基于混合高斯隐马尔科夫模型的滑坡发生时间预报∗

Landslide Disaster Occurrence Time Prediction Algorithm Based on MOG⁃HMM

摘要:

滑坡发生时间预报在防灾减灾工作中非常重要，准确的预报能够有效预防灾害发生可能造成的灾难性结果。 为解决当前滑坡预报中仅仅实现对滑坡位移等相关参数的预测和估计，而未最终计算出滑坡发生时间的问题，提出采用混合高斯隐马尔科夫模型(MOG?HMM)建立滑坡发生时间预报模型，即对滑坡灾害演化过程全周期数据利用混合高斯算法计算出宏观信息预报判据，与隐马尔科夫模型中的状态相匹配，建立滑坡演化状态模型，该模型能够反映全周期数据的多个状态，当需要对实时采集的位移数据进行时间预报时，首先利用解码算法对当前数据解码，即计算出其属于滑坡的哪个状态，然后利用 Dijkstra 最优路径规划算法，计算出从当前状态到达滑坡发生状态的时间，实现滑坡发生时间预报。通过对新滩滑坡和卧龙寺滑坡灾害全周期数据进行仿真验证，结果表明，本文方法能够比较准确地计算出滑坡发生的时间，同时利用评价指标对预报的结果进行测试，符合预报指标精度要求。

Abstract:

The forecast of landslide occurrence time is very important in disaster prevention and mitiga? tion. Accurate forecasting can effectively prevent the catastrophic consequences of disasters. In order to solve the problem that only the landslide displacement and other related parameters instead of the landslide occurrence time are predicted and estimated in the current landslide forecast method， a mixed Gaussian hidden Markov model (MOG-HMM) is proposed to establish the landslide occur? rence time calculation model. The macro-information prediction criterion is calculated by the mixed Gaussian algorithm based on the full-cycle data of the landslide hazard evolution process and matched with the state in the hidden Markov model to establish the landslide evolution state model， which can reflect multiple states of the full-cycle data. When the displacement data collected in real time is need?ed for time forecasting， the current data is first decoded using the decoding algorithm， that is， to calcu? late which state the landslide belongs to， then the time from the current state to the occurrence of the landslide is predicted using Dijkstra optimal path planning algorithm， achieving the prediction of the occurrence time of the landslide. The simulation results of the Xintan landslide and the Wolongsi land? slide show that the presented method can accurately calculate the landslide occurrence time. At the same time， the validation of the prediction results by using prediction indicators indicates that the re? sults meet the accuracy requirement of the forecast indicators.