Probabilistic modeling and uncertainty estimation of urban water consumption under an incompletely informational circumstance

Abstract

With a booming development characterized by new urbanization in current China, urban water consumption attracts growing concerns. An efficient and probabilistic prediction of urban water consumption plays a vital role for urban planning toward sustainable development, especially in megacities limited by water resources. However, the data insufficiency issue commonly exists nowadays and seriously restricts further development of urban water simulation. In this article, we proposed a consolidated framework for probabilistic prediction of water consumption under an incompletely informational circumstance to deal with the challenge. The model was constructed based on a state-of-the-art Bayesian neural networks (BNNs) technique. Three dominated influencing factors were identified and included into the BNN model. Future impact factors were generated by using a variety of methods including a quadratic polynomial model, a regression and auto-regressive moving average combination model and a Grey Verhulst model. Thereafter, water consumption projection (2013–2020) and uncertainty estimates was done. Results showed that the model matched well with observations. Through reducing the dependence on large amount of information and constructing a probabilistic means incorporating uncertainty estimation, the new approach can work better than conventional means in support of water resources planning and management under an incompletely informational circumstance.