|
OBJECTIVE
The aim of the POWADIMA research project is to establish the feasibility of
introducing optimal, real-time control of water distribution networks. Since
demand is fluctuating continually, it is necessary to adjust the control
apparatus frequently if optimal or near-optimal control is to be achieved.
The objective function is to meet the forecast demands at minimal operating
cost, subject to operational constraints such as statutory minimum pressure,
minimum acceptable flow to avoid stagnation etc. This implies minimizing
pumping costs which are the largest component of the operating cost (hence
the acronym POWADIMA).
DIFFICULTIES
In developing an optimal, real-time control system for water distribution,
the difficulties that have to be overcome include:
- the size and
complexity of water-distribution networks;
- existing networks
have not been designed with optimal control in mind;
- an uncontrolled demand
which is highly variable;
- the short time
available between successive changes in control settings;
- the enormous numbers
of possible combinations of valve and pump settings;
- the computational
time and memory requirements for large-scale networks;
- the need to include
complicated energy tariff structures.
When formulating the research programme, all of these
issues have been considered, many of them for the first time.
APPROACH ADOPTED
It will be appreciated that the use of a conventional hydraulic simulation
model for optimal real-time control is impractical for large networks because
of the excessive computational burden optimization imposes. If it were
necessary to run a hydraulic simulation model to evaluate the impact of
different combinations of pump/valve settings, it would not be possible to
identify the optimal setting in the time available before the next update.
Nevertheless, a hydraulic simulation model is required to estimate the
consequences of different valve/pump settings. Therefore, the approach
adopted has been to capture the knowledge-base of a complex hydraulic
simulation model of the network in a far more efficient form using an
artificial neural network (ANN). Thereafter, the ANN-predictor is embedded in
a Genetic Algorithm (GA)- optimizer which has been specifically developed for
adaptive, real-time control. The advantage of this approach is the vast
improvement in computational efficiency which enables the optimal control
settings for the current and all future time steps up until the operating
horizon, to be completed quickly. This is repeated at the next time interval,
following the latest update of the system state from the Supervisory Control
and Data Acquisition (SCADA) scheme. Whilst initially the resulting control
system would be advisory to experienced operational staff, this does not
preclude the possibility of closed-loop control in the longer term.
|
