FGD Process Identification: An Information-theoretic Convolution Network Learning Approach

Nowadays flue gas desulphurization (FGD) technologies have been extensively applied in the coal-fired electricity-generating plants. As emission standards for sulfur dioxide (SO2) have become more stringent in recent years, there is a real need to develop more advanced modeling techniques such that the FGD process can be identified accurately, which in turn provides a reliable foundation for FGD process control and optimization. However, FGD process has characteristics of large time delay, strongly nonlinear dynamics, making traditional identification models become ineffectual. To address this issue, this paper proposes a hybrid model consisting of information-theoretic technique and convolution neural network (CNN). To verify the proposed identification approach, a real FGD system of a 600MW coal-fired power station is selected as case analysis. Experimental results indicate that our model achieves satisfactory identification result and performs better than other popular FGD models appeared in previous studies, demonstrating the effectiveness and superiority of the proposed MI-CNN model.

I. Introduction

SO₂ pollution has drawn increasing concern due to its detrimental effects on ambient atmosphere and human health [1]. One of major contributors to SO₂ emissions are various types of fossil fuels-based power plants, among which the coal-fired power plant is found most commonly in practice. Due to this fact, the ultra-low SO₂ emission concentration of 35mg/m3 is required by the Chinese government, which is the most stringent SO₂ emission standard in history [2]. To achieve this target, FGD technologies are widely employed in SO₂ emitting processes, where alkaline absorbents such as sodium hydroxide (NaOH), calcium oxide (CaO) and limestone (CaCO₃), etc. are used to react with sulfur dioxide exists in the released flue gas. Among all FGD technologies, the limestone-based wet FGD plays a dominant role due to its cost-effectiveness and high reliability [3]. As the operating status of FGD system has a substantial effect on the desulfurization performance, it is essential to well understand the dynamic behavior of the FGD process. Consequently, a lot of research interests have been in the FGD process modeling; however FGD is a nonlinear multivariable process with high time delay, rendering the modeling task a real challenge. The modeling approaches can be separated into two categories: mechanism-based approaches and data driven approaches.