Decentralized model predictive control of urban drainage systems

摘要

Using real time control (RTC) techniques to improve urban drainage system performance is proven to be an effective solution for alleviating urban flooding. Many modeling methods of urban drainage systems have been introduced in existing literatures to accommodate different situations and scenarios. Nonlinear hydrologic models are useful for detailed simulations in pipes and sewers. However, to utilize RTC requires users to establish suitable models that both reflect physical characteristics of the system while not over complicating with unnecessary details. A mixed integer system was proposed where hybrid Model Predictive Control can be used to compute control actions in previous literature. However, the complexity of solving associated optimization problem grows exponentially with the size of the system and therefore, the computation time renders direct application of such method infeasible. This paper investigates the possibility of partitioning the system into several subsystems with communications and instead of computing solutions in centralized framework, the control actions are obtained distributedly by individual subsystems. The performance of decentralized schemes is demonstrated with numerical simulations on a fictional sewage system composed of 13 tanks and 12 control follows under 4 rain scenarios corresponding to different rain intensities. Decentralized Model Predictive Control is shown to have comparable performance compared with the centralized framework while having significantly improved computation time. Two methods are also presented to reduce pumping energy costs by harvesting rainfall energy.