一种基于遗传算法的回归模型寻优方法
A Regression Model Optimization Method Based on Genetic Algorithm

回归分析是数据分析和建模的重要工具，主要用于数据的预测和拟合。回归分析通常需要人工干预给定参考模型，再进行参数回归。然而，在多数情况下，用户难以给出参考模型，或者给出模型具有较大的误差。本文提出一种基于遗传算法得出回归模型的方法，主要利用遗传进化的思想，首先随机产生初始模型的种群；然后不断迭代的进行选择、交叉、变异操作，在解空间中动态地进行全局寻优，找出一个较优的模型；为了确定模型的参数，又利用梯度下降法对该模型进行参数估算。最后，将本文得出的模型与最小二乘法回归分析得出的模型进行对比，结果表明，在进行预测时，前者的误差比后者有显著减小，由14.24%减少到9.59%。
Regression analysis is an important tool for data analysis and modeling, mainly used for data pre-diction and fitting. Regression analysis usually requires manual intervention of a given reference model followed by parametric regression. However, in most cases, it was difficult for the user to given a reference model or given the model a large error. It proposed a method based on genetic algorithm to obtain regression model. It mainly used the idea of genetic evolution to first randomly generated an initial model populations; then iteratively selected, crossed, and mutated operations, perform global optimization dynamically in the solution space to find a better model; in order to determine the parameters of the model, the gradient descent method is used to estimate the parameters of the model. Finally, the model obtained in this paper is compared with the model obtained by least squares regression analysis. The results show that the error of the former is sig-nificantly reduced from the previous one, from 14.24% to 9.59%.