Identification of NPPs Transients Using Transductive Semi-supervised Learning Algorithm

Abstract

In this study, an identifier for NPPs transients based on semi-supervised learning (SSL) algorithm is developed. Modular identifier using transudative support vector machine (TSVM) model classifies the type of transients. This identifier versus unsupervised learning algorithms has the advantage of using the collected information. Moreover, the proposed identifier theoretically can measure the proximity between labeled and unlabeled patterns making it probably more efficient than supervised techniques. The developed identifier is examined by the Iris flower dataset as a benchmark test problem. Transients of the Bushehr nuclear power plant (BNPP) are studied as a case study. Results show good performance of the identifier. Recognition of unknown transients as don’t know, identification of transients in presence of noise, distinctive identification of transients, and training of the identifier by independent features are advantages of the proposed identifier. SVM is a supervised classifier that can find auto-correlation and detect cross-correlation of input data. SSL is trained on labeled and unlabeled patterns and makes it possible to measure similarity between new transients and trained ones.