Teaching-Learning-Based Optimization of Neural Networks for Water Supply Pipe Condition Prediction

Abstract

The bulk of water pipes experience major degradation and deterioration problems. This research aims at estimating the condition of water pipes in Shattora and Shaker Al-Bahery’s water distribution networks, in Egypt. The developed models involve training the Elman neural network (ENN) and feed-forward neural network (FFNN) coupled with particle swarm optimization (PSO), genetic algorithms (GA), the sine cosine algorithm (SCA), and the teaching-learning-based optimization (TLBO) algorithm. For the Shattora network, the inputs to these models are pipe characteristics such as length, wall thickness, diameter, material, lining and coating, surface type, traffic distribution, cathodic protection, flow velocity, and c-factor. For the Shaker Al-Bahery network, the data gathered include length, material, age, diameter, depth, and wall thickness. Three assessment criteria are used to evaluate the suggested machine learning models, namely index of agreement (IOA), correlation coefficient (R), and root mean squared error (RMSE). The results reveal that coupling FFNN with the TLBO algorithm outperforms other prediction models. Therefore, the FFNN-TLBO model can be a valuable tool for simulating the water network pipe condition. This study could help the water municipality allocate the available budget effectively and plan the required maintenance and rehabilitation actions.