单沟泥石流危险性评价模型相关问题研究

基于云南地区、黄河积石峡水库区、四川省的北川县和都江堰龙池地区等地的泥石流数据，以具有代表性的灰色关联分析（Grey Relation Analysis, GRA）和支持向量机（Support Vector Machine, SVM）泥石流评价模型为例，探讨了单沟泥石流危险性评价模型在参数选取、样本数据的不均衡、泛化能力和泥石流系统的空间变异性等方面存在的问题。结果表明：寻优算法能够提高模型参数选取的效率和预测精度；样本扩充在一定程度上能够处理样本数据不均衡问题；泛化能力为模型固有属性，难以通过样本扩充得到提升；空间变异性通过控制指标的重要程度进而影响模型的精度。研究过程为单沟泥石流危险性评价模型相关问题的研究提供了新的思路，所得结论将为今后各类泥石流危险性评价模型运用提供指导。

Related Problems of Single Gully Debris Flow Risk Assessment Model

In recent years, a variety of debris flow risk assessment models have been put forward. But due to the effect of various factors on the occurrence mechanism of debris flow, those evaluation models inevitably have problems in the prediction process. In order to improve the accuracy of single gully debris flow risk assessment and promote the application of various evaluation models in the field of debris flow risk assessment, based on the debris flow data of Yunnan area, Jishixia Reservoir area in the Yellow River, Beichuan County and Longchi area of Dujiangyan in Sichuan Province, this paper takes the two representative debris flow evaluation model, Grey Relation Analysis (GRA) model and Support Vector Machine (SVM) model, as an example. And then this paper discusses the problems in the selection of parameters, the imbalance of sample data, the generalization ability and the spatial variability of debris flow system in the single gully debris flow risk assessment models. The results show that: the optimization algorithm can improve the efficiency of model parameter selection and prediction accuracy; the generalization ability is an inherent property of the model, which is difficult to be improved through sample expansion; spatial variability affects the accuracy of the assessment model by controlling the importance of the evaluation index. The research process of this paper provides a new way to deal with the related problems of single gully debris flow risk assessment model, and the conclusion will provide guidance for the application of various debris flow risk assessment models in the future.