Minimization of water pumps' electricity usage: a machine-learning approach
Due to pervasive deployment of electricity-propelled water-pumps, water distribution systems (WDSs) are energy-intensive technologies which are largely operated and controlled by engineers based on their judgments and discretions. Hence energy efficiency in the water sector is a serious concern. To this end, this study is dedicated to the optimal operation of the WDS which is articulated as minimization of the pumps’ energy consumption while maintaining flow, pressure, and tank water levels at a minimum level, also known as pump scheduling problem (PSP). This problem is proved to be of the most difficult problem (namely NP-hard). To this end, we develop a hybrid methodology consists of machine learning techniques as well as optimization methods, that is to address real life and large sized WDSs. Other main contributions of this research are (i) also, variable speed pumps can be modeled and optimally controlled, (ii) operational rules such as water allocation rules can also be explicitly considered in the methodology. This methodology is tested using a large sized dataset in which the results are found to be highly promising. This methodology has been coded as a user-friendly software composed of MS-Excel (as a user interface), MS-Access (a database), MATLAB (for machine learning), GAMS (with CPLEX solver for solving optimization problem) and EPANET (to solve hydraulic models).