Aiming at the problem that the prediction results of the residual deformation data series in the old goaf,which has obvious nonlinear and nonsmooth characteristics,are significantly affected by the model,this paper proposes a combined prediction method of residual deformation by variational modal decomposition and integrated enhancement.Firstly,the Newton interpolation method was used to obtain the equally spaced residual deformation data series.Then,the experimental data were decomposed into disturbance and trend terms using variational modal decomposition(VMD).Then the BP neural network and the long short-term memory neural network(LSTM)were trained iteratively using the adaptive boosting algorithm(AdaBoost).Finally,the integrated and enhanced BP-AdaBoost model was used to predict the perturbation term,the LSTM-AdaBoost model predicts the trend term,and the prediction results of the perturbation term and the trend term were combined with equal weights to complete the prediction of the residual deformation in the old goaf.The experimental results show that the AdaBoost integrated training can improve the prediction performance of a single model,while the VMD decomposition reduces the non-stationarity of the residual deformation data,and the combined constructed VMD-BPAda-LSTMAda model absorbs the advantages of each of the two strong prediction models,and weakens the influence of the prediction hysteresis of the BP and LSTM models,which results in the robust model prediction performance.
关键词
老采空区/残余变形/组合预测/变分模态分解/自适应提升算法
Key words
old goaf/residual deformation/combined prediction/variational mode decomposition/adaptive boosting
维普
万方数据