Optimization of electric power prediction of a combined cycle power plant using innovative machine learning technique

Abstract

Accurate prediction of electric power generation in combined cycle power plants is challenging yet crucial, especially when employing machine learning techniques like artificial neural networks. This research presents an advanced forecasting model based on the robust adaptive neuro‐fuzzy inference system to estimate electric power generation under full operating conditions. The research dataset comprises 9568 data points featuring four input parameters, including ambient temperature, ambient pressure, exhaust vacuum, and relative humidity, spanning 6 years of the publicly available UCI Machine Learning Repository. These data were partitioned into 70% for training, 30% for validation, and 0% for testing to ensure robustness. A hybrid approach is implemented for optimization, combining the least squares method and gradient descent. The first‐order Sugeno fuzzy model was adopted to defuzzification the entire fuzzy set, achieving optimal results with three membership functions assigned to each input variable. This configuration minimizes the training's root mean square error values and the checking error of 3.8395 and 3.7849 regarding the generalized bell‐shaped membership functions, improving computational efficiency. These values are optimum when employing the root mean square error performance metric of identical studies. The validation of the adaptive neuro‐fuzzy inference system method and an optimal data selection strategy for training should be considered for optimum outcomes in the energy field.