A comprehensive comparative analysis of deep learning tools for modeling failures in smart water taps. Issue 1 (17th August 2021)

• Record Type: Journal Article
• Authors: Offiong, N. M.; Wu, Y.; Muniandy, D.; Memon, F. A.
• Is Part Of: Water Supply. Volume 22:Issue 1(2022)
• Page Start: 424
• Page End: 436
• Publication Date: 2021-08-17
• Subjects: Bi-LSTM -- CNN -- deep learning -- failure prediction -- LSTM -- smart water taps -- time-series
• DOI: 10.2166/ws.2021.261
• Languages: English
• ISSNs: 1606-9749

Abstract

Abstract: Predicting early-stage failure in smart water taps (SWT) and selecting the most efficient tools to build failure prediction models are many challenges that water institutions face. In this study, three Deep Learning (DL) algorithms, i.e., the Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and Bi-directional Long Short-Term Memory (BiLSTM), were selected to analyse and determine the most appropriate among them for failure prediction in SWTs. This study uses a historical dataset acquired from smart water withdrawal taps to determine the most efficient DL neural network architecture for failure prediction in the SWT, leading to a hybrid model's development. After a comprehensive evaluation of the three ML models, findings show that a hybrid combination of the CNN and Bi-LSTM (CNN-BiLSTM) models is a better solution for investigating failures in the SWT. HIGHLIGHTS: Automated feature extraction from smart tap time series dataset. The development of a temporal dependent model to explore time series dataset. A comprehensive analysis of three applied deep learning models. The development of a proposed hybrid model (CNN-BiLSTM) for failure prediction in SWTs deployed in rural Africa. Model evaluation with real world dataset.