DeepGSA: Plant Data-Driven Global Sensitivity Analysis using Deep Learning

Data-driven modeling provides a viable alternative for process modeling especially in applications where mechanistic modeling falls short of explaining the underlying phenomena. The increasing amount of plant data collected through various sensors and lab tests lays the foundation for various data-driven modeling approaches such as Deep Neural Networks (DNN). In this work, we present a new software tool, named deepGSA, incorporating well-established variance-decomposition and derivative-based global sensitivity analysis (GSA) methods, such as Sobol sensitivity indices, with the plant datadriven deep learning modeling techniques. The deepGSA aims at enabling non-specialist practitioners to leverage DL-based models for GSA application purposes. The tool is successfully applied on a benchmark case study as well as the case of modeling liquid nitrous oxide concentration in a wastewater treatment plant to highlight its capabilities. The deepGSA toolbox, documentation, installation guide, and several examples are freely available on GitHub through the (link)[https://github.com/gsi-lab/deepGSA.]

Recommended citation: Aouichaoui, A. R., Al, R., & Sin, G. (2022). DeepGSA: Plant Data-Driven Global Sensitivity Analysis using Deep Learning. In Computer Aided Chemical Engineering (Vol. 49, pp. 1759-1764). Elsevier. https://doi.org/10.1016/B978-0-323-85159-6.50293-1