大坝监测数据的时序预测与补全

Time Series Forecasting and Imputation of Dam Physical Quantities

大坝安全数据的时间序列预测和补全是大坝安全监测中两个常见问题。在实际应用中,应对这两类问题已有相应的经验模型,但由于经验模型过于简单,经验模型在上述两类问题上的效果并不理想。采用机器学习中的决策树模型和神经网络模型,为上述两类问题提供全新的解决方案。基于某水电大坝实测数据,对不同坝段的大坝安全数据进行预测和补全,结果表明本文模型相比于传统经验模型预测与补全精度均显著提高,同时,该模型还可以对模型输出结果进行不确定性分析,增强了结果的可靠性。

The time series forecasting and imputation of dam physical quantities are two major tasks in dam safety monitoring engineering.In practical scenarios,there are empirical models for the two tasks,but the empirical models perform weak in that they are too simple.The solutions to these two problems are provides via decision trees and neural network models,which have achieved good results.Based on the actual dam monitoring data of a hydropower station,the machine learning and neural network models are used to forecast and impute the physical quantity series of different dam sections.The experiments show that the proposed model is significantly better than the empirical models and can provide uncertainties quantification for the predictions which improves the reliability of the model.