针对静态机器学习模型难以有效反映滑坡的动态演化特性，且存在时序过长时历史数据遗忘导致位移预测结果不稳定的问题。提出一种基于鲸鱼优化卷积神经网络（convolutional neural networks, CNN）和双向门控循环神经网络（bidirectional gated recurrent neural network, BiGRU）的滑坡位移动态预测方法。首先对滑坡影响因子进行特征筛选，构建数据集，建立CNN-BiGRU网络模型，使用鲸鱼优化算法（whale optimization algorithm, WOA）对模型进行超参数寻优，使用CNN网络模型从滑坡数据中提取潜在的特征向量，将特征向量以时间序列的形式输入BiGRU模型中，利用其处理时间序列数据的优势，完成滑坡位移预测。结果表明：所提出模型得到的滑坡位移预测精度较高，与未优化的CNN-BiGRU相比均方根误差（root mean square error, RMSE）下降了0.030 5 mm。

For the static machine learning model is difficult to reflect the dynamic evolution characteristics of landslide effectively, and there is the problem that the displacement prediction results are unstable due to the forgetfulness of historical data when the time series is too long. A dynamic landslide displacement prediction method based on whale-optimized convolutional neural networks (CNN) and bidirectional gated recurrent neural network (BiGRU) was proposed. Firstly, the landslide impact factors were filtered for features and the data set was constructed, the CNN-BiGRU network model was established and the hyperparameter search was performed using the whale optimization algorithm (WOA), the potential feature vectors were extracted from the landslide data using the CNN network model, and the feature vectors were input to the BiGRU model in the form of time series, and used its advantage of processing time series data to complete landslide displacement prediction. The results show that the landslide displacement prediction accuracy obtained by the proposed model is high, and the root mean square error (RMSE) decreases by 0.030 5 mm compared with the unoptimized CNN-BiGRU.