Prediction of NOx emissions from gas turbines of a combined cycle power plant using an ANFIS model optimized by GA

• Published in: Fuel (Guildford) 2022-08, Vol.321, p.124037, Article 124037
• Main Author: Dirik, Mahmut
• Format: Article
• Language: English
• Subjects: Adaptive systems; ANFIS; ANFISGA; Artificial neural networks; Combined cycle power generation; Correlation coefficient; Correlation coefficients; Criteria; Emissions; Environmental impact; Error reduction; Fuzzy logic; Gas turbine combined cycle; Gas turbines; Gas-fired power plants; Genetic algorithms; Hybrid Intelligence emission monitoring technique; Incinerators; Industrial plant emissions; Model accuracy; Natural gas; Nitrogen oxides; NOx; Power plants; Root-mean-square errors; Statistical analysis; Steam electric power generation; Steam turbines; Training; Turbines
• ISSN: 0016-2361; 1873-7153
• DOI: 10.1016/j.fuel.2022.124037

•A number of effective parameters on NOx were collected.•A hybrid intelligence system (ANFIS-GA) was developed to predict NOx.•The prediction performance of ANFIS-GA was compared with the ANFIS model.•Prediction of NOx emissions from gas turbines of a combined cycle power plant using ANFIS model optimized by GA.•The parameters affecting the NOx value measured as output (NOx = NO2 + NO) are correctly determined. Combined cycle power plants, which combine gas and steam turbines, have negative impacts on surrounding populations and structures. Control of NOx emissions is an important issue for these gas-fired power plants. Accurate estimation of NOx emissions is critical for developing incinerators and reducing the environmental impact of existing plants. The objective of this study is to model ANFISGA and estimate NOx emissions from a natural gas-fired combined cycle power plant using emission monitoring system (PEMS) data. First, Adaptive Neuro Fuzzy Inference System (ANFIS) models were developed using fuzzy C-Means (FCM). Then, the parameters were optimized using a genetic algorithm (GA) to reduce the error. The proposed ANFISGA system was created, trained, and tested with PEMS datasets. The developed models were compared using several statistical performance criteria, including correlation coefficient (R2), mean squared error (MSE), error mean (EM), root mean square error (RMSE), standard deviation of error (STD), and mean absolute percentage error (MAPE). The obtained results show that the coefficient of determination varies between 0.79933 and 0.90363 for the data separated into test and training data with different rates. The minimum values of the criteria MSE, RMSE, EM, STD, and MAPE were found to be 24.8379, 4.9838, 3.4625e-05, 4.9839, and 5.1660, respectively, for the training data. The minimum values of these criteria for the test data were 26.5961, 5.1571, 0.065696, 5.157, and 5.3695, respectively. The collected results show that the proposed ANFISGA models have high potential for NOx prediction. Thus, the results show that GA has a great impact on the performance of ANFIS training and significantly improves the predictive accuracy of the model.