Stimulation Initiative for European Neural Applications
Esprit Project 9811
Drinking Water Supply Management
Company background
Canal de Isabel II (CYII) is a public company responsible for regulation
,supply, treatment and distribution of drinking water over the whole are of
Comunidad de Madrid (Madrid Region). Drinking water treated and distributed
in Madrid in 1993 was of around 498.03 Hm3. In addition, Canal de Isable II
has an annual capacity of waste water treatment of 202 Hm3. Canal de Isabel
II high technology input in all company role is considerable, especially
concerning water networks management.
Description of the problem
CYII has installed hundred of sensors in different autonomous station
to measure water consumption. These measures are
transmitted to the control centre to be processed.
With these data the water requirements in different zones are
calculated. However, within this scheme they want to solve
three different problems. The first one is
sensor validation that leads to missing values in the measurements
and requires technical verification. Therefore, a sensor validation
system is required to minimise this maintenance task.
The second problem consist of estimating these missing values
as a function of other variables that were actually measured.
This can be done using a model
that relates these variables and that could give an estimation of
one variable as a function of the others. Finally, the third problem
consists of estimating the future consumption using
the past history of the variable along with another important variables
such as temperature or rainfall measurements.
Neural Network Application
In order to give a solution to the above mentioned problems a
neural network approach has been used due to its flexibility
for modelling non-linear systems. Fuzzy logic technique was used
for sensor validation while neural network was chosen for subsystem
modelling and water consumption forecasting. Sensor validation gives
a confidence level of sensor measurements which is
used by the control engineers as an indication of possible fails.
On the other hand, models of the different subsystems of
the network is used not only as a second validation of the
sensor measurements but also as a way to estimate sensor values
when they are out of service. Neural networks are
also used along with state space reconstruction techniques
to make predictions of water consumption. Short term range
forecasting (2 hours) along with long term range prediction
(24 hours) are given so that engineers have an estimation
of the future network charge and they can take control
decision on the system.
Benefits
The IIC-system provides not only confidence on sensor measurements
that ensure control engineers the current state of
the network but also estimates future values of the network.
Of course, not all variables can be predicted or modelled, because
there are variables whose evolution do not follow any underlying model.
However, neural network models increases the confidence on
the measurements and therefore it leads to a saving in time
and money for CYII.
Generalisation
Although, the sensor validation system along with subsystem
modelling and variable forecasting have been installed in a
water supplier company the technology used in this project can be
used without minor changes in other industrial sectors where
the state of the plant depends on different sensor measurements
such as Chemical Reactors or industrial environments.
Contact persons
Francisco Cubillo, Canal de Isabel II - Santa Engracia, 125 - 28003 Madrid -
Spain - Phone: +34 1 445 10 00; Fax: +34 1 446 31 01.
Alberto Pérez, Instituto de Ingeniería
del Conocimiento - IIC Unversidad
Autónoma de Madrid - Módulo C-XVI planta 2
- 28049 Madrid - Spain -
Phone: +34 1 397 39 73; Fax: +34 1 397 39 72;
alberto@irene.iic.uam.es.
![[ ]](../logo_small.gif){kind=logo}
SIENA