature convergence problem of particle swarm optimization, two novel methods are introduced to adjust the inertia weight in parallel according to different fitness values of two dynamic sub-swarms. When fitness values are better than or equal to the average, two types of dynamic nonlinear equations are proposed to adjust the inertia weight in a continuous convex area which can retain the favorable conditions and achieve a good balance between global exploration and local exploitation. On the contrary, when fitness values are worse than the average, the inertia weight is adjusted by the introduced dynamic logistic chaotic map which can make local-optima trapped particles break away from the bad conditions, search global optima dynamically and avoid premature convergence. Experiments and comparisons demonstrated that the new proposed methods outperformed several other well-known improved particle swarm optimization algorithms on many famous benchmark problems in all cases.