Abstract:
Electricity has changed everyone's life since the day it was discovered. It is produced all over the world in power plants that uses different types of energy sources like fossil fuel, water falls, nuclear and wind Energy. The amount of electricity produced is the fundamental goal for any power plant. Therefor accurate prediction of this amount is very important for planning and operation activities of the power plant. This study aims to identify the attributes that influence the amount of generated power from a thermal power plant, and to accurately predict that amount. Datasets were prepared form real data that had been collected by SCADA over two years period, from two different units in a thermal power plant of 190 MW capacity. Feature selection was done using wrapper method, and power prediction was done using all available attributes, to give a ranking of the selected attributes, and show the influence of each parameter in the amount of generated power. For power prediction, only controllable parameters like pressure, temperature and steam flow at turbine inlet were used as predictors. Sixteen different algorithms were tested for each dataset, the algorithm that showed higher correlation coefficient and minimum error was selected to build the modelThe predicted amount using data mining was found to be more accurate than manufacturers expectations and thermodynamic laws. Models evaluation was done using separate dataset, and cross validation in case of small datasets. Moreover comparison between the predicted and the actual observed amounts was presented in a graph, to visualize the accuracy of the models.
