Research On Transfer Mechanism Of Multi-Tasking Evolutionary Algorithms And Its Application

In recent years,the concept of multi-tasking optimization has opened up a new research direction for intelligent optimization algorithms.Different from the traditional single-tasking evolutionary algorithm,the core idea of multi-tasking evolutionary algorithm is to use the potential correlation between tasks to transfer the potential beneficial information of other tasks to the target optimization problem through the knowledge transfer mechanism,so as to improve the performance of the algorithm.With the deepening of the research,the potential problems of the algorithm have been discovered.For example,the effect of solution variables change on the multi-tasking transfer mechanism is not clear;For another example,the knowledge transfer mechanism of original multi-tasking evolution algorithm may produce negative transfer effect,which leads to the low efficiency of the algorithm.To solve the above two problems,on the one hand,this paper explores and improves the original multi-tasking evolutionary algorithm.On the other hand,the application of multi-tasking evolutionary algorithm in the optimization of receiver operating characteristic curve is discussed by referring to the latest findings in the field of machine learning.Specifically,the research work of this thesis mainly includes the following three aspects.(1)In view of the unclear research status of the effect of solution variable order transformation on the multi-tasking transfer mechanism,combined with the original multi-tasking evolutionary algorithm framework,a qualitative analysis was carried out on the sequence characteristics of solution variables,and three kinds of solution variables with increasing complexity were designed to explore the impact of solution variable order transformation on the performance of the multi-tasking evolutionary algorithm.The experimental results show that the transformation of solution variable order has no significant effect on the multi-tasking transfer mechanism,while the multi-tasking optimization problem with high similarity and intersection degree is more sensitive to the transformation of complex solution variable order,and is more susceptible to the influence of solution variable order transformation.(2)Aiming at the problem of negative knowledge transfer in the multi-tasking evolutionary algorithm,this paper proposes a multi-tasking differential evolution algorithm based on the adaptive knowledge transfer mechanism.In this algorithm,the transfer probability between tasks is adjusted according to the similarity between tasks,so as to improve the forward knowledge transfer between tasks.Meanwhile,a differential transfer strategy based on super-particle guidance is proposed to increase the efficiency of knowledge transfer between tasks.In the experimental part,the effectiveness of the two strategies is verified independently,and compared with other algorithms,the effectiveness of the proposed algorithm is proved.(3)In view of the shortcomings of the traditional evolutionary algorithm in solving the receiver operating characteristic curve optimization problem,such as long computation time and unsatisfactory solution quality,combined with the previous research,this paper proposes a new multi-tasking evolutionary algorithm framework to solve the optimization problem.In the proposed algorithm,a multi-tasking receiver operating characteristic curve optimization environment is constructed based on the original receiver operating characteristic curve optimization problem,and then combined with the variable order transformation strategy,the proposed adaptive knowledge transfer mechanism multi-tasking differential evolution algorithm is used to optimize the environment.The experimental results show that the proposed algorithm can provide a competitive solution for the receiver operating characteristic curve optimization in neural networks.To sum up,this paper conducts an in-depth study on the transfer mechanism of multi-tasking evolutionary algorithms.First,it explores the influence of variable order transformation on the transfer mechanism of multi-tasking.Then,an adaptive transfer strategy and a differential transfer operator are proposed to improve the multi-tasking transfer mechanism,so as to improve the performance of the multi-tasking evolutionary algorithm.Finally,the proposed multi-tasking evolutionary algorithm is used to solve the actual optimization problem in the field of machine learning.