为了准确预测尾矿坝浸润线的位置变化，结合浸润线埋深非稳定、非线性的时间序列以及动态变化的特点，利用小波分解与重构，提出基于小波分解的时间序列指数平滑法和BP神经网络法，采用时间序列的指数平滑法和BP神经网络方法分别对多个细节信号序列和逼近信号序列进行拟合预测，并对其拟合结果进行叠加，实现对尾矿坝浸润线的预测。将预测结果与实际监测数据进行对比，结果表明小波分解预测方法的预测结果与传统单一的指数平滑法和神经网络法预测结果相比，在预测精确度和拟合度方面：小波分解>指数平滑>神经网络。

The seepage line is the safe lifeline of tailings dam, and its location change directly reflects the seepage characteristics inside the dam body.In order to accurately predict the location change of the seepage line of tailings dam in flood season, analyze the law of change,and predict the future change of seepage line,so as to ensure production, reduce accidents such as dam break, and ensure the safety of people’s life and property. Combined with the characteristics of unstable, nonlinear time series and dynamic change of the submerged line of the tailings dam, using wavelet decomposition and reconstruction, the exponential smoothing analysis method of time series based on wavelet decomposition and the analysis method of BP neural network are proposed. Firstly, the non-stationary time series s is decomposed into five detail signal sequences and one approximate signal sequence by wavelet decomposition. The d₁~d₅ of the detailed signal sequence is predicted by exponential smoothing method of time series. The BP neural network method is used to predict the approximate signal sequence A5. Finally, wavelet reconstruction is used to predict the approximate signal sequence A5 based on MATLAB programming, and the fitting results are superimposed to predict the soakage line of the tailings dam. In this paper, a metal tailings dam with more precipitation and humid climate is selected as the research object, and the data of the buried depth of the infiltration line in the flood season for 100 days are selected for modeling and analysis. The time series exponential smoothing analysis method based on wavelet decomposition and the BP neural network analysis method are used to predict the development trend of the buried depth of the infiltrating line in the next 10 days, and the predicted results are compared with the actual monitoring data.The results show that the prediction results of wavelet decomposition method are compared with those of traditional single exponential smoothing and neural network prediction methods. The prediction accuracy is as follows: wavelet decomposition (0.9287) > exponential smoothing (0.9038) > neural network (0.8725). The prediction error of wavelet decomposition is the smallest and the accuracy is the highest. In terms of fitting degree, wavelet decomposition (0.8837) > exponential smoothing (0.8573) > neural network (0.8462). The fitting result of wave decomposition prediction method is almost consistent with the whole development trend of original time series, and the degree of coincidence is high. Therefore, This method has good applicability and superiority in predicting the buried depth of tailing dam in flood season.