A Debris Flow Prediction Model Based on the Improved KPCA and Mixed Kernel Function LSSVR

In this study, we propose a debris flow prediction method based on the improved kernel principal component analysis (KPCA) and mixed kernel function least-squares support vector regression (LSSVR) to address the curse of prediction dimensionality when considering the multiple factors that influence the debris flow disaster and to tackle the problem of partial defects of model training performance caused by the selection of a single kernel function in the LSSVR model. First, we weight seven initial factors affecting the debris flow and select three principal-component-influencing factors as model inputs using the weighted KPCA method. Second, we use the LSSVR model by combining the local kernel function with the global kernel function to predict the debris flow probability and for balancing the learning ability and generalization ability of the samples. Third, we use the fruit fly optimization algorithm to obtain the optimal parameters of the model. Finally, we conducted a simulation experiment based on the monitoring data of a mill groove. The results denote that the proposed method can effectively reduce the curse of dimensionality and improve the accuracy of the prediction model. The probability value of debris flow within the allowable error and the corresponding warning level are predicted, which can provide certain reference for the relevant decision-making departments.