Res-BiLSTMs model based on multi-task attention for real-time measurement of the free calcium oxide content

Abstract

Abstract The free calcium oxide (f-CaO) content of cement during the firing process is the main economic indicators for evaluating cement quality. Real-time monitoring of the f-CaO level is of crucial to ensure the scientific production of cement. In allusion to the properties of time series coupling, dynamic nonlinearity, and limited labeled data in the cement clinker production process, this paper proposes a residual bidirectional long-short-term memory network model (Res-BiLSTMs) based on multi-task attention mechanism for online monitoring of the f-CaO content.The model takes the Bi-LSTM as the basic component, and combines residual network to construct the Res-BiLSTMs coding structure, which aims to summarize the multi-level characteristic information of the input sequence. Besides, a multi-task attention is proposed, which combines the attention mechanism with the idea of semi-supervision.Under the effect of quality supervision,the control coupling relationship and the data coupling relationship between the devices and between the variables are further extracted. Finally, through experimental comparison, the proposed model gives better measurement results under the condition of limited label samples.