水利设施形变预测可以有效地判断水利设施的运行状态。水利设施安全监测数据是时间序列数据,既有趋势性又有季节性。为了获得更准确的预测结果,文中提出一种基于季节自回归差分移动平均(SARIMA)模型和双向长短时记忆(BiLSTM)网络的预测模型,以解决无法充分挖掘数据中正向与反向的关联进行预测的问题。该模型采用SARIMA 模型预测变形数据中的线性分量,采用BiLSTM 模型预测变形数据中的非线性分量,使得模型能够更好地提取历史数据中的非线性关系以及正向与反向关系从而提高预测准确度。结合某水电站4#引水涵洞监测数据,使用SARIMA-BiLSTM 模型对裂缝计开合度时间序列进行了预测,并与反向传播神经网络模型、SARIMA 模型和SARIMA-LSTM 模型的预测结果进行对比,比对结果证明所提方法有效地提高了预测精度。

Prediction of water conservancy facilities deformation can effectively judge their operation state. Water conservancy facilities safety monitoring data is time series data, which has both tendency and seasonality. In order to obtain more accurate prediction results, a prediction model based on seasonal autoregressive differential moving average (SARIMA) model and bidirectional long and short time memory (BiLSTM) network is proposed in this paper, this model solves the problem that the correlation between forward and backward in data cannot be fully utilized for prediction. In this model, SARIMA model is used to predict the linear components of deformation data, BiLSTM model is used to predict the nonlinear components of deformation data. The model can better extract the nonlinear relationships in historical data and improve the prediction accuracy. SARIMA-BiLSTM model is established based on the monitoring data of 4# diversion culvert of a hydropower station, then the model is used to predict the time series of crack meter opening and closing gap. The prediction result of this model is compared with results of back propagation (BP) neural network model, SARIMA model and SARIMA-LSTM model. The comparison results prove that the prediction accuracy is effectively improved by the proposed model.

TP391; TP181

基金项目:国家重点研发资助项目(2019YFB1310504);四川省自然科学基金资助项目(2022NSFSC0542)