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 for improving the performance of wastewater management and pumping system is proposed, which is getting knowledge from data mining methods as the input parameters of optimization problem to be solved in nonlinear programming environment. As the first step, we used CART classifier decision tree to classify the operation mode -number of active pumps- based on the historical data of the Austin-Texas infrastructure. Then SOM is applied for clustering customers and selecting the most important features that might have effect on consumption pattern. Furthermore, the extracted features will be fed to Levenberg-Marquardt (LM) neural network which will predict the required outflow rate of the period for each operation mode, classified by CART. The result show that F-measure of the prediction is 90%, 88%, 84% for each operation mode 1,2,3, respectively. Finally, the nonlinear optimization problem is developed based on the data and features extracted from previous steps, and it is solved by artificial immune algorithm. We have compared the result of the optimization model with observed data, and it shows that our model can save up to 2%-8% of outflow rate and wastewater, which is significant improvement in the performance of pumping system.