A data-mining approach is proposed to model a pumping system in a wastewater treatment plant. Two parameters, energy consumption and wastewater flow rate after the pumping system, are used to evaluate the performance of 27 scenarios while the pump was operating. Five data-mining algorithms are applied to identify the relationships between the outputs (energy consumption and wastewater flow rate) and the inputs (elevation level of the wet well and the speed of the pumps). The accuracy of the flow rate and energy consumption models exceeded 90%. The derived models are deployed to optimize the pump system. The computational results obtained with the proposed models are discussed.
Wastewater treatment plants (WWTPs) are energy-consuming facilities, and it has been reported that they use 4% of the nation's electricity to move and treat water/wastewater (Clean energy opportunities in water and wastewater treatment facilities background and resources,, et al., 2006 and Goldstein and Smith, 2002). Energy costs constitute 25–30% of the operational and maintenance (O&M) costs of water and wastewater facilities. Furthermore, the demand for electricity by WWTPs is expected to grow by approximately 20% over the next 15 years as population grows and environmental requirements become more stringent. Thus, conserving energy at WWTPs is a significant issue. Considerable energy savings can be achieved by optimally operating and monitoring wastewater treatment processes.
Since pump and blower motors account for more than 80% of WWTPs' energy costs (Wastewater management fact sheet, 2006), designing effective operational strategies for pumping systems can greatly reduce these costs. Research on pumping systems has been reported in the literature. Ormsbee and Lansey (1994) reviewed different models and approaches proposed for water-supply pumping systems. Bechwith and Wong (1995) used a genetic algorithm to solve the pump scheduling problem in a multi-source water supply system with multiple tanks. Barán et al. (2005) utilized a mass balance model and evolutionary computational algorithms to solve a multi-objective, pump-scheduling problem by minimizing four types of costs while satisfying the water demand and other constraints. More recently, Yang and Børsing (2010) developed a mixed-integer, non-linear, programming model for a simple, multi-pump, boosting system that included three variables: speed pumps, a simple water circular loop, and a storage tank. Wang et al. (2009) modeled pump scheduling in a water distribution system as a bi-objective optimization problem by taking into account pump operational costs and the land subsidence issue to reduce costs and address environmental concerns. The results obtained by the proposed genetic algorithm-based method have resulted in a wide range of schedules.
Modeling approaches investigated for pumping systems to date have been based predominantly on physics and mathematical programming. The assumptions that have been made limit the applicability of these models in industry. For example, Barán et al. (2005) assumed that the capacity of each pump, the pump discharge, electric energy consumption, and the power consumption of each pump combination were fixed for a period of 1 h. These assumptions neglected the dynamic characteristics of the pumping process and impeded the applicability of such models in industrial practice. Assuming static models of the pumps, Wang et al. (2009) demonstrated a potential for improvement of the efficiency of multi-pump systems.
In this paper, a data-mining approach for modeling pumping systems is introduced. As an emerging science with an abundance of successful applications in wind energy (Kusiak and Zhang, 2011, Zhang and Kusiak, 2012 and Verma and Kusiak, 2012), HVAC systems (Kusiak et al., 2011, Kusiak and Li, 2010 and Kusiak et al., 2010), and other areas (Berry and Linoff, 2004, Harding et al., 2006, Shah et al., 2006 and Siegelmann and Sontag, 1994), data mining has proven to be a promising approach for modeling complex, dynamic, and non-linear systems. However, the research literature on utilizing data mining in modeling and optimizing pumping systems is sparse. In this research, data-mining approaches are applied to model the energy consumption and hydraulic workload (water flow rate after pumping) of a pumping system based on 15-min data collected by a municipal WWTP. Effectiveness of data-mining approaches in modeling pumping system is demonstrated. The developed models are applicable to scheduling operations of a pumping system.
In the research reported in this paper, performance models of a pumping system in a wastewater treatment process were investigated. Two metrics, i.e., energy consumption and wastewater flow rate after the pumping system, were considered to assess the performances of 27 pump combinations. Five data-mining algorithms were selected as candidates to develop the energy consumption and water flow rate models. The energy consumption models achieved 90% accuracy, but the accuracy of the flow rate models was less than 90% for some pump-combination scenarios because of the time delay between the output and the input. The use of time series was incorporated with MLP to develop better models for those pump-combination scenarios. The experimental results showed that significant improvements were achieved.
Further research will focus on refining the models for optimization of the pump system. The models will generate optimal pump configurations and the corresponding pump speeds.
