Work

Modelling Multi-Component Predictive Systems as Petri Nets

Year: 2017
Type: conference-paper
Abstract: Building reliable data-driven predictive systems requires \na considerable amount of human effort, especially in the \ndata preparation and cleaning phase. In many application \ndomains, multiple preprocessing steps need to be \napplied in sequence, constituting a ‘workflow’ and facilitating \nreproducibility. The concatenation of such work- \nflow with a predictive model forms a Multi-Component \nPredictive System (MCPS). Automatic MCPS composition \ncan speed up this process by taking the human \nout of the loop, at the cost of model transparency. In \nthis paper, we adopt and suitably re-define the Wellhandled \nwith Regular Iterations Work Flow (WRI-WF) \nPetri nets to represent MCPSs. The use of such WRIWF \nnets helps to increase the transparency of MCPSs \nrequired in industrial applications and make it possible \nto automatically verify the composed workflows. We \nalso present our experience and results of applying this \nrepresentation to model soft sensors in chemical production \nplants.
Language: English
Open Access status: green
OpenAlex ID: W2964191388
FWCI: 0.72
Cites: 40
Cited by: 13
Related to: 10