Smart Water Leakage Detection Using a Hybrid ANN-SVM Machine Learning Framework
Abstract
The efficient distribution of water and opportunities to detect leaks in time are issues of great concern in the infrastructure systems of the modern period when the losses that remain unnoticed can result in great economic and environmental consequences. This paper proposes, a data-based framework of smart monitoring of water distribution networks was proposed based on a hybrid machine learning approach. The system proposed takes advantage of various operational parameters that include: pressure, flowrate, and vibration to identify anomalies in pipeline systems. An interaction mechanism, feature interaction, is added to model complex variables relationships, improving prediction power. A controlled stochastic augmentation method is used to overcome the draw back of the limited labeled data to enhance the generalization of the model. The experimental evaluation was performed using a set of water distribution data that is augmented with sensor data like pressure, flow rate, vibration, temperature and pipe characteristics. The novelty of the proposed framework is the weighted fusion of ANN and SVM models along with the feature interaction engineering and controlled stochastic augmentation for better leak prediction robustness. The predictive architecture is an Artificial Neural Network (ANN) and a Support Vector Machine (SVM) whose outputs are combined via a weighted decision strategy to boost the classification robustness. Experimental results demonstrate that the proposed hybrid model has reached classification accuracy of 92% in the noisy and fluctuant operating conditions and it is better than using ANN alone and SVM alone. It has presented a scalable and powerful smart water management system for water saving and stabilization of operation.