Adaptive Real-time Leak Detection in Water Distribution Systems Using Online Learning

Abstract

In this paper, we address the critical challenge of real-time leak detection in water distribution systems using online learning algorithms. The data collected by accelerometers was exploited to identify distinctive characteristics of leaks. Our study focuses exclusively on the Online Gradient Boosting Machines (Online GBM) method following data preprocessing. The analysis reveals that the Online GBM model, optimised through random search for its hyperparameters, excels in leak detection, achieving an accuracy of 92.30%. These results, obtained on a test set, demonstrate the effectiveness of Online GBM in managing large data sets and its reliability as a rapid detection tool. The article highlights the significant potential of online learning techniques, particularly Online GBM, in enhancing water resource management and effectively reducing losses due to leaks. The results of this research offer a promising path towards improving the monitoring and maintenance of water distribution infrastructure.