基于极限学习机和熵值法的岩土灾变预警

滑坡的监测和预测是降低滑坡灾害的有效手段和可行方法。传统监测手段时效性差,采用统计预报模型、确定性模型等模型建立的预测模型精度相对较低。为了改善此问题,提出了一种基于综合测量、以极限学习机与熵值法结合的滑坡预警研究方法。通过分析滑坡的影响因子,搭建滑坡模拟监测平台,由多传感器实时监测到雨量、土壤浅层含水率、土壤深层含水率、下滑应力、地下位移、地表位移等影响滑坡的综合因子。将熵值法用在滑坡的评价中,将其综合评分作为危险性参数及综合测量参数作为训练样本,搭建极限学习机模型。结果表明:在综合测量方法下,将熵值法与极限学习机算法结合的预警模型得到的结果与实际情况一致,预测值与测量值吻合;其精度为98.48%,比BP神经网络精度更高;且网络的学习速度明显提高。可见该方法对滑坡预测的可行性,适用于复杂非线性的滑坡预测中,为建立滑坡预警模型提供了一种可行方法。

Rock and Soil Catastrophe Early Warning Research Based on Extreme Learning Machine and Entropy Method

Abstract] Monitoring and forecasting of landslides is an effective means and feasible method to reduce landslide disasters. The traditional monitoring methods have poor timeliness, and the prediction models established by using statistical forecasting models and deterministic models are relatively inaccurate. In order to improve this problem, a landslide early warning research method based on comprehensive measurement and combination of extreme learning machine and entropy method is proposed. Analyze the impact factors of landslide, a landslide simulation monitoring platform was build, and monitor the comprehensive parameters affecting landslide by rainfall, soil shallow water content, soil deep water content, sliding stress, underground displacement, etc. Entropy method is applied to the evaluation of landslide, and its comprehensive score is taken as the risk parameter and the comprehensive measurement parameter as the training samples to build an extreme learning machine (Extreme Learning Machine). The results show that under the comprehensive measurement method, the results of the early warning model which combines the entropy method with the extreme learning machine algorithm are consistent with the actual situation, and the predicted values are same as the measured values; the lowest accuracy of the model is 98.48%, higher than that of the BP neural network; and the learning speed of the network is obviously improved. It is concluded that this method is feasible for landslide prediction, which is suitable for complex and non-linear landslide prediction, and provides a feasible method for landslide early warning model.