Research On Multi-task Evolutionary Algorithm And Its Applicatio

Evolutionary algorithm is an "algorithm cluster" designed based on the population evolution pattern in nature,in which operators rely on random selection,mutation and reorganization to make the population evolve to a better region in the candidate solution space.Multi factor optimization(MFO)solves multiple optimization problems at the same time through evolution in a single search space.In MFO,the differences of common knowledge between these optimization problems are dynamically used to help these different tasks and achieve the purpose of optimization through valuable knowledge exchange.Multi factor evolutionary algorithm(MFEA)is the first MFO algorithm to realize multi task evolution.In the process of population evolution of MFEA,we use classified mating and vertical culture transmission to produce offspring,so as to realize knowledge transmission and optimize by using the tacit knowledge in these multi tasks.The concept of classified mating ensures that individuals from the same task have a high probability of mating and producing offspring.This evolutionary strategy has been successfully applied to single objective and multi-objective optimization problems.In multi-objective optimization problems,its variant is MOMFEA.In this paper,the MFEA algorithm for single objective problem and the MOMFEA algorithm for multi-objective problem are fully studied and analyzed.Specifically,this topic studies the multi task evolutionary algorithm from the following three aspects:(1)For the MFEA algorithm of single objective problem,the cross mode of Lévy flight search is introduced in the process of evolution,and the MFEA-LF algorithm is proposed.The crossover mode of Lévy flight search is to use Lévy flight distribution to effectively explore information and optimize the crossover process,which promotes the crossover of effective information between individuals and the effective knowledge transfer between tasks,and more effectively improves the population quality and optimization effect of offspring.In the experimental part,the MFEA-LF algorithm is tested in the single objective problem set with different characteristics,and the results are compared with many classical multi task evolutionary algorithms such as MFEA and MFEA-Ⅱ to verify its performance improvement in the single objective multi task evolutionary problem.Then,the Lévy flight strategy is applied to the multi-objective problem to derive the MOMFEA-LF algorithm,and compared with MOMFEA and MOMFEA-Ⅱ on the multi-objective optimization problem set to verify its performance improvement in the multi-objective and MFO problem.(2)When MFEA deals with multi-objective problems,Gupta et al.Proposed MOMFEA algorithm,which uses NSGA-Ⅱ as the non-dominated sorting basis of evolutionary algorithm.This topic has fully studied the current advanced non-dominated sorting algorithm,found an efficient non-dominated sorting strategy that can reasonably replace NSGA-Ⅱ in MOMFEA,that is,non-dominated sorting based on sequential search strategy,The operation efficiency of MOMFEA is greatly improved.This optimized algorithm is called FMOMFEA.(3)Due to the multilayer nature of real-world systems,the problem of inferring multilayer network structure from nonlinear and complex dynamic systems is very important in many fields,including engineering,biology,physics and computer science.In order to solve this problem,many network reconfiguration methods have been proposed.Considering the similarity between network reconfiguration tasks of different component layers,inspired by the topological Association and dynamic coupling between different component layers,the multi-layer network reconfiguration problem is first abstracted as a multi-task multi-layer network reconfiguration problem,in which the goal of each task is to reconstruct the network structure of the component layer.In this paper,MFEA-LF framework is used to evolve multiple multi-layer network structure models at the same time to improve the reconstruction performance.The results show that this algorithm can improve the success rate of reconstruction and accelerate the convergence speed of multi-layer network reconstruction.