Development of FIR Models for an Estimation of the Water Demand of a Region

Introduction

In order to reliably supply a population center with water, local authorities must be able to estimate the water demand about 24 hours in advance, so that the sluices of the reservoirs in the mountains surrounding the population center can be adequately controlled. Especially in water-poor areas, it is important, not to supply too much water to any region, because otherwise, the surplus water will flow unused into the sea.

The water demand of a region is a variable with a strongly cyclic character. Tomorrow's water demand depends heavily on today's water consumption (daily cycle) as well as on the water consumption six days ago (weekly cycle). In addition, it may make sense to also take special public holidays as well as the beginnings and endings of school vacation periods into account (annual cycle).

Although the influencing factors could be treated as external control signals, the project described here did not treat them in this fashion. Instead, water consuption of a region was treated as a univariant time series. Consequently, we identified a FIR model that predicts future water consumption from observations of its own past.

Water consumption models were derived for various regions in collaboration with the respective local authorities. In particular, water demand models were identified for the region of Barcelona (Spain), the region of Lisbon (Portugal), as well as for a sub-region of the city of Rotterdam (The Netherlands). The regional authorities are not being listed below as sponsors of this project, because the corresponding research grants were received for other projects. The task of identifying the water demand of these regions by means of FIR was financed indirectly through several CICYT projects.

The demand prediction offered excellent results for the region of Barcelona, acceptably good results for the region of Lisbon, but didn't work well at all in the case of the region of Rotterdam.

Different modeling approaches were compared with each other, including ARIMA, but also NARMAX, and ANN. All modeling techniques produced the best predictions for Barcelona and the worst for Rotterdam. The predictions generated by FIR were always superior to those obtained with other modeling approaches. The time-series approach to modeling water demand doesn't seem to work well at all for the region of Rotterdam.


Most Important Publications

  1. López, J., G. Cembrano, and F.E. Cellier (1996), Time Series Prediction Using Fuzzy Inductive Reasoning: A Case Study, Proc. ESM'96, European Simulation MultiConference, Budapest, Hungary, pp.765-770.

  2. López, J. (1999), Time Series Prediction Using Inductive Reasoning Techniques, Ph.D. dissertation, Organització i Control de Sistemes Industrials, Universitat Politècnica de Catalunya, Barcelona, Spain.

  3. Escobet, A., R.M. Huber, A. Nebot, and F.E. Cellier (2000), Enhanced Equal Frequency Partition Method for the Identification of a Water Demand System, Proc. AI, Simulation and Planning in High Autonomy Systems, Tucson, Arizona, pp.209-215.

Sponsors


Deutsche Version
Homepage


Last modified: July 16, 2005 -- © François Cellier