不同方法在感潮河段ADCP在线测流系统中应用的比较分析

在水力因素多变的长江下游感潮河段建立ADCP在线测流系统实时采集指标流速,可选用合适的方法推求断面平均流速,从而实现流量的实时报汛和整编。多元线性回归分析和BP神经网络具有原理明晰、实现便捷等特点,为比较以上两种方法在断面平均流速计算中的优劣,以南京水文实验站2014年以来实测数据为例,分析不同情况下两种模型的拟合精度和预测精度。结果表明,两种模型均具有较好的有效性、精确性和稳定性,且拟合精度与模型选用的监测指标有关;对于只采用单一指标流速而言,BP神经网络模型的结果明显优于多元线性回归模型。同时,两种模型都能较好的预测断面平均流速,其中BP神经网络适应更好。

Comparative Analysis of Different Methods in Application of ADCP Online Flow GaugingSystem for Tidal Reach

ADCP online flow gauging system can be established for the tidal reach in the lower Yangtze River with changeable hydraulicfactors to collect index velocity in real time, and determine mean velocity at a cross-section by using appropriate method,so as to realize flood reporting and reorganizing the flow in time. Multiple linear regression analysis and BP neural network havethe characteristics of clear principle and convenient application. The fitting precision and prediction accuracy of the two models un -der different circumstances were analyzed based on the observed data at the Nanjing Hydrological Experimental Station since 2014to compare the advantages and disadvantages of the two methods above in the calculation of mean section velocity. The resultsshow that both of the two models have good effectiveness, accuracy and stability, and fitting precision is related to the monitoringindex selected for the model. If the single index velocity is adopted only, the results of BP neural network model are obviously su -perior to those of multiple linear regression model. Meanwhile, both of the two models can forecast mean section velocity well, andBP neural network is more adaptable.