Research On Improved Multi-Task Optimization Algorithm Based On Multifactorial And Their Applications

There are different approaches to solve optimization problems,some of which do not assume any prior knowledge about the task under consideration.However,real-world problems rarely exist in isolation,and humans can effectively manage and perform multiple tasks simultaneously.Inspired by the parallel mechanism of human brain,Evolutionary Multi-task Optimization(EMTO)paradigm is proposed.Compared with the traditional evolutionary single-task optimization,EMTO algorithms processes multiple optimization tasks in parallel by search operators(crossover,variation,etc.),distribution estimation and other methods,taking advantage of the potential complementarity between multiple tasks.The potential complementarity is reflected in the positive knowledge transfer,which improves the quality of task solutions by sharing excellent knowledge between tasks.Therefore,a key problem for the EMTO algorithms is how to facilitate the positive transfer of knowledge between tasks to improve the efficiency of multiple task.Based on Multifactorial Optimization(MFO),this paper conducts an in-depth study of this problem,proposes two specific algorithms,and verifies the performance of the proposed algorithms for practical applications in network reconstruction problems.Specific contents and innovations are as follows:1.A memory-based multidimensional evolutionary multi-task optimization is proposed:This paper focuses on the research of assisted task selection and effective knowledge transfer of EMTO from two perspectives of time series and individual multidimensional,and combines the idea of population optimization.a memory-based assisted task selection strategy,an individual dimension-based knowledge learning strategy and a population layered optimization strategy are constructed,and then a memory-based multidimensional evolutionary multitask optimization algorithm is proposed.(1)The memory-based assisted task selection strategy measures the utilization rate of a task by remembering the number of times for the task is selected as an assisted task at historical moments,matching the appropriate assisted task for the target task.This strategy ensures the structural unity of the knowledge transferred from the assisted tasks at different moments.(2)The individual dimension-based knowledge learning strategy focuses on each dimension of the individuals,and strengthens the beneficial knowledge of each dimension of the individual by calculating the fitness value proportion of the excellent individuals in the assisted task,which promotes the positive knowledge transfer between tasks.(3)The population layered optimization strategy divides the population into an exploitation layer,a diversity layer and an exploration layer according to the ranking of individual fitness value.The individuals in the three layers coevolve to achieve a trade-off between exploration and exploitation.The performance of the proposed algorithm is evaluated on two benchmark test suites,The experimental results show that the proposed algorithm is highly competitive in promoting positive knowledge transfer between tasks.2.A evolutionary multi-task optimization algorithm with influence control of transfer knowledge is constructed:This paper studies the intensity of knowledge transfer between tasks and the real effectiveness of knowledge transfer,and proposes an evolutionary multi-task optimization algorithm with knowledge transfer influence control,which includes knowledge transfer parameter adjustment mechanism,knowledge transfer influence control strategy and whale optimization strategy.(1)In order to prevent the saturation of knowledge transfer between tasks,a knowledge transfer parameter adjustment mechanism is established,which makes the random matching probability controlling knowledge transfer fit different evolutionary stages of the population by introducing the number of iterations into the update method of the random matching probability.(2)In order to make full use of inter-task transfer knowledge,a transfer knowledge influence control strategy is conceived,which configures an influence controller for inter-task transfer knowledge,and if the transfer knowledge has no influence to lead the evolution of multiple tasks,new transfer knowledge will be updated in time to achieve inter-task knowledge sharing.(3)To balance the exploration and exploitation capabilities of the population,a whale optimization strategy is introduced to improve the efficiency and precision of the population optimization.An empirical study on the effectiveness of the proposed algorithm is carried out through two comprehensive benchmark test suite.The experimental results show that,compared with the latest EMTO research,the proposed algorithm has superior performance in promoting the positive knowledge transfer between tasks.3.The proposed evolutionary multi-task optimization algorithm is applied to the network reconstruction problem:To verify the performance of the improved algorithm for practical applications,the memory-based multidimensional evolutionary multi-task optimization and the evolutionary multi-task optimization algorithm with influence control of transfer knowledge are used to optimize the dual network reconstruction task.According to the superior performance of the two algorithms in facilitating positive knowledge transfer between tasks,the correlation between network reconstruction tasks is exploited to improve the efficiency of solving the dual network reconstruction problem.The experimental results show that both proposed algorithms outperform the latest research results in solving the network reconstruction problem.