Forecasting Daily Urban Water Demand Using Dynamic Gaussian Bayesian Network

Abstract

The objective of the presented research is to create effective forecasting system for daily urban water demand. The addressed problem is crucial for cost-effective, sustainable management and optimization of water distribution systems. In this paper, a dynamic Gaussian Bayesian network (DGBN) predictive model is proposed to be applied for the forecasting of a hydrological time series. Different types of DGBNs are compared with respect to their structure and the corresponding effectiveness of prediction. First, it has been found that models based on the automatic learning of network structure are not the most effective, and they are outperformed by models with the designed structure. Second, this paper proposes a simple but effective structure of DGBN. The presented comparative experiments provide evidence for the superiority of the designed model, which outperforms not only other DGBNs but also other state-of-the-art forecasting models.