Dynamic neural networks for real-time water level predictions of sewerage systems-covering gauged and ungauged sites

Abstract

Abstract. In this research, we propose recurrent neural networks (RNNs) to build a relationship between rainfalls and water level patterns of an urban sewerage system based on historical torrential rain/storm events. The RNN allows signals to propagate in both forward and backward directions, which offers the network dynamic memories. Besides, the information at the current time-step with a feedback operation can yield a time-delay unit that provides internal input information at the next time-step to effectively deal with time-varying systems. The RNN is implemented at both gauged and ungauged sites for 5-, 10-, 15-, and 20-min-ahead water level predictions. The results show that the RNN is capable of learning the nonlinear sewerage system and producing satisfactory predictions at the gauged sites. Concerning the ungauged sites, there are no historical data of water level to support prediction. In order to overcome such problem, a set of synthetic data, generated from a storm water management model (SWMM) under cautious verification process of applicability based on the data from nearby gauging stations, are introduced as the learning target to the training procedure of the RNN and moreover evaluating the performance of the RNN at the ungauged sites. The results demonstrate that the potential role of the SWMM coupled with nearby rainfall and water level information can be of great use in enhancing the capability of the RNN at the ungauged sites. Hence we can conclude that the RNN is an effective and suitable model for successfully predicting the water levels at both gauged and ungauged sites in urban sewerage systems.

Machine Learning‐Based Surrogate Modeling for Urban Water Networks: Review and Future Research Directions

“…For example, Pasha and Lansey (2014) used support vector machines (SVMs) for improving the real-time estimation of water tank levels and thus decreasing pump energy consumption in a WDS. In UDSs, Chiang et al (2010) implemented an early form of recurrent neural network (RNN) for water level predictions at gauged and ungauged sites. Their decision of using this architecture was motivated by its increase in performance.…”

“…Real-time operation uses the current state of the system to modify its behavior and improve its functioning in future time steps. In the case of UDSs, they are usually designed to retain stormwater for a certain period, to avoid combined sewer and stormwater outflows (Rosin et al, 2021;She & You, 2019) or to reduce flooding (Berkhahn et al, 2019;Chiang et al, 2010;Keum et al, 2020;Kim et al, 2019;. Whereas, in WDSs, the objective is to deliver high-quality drinking water while minimizing pumping costs (Pasha & Lansey, 2014;Rao & Alvarruiz, 2007;.…”

Online multistep-ahead inundation depth forecasts by recurrent NARX networks

“…For instance, the simple recurrent network (Elman, 1990) has the outputs of the hidden layer fed back to the input layer, and demonstrates its great ability in extracting dynamic time variation characteristics. In recent years, RNNs have been applied to the field of hydrological modeling (Chang et al, 2002, Chang et al, 2004Coulibaly and Baldwin, 2005;Besaw et al, 2010;Chiang et al, 2010).…”

On-line multistep-ahead inundation depth forecasts by recurrent NARX networks

“…For instance, the simple recurrent network (Elman, 1990) has the outputs of the hidden layer fed back to the input layer, and demonstrates its great ability in extracting dynamic time variation characteristics. In recent years, RNNs have been applied to the field of hydrological modeling (Chang et al, 2002, Chang et al, 2004Coulibaly and Baldwin, 2005;Besaw et al, 2010;Chiang et al, 2010).…”