Meta-heuristic versus gradient-type methods for real time demand calibration in water distribution system

Water distribution system modeling in real time involves many challenges for water utilities. Calibration of models to match measurements coming from a water network with calculation resulting from models is a necessary step to make models reliable. There are many variables involved in the analysis of a water distribution system and each of them contains a determined level of uncertainties. Nevertheless, in a real time scenario it is significant the impact that water demands at nodes could have compared with other variables. Demands, in contrast to pipe roughness for example, are much more sensitive to changes in time. A better estimation of how much water is being consumed at each point can help a lot to assess properly the current state of the network. This information is crucial for supporting decision making processes based on models working online and can be used as a starting point for decisions involving a short term forecasting of the network behavior. Among the challenges of the presented research it should be mentioned the short available time for providing calibration results in a “real time” context. The methods used in this paper are not only trying to achieve good results but to achieve results in a very short period of time. This idea has been accomplished by combining different heuristics, data mining and optimization techniques. Results presented in an online context are calculated based in the analysis of historical data and the analysis of the current information coming from water networks. For testing purposes, this research also includes the development of a software component for emulating the data transmission from the water network to an OPC server and from there to the algorithms in charge of demand estimation. A conclusion is also given about merit of the Meta-heuristic versus Gradient-type calibration methods