论文部分内容阅读
【目的】比较神经网络算法和传统统计建模方法对土壤盐渍化预测模型的效果。【方法】对渭干河流域多年土壤盐渍化和其影响因子进行分析的基础上,采用BP网络的3种算法,建立基于BP神经网络土壤盐渍化预测模型。将预测结果与多元线性回归模型预测结果进行对比分析,讨论线性和非线性方法用于土壤盐渍化预测模型。【结果】与传统的统计建模方法相比BP神经网络结构简单、快捷,预测精度高,很好地再现了土壤盐渍化与其影响因素之间复杂的非线性函数关系;三种BP算法中,基于trainlm算法建立的壤盐渍化预测模型具有较好的推广能力。【结论】BP神经网络的土壤盐渍化预测性能良好,用来可以预测土壤盐渍化情况。
【Objective】 The objective of this paper is to compare the effects of neural network and traditional statistical modeling on soil salinization prediction model. 【Method】 Based on the analysis of soil salinization and its influencing factors in Weigan River basin for many years, three models of BP neural network were used to establish the soil salinization prediction model based on BP neural network. The prediction results are compared with the results of multivariate linear regression model to discuss the linear and nonlinear methods for soil salinization prediction model. 【Result】 Compared with the traditional statistical modeling methods, BP neural network has the advantages of simple and fast structure, high prediction accuracy and well reproduced complex nonlinear function relationship between soil salinization and its influencing factors. Among the three BP algorithms , The soil salinization prediction model based on the trainlm algorithm has a good promotion ability. 【Conclusion】 The prediction of soil salinization of BP neural network is good and can be used to predict the salinization of soil.