Research On Multi-task Teaching-learning-based Optimization Algorithm

In recent years,evolutionary multitask optimization(EMTO)has received wide attention in the field of research.It makes full use of the potential parallelism of populationbased search to achieve cross-domain optimization of multiple optimization problems,and makes knowledge transfer between different optimization problems possible.In the real world,there are often optimization problems with similar attributes,which are mostly related.Compared with traditional optimization methods,EMTO have better performance in terms of convergence speed and solution quality than the optimization of a ask.At the same time,due to the advantages of easy to understand and few factors,teaching-learning-based optimization(TLBO)is often used to solve all kinds of problems.According to the similarity and complementary nature of the problems,the improved TLBO algorithm will be used to solve multitask problem,and three improved methods are proposed:(1)To improve the utilization of effective information between tasks.A new easy-toimplement teaching-learning-based optimization algorithm for multitask optimization(MTTLBO)is designed,in which students can accept knowledge transfer from teachers with the same or different attributes.It takes full advantage of knowledge transfer between different optimization problems,and improves the performance and efficiency of solving.The MTTLBO algorithm is compared with some more advanced evolutionary multitask algorithms.The experimental results show that the MTTLBO algorithm has good performance in solving single objective multitask problems.(2)To effectively reduce knowledge transfer of negative information and improve the fault tolerance rate.An improved teaching-learning-based optimization algorithm with multilearning strategy and ranking-based selection for multitask optimization(MTTLBO-MR)is further proposed.First,in the teaching stage,there are three learning strategies for students to choose.Second,in the learning stage,each student chooses from two strategies according to his or her situation.Third,unlike the greedy selection strategy of the TLBO algorithm,offspring selection is based on individual fitness values and diversity.The test results show that MTTLBO-MR algorithm had improved its knowledge through appropriate learning strategies at different stages of its improvement.The advantages of each strategy had been fully displayed.The search ability of TLBO algorithm and the diversity of the population has been improved.The MTTLBO-MR algorithm have superior performance in dealing with single-objective multitask problems.(3)To solve the problem of multi-objective multitask optimization,an improved teaching-learning-based optimization algorithm for multi-objective multitask optimization(MOMTTLBO)is proposed.The new groups are selected by a sorting method combining non-dominant rank and crowding distance.The teachers were randomly selected from the pareto optimal solution in each generation.The MOMTTLBO algorithm has been verified by simulation experiments and obtained good results.