(399e) Data-Driven Optimization of Manufacturing Systems Under Supply and Demand Volatilities in an Industrial Symbiosis Network | AIChE

(399e) Data-Driven Optimization of Manufacturing Systems Under Supply and Demand Volatilities in an Industrial Symbiosis Network


Suvarna, M. - Presenter, National University of Singapore
P S, P., National University of Singapore
Wang, X., National University of Singapore
The concept of industrial symbiosis (IS) draws inspiration from circular economy, and is a promising approach for resource exchange, energy conservation and emission reduction in the global manufacturing industry (1). However, uncertainties in supply, demand and production environments render significant complexities for decision support systems (DSS) within individual entities of the IS network (2), thereby introducing the risk of suboptimal decisions when operations are performed deterministically . Traditionally, such problems have been solved using the classical stochastic or chance-constrained optimization under uncertainty.

In contrast to previous studies, this research investigates the application of data-driven optimization by coupling the concepts of machine learning and conventional mathematical optimization (3), to devise a DSS. Specifically, we consider uncertainties associated with the supply of resources and demand – applied to a real-world wood manufacturing company in an IS network. For its production requirements, the company primarily depends on the supply of waste-wood (material resource) from other entities of the IS network, which are highly volatile in terms of quantity and quality. The company also relies on a hybrid power system (energy resource) comprising of solar panels in addition to the main power grid to meet the energy demands, wherein the volatility in weather conditions and time-of-use power pricing significantly impacts the production.

To this aim, based on historical data, we first develop hybrid machine learning (4) models to predict supply of material and energy resources and incorporate the concept of prediction interval (PI) to the outcome of the model to quantify the degree of uncertainty associated with the respective volatile resources. This prediction interval serves as an ambiguity set for an optimization formulation that accounts for the optimal utilization of wood-waste as well as optimal combination of power sources to meet the production demands; with multi-objectives to minimize the overall production cost, emissions and optimize inventory utilization. An optimal DSS based on data-driven optimization framework as described in this study, is computationally less intensive (5) and offers greater flexibility in decision making in the face of uncertainties (5) to aid the overall productivity, profitability and sustainability of individual networks in an IS network.


  1. Valenzuela-Venegas G, Vera-Hofmann G, Díaz-Alvarado FA. Design of sustainable and resilient eco-industrial parks: Planning the flows integration network through multi-objective optimization. Journal of Cleaner Production. 2020 Jan 10;243:118610.
  2. Cao X, Wen Z, Xu J, De Clercq D, Wang Y, Tao Y. Many-objective optimization of technology implementation in the industrial symbiosis system based on a modified NSGA-III. Journal of Cleaner Production. 2020 Feb 1;245:118810.
  3. Ning C, You F. Data-driven Wasserstein distributionally robust optimization for biomass with agricultural waste-to-energy network design under uncertainty. Applied Energy. 2019 Dec 1;255:113857.
  4. Chandra R, Ong Y-S, Goh C-K. Co-evolutionary multi-task learning for dynamic time series prediction. Applied Soft Computing. 2018 Sep 1;70:576–89.
  5. Ning C, You F. Optimization under uncertainty in the era of big data and deep learning: When machine learning meets mathematical programming. Computers & Chemical Engineering. 2019 Jun 9;125:434–48.