Research On Differential Evolution Algorithm And Its Applications In Wireless Sensor Networks

The Differential Evolution(DE)algorithm is a kind of evolutionary algorithm with strong global optimization ability,and it has been widely used in many fields of engineering practice.However,it still has some limitations when solving the problems with high coupling degree of decision variables,complex nonlinear constraints,complex irregular Pareto Front(PF).At the same time,the existing DE algorithm and its improvement are mainly based on the random iteration pattern to search the optimal solution,which lacks the analysis and utilization of the problem feature,and rarely designs the problem-oriented operating mechanism,thus making the search process random.Therefore,this thesis focuses on several typical complex optimization problems such as high coupling degree of decision variables,complex constraints,irregular Pareto frontier,and designs a targeted mechanism to embed differential evolution algorithm according to the characteristics of the problems,so as to improve the performance of differential evolution algorithm in solving complex optimization problems.The main research work are as follows:(1)An adaptive DE algorithm guided by construction learning strategies is proposed for single objective optimization problems with coupled decision variables.To avoid the negative impact of the poor quality dimension in the current population optimal solution,and to balance the exploration with exploitation,the proposed algorithm constructs the optimal solution of the current population with the excellent dimension combination of each subproblem,and guides the population to learn from the constructed optimal solution by taking advantage of potentially useful information of multiple excellent individuals after decomposing the problem.Meanwhile,an adaptive mutation strategy guided by search preference knowledge is designed.The prior search preference knowledge is applied to the evolution process of the population,and the diversity and convergence of the population are given different degrees of importance at different evolution stages,and the global detection and local development are adaptively switched.The comparison results on small scale and large scale test problems verify the effectiveness of the proposed algorithm.(2)A constrained DE algorithm based on multi-stage strategy is proposed for constrained single objective optimization problem.Due to the effect of constraints,the evolutionary algorithm for solving constrained optimization problems has a distinct stage characteristic in the evolution process.Therefore,the proposed algorithm divides the evolutionary process into three stages,namely unfeasible stage,semi-feasible stage and feasible stage,based on the feasible solution ratio of the population.Then,a targeted mechanism is designed according to constraint processing mechanism and search strategy.In terms of constraint processing mechanism,constraint violation degree and objective function comprehensive ranking method are used to process unfeasible solutions at unfeasible and semifeasible stages,respectively.In terms of search mechanisms,the best combination of complementary search strategies are selected for each stage from the four strategies,i.e.“DE/current-to-rand/1”(more emphasis on diversity/random perturbation),“DE/current-topbest/1”(more emphasis on convergence/objective function),“DE/rand-to-best/1”(more emphasis on feasibility/constraints),and “DE/rand/1/levy”(more emphasis on searching over a large area/jumping out of a locally optimal area).Experimental results show that the proposed algorithm performs well in both accuracy and feasibility of the optimal solution,and it can effectively handle constraints while optimizing the objective function.(3)A multi-objective DE algorithm based on Bi-level environment selection is proposed for multi-objective optimization problems with complex and irregular PF.Complex and irregular Pareto fronts greatly increase the difficulty for the optimal solution set to completely cover the PF and uniformly distribute,making it impossible to simultaneously consider convergence and diversity.To solve this problem,the proposed algorithm uses“DE/current to rand/1” and “DE/rand to test/1” mutation strategies to generate two temporary populations with diversity priority and convergence priority,respectively.For the collection of temporary populations,the first level environmental selection is performed through fast nondominated sorting and crowding distance sorting to select individuals with good convergence and diversity.Then,a single point cyclic clipping method based on Euclidean distance with better diversity is employed to perform the second level environmental selection on the results of the first level environmental selection and the parent individuals,and the selection results enter the next generation population.Experimental results on multiple test sets such as UF,WFG,and Ma F show that the algorithm performs well on test problems with complex PF.(4)A constrained multi-objective DE algorithm based on dynamic selection strategy is proposed to solve the constrained multi-objective optimization problem with complex objective space.Constrained multi-objective optimization problems with complex objective spaces such as small or discontinuous feasible regions may cause the optimal solution set to fail to converge well to the constrained PF,or fall into local optimization.To solve the above problems,the proposed algorithm designs two indicators,objective function priority and constraint condition priority,to dynamically control the selection of individuals,ensuring that the population spans unfeasible regions while considering the objective function.At the same time,the generation strategy of new solutions is selected according to the proportion of feasible solutions to adjust the search direction and evolution trend of the algorithm.The experimental results on MW,DASCMOP and LIRCMOP test sets show that the proposed algorithm can effectively balance the feasibility,diversity and convergence of the population,and successfully overcome the constrained multi-objective optimization problem with complex objective space.(5)A problem feature driven multi-objective DE algorithm based on two-level environment selection framework is proposed to solve the cluster head selection problem in practical applications of wireless sensor network(WSN).First,the characteristics of WSN applications such as “optimal cluster head ratio,correlation between targets” are analyzed.Then,the algorithm is initialized based on the optimal cluster head ratio to generate a high-quality initial population and accelerate the convergence of the algorithm.Finally,nonlinear correlation information entropy is used to detect the correlation between objective functions,reducing redundant objectives with high correlation to decrease the dimension and difficulty of solving the problem.Experimental results show that the proposed algorithm performs better than the standard multi-objective optimization algorithm in WSN life cycle and node energy consumption,and can well deal with WSN application problems.