Research On Multitasking Genetic Programming Algorithm For Symbolic Regression Problems

At present,the process of analyzing and establishing the mapping relationship between different variables based on data is called data-driven modeling.As a main data modeling method,symbolic regression searches the optimal form of function and parameter set of the given problem at the same time.When there is little prior knowledge of model structure or data distribution,symbolic regression is a powerful regression technology and has flexible expression,which is widely used in industrial production,scientific research and other fields.Genetic programming algorithm is the mainstream method to solve the symbolic regression problems.However,traditional genetic programming can only solve one task at a time,which is inefficient when multiple tasks need to be solved at the same time.In this case,multifactorial optimization is proposed as a new evolutionary multi-task processing paradigm.It attempts to search multiple tasks simultaneously in a single run.At the same time,we should also note that in practical application scenarios,problems rarely exist in isolation,and problems often contain useful information.If this property is properly used,when another related problem is encountered,the problem-solving process can be effectively enhanced.To solve the above problems,an adaptive multifactorial genetic programming algorithm is proposed to solve multiple symbolic regression tasks simultaneously.In order to take into account the knowledge transfer between tasks and improve the quality of solutions in the process of evolution,an adaptive control strategy is adopted for the control parameters in the cross-recombination operation to adjust the evolutionary behavior.This paper also designs an adaptive computing resource reallocation mechanism to balance different tasks,so that tasks with relatively poor performance can get more computing resource,so that all concurrent tasks can get the optimal solution as much as possible.In addition,most of the current research on multi-task optimization in genetic programming only studies the concurrent search of two or three tasks,and lacks the exploration of simultaneous concurrency of more tasks.In the experiment part,we will further explore the performance of multifactorial optimization in the case of more concurrent tasks.In order to verify the effectiveness of the proposed method,a series of experiments are designed based on the complex symbolic regression benchmark problem to verify the effectiveness of the proposed algorithm.