Abstract: The deformation of abutment slope is difficult to be estimated as it is uncertain and random under the influence of many factors. We present a prediction model for the abutment slope deformation by coupling ensemble empirical mode decomposition (EEMD), sample entropy (SE), improved particle swarm optimization (IPSO), and least square support vector machine (LSSVM). First of all, the time series of abutment slope deformation is decomposed by the EEMD into several subsequences with different complexity. Secondly, the complexity of each subsequence is determined using SE, and similar subsequences are merged to reduce the computational scale. Subsequently, the prediction model based on IPSO-LSSVM is established for each newly merged subsequence. The final prediction value of abutment slope deformation is obtained through superimposing and reconstructing each component. The model is applied to predicting the abutment slope deformation on the left bank of Miaowei Hydropower Station on Lancang River as a case study, and the result is compared with those of BPNN, RBFNN, LSSVM, EEMD-SE-LSSVM and EEMD-SE-PSO-LSSVM models. The comparison demonstrates that the present model has higher accuracy and better stability than the other neural network models.

Key words: abutment slope, deformation prediction, ensemble empirical mode decomposition (EEMD), sample entropy (SE), improved particle swarm optimization (IPSO), LSSVM