Author(s): Hokuto Okabe; Mariko Suzuki; Kazuya Inoue

Linked Author(s): Kazuya Inoue

Keywords: Dam Seepage Machine Learning Random Forest Gradient Boosting Decision Tree

Abstract: This study proposes a machine learning-based estimation method for dam management to achieve low-cost and rapid dam diagnosis based on observation data related to the seepage rate. This study employs random forest (RF) and gradient boosting decision tree (GBDT), which are some of the machine learning techniques derived from the decision trees, as machine learning algorithms. In this study, transitionally observed seepage rates at actual rock-fill dams located in Kyushu province in Japan were used to build the training data and testing data. In addition to the measured seepage rate, the rainfall data and the water level in the reservoir from 1 day through 7 days before were employed since the seepage rate may be influenced not only by the rainfall event of the day but also by the status of the water level in the reservoir. As a result, it was found that with approximately 4 years of training data, high estimation accuracy could be achieved, and that GBDT had a smaller estimation error compared to RF. Analyzing the difference between estimated and observed seepage rate with sufficient data suggests potential for detecting anomalies in the dam's behavior.

DOI:

Year: 2025