DefinePK

Abstract

In this research, a multilayer perceptron neural network model with multiactivation function called (MLP-MAF) model has been developed for municipal water demand forecasting. The developed model uses different activation functions in the hidden layer neurons. Different combinations of the linear, logistic, tangent hyperbolic, exponential, sine and cosine activation functions were used in the hidden layer neurons. In order to assess the credibility of developed model results, the model was run over the available data which include the time series of daily and monthly municipal water consumption for fourteen years (1/1/1992-31/12/2004) of Tampa city, USA. Each time series was divided into two subsets: the estimate subset for fitting the model and the holdout subset for evaluating the forecasting ability of the model. Additionally, three statistical measurements, namely the coefficient of determination (R²), the root mean square error (RMSE) and the mean absolute percent error (MAPE) and two hypothesis tests, namely the <i>t</i>-test and <i>F</i>-test have been reported for examining the forecasting accuracy of the developed model. The results show that the combination of linear, sine and cosine functions is better than other combinations. Furthermore, the effectiveness assessment of this model shows that this approach is considerably more accurate and performs better than the traditional multilayer perceptron (MLP) and radial basis function (RBF) neural networks.