Abstract:
Time to time, many researchers have suggested modifications to the standard parti cle swarm optimization to find good solutions faster than the evolutionary algorithms.
However, it may fall into local optimum (premature convergence) when solving complex
multimodal problems. To overcoming such problems, in this thesis accelerating com ponents information based particle swarm optimization with time varying acceleration
coefficients algorithm is proposed. In the proposed algorithm, new combination compo nent was introduced based on the average of all individual best position and global best
position. Such mechanism can help the algorithm to escape local optimum by combining
the learning experience of particle itself and other particles experiences. The global best
position plays an important role in improving the convergence rate but reduce the diver sity of population. If global optimum is not close to the best particle the particles may
trap in local minima. At the same time the combination component information can
lead the particles exploring to a better location by balancing their positions and weaken
the attraction of the best particle. Therefore, every swarm updates its position based
upon cognitive, social and combination component information and to proper control
of these three components during the initial and latter part of the search, time varying
acceleration coefficients are employed. The proposed algorithm is coded in MATLAB
R2019a and twenty four benchmark test functions which have different characteristics
are optimized. In terms of quality solution, convergence speed and robustness, the exper imental results show that the proposed algorithm has a superior performance compared
to previously reported results of other three particle swarm optimization algorithms.
This suggests that the proposed algorithm has the ability to escape local optima solu tions and achieving the global optimum efficiently.
