(437b) Discovering Heuristics for Sustainable Design By Multiobjective Evolutionary Optimization and Machine Learning

In sustainable process design (SPD), it is increasingly common to use multiobjective optimization for understanding the trade-off between monetary and life cycle environmental objectives. While these Pareto fronts are useful, what would be more valuable is the knowledge about what makes the solutions optimal. For example, it would be useful to know if there are specific decisions that must be made or avoided for ensuring optimality. Such knowledge can result in heurisitics that can be used to guide other SPD projects. The benefit of such knowledge for guiding conventional engineering design is well-known and widely used by practitioners. However, heuristics for SPD are not yet available. In order to obtain such insight, additional analyses are required to go beyond just finding the Pareto front.

Nonetheless, most current SPD generally neglect the ecological carrying capacity. The ignorance of these limits while designing systems has led to unintended harm, such as ecological degradation. Techno-Ecological Synergy (TES) framework [1] was developed to fill this gap to account for ecological carrying capacity by quantifying the demand (i.e. resource use and emissions) and supply (i.e. capacity of nature) of ecosystem services (ES). The sustainability metric has been defined based on the the difference between the supply and demand for ES.

This work integrates the recently developed approach of techno-ecological synergy (TES) design with multi-objective evolutionary optimization and classification and regression tree (CART) machine learning algorithm, with the main focus on obtaining insights to explain optimality. First, evolutionary optimization has been applied to obtain the Pareto fronts. Optimal points on the pareto fronts and suboptimal points in the solution space can be generated simultaneously with the optimization. The CART algorithm has been applied to classify the solutions in the decision space, which could potentially lead to the discovery of general heuristics about TES sustainability designs. The CART algorithm has the ability to detect interactions and identify groups that have similar outcomes along with the associated predictor variables[2]. The predictor variables that contribute to a more thorough classification can thus be determined and identified as a design heuristic.

The methodology has been applied to a case study describing the design of a residential system accounting for ecosystems like trees, lawn and a vegetable garden[3]. Technological, ecological and behavioural variables are considered in the design to simultaneously minimizing cost while maximizing environmental benefits. This multi-objective optimization problem is formulated as a simulation-based optimization by integrating EnergyPlus simulation software and evolutionary algorithm. The results show that the solutions obtained from TES-designed systems are economically superior compared to techno-centric solutions; they also have environmental benefits in terms of reducing overshoot, thus leading to a “win-win” scenario. In addition, based on initial work, the existence of shading trees has been identified as a design heuristic that will be “win-win”. The discovery of such heuristics will be useful in guiding the future design of similar systems.

References

[1] Bakshi, Bhavik R., Guy Ziv, and Michael D. Lepech. "Techno-ecological synergy: A framework for sustainable engineering." Environmental science & technology 49.3 (2015): 1752-1760.

[2] Neville, Padraic G. "Decision trees for predictive modeling." SAS Institute Inc 4 (1999).

[3] Urban, Robert A., and Bhavik R. Bakshi. "Techno-ecological synergy as a path toward sustainability of a North American residential system." Environmental science & technology 47.4 (2013): 1985-1993.