Model-based Multi-objective Algorithm And Its Application In Gene Networks

Optimization problems wildely arise in many fileds,such as engineering design,manufacturing and science research.There is a kind of optimization problem that needs to optimize serveral objectives at the same time,which is called multi-objective optimization problems(MOPs).MOP includes several objectives that each objective tends to have nonlinear,non-differentiable characteristics and each objective may conflict with others.Therefore,the traditional single-objective optimization method is no longer suitable for dealing with multiobjective optimization problems.The most widely used method for dealing with such problems at present is the multi-objective evolutionary algorithm(MOEA).It is designed based on the biological evolutionism.The algorithm mainly includes three steps of selection,crossover and mutation.In recent years,many related researches have appeared based on these three steps.This thesis first analyzes the traditional operation operators(Simulated Binary Crossover,Differential Evolution),and then uses the mathematical statistics knowledge to model the crossover operators,changing the view that the traditional crossover operators have been questioned without the theoretical basis of mathematics.The method mainly uses the analyzed search mode to reconstruct the search mode of the operator using the sampling points in the mathematical statistics,and finally uses the built search model to generate the solution.The modeled operator can not only ensure the original search characteristics of the operator,but also increase the diversity of the operator.Then,this thesis introduces Gaussian operators based on probability models and summarizes the current research methods using Gaussian operators.An evolutionary algorithm(MOEA/D-AMG)that adaptively uses Gaussian operators is proposed.The algorithm avoids the Gaussian model using only one(0,1)distribution.Moreover,in order to ensure that individual information is not lost,we use the individual in the neighbors of the MOEA/D framework to complete the modeling process.Finally,this thesis uses the idea of multi-objective algorithm to solve the practical problems of genetic network(such as: protein-protein interaction network)in real life.The PPI network is modeled into two optimized targets(sequence similarity and topological similarity),and then a MOEA/D-Net algorithm is proposed based on the decomposition idea in the multiobjective algorithm.For the first time,the algorithm integrates the decomposition ideas in the multi-objective algorithm,and proposes to adjust the relationship between the edge and the node in the initialization stage to obtain the initial population with strong diversity;then in the intersection stage,the greedy algorithm is used to adjust the node comparison relationship between the node and its corresponding domain,so that the solution corresponding to each subproblem converges toward the direction of the weight vector,and a candidate solution set with strong convergence and diversity is obtained.