Research On Quasi-Affine Transformation Evolutionary(Quatre)Algorithm With Cooperative Structure

There are many optimization demands for tough optimization problems in our lives nowadays and different optimization algorithms are proposed to tackle these specific complex problems in different areas.The common approach to tackle these optimization problems usually begins with designing their objective functions that can model the problems or the systems while incorporating some constraints.If these objectives are continuously differentiable or even second-order differentiable,Quasi-Newton methods or Newton's method can be of the good choices.If these objectives are non-differentiable or they are black-box ones or even of noises,these derivative based methods become unsuccessful,then optimization algorithms in Evolutionary Computation(EC)come to take effect for these situations.The paper mainly focused on Evolutionary Computation algorithms especially for those algorithms in the field of Particle Swarm Optimization(PSO)and Differential Evolution(DE),which are two of the main branches in EC domain.Furthermore,deeper researches were conducted to examine the up-to-date status and still existing weaknesses in the two branches.Then a new framework called QUasi-Affine TRansformation Evolution(QUATRE)framework as well as its corresponding algorithm,the QUasi-Affine TRansformation Evolutionary(QUATRE)algorithm,is proposed to tackle the weaknesses existing in state-of-the-art PSO and DE variants.The paper also depicted some further developments of QUATRE algorithm and proposed a new External ARchive based QUATRE algorithm with cooperative structure,called QUATRE-EAR algorithm.Finally,a new parameter-independent tool aiming at tackling real-parameter non-linear non-convex black box optimization problems was developed,and the main body of the tool was the new proposed QUATRE-EAR algorithm.The main focuses and contributions of the research are listed as follows.In the thesis,an ebb-tide-fish algorithm was firstly proposed to tackle lower dimensional real-parameter single-objective optimization problems,and then a novel memetic Monkey King Evolution(MKE)algorithm was proposed to enhance the overall performance of it on complex higher dimensional optimizations.The ebb-tide-fish algorithm aimed at tackling the slow convergence speed weakness of PSO variants on lower dimensional optimizations,and the memetic MKE algorithm further enhanced the optimization capacity of the ebb-tide-fish algorithm.There were three versions of MKE algorithm,the first version MKE algorithm enhanced the global search capacity of ebb-tide-fish algorithm by incorporating a scale factor,nevertheless,the higher dimensional optimization performance of the first version MKE algorithm was not satisfying enough.The second version MKE algorithm enhanced the first version on its local search capacity by incorporating difference vectors and then a better performance on higher dimensional optimization was obtained by this improvement.However,individuals in the above mentioned algorithms had two search modes,one was local search mode and the other was global search mode,both of which resulted in a less cooperative search of individuals.Therefore,the final version MKE algorithm was proposed,in which each individual was upgraded to be the most powerful Monkey King individual,and a cooperative evolution matrix was incorporated into it as well.By incorporating these two improvements,the final version MKE algorithm maintained a more cooperative search of individuals and consequently secured an overall better performance on high dimensional complex optimization problems.In the thesis,a new state-of-the-art Parameter with Adaptive Learning Mechanism Differential Evolution(PALMDE)was advanced to tackle the weaknesses existing in several former state-of-the-art DE variants.The new proposed PALMDE algorithm tackled the mis-interaction weaknesses between parameters of DE variants by separating them into different groups,and each of these control parameters in PALMDE algorithm was renewed in independent manner.Moreover,a time stamp mechanism was also advanced to enhance the diversity of mutant vectors while voiding too old inferior solutions being archive-residents in the external archive during the whole evolution.The new proposed PALMDE algorithm was verified on a test suite containing a large number of benchmark functions and experiment results showed that the PALMDE algorithm was competitive with the other contrasted state-of-the-art DE variants.In the thesis,a new cooperative QUasi-Affine TRansformation Evolution(QUATRE)framework was proposed for optimization problems as well as its corresponding algorithm,the QUasi-Affine TRansformation Evolutionary(QUATRE)algorithm.From the “evolution mode” perspective,the QUATRE algorithm originated with the final version memetic MKE algorithm,therefore,it also tackled the “two steps forward,one step back” weakness which resulted in a slow convergence speed in PSO variants.From the“motion of individuals” perspective,the QUATRE algorithm also can be considered as a parameter-reduced DE algorithm,and it tackled the representational bias from high dimensional perspective of view and consequently achieved a more reasonable search of the solution space.Moreover,a new External ARchive based QUATRE algorithm with cooperative structure,the QUATRE-EAR algorithm,was proposed to enhance the overall optimization performance of QUATRE algorithm.The Parameters with Adaptive Learning Mechanism and the mutation strategy with time stamp scheme proposed in PALMDE algorithm were also incorporated into the proposed QUATRE-EAR algorithm.An adaptive initialization scheme for evolution matrix in each generation was advanced as well in the algorithm in order to get a better perception of the landscape of optimization objectives,all of which constituted the much more powerful and scientific evolutionary algorithm.The QUATRE-EAR algorithm was also verified on a test suite containing a large number of benchmarks and experiment results revealed that the new proposed QUATRE-EAR algorithm secured an overall better performance than the other contrasted algorithms.To summarize,the proposed memetic Monkey King Evolution algorithm was to tackle the weaknesses existing in the state-of-the-art PSO variants,and the proposed Parameter with Adaptive Learning Mechanism Differential Evolution was to tackle the weaknesses existing in state-of-the-art DE variants.Furthermore,the External ARchive based QUasi-Affine TRansformation Evolutionary algorithm was proposed by incorporating the advantageous of the memetic MKE algorithm and PALMDE algorithm,and the proposed QUATRE-EAR algorithm secured an overall better performance on realparameter single-objective complex non-linear non-convex optimization problems.All the parameters in the proposed QUATRE-EAR algorithm did not need to be tuned manually,therefore,it was very convenient for the users due to the parameter independent characteristic.The black-box optimization tool based on the powerful QUATRE-EAR algorithm that the thesis mainly focused on achieved an excellent performance on tackling high dimensional complex optimizations,and an overall better performance was obtained by this tool in comparison with other contrasted state-of-the-art evolutionary algorithms.