Improved Bare-bones Particle Swarm Optimization Algorithm And Its Application

The problems of optimization in the field of scientific research and engineering have many characteristics such as multi-constraint,non-linearity and high dimensions.The classical numerical method is difficult to obtain the optimal solution.Bare-bones particle swarm optimization(BPSO)algorithm has been highly concerned by experts and scholars from all over the world due to its simple structure,easy to implement,fast convergence,etc.BPSO has been widely used in industry optimization,image process,data mining and other fields.However,BPSO algorithm has the same phenomenon of premature convergence as the standard particle swarm optimization(PSO)algorithm in solving complex optimization problems.In this paper,an improved BPSO algorithm is proposed for solving the phenomenon of premature convergence of static nonlinear optimization,constrained optimization and dynamic optimization,and it is applied to solve the problem of image segmentation.The main works in this paper are following:(1)A self-learning BPSO(SLBPSO)algorithm is proposed for static nonlinear optimization problem.Firstly,a self-learning strategy is proposed by analyzing the reasons of premature convergence.The expectation of Gaussian distribution in the updating equation is controlled by an adaptive factor,which balances the exploration in earlier stage and the convergence in later stage.Secondly,SLBPSO adopts a novel mutation to the personal best position(Pb)and the global best position(Gb),which helps the algorithm jump out of the local optimum.Simulations experiment show that SLBPSO has highly competitive on the static optimization problems.(2)A novel BPSO algorithm is proposed to improve the convergence accuracy in solving the contrained optimization problems(COPs).Firstly,a time-varying constraint parameter ? is introduced to make full use the effective information of infeasible solutions in early stage.Then,in order to block premature convergence,dynamic-learning strategy is presented in which particles can randomly learn from the excellent individual,and the adaptive learning weight is used to achieve the abilities of the swarm from global search to local search over time.Finally,the gradient mutation strategy is probabilistically used to make the particles in infeasible region swarm into the feasible region.Simulation results on 36 benchmark test functions show the proposed BPSO algorithm is competitive with other state-of-the-art optimization algorithms in solving constrained optimization problems.(3)To solve the challenges of outdated memory and diversity loss in dynamic optimization problems(DOPs),this paper proposes multi-swarms BPSO(MBPSO)algorithm.First of all,the particles used for environment survey is set to detect timely the change of environment in MBPSO,which avoids incorrect information guiding the direction of swarms' evolution.After the change of environment,MBPSO reinitializes all swarms by using the information which every swarm explores in last environment which enhances fast tracking ability of the excellent solution to the current environment.In addition,MBPSO designs newly methods to enhance particles' activation and use the multi-swarms measure to maintain the whole swarm's diversity when the swarm falls into a standstill.The simulation experiment results show that MBPSO has stronger competitiveness in dynamic environment.(4)The SLBPSO algorithm is applied to multi-thresholds image segmentation,which overcomes the problem of large computational complexity in the image segmentation.Experiments have been performed on all kinds of images using various numbers of thresholds.The experimental results are compared with exhaustive method,PSO,ABC and electromagnetism-like mechanism(EM).Experimental results show that the speeds of the proposed method in the three thresholds segmentation and the four thresholds segmentation are 89.74 and 5368.70 times faster than those of exhaustive method,respectively.Compared with PSO,ABC and EM,SLBPSO algorithm has obvious superiority in improving image segmentation quality and efficiency.