Multiscale Decision-Making for Enterprise-Wide Operations Incorporating Clustering of High-Dimensional Attributes and Big Data Analytics: Applications to Energy Hub

Abstract

In modern systems, there is a tendency to model issues more accurately with low computational cost and considering multiscale decision-making which increases the complexity of the optimization. Therefore, it is necessary to develop tools to cope with these new challenges. Supply chain management of enterprise-wide operations usually involves three decision levels: strategic, tactical, and operational. These decision levels depend on each other involving different time scales. Accordingly, their integration usually leads to multiscale models that are computationally intractable. In this work, the aim is to develop novel clustering methods with multiple attributes to tackle the integrated problem. As a result, a clustering structure is proposed in the form of a mixed integer non-linear program (MINLP) later converted into a mixed integer linear program (MILP) for clustering shape-based time series data with multiple attributes through a multi-objective optimization approach (since different attributes have different scales or units) and minimize the computational complexity of multiscale decision problems. The results show that normal clustering is closer to the optimal case (full-scale model) compared with sequence clustering. Additionally, it provides improved solution quality due to flexibility in terms of sequence restrictions. The developed clustering algorithms can work with any two-dimensional datasets and simultaneous demand patterns. The most suitable applications of the clustering algorithms are long-term planning and integrated scheduling and planning problems. To show the performance of the proposed method, it is investigated on an energy hub as a case study, the results show a significant reduction in computational cost with accuracies ranging from 95.8% to 98.3%.