Intelligent Fitting Method Based On Evolutionary Multi-objective Optimization

Aiming to find out expressions between variables among data,data fitting is often used to analyze and predict the experimental data generated in production and science.Data fitting methods can be divided into traditional fitting and intelligent fitting.The traditional fitting method usually employs interpolation or least square method to determine the parameters in the expression with given form,which obtains only one expression with fixed form and may result in low accuracy.Intelligent fitting method can evolve the expressions either form or parameters using genetic algorithms,which can obtain many different expressions to fit a given dataset.However,the process of intelligent fitting usually suffers blindness,bloat,overfitting and lack of diversity.This thesis aims to study the intelligent fitting method based on multi-objective evolutionary optimization from the perspective of designing efficient intelligent fitting algorithm.The main works of this thesis can be summarized as follows.Firstly,this thesis introduces the principle and process of the traditional fitting and the intelligent fitting method,meanwhile analyzes the advantages and disadvantages of these methods.Respectively,the traditional fitting needs to determine the form of expression in advance and just gets one expression,which can be solved in some way in the intelligent fitting method,but the intelligent fitting search process usually suffers blindness,bloat,overfitting and lack of diversity.Secondly,in order to reduce blindness of intelligent fitting,and guide the evolution direction,this thesis improves the intelligent fitting by the combination of traditional fitting and intelligent fitting,and designs two strategies,adding the expressions of traditional fitting only in the initial population and adding the expressions of traditional fitting in the evolutionary process.Thirdly,in order to reduce the running time of the intelligent fitting,this thesis presents a novel intelligent fitting algorithm based on decomposition.Original Intelligent fitting method fits the original data directly,and calculates the accuracy of the expression using all the data.If there are a large amount of data,the calculation of all individuals will spend a lot of time.This thesis treats the fitting data decomposition,directly using decomposition of the data to fit,to reduce time of the calculation of expressions,thereby reduce the running time of the intelligent fitting.Fourth,this thesis has evaluated and verified the proposed algorithm using the experimental data,and compared with other algorithms from the expression of the three aspects of accuracy,prediction accuracy and time consuming.The experimental results show that the proposed algorithm has good performance.