(537f) Multi-Objective Stochastic Optimization for Hybrid Renewable Energy-Based Industrial Symbiosis Network

Industrial Symbiosis (IS) is a collaborative approach that optimizes the resource use through the exchange of materials, energy, water and manpower across various industries thereby generating technical, ecological, social and economic benefits. A significant amount of work reported in the literature on IS have focussed on material-based IS synergies; while less attention has been devoted to the energy-based counterpart. Renewable energy-based IS mainly aims at (i) reducing the amount of energy requirement from outside industrial systems and (ii) reducing the use of traditional fossil fuel based energy production thereby promoting a circular energy transition [1]. In this study, an enterprise-wide energy optimization framework in an IS network is proposed that decides the optimal dispatch/exchange of electricity among the available resources taking into account both environmental and economic objectives.

We demonstrate the aforementioned approach of an energy-based IS by adopting a real-world wood processing industry as the case study. For the production requirements, the company relies on waste wood supplied from horticulture waste, furniture waste and municipal waste, which are highly volatile in terms of both quantity and quality. For the energy requirements, the industry has a grid connected inbuilt hybrid renewable energy system comprising of solar PV panels and wind turbines in addition to a co-generation industry that generates power by utilising the waste wood purchased from the wood processing industry.

As the industry faces several challenges due to uncertainty in solar/wind availability, time of use (TOU) power price from the grid and the availability of cogeneration power, a multi-objective stochastic optimization problem is formulated as a mixed integer linear programming (MILP) problem with the objectives to (i) minimize the cost of overall power usage, (ii) minimize the penalty cost for unmet wood demand, (iii) minimize the overall carbon emissions and (iv) maximize the revenue for power exported to the grid. The uncertainty and the temporal correlation among the above-mentioned multiple variables are modelled using concepts of machine learning including clustering and general adversarial networks (GANS) by making use of historical data [2]. This aids in the development of uncertainty set for scenario generations. The MILP problem is then solved using a data-driven modified version of Non-dominated Sorting Genetic Algorithm (NSGA) which automates transfer of information across the multiple objectives that may share underlying similarities, thereby facilitating improved convergence characteristics and simultaneous convergence toward the Pareto front [3]. The efficacy of applying the modified NSGA to the MILP formulated in this study is benchmarked with state-of-the-art solvers including CPLEX and GUROBI. This study provides model optimization, algorithm performance evaluation, and decision-making analysis capabilities for optimizing hybrid renewable energy system implementation in an IS network under practical setting using real-world data.

References

[1] M. Butturi, F. Lolli, M. Sellitto, E. Balugani, R. Gamberini, B. Rimini, Renewable energy in eco-industrial parks and urban-industrial symbiosis: A literature review and a conceptual synthesis, Applied Energy 255, 113825, 2019.

[2] C. Ning, F. You, Optimization under uncertainty in the era of big data and deep learning: When machine learning meets mathematical programming, Computers & Chemical Engineering 125, 434–448, 2019.

[3] A. Gupta, Y. Ong, L. Feng and K. C. Tan, "Multiobjective Multifactorial Optimization in Evolutionary Multitasking," in IEEE Transactions on Cybernetics, vol. 47, no. 7, 1652-1665, 2017.