An Analysis on Machine Learning Based Predictive Maintenance for Power Plants Sector

Electricity and energy systems experts have a significant issue when trying to forecast output from combined cycle power plants. This research aims at application of artificial intelligence methods for improving predictions for maintenance and power generation in nuclear power plants. In this analysis, different Machine learning (ML) models that are DNN, MLRP, Gradient Boosting, and RNN, are explored using the dataset of the CCPP with features including the exhaust vacuum, ambient temperature, ambient pressure, relative humidity, and net electrical energy output that amounts to 9,568 hourly measurements. The models’ performances are evaluated using statistics like RMSE, MAE, and the coefficient of determination of the models (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>). Findings show that the RNN model has the overall better fit by being able to achieve the highest R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0. 999 in the current analysis and hence better predictive ability than all the models being compared as well as the MLRP model with the lowest R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0. 016. It provides an opportunity for a comparative analysis of how ML contributes to improving the predictive maintenance and operation of nuclear power systems and overcoming issues related to the handling of high volumes of data and the stability of station operations.