基于PCA-LSTM的多变量矿山排土场滑坡预警研究

Early Warning of Landslide in Mined Mine Dumping Site Based on PCA-LSTM

针对矿山排土场滑坡的过程是一个动态、大延迟、情况复杂的特性问题,影响矿山排土场滑坡的因素众多,各个特性指标间相互影响,而关于排土场滑坡预警并没有严格的划分标准,本文提出主成分关联长短期记忆(PCA-LSTM)网络模型,利用主成分分析和关联性分析,挖掘出所有特性指标当中的第一主成分,与第一主成分关联性较强的其它特性指标,将得到的其它特性指标和第一主成分作为预测排土场滑坡的主要特性指标,利用长短期记忆网络(LSTM)在处理时间序列问题能够将现有输入的信息与历史信息相互结合的特点;采用LSTM网络模型通过多个其它特性指标对第一主成分地表位移指标进行预测,并取得了较好的效果.

The process of landslides for mine dumps is a dynamic, large-delay, and complex situational problem. There are many factors that affect the landslide in mine dumps, and each characteristic index influences each other. However,there is no strict categorizing standard for index of landslide warning for dumping sites. This study proposes Principal Component Analysis Long-Term and Short-Term Memory network（PCA-LSTM） model, using PCA and correlation analysis, mining the first principal component among all the characteristic indicators, and the other indicators with strong correlation with the first principal component. The obtained other characteristic indexes and the first principal component are used as the main characteristic indicators to predict the dumping landslide, and the LSTM is used to combine the existing input information and the historical information when dealing with time series problems. The LSTM model predicts the displacement of the first principal component through a number of other characteristic indicators and has obtained sound results.