Model Development For Turbine Energy Yield (TEY), Carbon Monoxide (CO) And Nitrogen Oxide (NOx) From Gas Turbine Power Plant
Loading...
Date
2021-06-01
Authors
Wan Anuar, Wan Ahmad Aizat
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
In order to combat the environmental issues that have been constantly rising
since the start of the first Industrial Revolution in the 18th century, many solutions have
been introduced and been applied around the world. One of the approaches for
overcome air pollution issues from greenhouse gas emissions is by monitoring their
release from its most abundant sources, for example, gas turbine power plants.
Predictive emission monitoring system (PEMS) is one of the methods for monitoring
these greenhouse gas emissions. It is powered by an artificial neural network (ANN)
by taking into account the collected data from Kaya et al. (2019) such as ambient
temperature, ambient pressure, ambient humidity and many more from selected gas
turbine power plants for the emission prediction purpose. Several models will be
developed and will be classified according to their responding outputs. Multi input
single output (MISO), where carbon monoxide (CO), nitrogen oxide (NOx) and turbine
energy yield (TEY) will be operated as separate output and multiple inputs multiple
outputs where CO, NOx and TEY will be its output simultaneously. For each model’s
type, it will be further classified into the model with input selection and the model
without input selection. The performance of the model will be demonstrated by the
value of its respective mean squared error (MSE), R and R2. The model with input
selection is having almost the same performance as the model without input selection
although having fewer input variables compared to the latter. R2 values for each
training model with input selection are 0.5094, 0.8260, 0.7573 and 0.6922 for the
model with CO, NOx, TEY as output and MIMO model respectively compare to the
R2 values for each training model without input selection are 0.5382, 0.8278, 0.7627
and 0.6950 for model with CO, NOx, TEY as output and MIMO model respectively.
MIMO model is the better model compared to MISO, even though it combines 3
outputs and could be more complex, but ANN still able to predict accurately.
Therefore, developing MIMO model could be better than developing MISO model as
it will reduce model times (one model for 3 outputs rather than a separate model for
each output).