基于时序分解和SSA-LSTM-Attention模型的尾矿坝位移预测

针对尾矿坝位移变形的动态特性和传统预测模型在进行尾矿坝位移预测中的不足，提出了一种基于时序分解和麻雀搜索算法-长短时记忆-注意力机制(sparrow search algorithm-long short-term memory-attention mechanism, SSA-LSTM-Attention)模型的尾矿坝位移预测方法。首先，通过改进的自适应噪声完备集合经验模态分解算法(improved complete ensemble empirical mode decomposition with adaptive noise, ICEEMDAN)将尾矿坝位移监测数据进行分解为趋势项和波动项；其次，一方面采用高斯拟合方法对趋势项进行拟合预测，另一方面通过灰色关联度进行波动项相关影响因子筛选，并将注意力机制与LSTM相结合，建立了基于注意力机制及LSTM的波动项位移预测模型，同时利用SSA对该模型的超参数寻优；最后，将趋势项与波动项叠加得到总的位移预测值。以攀西地区尾矿库为例对模型性能进行了验证，并与反向传播(back propagation, BP)、LSTM、LSTM-Atte...

Study on tailings dam displacement prediction based on time-series decomposition and SSA-LSTM-Attention model

Aiming at the dynamic characteristics of tailings dam displacement and deformation and the shortcomings of traditional prediction models in tailings dam displacement prediction, a tailings dam displacement prediction method based on time series decomposition reconstruction and SSA-LSTM-Attention model is proposed. First, the tailings dam displacement monitoring data is decomposed and reconstructed into trend terms and fluctuation terms through ICEEMDAN; secondly, on the one hand, the Gaussian fitting method is used to fit and predict the trend terms, and on the other hand, the fluctuation terms are calculated by the gray correlation degree. The relevant influencing factors were screened, and the attention mechanism was combined with LSTM to establish a fluctuation term displacement prediction model based on the attention mechanism and LSTM. At the same time, SSA was used to optimize the hyperparameters of the model. Finally, the trend term and the fluctuation term were combined. The superposition gives the total displacement prediction. Taking a tailings pond in the Panxi area as an example to verify the performance of the model, and compared with BP, LSTM, LSTM-Attention, and other models, the results show that the root means square error, mean absolute error and coefficient of determination obtained by this method are respectively 0.742mm, 0.553mm and 0.994, the proposed method can greatly improve the prediction accuracy of tailings dam displacement and deformation.