基于随机森林的区域水资源可持续利用评价

针对区域水资源可持续利用评价中指标多、噪声复杂和非线性的特点以及传统方法缺乏可操作性、难以解决稳健性低和过学习等问题,介绍了一种稳健性较高的智能学习方法——随机森林,将其应用于区域水资源可持续利用评价中,并以汉中盆地平坝区为例,对该方法的评价效果进行了验证。结果表明,与SP插值、人工神经网络(ANN)和支持向量机(SVM)模型评价结果相比,本文方法实用性强、稳健性较高、泛化性能高,在分类预测阶段和交叉验证阶段分类准确率均高达100%;同时可知,在影响区域水资源可持续利用的各评价指标中,水资源利用率和人均供水量的影响较为重要。

Assessment of Sustainable Utilization of Regional Water Resources Based on Random Forest

In connection with the problems of the characteristics of more indicators, complex noise and nonlinear in the sustainable utilization level assessment of regional water resources, as well as lacking maneuverability, poor robustness and over-fitting of the traditional assessment methods, an new assessment model based on random forest ( RF) which is more robust intelligent learning method was put forward in this paper and applied to assess sustainable utilization of regional water resources in Hanzhong basin in China. Compared with assessment results of the SP method, artificial neural networks and support vector machine model, it shows that the new method was more practical, stronger robustness and generalization. Especially in the model classification prediction phase and cross validation phase ,the classification accuracy rates are up to 100%. And we also draw a conclusion that among all explanatory variable which affect sustainable utilization of regional water resources, the factors of the utilization of water resource and the supply of water resources per person are more important.