Combined Sewer Flow Prediction Using Hybrid Wavelet Artificial Neural Network Model

Abstract

This study presents a hybrid model for the prediction of dry and wet-weather flows in a combined sewer system. For this aim, a model based on wavelet transformation and artificial neural network (WANN) is developed. A high-resolution data set of rainfall and combined sewer flow from a catchment in Germany is used in the hybrid model. To simulate the combined sewer flow, the dry-weather flow is firstly modelled employing Artificial Neural Network (ANN). Subsequently, another ANN is applied and fed with rainfall time series, dry-weather flow simulated in the previous step and lagged combined sewer flow time series as the main input variables to simulate the combined sewer flow. In modeling both the dry-weather flow and the combined sewer flow, the wavelet transformation is firstly applied to extract the temporal and the spectral features of the measured sewer flow time series before using them in the ANN. To improve the WANN hybrid model result, different mother wavelet functions and decomposition levels, various lagged values for input variables, several training functions and network structures are implemented in the model and their influence on the hybrid model is investigated. According to this study, the proposed hybrid model can identify the complicated and dynamic nature of the combined sewer systems and thus provide accurate results.