(371b) Ontology Controlled Optimization in Process Synthesis Applications

Previous work has shown that knowledge-based optimization models in process synthesis applications are more robust in providing final outputs while improving its computational performance. These models are acting as self-supervised systems mainly using production rules for representing knowledge formulated according to analytical results collected during the optimization process and with a view to deepen search in high performance regions. However, most of these on-line analytical approaches have not been upgraded to general knowledge level and interpreted in human languages, so non-experts are hard to understand detailed procedures of optimization. Fortunately, ontology provides a possible shared vocabulary to model a domain of knowledge and enables knowledge to be represented in advanced formats such as RDF (Resources Definition Framework) or OWL (Ontology Web Language) for programming and sharing. Defined as ?an explicit specification of a conceptualization?, ontology has found applications in the field of knowledge representation and was successfully used in other areas of science, such as biology and geosciences. It contains the following 4 types of components: i) classes representing concepts in a domain, ii) instances or individuals representing respective the objects, iii) properties which are attributes describing classes, hence binary relations on individuals, and iv) restrictions expressing constraints on the values of properties. This paper presents an attempt to create ontology for the domain of optimization of process synthesis applications. It is then used to improve the optimization process by guiding the whole process towards more promising regions together presenting new extracted knowledge at optimization run-time, together with serving as an explanation facility to the user. We use an OWL ontology modeling approach which extracts the knowledge from solution pool at every step of optimization process, as shown in the Figure. A benchmark reactor network design synthesis case is studied with an ontology-based optimization system that is automated with the assist of a stand-alone Java application. The Java application continuously collects data from a rule-based optimization model at run-time, analyzes them and populates new findings in ontology by adding instances to a pre-defined class. Production rules are formulated based on the new extracted knowledge and fire to direct optimization search in high performance regions. Through investigating the results it is evident that not only can ontology-based optimization system improve robustness of solutions and computational performance, but also gain a more accurate understanding of the process synthesis procedures and present new extracted knowledge in a decent format.