Quantum-behaved Particle Swarm Optimization With Collaborative Learning And Cultural Evolution Mechanism

Quantum-behaved particle swarm optimization(QPSO)is a swarm intelligence optimization algorithm,which is based on quantum model.Because of its simplicity and easy implementation,QPSO has been widely used to solve different kinds of optimization problems.In this thesis,QPSO algorithms with different optimization contexts are studied systematically.According to the characteristics of different optimization problems,specific QPSO-based algorithms are presented.In QPSO,the critical step is to obtain an appropriate attractor for each particle.In traditional QPSO,the attractor of one particle is obtained as the weighted sum of its personal and global best positions.However,some theoretical analysis demonstrates the traditional strategy is not an efficient way to get attractors for particles.In this thesis,a QPSO algorithm with collaborative learning strategy to get attractors is proposed.In real life,many problems have multiple or dynamic optimization objectives.For multiobjective optimization problems(MOPs),it is difficult to find a solution which can optimize all the objectives simultaneously.So,how to find the appropriate personal and global best positions for each particle is very important in extending QPSO to multiobjective context.For dynamic optimization problems(DOPs),the challenge is how to balance the convergence and diversity in QPSO.In this thesis,cultural evolution mechanism,which extracts different kinds of knowledge from population and then adopts the extracted knowledge to guide the evolution process of the individuals in population,is introduced into QPSO.The effectiveness of QPSO with cultural evolution mechanism in solving MOPs and DOPs are demonstrated experimentally and theoretically.The main contributions of this thesis are listed as follows:1.In QPSO,the traditional strategy to get attractors usually loses useful information hidden in particles' personal and global best positions.Moreover,the traditional strategy makes it difficult for QPSO to jump out of local optima in the last runs.To address these problems,a QPSO algorithm with collaborative learning(CL)strategy to obtain attractors is proposed.CL strategy contains orthogonal and comparison operators.The two operators are controlled by a probability parameter in order to balance the exploration and exploitation in CL strategy.The experimental results tested on CEC2014 complex single-objective functions prove the effectiveness of the proposed CL strategy.2.In multiobjective QPSO,the difficulty is to obtain the appropriate personal and global best positions.To address this problem,a QPSO based on cultural evolution mechanism,namely Cultural MOQPSO,is proposed to solve MOPs.Cultural MOQPSO has two spaces: population space and belief space.The particle swarm in population space evolves according to the knowledge stored in belief space.In Cultural MOQPSO,a local search strategy,which is based on the situational knowledge in belief space,is adopted to generate personal best position for each particle.Moreover,a combination-based update operator,which is based on the history knowledge in belief space,is adopted to help the algorithm get even-distributed Pareto fronts.Experimental results demonstrate the effectiveness of the proposed algorithm.3.In order to solve the environmental/economic dispatch(EED)problem,a multiobjective QPSO algorithm with cultural evolution mechanism and multiple measurements strategy is proposed.In this algorithm,each particle is measured several times.The personal and global best positions for the several measurements can be obtained according to the situational and topographical knowledge stored in belief space.Moreover,an adaptive mutation operator,which is based on the history knowledge,is introduced to enhance the algorithm's ability of jumping out of local optima.The experimental results tested on two EED systems show that the proposed algorithm can achieve promising performance.4.For single-objective optimization problems in dynamic environments,one challenging subject is how to locate the new global optimum after the environment changes.To address this problem,a dynamic QPSO with memory enhanced learning strategy is proposed.In the proposed algorithm,multiple sub-populations are adopted,and each sub-population focuses on one peak respectively.Three memory archives,namely personal memory archive,global memory archive and temporary peak memory archive,are adopted.Personal and global memory archives are used to obtain personal and global best positions,respectively.A re-initialization strategy which is based on the temporary peak memory archive is proposed to help algorithm adapt well to the changing environment.The experimental results tested on single-objective functions in dynamic environments show the effectiveness and superiority of the proposed algorithm.5.In order to balance the diversity and convergence in MOPs with dynamic environments,a multiobjective QPSO,which is based on cultural evolution mechanism,is proposed.To maintain population diversity,a multiple population strategy is adopted.In this strategy,one sub-population optimizes all the objectives simultaneously and the global best positions of particles in this sub-population are obtained according to the topographical knowledge stored in belief space.Each of the rest sub-populations focuses on one of the multiple objectives respectively.Moreover,a prediction strategy is proposed to increase the searching efficiency and a re-initialization strategy is proposed to help the algorithm locate new Pareto fronts in changing environments.Experimental results demonstrate that the proposed strategies do help improve the performance of QPSO in solving MOPs with dynamic environments.