Long-term hybrid prediction method based on multiscale decomposition and granular computing for oxygen supply network

Abstract An accurate long-term prediction of oxygen demand can provide guidance for planning and scheduling the real production of the steel industry. In this paper, a polynomial-feature granulation prediction method based on long short-term memory network (GLSTM) is first developed and uses quadratic polynomials to form linguistic descriptors for data granules and perform prediction with LSTM. Considering the multiscale characteristics of oxygen flow data, a long-term oxygen demand prediction method based on multiscale decomposition and granular computing for oxygen supply network is further proposed. In this method, oxygen flow data are decomposed into low-, middle-, and high-frequency sublayers by empirical wavelet transformation. After removing the high-frequency noise sublayers, the GLSTM algorithm is employed to predict the low-frequency sublayers, while the k-SVM is employed to predict the middle-frequency sublayers. Finally, the effectiveness and performance of the proposed method is demonstrated in an industrial case study.