Research On Intelligent Optimization Algorithms Based On Dual-population Co-evolution And Their Applications

Optimization problems widely exist in scientific research and engineering practice,which have been one of the research hotspots and difficulties in intelligent computing.Inspired by biological evolution,many scholars have put forward lots of intelligent optimization algorithms which are able to solve most of the optimization problems efficiently,including genetic algorithm,differential evolution(DE)algorithm,particle swarm optimization(PSO),ant colony algorithm,and teaching and learning–based optimization(TLBO)algorithm.However,these single swarm intelligence optimization algorithms often show such defects as searching stagnation,low accuracy of convergence,part optimum and poor generalization ability when facing the increasingly sophisticated optimization problems.By setting up two populations and establishing a competitive or cooperative relationship between them,dual-population co-evolution algorithm improves their performance respectively and complements each other's advantages through the interaction of the two populations,and further improves the performance of the algorithm in solving various sophisticated optimization problems.The dissertation selects the well-performed DE algorithm and TLBO algorithm as the key research objects and establishes a model of double-population co-evolutionary algorithm(DPCE Model),designs three algorithms based on DPCE Model combined with multi-constraints optimization,multi-objective optimization and parameter estimation of nonlinear complex optimization problems: DPCEDE,DPCETLBO and DE and TLBO-based DPCE algorithm(DPCEDT),and applies these algorithms to different practical complex optimization problems.The main research work done in this dissertation is framed as follows:(1)In this dissertation,the basic principles of DE and TLBO algorithm are studied,and the current research situation of these two algorithms is analyzed.According to the DE algorithm in solving multimodal function optimization problems with premature convergence,a multipopulation parallel DE algorithm(MPDE)based on even partition is introduced to solve this problem.Several sub-populations which have no overlaps and reflect whole characters of the functions are gained through even partition.These sub-populations realize parallel optimum search in each sub-population and collect excellent individuals after every evolution,then search the optimum solution in whole population.Experimental results show the effectiveness of the MPDE algorithm.In the research of TLBO algorithm,this dissertation puts forward a kind of modified teaching and learning–based optimization algorithm(MTLBO),applies MTLBO into the parameter optimization of the Van Genuchten equation to verify the effectiveness and superiority of the MTLBO algorithm.(2)Through experimental analysis of four cases of population setting,sub-population evolutionary strategy design based on the DE and TLBO algorithm,in the improved algorithm,it can be concluded that the design of multiple populations evolution can only improve the performance in a certain extent by simply adding the number of individual population,what's worse,it would also decline the performance because of the separating communication between the sub-populations.While selecting the strategies of the sub-populations,taking different strategies of the similar evoluation would efficiently improve the algorithm performance.The performance will be better if the sub-populations choose complementary evoluation strategies.At the same time,this dissertation bulids a universal dual-population co-evolutionary algorithm model(DECE)based on combining dual-population size setting with co-evolution mechanism through further reflection on the setting of dual-population size and the character of intelligent optmizing algorithm.The dynamics and convergence of the DPCE model are analyzed,and the algorithm design steps under the guidance of the DPCE model are given.Under the guidance of DPCE model,only by appropriately adjusting related strategies,it is easy to design an algorithm that can be used to solve different types of optimization problems,which has significant meaning in guiding the design of sophisticated optimization algorithm.(3)Directed against such defects as local optimum,low convergence,premature and large computational cost of the original DE algorithm,and utilizing DPCE model,the dissertation transforms the single-population independent evolution model to dual-population co-evolution model,and proposes a dual-population co-evolution differential evolution(DPCEDE).In the DPCEDE,one sub-population uses crossover adaptive mutation operators which are used for evolutionary operation and extensive search in solution space,so as to improve the convergence of the algorithm,the other sub-population uses the crossover operators of DE and biogeography-based optimization(BBO)to operate,on the one hand,they can make effective utilization of the information among sub-populations through shift operators,on the other hand they can balance the exploitation capabilities of the algorithm to avoid premature.Meanwhile,niche elimination mechanism is intrudued to regroup the two populations and achieve the information sharing and exchange between the populations,so as to improve the overall performance of the algorithm.By using 5 sets of standard test functions,DPCEDE is compared with DE and DPDE through simulation.The experimental results show that DPCEDE has presented some advantages in both global searching ability and convergence than DE and DPDE.At the smae time the DPCEDE is applied to sovle the optimization of Flexible Job Shop Scheduling Problem(FJSP),and achieves the effective allocation of resources in flexible job shop which can effectively balance the relationship between workshop,equipment and process.In order to prove the validity and practicability of the proposed DPCEDE algorithm in sovling FJSP,the performace experiment of comparing with the basic DE algorithm and the basic PSO algorithm in sovling FJSP is done.(4)Directed against such defects as local optimum,low convergence,premature and large computational cost of the original TLBO algorithm,and under the guidance of DPCE model,this dissertation builds a dual-population co-evolutionary teaching-learning-based optimization algorithm(DPCETLBO).The two sub populations are divided and interflowed in the DPCETLBO.The sub population with greater adaptive value evolutes by using the TLBO with adaptive learning factors,and makes non-convex multi-parent crossover to elite individuals after the completion of learning process,so as to speed up convergence of the algorithm.The sub population with smaller adaptive value evolutes with the biased opposition-based learning TLBO(OLP_TLBO)algorithm to improve the global searching ability.After each round of iteration,the two sub populations merge together,regroup to two populations sorted by adaptive values,and keep the diversity of the populations to obtain a better solution quality.By using 2 sets of standard test functions,DPCETLBO is compared with TLBO and ETLBO through simulation.The experimental results show that DPCETLBO has presents obvious advantages in convergence precision and diversity.Finally,the DPCETLBO is applied to the optimization of PID controller parameters,and achieves the optimal solution of convergence after 22 times of iteration.Observing its step response signals through substituting the optimal solution parameters into the Simulink model of the PID controller,it can be found that the step response signals stabilize quickly.At the same time,compared with the basic TLBO algorithm and the basic PSO algorithm in solving the PID controller parameter optimization,this proposed algorithm is further proved to be practical.(5)Aiming at the advantages and disadvantages of DE and TLBO,this dissertation exquisitely combines DE and TLBO to build a dual-population co-evolutionary algorithm based on DE and TLBO(DPCEDT)by making best use of the advantages and bypassing the disadvantages under DECE model.In DPCEDT,dual-population independent evolution is set up.One sub population uses DE for evolution,the other sub population uses TLBO.In the evolutionary process of each generation,a composite individual is generated by combining the current best individuals in the two sub populations.Then,the composite individuals are used to further guide the co-evolution of the two sub populations.By this way,the excellent information of the two sub populations is integrated.On the one hand,the proposed algorithm maintains the diversity of the populations in searching process;on the other hand,it achieves the complementary advantages of DE and TLBO.In order to test the performance of DECEDT,13 sets of standard test functions are used to make simulation experiments in different dimensions compared DECEDT and jDE,TLBO and TLBO-DE,and two-tailed test is taken on the experimental results.The experimental results show that DPCEDT has higher optimization efficiency and searching precision.Finally,DPCEDT is applied to power battery SOC estimation,and compared with DPCEDE,DPCETLBO methods.The simulation experiment shows that DPCEDT has a good convergence in power battery SOC estimation,with estimation accuracy being raised by 2.54% than conventional method,which is a further proof of the practicability and superiority of DPCEDT.In summary,this dissertation studies on intelligent optimization algorithm from such multiple aspects as mechanism modeling,algorithm design and practical application.The DPCE Model proposed in this dissertation has strenghened a strong versatility.All of the three dual-population co-evolutionary algorithms design based on DPCE Model present their effectiveness and superiority in the simulation and practical application,and have a certain theoretical value and practical value.