【摘要】 鉴于传统神经网络模型难以充分利用大坝变形监测时间序列数据前后信息的拓扑关系，而双向长短时记忆神经网络(BiLSTM)能够有效地学习前后向信息，提出一种基于秃鹰搜索算法优化双向长短时记忆神经网络的大坝变形预测组合模型BES-BiLSTM,即首先采用秃鹰搜索算法捕获模型参数最优值；其次利用BiLSTM双向学习的特性进行模型训练增强数据间的相关性；然后以某混凝土坝水电站沉降值为实例1基于BES-BiLSTM模型进行大坝变形预测，以另一混凝土坝坝体水平位移值为实例2辅助验证模型性能；最后将BES-BiLSTM模型预测结果与传统长短时记忆神经网络(LSTM)模型和BiLSTM模型预测结果进行了对比研究。结果表明，BES-BiLSTM模型较单一传统LSTM、BiLSTM模型拥有更强的拟合能力和预测能力，可用于混凝土坝、边坡等变形预测中。更多还原

【Abstract】 In view of the fact that traditional neural network models can hardly make full use of the topological relationship of the backward forward information of dam deformation monitoring time series data, while the bidirectional long short-term memory(BiLSTM) can effectively learn the backward forward information, a combined dam deformation prediction model BES-BiLSTM was proposed based on bald eagle search algorithm optimized bidirectional long short-term memory neural network. Firstly, the bald eagle search algorithm was used to optimize parameters of the model. Secondly, the bidirectional learning feature of BiLSTM was used to train the model to enhance the correlation between the data. Then, the settlement value of a concrete dam hydropower station was taken as example 1 for dam deformation prediction based on the BES-BiLSTM model. Another concrete dam horizontal displacement value was taken as example 2 to verify the model performance. Finally, the prediction results of the BES-BiLSTM model were studied in comparison with those of the traditional long and short term memory neural network(LSTM) model and the BiLSTM model. The results show that the BES-BiLSTM model has stronger fitting and prediction capabilities than the single traditional LSTM and BiLSTM models, which can be used for deformation prediction of concrete dams and slopes.更多还原