Modeling of Liquid Level Based on Combined Dynamic Neural Network

This paper addresses the challenges of modeling the nonlinear dynamic behavior of a wave-washing scrubber in the flue gas acidification process, essential for sulfuric acid production. Due to the process’s complexity, accurate system identification is critical for effective controller design. A double-tank water model is developed, and a Combined Dynamic Neural Network (CDNN) approach is proposed for liquid level identification. The CDNN integrates multilayer perceptrons with dynamic operators based on differential equations, enhancing the model’s ability to capture the system’s dynamics. Validation using real-world data from a steel plant’s sulfuric acid purification section shows that the CDNN model accurately replicates the scrubber’s behavior, confirming its industrial applicability.