Predicting the urban water demand by equipping intelligent-based methods with discrete wavelet transform function

To satisfy the consumer demand, urban infrastructures are generally designed. The water distribution network (WDN) is one of the most important urban infrastructures in which optimal design and operation of it is essential during the operation period. For this purpose, in this research, artificial intelligence and data mining methods, including genetic programming (GP), gene expression programming (GEP), artificial neural network (ANN), and discrete wavelet transform function, are used to predict the daily drinking water consumption values of WDN. For this purpose, a dataset of temperature, precipitation, humidity, and daily water value of Najaf-Abad city in Iran is used during year 2014 to 2019. Here, hybrid models named W-GEP, W-GP, W-ANN, are proposed by equipping GEP, GP, and ANN with a wavelet transform function. In addition, two formulations are proposed for each model. Performance of proposed methods is investigated by determining R2, RMSE, and NSE statistical indices. For the training data of the W-GP model, the RMSE, NSE, and R2 values are 2810.46 (m3/day), 0.85, and 0.85, respectively, while for test and validation data these values are 2638.92 (m3/day), 0.87, and 0.87, respectively. Results show the good performance of proposed methods. In addition, the discrete wavelet transform function improves the models’ performance, in which the best results obtained by using the W-GEP model.