The Prediction of Dam Displacement Time Series Using STL, Extra-Trees, and Stacked LSTM Neural Network

Abstract

Dam behavior prediction is a fundamental component of dam structural health monitoring. By comparing the predictions and the observations, anomalies can be detected, and then the remedial measures can be executed in time. As the most intuitive monitoring indicators, deformation is often used to evaluate dam structural health status. In this research, we propose a novel combined model for predicting the dam displacement time series. First, the seasonal-trend decomposition based on Loess (STL)method is utilized to decompose the dam displacement time series into seasonal, trend, and remainder components. Then the extremely randomized trees(extra-trees) model is used to predict seasonal components based on the causal models and influencing factors, whereas the stacked Long-Short Term Memory (LSTM)model is used to predict trend and remainder components based on the numerical models and historical observation data. Finally, the predicted results of the three components are aggregated to obtain the total predicted dam displacement. Seven state-of-the-art methods are introduced as benchmark methods to verify the effectiveness and feasibility of the proposed model. To quantitatively evaluate and compare the prediction results, three evaluation indicators, and a statistic test method are introduced. The experimental results show that the proposed model is the best-performing method compared with other benchmark methods both in prediction accuracy and stability. This indicates the proposed novel combined model STL-extra-trees-LSTM is a promising method for predicting displacement time series.

Keywords

• Dam behavior prediction
• time series decomposition
• deep learning
• ensemble learning