By Mario Arquilla and Fabio Pasut
The need to improve efficiency in power plant management is taking a
leading role in the hydroelectric field. STE Energy has developed an
energy production forecasting tool for run-of-river hydro plants that
can be integrated within monitoring and supervision platforms.
To collect power plant data, the web-based STE-Monitor and
STE-Guardian software systems were used. STE-Monitor, designed for the
remote control of production facilities, continuously monitors the main
operating parameters and, for this study, data considered are energy
production profiles and alarms and faults lists. STE-Guardian is
designed to plan maintenance activities, coordinate maintenance
personnel and automatically send reports to the customer. Data
considered are cataloging of plant shutdowns and the duration of plant
shutdowns. For weather data, the web application provided by the Italian
Regional Agency for Environmental Prevention and Protection was used.
Development of the algorithm
Energy production of a run-of-river plant depends on the amount of
precipitation and thus the available flow. But other factors - such as
programmed plant shutdowns, internal faults, electrical grid
disturbances and extreme weather conditions - significantly affect
energy production. For these reasons, the collection of power plant
operation data is not enough; data must be analyzed, reprocessed and
understood.
For this solution, the potential of artificial neural networks (ANN)
was exploited. An algorithm in Matlab Simulink was developed; it
receives historical data relating to plant operation together with
meteorological data and outputs the energy production profile. A neural
network is formed by several layers: an input layer, several hidden
layers composed of neurons, and an output layer.
During the “learning phase,” the weights associated with the various
neurons of the hidden layers are iteratively changed, to obtain an
output as close as possible to the real data used for comparison. A
back-propagation approach with decreasing gradient was adopted to
minimize the deviation between the expected output and simulation
results. The algorithm, used to update the coefficients of the network, is
called a back-propagation algorithm for the fact that the offset
recorded in correspondence of a certain piece of data is propagated
backwards in the network to obtain the updating formulas of network
coefficients. The output error is calculated using the mean square error
approach.
Case studies
The forecast algorithm was tested on two Italian run-of-river plants,
each with a bulb turbine with rated power of 1.031 MW, rated discharge
of 26 m3/sec of and head of 4.2 to 5 m.
Scenario 1 - Medium-term forecast for Genivolta plant
Two years of historical data (operation and meteorological) were
used, then the production trend over one year was forecast, giving as
input the annual rainfall profile and programmed plant shutdowns. The
results show there are significant divergences between the prediction
and actual production during the year. Overall, however, an estimation
of annual energy production with an error of less than 1% was obtained;
this result is strongly dependent on the accuracy of the rainfall
forecast.
Scenario 2 - Short-term forecast for Cassano plant
A year and a half of historical data (operation and meteorological)
was used, and the production trend over a month was forecast, giving as
input the monthly rainfall profile and the programmed plant shutdowns.
The results show that there are small divergences between the prediction
and actual production. The maximum daily error on the average power is
24% and the maximum monthly error on the energy production estimation is
6%. Again, the results strongly depend on the accuracy of rainfall
forecast.
http://www.renewableenergyworld.com/articles/print/hrhrw/volume-23/issue-5/features/case-study-energy-production-forecast-for-run-of-river-plants.html
No comments:
Post a Comment