基于小波去噪和神经网络的滑坡变形组合预测研究

基于滑坡的变形监测数据,首先探讨了小波去噪过程中各参数对去噪效果的影响规律,选取最优的小波去噪数据作为趋势项序列和误差项序列的分解依据,再利用BP神经网络和RBF神经网络对两序列进行预测,并与传统预测进行对比分析,最后对组合预测的效果进行探讨研究。结果表明:在滑坡变形数据的去噪过程中,以采用sym 4小波函数、固定式阈值、硬阈值选取阈值和7层小波分解时的去噪效果最好,并由后期预测结果可知其分项预测的效果要优于传统单项预测的效果,且线性组合预测对误差精度的提高有限,而非线性组合预测对误差精度的提高较大。通过上述研究,为滑坡的变形组合预测研究提供了一种良好的方法。

Research on Combined Forecast of Landslide Deformation Based on Wavelet De-noising and Neural Networks

Based on the deformation monitoring data of landslide, this paper firstly discusses the influence rule of various parameters on the de-noising effect in the wavelet de-noising process, selects the optimal wavelet de-noising data as the decomposition basis for trend item sequence and error item sequence, uses BP neural network and RBF neural network to predict two sequences, makes a comparative analysis with traditional forecast, and finally discusses the effect of combined prediction. The results show that: in the de-noising process of landslide deformation data, the de-noising effect by adopting sym 4 wavelet function, fixed threshold value, hard threshold and seven-layer wavelet decomposition is the best, and later prediction results show that the effect of itemized forecast is better than the effect of traditional single forecast results. The linear combined forecast improves the error accuracy with limited effect, but the nonlinear combined forecast improves the error accuracy to a great extent. The above-mentioned studies provide a good method for combined forecast research on deformation of landslide.