This study focuses on data-driven approaches for burst detection and classifies them into three categories: classification method, prediction-classification method and statistical method. The performance of these methods is discussed. By analysing uncertainty in burst detection, this paper revealed that non-stationary monitoring data and limitations present in these methods challenge the reliability of detection results. Data pre-processing and probabilistic solutions to deal with the uncertainty are summarised. From these findings and discussions, this paper concludes and recommends that: a) data-driven approaches are promising in real-life burst detection and reducing false alarms is an important issue; b) more comprehensive performance evaluation might be necessary, in particular regarding detectable burst size; c) further research on new methods employing multivariate analysis and a new category based on clustering analysis would be beneficial to tackle uncertainty; d) more focus on the use of pressure data might facilitate burst location and reduce investment in burst detection.

Short-time water demand forecasting is essential for optimal control in a water distribution system (WDS). Current methods (e.g., time-series models and conventional artificial neural networks) have limited power in practice due to the nonlinear nature of changes in water demand. In particular, 15-min time-step forecasting may not be accurate when using conventional models. To tackle this problem, this paper investigates the potential of deep learning in short-term water demand forecasting, developing a gated recurrent unit network (GRUN) model to forecast water demand 15 min and 24 h into the future with a 15-min time step. The performance of GRUN was compared with a conventional artificial neural network (ANN) model and seasonal autoregressive integrated moving average (SARIMA) model. A correction module was used to reduce the cumulative error. The results show that the deep learning method improves the performance of water demand prediction. The correction module enhances the performance of ANN and GRUN models. In general, deep neural network models like GRUN outperform the ANN and SARIMA models for both 15-min and 24-h forecasts. These findings can provide more flexible and effective solutions for water demand forecasting.