A DWT-LSTM Hybrid Model for Temperature Prediction in the Cement Kiln

Accurate measurement of the temperature within the calcination zone of a cement kiln is crucial for stabilizing clinker quality and achieving efficient energy control. However, direct temperature measurement is extremely difficult due to the high temperature, intense radiation, and dust inside the kiln. To solve this problem, this paper presents a hybrid model combining discrete wavelet transform and long short-term memory networks (DWT-LSTM) to predict the calcination zone temperature. First, the temperature sequence is decomposed by wavelet transform into one low-frequency and several high-frequency sub-sequences. Secondly, each reconstructed sub-sequence is separately predicted by an LSTM network. Finally, the individual forecasts are aggregated to obtain the final temperature prediction. The proposed hybrid model is applied to predict the calcination zone temperature from a real cement clinker calcination process. Experimental findings indicate that the hybrid model delivers superior performance in temperature prediction.