机理和数据混合驱动的排水系统控制模型构建方法

针对城市排水系统的优化控制,提出一种机理和数据混合驱动的控制模型构建方法。该方法根据与控制目标的相关程度划分排水系统,并利用长短时记忆神经网络模型和圣维南方程刻画不同分区,可为系统的模型预测控制提供工具支撑。为验证方法的有效性,选取A市某污水处理厂服务片区为研究对象,运用所提出方法,基于机理模型构建简化控制模型,并在实际降雨下将该模型与水库模型和纯数据驱动模型进行对比。结果表明:所构建的控制模型在2个溢流口的模拟准确性相对水库模型提升了3.85%和22.86%,相对纯数据驱动模型提升了5.66%和3.57%(以均方根误差的均值计);模拟效率较机理模型提高了98.7%。该控制模型构建方法可在一定程度上平衡模拟准确性和模拟效率的矛盾,从而可更好地支撑城市排水系统的优化运行。研究结果可为实施排水系统实时控制提供参考。

A HYBRID MODELING STRATEGY FOR CONTROL SIMULATOR OF URBAN DRAINAGE SYSTEMS BASED ON DATA-DRIVEN AND MECHANISM-DRIVEN METHOD

A hybrid control modelling method was proposed for optimal control of urban drainage systems,which adopted both data-driven and mechanism-driven simplification strategies.The method divided urban drainage systems into different regions,according to connected degree to the control target.These regions were modeled by LSTM model and Saint Venant equation separately.The method was verified within the service area of a wastewater treatment plant in City A in China.The proposed method established a surrogate control model based on a detailed hydraulic model.Compared with tank model,simulation accuracy in two CSO outfalls was improved by 3.85% and 22.86%,and improved by 5.66% and 3.57% compared to the LSTM model (measured in an average root mean value).The simulation time could be reduced by 98.7% in comparison to the detailed hydraulic model.Due to the advantage in accuracy and efficiency,the modelling strategy can provide references for the implementation of real-time control in urban drainage systems.