基于BP神经网络泥石流沟发育阶段的判定——以成昆铁路四川段和昆东铁路为例

DETERMINATION ON EVOLUTION STAGE OF DEBRIS FLOW GULLY BASED ON BP NEURAL NETWORK

泥石流沟谷发育阶段的判定是进行泥石流预测、评价和防治的第一步，它影响着泥石流发生的规模与频度。利用人工智能BP神经网络，选取流域地貌上的面积、主沟长、沟床比降、山坡平均坡度、相对高差、圆状率和相对切割程度等这7个因素作为泥石流沟谷发育阶段评价的指标，并将泥石流沟发育阶段分为4个等级——幼年期、发展期、活跃期和衰退期。利用成昆铁路80条泥石流沟资料，将数据进行预处理（系统分类、标准化），对其进行网络训练建立预测模型，通过模拟，平均相对误差为8．22％，模拟结果达到要求。并利用该模型对昆东铁路沿线的6条泥石流沟进行了泥石流沟发育阶段的判定，结果表明这6条泥石流沟均处于发育的活跃阶段，其发育阶段值为3．0～3．5，因此要加强对这6条泥石流沟的监测和预防工作，避免引起灾难的发生。结果可为泥石流预测、评价和防治提供理论基础和技术支撑。

The determination of evolution stage of debris flow gully is the first step of forecasting, evalua- tion and control of debris flow and the scope and frequency of the debris flow. Utilizing artificial three-ply intelligence-BP neural network model, then selecting catchment area, main groove length, groove gradient ratio, average aspect, relative height difference, round ratio and relative cutting degree of the catchment evolution geomorphology as assessment index of the evolution stage of debris flow gully, the stage of debris flow gully was divided into four stages, young stage.developing stage, active period and decline stage. Authors pre-treated the 80 debris flow data along Chengkun railway in Sichuan province （systematic classifi- cation.standardization） in order to avoid artificial error, secondly, network training the 80% of the data, and then built forecasting model,the simulation of the residual 20 % of the data shows that the average relative error is 8.22% with satisfactory result. The evolution stage of the 6 debris flow gullies along Kundong railway was determined based on the model, the result showed that：the 6 debris flow gullies are in active period and the evolution score is between 3 -3.5, so the monitor and forecast should be strengthened in these six debris flow gullies in order to avoid disasters. The intelligence-BP neural network model can be used as an advantageous method to determinate the evolution stage of debris flow gully. The result can provide theoretical basis and technical support for debris flow forecasting, evaluation and control.