Mathematical optimization and learning models to address uncertainties and sustainability of supply chain management

As concerns about climate change, biodiversity loss, and pollution have become more widespread, new worldwide challenges deal with the protection of the environment and the conservation of natural resources. Thus, in order to empower sustainability and circular economy ambitions, the world has shifted to embrace sustainable practices and policies. This is carried out, primarily, through the implementation of sustainable business practices and increased investments in green technology. Advanced information systems, digital technologies and mathematical models are required to respond to the demanding targets of the sustainability paradigm. This trend is expanding with the growing interest in production and services sustainability in order to achieve economic growth and development while preventing their negative impact on the environment. A significant step forward in this direction is enabled by Supply Chain Management (SCM) practices that exploit mathematical and statistical modeling to better support decisions affecting both profitability and sustainability targets. Indeed, these targets should not be approached as competing goals, but rather addressed simultaneously within a comprehensive vision that responds adequately to both of them. Accordingly, Green Supply Chain Management (GSCM) can achieve its goals through innovative management approaches that consider sustainable efficiency and profitability to be clearly linked by the savings that result from applying optimization techniques. To confirm the above, there is a growing trend of applying mathematical optimization models for enhancing decision-making in pursuit of both environmental and profit performance. Indeed, GSCM takes into account many decision problems, such as facility location, capacity allocation, production planning and vehicle routing. Besides sustainability, uncertainty is another critical issue in Supply Chain Management (SCM). Considering a deterministic approach would definitely fail to provide concrete decision support when modeling those kinds of scenarios. According to various hypothesis and strategies, uncertainties can be addressed by exploiting several modeling approaches arising from statistics, statistical learning and mathematical programming. While statistical and learning models accounts variability by definition, Robust Optimization (RO) is a particular modeling approach that is commonly applied in solving mathematical programming problems where a certain set of parameters are subject to uncertainty. In this dissertation, mathematical and learning models are exploited according to different approaches and models combinations, providing new formulations and frameworks to address strategic and operational problems of GSCM under uncertainty. All models and frameworks presented in this dissertation are tested and validated on real-case instances.