Comparison of adaptive neuro-fuzzy inference system (ANFIS) and machine learning algorithms for electricity production forecasting

• Published in: Revista IEEE América Latina 2022-10, Vol.20 (10), p.2288-2294
• Main Authors: Aguirre Rodriguez, Elen Yanina, Rodriguez Gamboa, Alexander Alberto, Aguirre Rodriguez, Elias Carlos, Da Silva, Aneirson Francisco, Silva Rocha Rizol, Paloma Maria, Silva Marins, Fernando Augusto
• Format: Article
• Language: eng ; por
• Subjects: Adaptation models; Adaptive systems; Algorithms; Artificial neural networks; Combined cycle power generation; Economic forecasting; Electricity; Fuzzy logic; Fuzzy neural networks; Gas turbines; Inference; Machine learning; Mathematical models; Multilayer perceptrons; Power generation; Power plants; Predictive models; Pressure; Production; Relative humidity; Steam turbines; Thermodynamic cycles; Turbines
• ISSN: 1548-0992; 1548-0992
• DOI: 10.1109/TLA.2022.9885166

Combined cycle power plants (CCPP) are popular in the energy sector for the production of electricity, and are the union of two thermodynamic cycles, corresponding to the steam turbine and the gas turbine. This paper presents the application of several machine learning (ML) techniques and the adaptive neuro-fuzzy inference system (ANFIS) to predict the hourly electricity production in a CCPP. The models were developed using 5-fold cross-validation with the collected features of temperature, exhaust pressure, relative humidity, ambient pressure, and electricity production per hour (the target feature). The hyperparameters of the tested models were optimized. The correlation and determination coefficients of the models were higher than 92%, showing a significant performance. The ANFIS (r = 98% e R2 = 95%) model shows the lowest values in the evaluated error metrics, compared to the other ML models. Finally, the results showed the effectiveness of ANFIS in predicting the hourly production of electricity in CCPP.