Monthly water consumption prediction using season algorithm and wavelet transform-based models

Accurate prediction of water demand is essential for optimum management of water resources and sustainable growth and development. Recently, models based on artificial neural networks (ANNs), in combination with data preprocessing techniques, have been used for water demand prediction due to their ability to handle large amounts of complex nonlinear data. Discrete wavelet transform (DWT) is one of the most widely employed data preprocessing techniques, and is used in combination with ANNs to improve prediction accuracy and extend prediction lead time. However, DWT is known to have serious drawbacks, and the accuracy and prediction lead times of the models have not been satisfactory. In this study, multiplicative season algorithm (MSA) is applied for the first time as an alternative data preprocessing technique in the area of hydrology and its performance is compared with DWT. The outputs of MSA and DWT are used as inputs to a multilayer perceptron (MLP) in order to develop combined models called discrete wavelet transform-multilayer perceptron (DWT-MLP) and multiplicative season algorithm-multilayer perceptron (MSA-MLP), which are compared with the stand-alone MLP model. The results demonstrate that MSA is a better preprocessing technique than DWT and, thus, that MSA captures periodicity and converts nonstationary time series into stationary time series better than DWT.