An Uncertainty Based Predictive Analysis of Smart Water Distribution System Using Bayesian LSTM Approach

Abstract

A well-designed water distribution system is crucial for maintaining high service standards in any modern smart city. Moreover, as the population is sky-rocketing, the demand for energy and water is increasing more rapidly than a decade before. Therefore, ensuring a steady clean water supply with optimized energy and water consumption has become necessary. To accurately monitor water distribution systems, the accuracy of input data plays a vital role in determining how accurate the system’s status estimations are. There must be a way for system operators to know what is going on at any given time to make practical decisions about how reliable the data they are receiving is. The input data uncertainty can induce flow and pressure calculation inaccuracies, which can be fatal while planning for future demands and needs to be quantified.Knowing the degree of uncertainty in predicting the water distribution system’s capacity or load can help people better prepare for future capacity or load predictions. Accurate uncertainty calculations are critical to time series forecasting. Probabilistic formulae are widely employed with classical time series models to estimate uncertainty. But incorporating new data and fine-tuning these models is a challenging task. This research paper presents a Bayesian LSTM network that computes both time series prediction and uncertainty assessment at the same time. In this paper, a real-time data set from VCU’s OpenCity test bed is employed to evaluate the efficacy of the suggested strategy.