The FLOAT Project
Saving freight costs by predicting Danube water levels
Inland shipping companies do not fully load their barges in order to
increase their chances of navigating all sections of the waterway they
are travelling without delay. If water levels drop below expectations
and barges waste time waiting to pass a shallow stretch of river or
must lighten their load via additional transshipment, transport costs
are increased. It is only from April to July that ship operators may
not expect problems with the water levels at the critical sections of the waterway. The
critical sections in Austria are: Wachau and Wien-Bratislava.
Danube Map courtesy of ÖIR
This has a strong negative impact on the competitiveness of inland
shipping. On one of the rivers most prone to this kind of problem -
the Danube. However this mode of transport is an important alternative
to land freighting, especially - in view of the growth in trade with
Central and Eastern European Countries.
Given access to sufficiently accurate prediction of water levels on
the Danube, in particular on critical sections of this waterway,
carriers could adjust schedules and levels of cargo to the given water
situation. This would reduce or eliminate waiting times, thus
improving reliability of delivery, and raise the average load
transported, lowering the freight costs. This applies to many European
waterways and their users, but is of extreme relevance for Danube
The project goals:
The goal of this project was to determine whether there is a need for
HPCN techniques in order to create a water level prediction model
which could be used to determine water levels at critical sections of
a route one day in advance.
- perform a feasibility study as a first step of a model to
predict water levels on inland waterways
- assess the potential for parallelisation of the selected software (EVIS)
- help increase the attractiveness of inland shipping by enabling companies to optimise their schedules and load.
ÖIR selected some hydrological data from 1991-1994 and forwarded
these to EVIS:
EVIS, a machine learning software product based on applying genetic
algorithms to symbolic computation, was selected as the basis for the
- daily mean water levels of the Danube in Achleiten, Grein,
Kienstock and Hainburg
- daily amount of precipitation at the stations Zell/See, Flauchau
- daily mean temperature at the measuring stations Zell/See and
Water level prediction model:
Based on the above mentioned hydrological and meteorological data the
software provided by EVIS was trained to evolve a water level
prediction model for one day in advance, using the data from
How the EVIS Software works:
Why HPCN is used:
- At first the software starts with a set of randomly generated
possible solutions (=Formula) to estimate the water levels of the
- All solutions are evaluated according to a fitness criterion and
the best ones are selected.
- The selected solutions are combined (application of genetic
operators) to obtain an improved estimate of the water level.
- Loop through the last two steps until a satisfactory solution to predict the future water levels is found.
- Despite to the small amount of input data, for the one day
prediction the EVIS Software runs on a single workstation for
approximately 7 hours.
- Shipping companies need at least a reliable prediction (14 days
would be optimal).
- To achieve the necessary accuracy for a 14 days prognosis,
additional variables will be needed like meteorological satellite
data, weather forecast data, Europe-wide water level monitoring
stations on the Danube, and data supplied from hydroelectric power
stations. This augmented data set will increase the demand in
- This computing time could be reduced by running the EVIS Software on several computers concurrently.
To evaluate the quality of the prediction model, the
results have been compared with the observed water level data of the
next day for a one year period (4th year). The mean
absolute deviation was 8,6 cm between the predicted and the actual
The accuracy for the test prediction based on the training data set
previously mentioned, was very promising for the 1 day
prognosis. Although only a very small amount of input data was used
for the prediction we obtained a very accurate prognosis:
85% of all predicted water levels for one year deviated no more than
+/- 15 cm from the actual value.
Since the most important information for shipping companies is the
prediction of low water levels we concentrated on the prediction of
those data. Based on the small input data set first attempts were made
to extend the prediction model to 7 days.
It is possible to parallelize the EVIS Software. Doubling the amount
of computing resources can nearly halve the execution time. In order
to calculate a water level prediction up to 14 days in advance - as
demanded from the shipping companies with EVIS, the amount of
data as well as the demand of computational power will raise. By using
of HPCN an accurate prediction of water levels within reasonable time
can be achieved.
FLOAT was integrated into activities carried out by the ÖIR as partner
in EUDET, funded by the EC/DGVII under the IVth Framework Programme.
Please find further information at:
Radio sound file (MS wave
format), Copyright by ORF 1998