Predictive control of hydraulic pumps in water distribution systems via the optimization of dynamic models

Abstract

This project proposes the development of a software for the control of pumps in water distribution systems using the optimization of dynamic models, with parameter and state estimation (offline and online, respectively). The optimization will make use of a hydraulic model, the operational constraints, the water demand curve and the energy costs to find the best control signals for a time horizon, planning actions through the prediction of the system future states. The control objectives will be the reduction of pressure (to reduce leakage and water loss) and energy costs. The complete model is composed by many components. The most important one is the network hydraulics. Its architecture depends on the real system to be controlled with parameters estimated offline using the same optimization technology used in the control. The model must also include the demand curve for the users of the water network, which can be extracted from historical data or predicted in real-time using time-series. The operational constraints are defined by the user (e.g., system operator) and restrict the optimization to search only for viable solutions. The model also needs the energy costs and pump efficiency curves. The existing products and academic work in this area (water distribution systems) usually focus on pump scheduling using stationary models, not in real-time control. Even though the optimization of dynamic models is harder, it encapsulates the optimization of static models, so scheduling problems are a subset of the more general problem. Real-time control using dynamic models, on the other hand, can deal with rapid variations and reject disturbances, which can be determined through state estimation, a problem that is analogous to control. One of the research focuses will be on how to solve the dynamic control problem fast enough, using either traditional algorithms (e.g., convex programming) or metaheuristics (e.g., genetic algorithms).Control, state and parameter estimation are all solved via optimization, but the optimization type (e.g., linear, convex, integer, etc.) depends on the hydraulic model. For example, a pump without variable frequency drive can only be turned on or off, thus the problem is combinatorial (binary). Solving techniques for each problem that can arise from common hydraulic models will be investigated in this research. The model architecture, the operational constraints (e.g., input minimum and maximum, objectives, etc.) and the control result will be configured and displayed through a graphical user interface. The software will communicate with the real system using SCADA (Supervisory Control And Data Acquisition), which must already be installed and configured on the water network. The communication with the SCADA system will be via OPC, a protocol for industrial networks. The software will be tested and benchmarked on the Van Zyl network, the only standard water control problem in the literature. (AU)