Abstract
A misusage of machine learning (ML) strategies is usually observed in the process systems engineering literature. This issue is even more evident when dynamic identification is performed. The root of this problem is the gradient explode and vanishing issue related to the recurrent neural networks training. However, after the advent of deep learning, these issues were mitigated. Furthermore, the problem of data structuration is often overlooked during the machine learning model identification in this field. In this scenario, this work proposes a guideline for identifying ML models for the different applications in process systems engineering, which are usually for simulation or prediction purposes. While using the proposed guideline, the work also identifies a virtual analyzer for a pressure swing adsorption unit. In these types of adsorption separations, it is usual that the measurement of the main properties is not done online. Therefore, the virtual analyzer is another contribution of this manuscript. The overall results demonstrate that even though the test provides good performance during the ML model identification, its quality might degenerate over the application domain if the model application is overlooked.
Highlights
While, from a first perspective, the concepts of prediction and simulation appear to be the same, they are different, and this difference must be emphasized
The machine learning models used to represent the nonlinear dynamic system of the pressure swing adsorption (PSA) will be based on deep neural networks, recurrent neural networks, and feedforward neural networks
For the cases of the deep neural network (DNN) and the feedforward neural network (FNN) models, it is possible to observe that their parity is randomly distributed around the diagonal line across the whole range of values, indicating that the residuals are random
Summary
From a first perspective, the concepts of prediction and simulation appear to be the same, they are different, and this difference must be emphasized. While the NARX can be utilized reasonably in certain cases, its use must be evaluated, as the cumulative error associated with its structure can lead to unrealistic results or instability of the model [1] In this scenario, this work addresses this open issue in the literature providing guidelines for the adequate use of machine learning models. Subraveti et al (2019) [12], they applied a feedforward structure to model a PSA unit for optimization purposes These works have important contributions to the PSA field, addressing the process optimization and modeling issues. These instruments require a significant amount of time to perform the measurement, during which the desired information is unknown In this scenario, this work proposes identifying a soft sensor to perform real-time predictions of the purities and recoveries of a PSA unit. Two specific contributions are the development of a soft sensor for a PSA unit and a simulator of the PSA process
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have