| 引用本文: | 张晨,王清,张文,谷复光.基于神经网络对泥石流危险范围的研究[J].哈尔滨工业大学学报,2010,42(10):1642.DOI:10.11918/j.issn.0367-6234.2010.10.026 |
| ZHANG Chen,WANG Qing,ZHANG Wen,GU FU-guang.Prediction on hazardous areas of debris flow based on neural network[J].Journal of Harbin Institute of Technology,2010,42(10):1642.DOI:10.11918/j.issn.0367-6234.2010.10.026 |
|
| 摘要: |
| 为了更加客观准确地预测泥石流危险范围,通过对云南金沙江流域的各类泥石流进行深入调查分析,提取出对泥石流危险范围有主要影响的几种因素的指标值.利用改进BP神经网络的学习能力分析几种影响因素对泥石流危险范围的敏感程度,提出以误差曲线的斜率作为敏感程度定量指标,以误差系数的形式对不同种类泥石流危险范围的影响因素进行定量的评定,从流体力学的角度深入解析结论,并以此为依据提出一个修正公式,对传统预测模型进行修正.在实例模拟中,修正后的模型得到了更加准确的预测结果,相对误差最多相差4.54. |
| 关键词: 神经网络 泥石流危险范围 影响因素 误差系数 |
| DOI：10.11918/j.issn.0367-6234.2010.10.026 |
| 分类号:P642.23 |
| 基金项目:国家自然科学基金资助项目(40872170);吉林大学985工程资助项目(450070021107) |
|
| Prediction on hazardous areas of debris flow based on neural network |
|
ZHANG Chen1, WANG Qing1, ZHANG Wen1, GU FU-guang1,2
|
|
1.College of Construction Engineering,Jilin University,Changchun 130026,China;2.School of Surveying and Prospecting Engineering,Jilin Architecture and Civil Engineering Institute,Changchun 130021,China)
|
| Abstract: |
| By in-depth investigation and analysis of debris flows in Jinsha River watershed,the index values of several factors mainly the affecting hazardous areas of debris flow are extracted to predict the hazardous areas more objectively.The capability learning of improved BP neural network is used to predict the sensitivity of these factors.The slope of error curve is presented as the quantitative indicator of sensitivity.The factors affecting hazardous areas of various types of debris flows are assessed with the error coefficient and the conclusions are analyzed from the view of fluid mechanics.A new formula is proposed to improve the traditional forecasting model.In the example simulation,the improved model gets more accurate forecasting results.The difference between relative errors calculated by the improved model and the traditional model is 4.54 at most. |
| Key words: neural networks risk areas of debris flow factors error coefficient |
