基于布谷鸟优化轻量梯度提升机的泥石流预测

摘 要： 针对山区环境中引发泥石流的影响因素复杂多样，影响因子之间易存在相互耦合，以及梯度提升树（Light Gradient Boosting Machine,LightGBM）预测模型易陷入局部最优问题，提出了核线性判别分析法（Kernel Linear Discriminant Analysis,KLDA）与经布谷鸟算法（Cuckoo Search,CS）寻优后的LightGBM预测模型。首先，对传感器采集到的原始数据进行清洗，并将“清洗”后得到的规范数据通过KLDA进行降维处理，得到相关性低且贡献率高的影响因子作为预测因子。采用随机取样的方法对降维后数据进行规划，选取70%的数据用于训练模型，剩余30%用于验证模型。然后，将训练数据作为输入，基于CS-LightGBM算法训练出最优预测模型。最后，结合鹅项沟监测数据进行仿真。结果证明，此方法能够将复杂的泥石流影响因子降维成利于建模的预测因子，使预测模型具有较好的预测准确度，为泥石流灾害预测方面的研究提供了新的思路。

A Debris Flow Prediction Model Based on the CS-LightGBM

Abstract] In view of the complex and diverse factors that cause debris flow in mountainous environments,cause influencing factors are easily coupled to each other,and light gradient boosting machine (LightGBM) is easy to fall into local optimal problems when a debris flow prediction model is preformed,this paper proposed kernel linear discriminant analysis (KLDA) and the LightGBM prediction model that was optimized by cuckoo search (CS).Firstly,the raw data collected by the sensor was cleaned , then it was sended through KLDA for dimensional degradation processing,and the influence factors with low correlation and high factor contribution rate was obtained as the predictors.the data after the degradation was planned by random sampling method, 70% of the data is selected for model training, and others of the data is used to validate the model.After that,The training data is used as input, and the optimal prediction model is trained based on CS-LightGBM algorithm. Finally, the experimental simulation is carried out with the monitoring data of Exianggou. Experiments show that this method can reduce the complex debris flow influence factors to the predictors of modeling, and provide the prediction model with good prediction accuracy, which can provide new idea for the research of debris flow disaster prediction.