Application of graph theory to resource distribution policy-based synthesis of industrial symbiosis networks.

Abstract

Industrial symbiosis (IS) involves the repurposing of waste and by-product streams from one chemical industry as feedstock to another. Given the growing environmental and economic concerns, it has become increasingly difficult for industries not to participate in IS. This has encouraged much research into the field, with IS network design being an important optimisation problem in the research space. However, challenges are associated with the creation of IS networks, with transportation costs and resource distribution being key factors. Furthermore, solution strategies are usually complex and neglect the structural features of the network. A possible solution is the use of graph theory for IS network creation. It was hypothesized that structural features of an IS network can evaluate the effect of distribution policies on IS networks created by graph matching algorithms. The Simplex method (SM), Edmonds-Karp algorithm (FF), and the Hungarian method (HM) were adapted to model IS networks, with the intention to establish a ranking in the suitability in creating IS networks. The adaption rendered the algorithms applicable to feasible IS network discovery under different distribution policies. This graph-based approach allowed for the seamless extraction of the network features as graph metrics. Rigorous testing of the adapted algorithms’ performance using graph metrics was done by simulating numerous IS scenarios. It was found that HM identified connections that, on average, minimised transportation costs to the greatest extent. The HM created networks with the smallest travelling distance than those of SM and FF, showing a 9 % and 6.06 % lower value than SM and FF, respectively. Furthermore, HM-IS networks created more stable and fair networks, which was inferred from the graph metrics. To confirm the HM’s apparent superiority in IS network creation, a case study was simulated with the defined distribution policies being modelled from the matching features. Each distribution policy was quantified as a cost from which it was found that HM-IS networks had a 72.5 % and 74.9 % lower overall distribution cost than FF-IS networks and SM-IS networks, respectively. It was concluded that HM is the most suited for IS network creation and that graph-based modelling of IS is a feasible approach.