Prioritizing water distribution pipelines rehabilitation using machine learning algorithms

Abstract

The majority of water pipelines are subjected to serious deterioration and degradation challenges. This research examines the application of optimized neural network models for estimating the condition of water pipelines in Shaker Al-Bahery, Egypt. The proposed hybrid models are compared against the classical neural network, adaptive neuro-fuzzy inference system, and group method of data handling using four evaluation metrics. These metrics are; Fraction of Prediction within a Factor of Two (FACT2), Willmott's index of agreement (WI), Root Mean Squared Error (RMSE), and Mean Bias Error (MBE). The results show that the neural network trained using Particle Swarm Optimization (PSO) algorithm (FACT2 = 0.93, WI = 0.96, RMSE = 0.09, and MBE = 0.05) outperforms other machine learning models. Furthermore, three multi-objective swarm intelligence algorithms are applied to determine the near-optimum intervention strategies, namely PSO salp swarm optimization, and grey wolf optimization. The performances of the aforementioned algorithms are evaluated using Generalized Spread (GS), Spread (Δ), and Generational Distance (GD). The results yield that the PSO algorithm (GS = 0.54, Δ = 0.82, and GD = 0.01) exhibits better results when compared to the other algorithms. The obtained near-optimum solutions are ranked using a new additive ratio assessment and grey relational analysis decision-making techniques. Finally, the overall ranking is obtained using a new approach based on the half-quadratic theory. This aggregated ranking obtains a consensus index and a trust level of 0.97.