Monthly water demand prediction using wavelet transform, first-order differencing and linear detrending techniques based on multilayer perceptron models

Abstract

In this study, combined Discrete Wavelet Transform-Multilayer Perceptron (DWT-MP), combined First-Order Differencing-Multilayer Perceptron (FOD-MP) and combined Linear Detrending-Multilayer Perceptron (LD-MP) were developed and compared with stand-alone Multilayer Perceptron (MP) model for predicting monthly water consumption of Istanbul. The performance of these models were assessed by using coefficient of determination (R²), root mean square error (RMSE) and the Nash-Sutcliffe coefficient of efficiency (CE) as evaluation criteria. The study showed that DWT-MP could be used for forecasting the monthly water demand of Istanbul for only up to prediction lead-time of 3 months. However, FOD-MP was found to perform very well up to 12 months. It can be concluded from the results of the study that First-Order Differencing (FOD) is a reliable pre-processing technique for monthly water demand prediction.

Keywords:

• First-order differencing
• Linear detrending
• Wavelet Transform