5th International GAP MATHEMATICS-ENGINEERING-SCIENCE AND HEALTH SCIENCES , Şanlıurfa, Turkey, 4 - 06 December 2020, pp.213-225
In
this investigation, adaptive neuro-fuzzy interference system (ANFIS) was
carried out to estimate the power capacity and rotational speed of a turbine
located in a power plant in order to demonstrate that the adaptive neuro-fuzzy interference
system is an acceptable algorithm to estimate turbine characteristics
parameters. For the development of forecasting models, the total of 743 data
records were collected and these data were divided into two subsets such as
training and testing data set. While the training data set includes in a total
of 527 records, the testing data set consists of 216 records obtained from the
turbine. In the current simulation, the fuzzy membership functions (MFs) are
set as trapmf and output of MF is adjusted to linear. The optimization method
of the Train FIS is the hybrid. Back propagation was also performed but hybrid
one demonstrated more accurate results. The values acquired by adaptive-neural fuzzy
interference system were compared with real data. The mean absolute percentage
error (MAPE) was found as 0.55% for turbine rotational speed and 2.7% for power
output of the turbine. On the other hand, while mean absolute deviation (MAD)
and root mean squared error (RMSE) were calculated as 0.81% and 10% for the
prediction of turbine rotational speed, respectively, they were 4.2% and 5.4%
for the estimation of power output of the turbine. The prediction of ANFIS method follows the
change of actual values almost exactly. It was stated that the correlation
between actual and predicted values are very high. Moreover, it can be said
that the performance values of power output and rotational speed such as mean
absolute percentage error (MAPE), mean absolute deviation (MAD) and root mean
squared error (RMSE) was found to be smaller for training case compared to the
testing case. Thus, this study shows that the rotational speed and power output
of a turbine can be accurately predicted by using the input data of temperature
and speed.