The Design of Pipeline Leak Location Prediction System with Incremental Extreme Learning Machine (I-ELM)

Water loss is a global concern faced by all nations. It stems from various factors, categorized into physical and non- physical losses. Physical loss, mainly attributed to pipe leaks, significantly contributes to total water loss, with 80% attributed to leaks. Thus, an Artificial Neural Network-based leak detection system using Incremental Extreme Learning Machine algorithm is designed. A District Metered Area (DMA) is selected for study. DMA specifications are used for modeling through EPANET software. After validating the DMA model, leak simulations employ emitter coefficients (0.5-1 LPS). Data from 17 junctions under different leak scenarios are collected. An I-ELM leak detection system is developed, using 17-feature pressure data and two target variables for training. Testing data, derived from the split training set, results in leak location predictions for pipeline sections with average deviations of 3.7577, 2.4310, and 0.5630. Leak size predictions have average deviations of 0.0807, 0.0585, and 0.0543. Overall, leak location prediction's RMSE is 0.0516, and leak size prediction's RMSE is 0.0696.