Desulfurization Process Modeling by Using Deep AttLSTM-based Network

The desulfurization process is a complex industrial process with high dynamic nonlinearity and large delay time. To solve the modeling problem of the flue gas desulfurization process, this paper designed an attention mechanism-based long short-term memory (AttLSTM) network, and an autoencoder based on this model to complete the detection of abnormal points in the desulfurization process. This paper proposed for the first time the use of a deep long short-term memory network autoencoder based on the attention mechanism to detect outliers in the desulfurization process, and the AttLSTM model is likewise applied to the identification task of the desulfurization process for the first time. In a deeper sense, this article attempted to use a deep learning model to identify complex systems. The AttLSTM identification model established has not appeared in the field of system identification before. The emergence of this model can enrich the selection of identification models. At the same time it provides artificial intelligence technology with new opportunities in the field of system identification, and proposes a reference for application and promotion in the field of system identification and control. Experimental results show that compared with the desulfurization process modeling methods appearing in previous literature, the proposed method has better performance in different performance indicators, which can prove the effectiveness of the deep AttLSTM model in desulfurization scenarios.