| Many problems in real life do not exist in isolation,but are usually related to other problems.If we combine these problems and make good use of this potential relationship,we will effectively improve the efficiency of solving problems.Evolutionary multitasking is an algorithm paradigm that solves multiple optimization tasks at one time,and uses the internal relationship between tasks to help solve optimization problems.In this thesis,two methods are designed to effectively utilize the internal relationship between tasks,thus improving the performance of the algorithm.The main methods are as follows:The knowledge transfer process of the existing EMT algorithm is unidirection,that is,only the source task moves to the target task.However,this way of knowledge transfer ignores the needs of the target task and blindly transfers individuals.To solve this problem,this thesis proposes a bidirection knowledge transfer strategy.This strategy can help the algorithm select valuable migration individuals in the source task according to the search preference information of the target task,thus helping the task to carry out effective knowledge transfer.In addition,the current EMT algorithms has a fixed intensity of knowledge transfer,which seems unreasonable in the face of many different characteristics.The adaptive survival strategy of migrating individuals proposed in this thesis can adaptively adjust the intensity of knowledge transfer according to the survival conditions of individuals carrying transferred knowledge to cope with different optimization problems.The adaptive survival strategy of migrating individuals improves the generalization ability of the algorithm.The experimental results show that the proposed algorithm has certain competitiveness.In the study,we found that individuals with different characteristics in the source task have different effects on the target task.Some characteristics can help the target task search,while others can cause negative knowledge transfer.In order to find individual migration that can help optimize the target task.The proposed Evolutionary Multitask Algorithm based on Population Classification and Feedback mechanism(EMT/PCF)firstly labels the population according to the individual convergence characteristics and target preference characteristics.Then the algorithm selects migration individuals in potential groups according to the situation of knowledge transfer between tasks to minimize the negative knowledge transfer between tasks: if the migration individuals can promote positive knowledge transfer,it indicates that the currently selected group is helpful for the optimization of the target task,so the next iteration still selects the migration individuals in the group;If the transferred individual causes negative knowledge transfer,it indicates that the currently selected group is not suitable for helping the target task.In the next iteration,select the transferred individual from other groups to reduce the impact of negative transfer on the target task.In addition,the reverse learning search strategy is also introduced.The strategy expands the search scope by shrinking the boundary of the search space to generate the inverse solution of the offspring.The reverse learning search strategy improves the search ability of the algorithm.The final results show that the performance of the proposed algorithm is better than other comparison algorithms in the problem of the absolute logarithm. |
