物理
管道运输
校准
表面光洁度
机械
表面粗糙度
海洋工程
机械工程
热力学
量子力学
工程类
作者
Lin Shi,Jian Zhang,Xiaodong Yu,Sheng Chen,Wei He,Nan Chen
出处
期刊:Physics of Fluids
[American Institute of Physics]
日期:2024-09-01
卷期号:36 (9)
被引量:1
摘要
Hydraulic models are essential for predicting, estimating, analyzing, and optimizing long-distance water supply systems. Accurate calibration of these models is crucial for effectively representing the behavior of such complex systems and achieving a comprehensive understanding. However, the limited availability of measured data in actual systems often leads to an ill-posed problem characterized by more variables than equations. This paper proposes a calibration framework based on artificial neural networks (ANN) and adaptive particle swarm optimization to address this challenge. The influence of the number of measurement points and the measurement errors on the calibration results is analyzed within an actual long-distance water supply system featuring multi-branch pipelines. Results showed that ANNs can accurately reconstruct flow rates and pressures at unmeasured nodes, even with very limited measurement data. Compared to conventional calibration methods, the proposed framework can reduce the influence of measurement data uncertainty on calibration results, achieving better calibration accuracy. Additionally, by introducing regularization into the loss function, the constraints of physical laws are incorporated into the neural network's training process, further enhancing the calibration accuracy of the model.
科研通智能强力驱动
Strongly Powered by AbleSci AI