Characterization of ANFIS-SC method on power peaking factor

• Main Authors: Mohd. Ali, Nur Syazwani, Hamzah, Khaidzir, Idris, Faridah, Sazali, Muhammad Arif, Sarkawi, Muhammad Syahir, Basri, Nor Afifah, Jamaluddin, Khairulnadzmi, Zainal, Jasman
• Format: Article
• Language: English
• Published: Areadiscover 2022
• Subjects: T Technology (General); TP Chemical technology
• Online Access: http://eprints.utm.my/104146/1/NurSyazwaniMohdAliKhaidzirHamzahFaridahIdris2022_CharacterizationofANFISSCMethodonPower.pdf; http://eprints.utm.my/104146/; https://www.jnrtmns.net/index.php/jnrt/article/view/206/198

Prediction of power peaking factor (PPF) using ANFIS method at TRIGA research reactor has been conducted in the previous study and resulted in good predictive performances. This method could be implemented as a real-time monitoring system for various reactor types. In this paper, the ANFIS-SC trained models will be characterized to investigate the generalization capability of the models against new input data. Three ANFIS-SC trained models adopted from the previous study with 0.45, 0.45, and 0.50 of the cluster radii were chosen for this characterization study. Based on the statistical analysis, the correlation coefficients of the trained models show a weak relation between predicted and actual output. The Means Absolute Error (MSE) and Root Means Square Error (RMSE) were near to zero in the range of 7.2112 x 10-7 – 9.4304 x 10-7. However, the average output of all the trained models was in the range of 1.8722 - 1.8724 while the average output of the actual PPF is 1.8728. This statistical result shows that the generalization capabilities of the ANFIS-SC method were excellent and could be improved further with a deep learning mechanism for exact prediction performances. Besides, the ANFIS-SC method also can be applied for PPF monitoring at the control room of the nuclear reactor for enhancing the reactor operation as well as for education and training.