A Novel Data-driven and Feature-based Forecasting Framework for Wastewater Optimization of Network Pressure Management System

In this paper, a novel data-driven approach to improving the performance of wastewater management and pumping system is proposed, in which necessary data are obtained by data mining methods as the input parameters of optimization problem to be solved in nonlinear programming environment. In this regard, first, CART classifier decision tree is used to classify the operation mode, or the number of active pumps, based on the historical data of Austin-Texas infrastructure. Then, SOM is utilized to classify the customers and select the most important features that might have effect on the consumption pattern. Further, the extracted features is fed to Levenberg-Marquardt (LM) neural network that predicts the required outflow rate of the period for each operation mode classified by CART. The results showed that the prediction F-measures were measured 90%, 88%, and 84% for each operation mode 1, 2, and 3, respectively. Finally, the nonlinear optimization problem is developed based on the data and features extracted from the previous steps solved by artificial immune algorithm. The results of the optimization model were compared with the observed data, showing that the proposed model could save up to 2%-8% of the outflow rate and wastewater, regarded as a significant improvement in the performance of pumping system.