IEEE Project Abstract

In this paper, a model identification method basedon a long short-term memory (LSTM) neural network composedof a network structure and training algorithm is used to builda thermal field model that accurately simulates the crystal growthprocess. The support vector machine (SVM) approach is thenadopted to identify model order and lag to determine networkinput and to improve precision. The thermal field model reflectingthe growth process in the Czochralski crystal furnace is simulated. Experimental results and comparative analysis results bothsuggest that the method proposed by this paper can build an efficient thermal field model which outperforms other methods interms of precision.