(434b) Enabling Real-Time Synergies in Techno-Ecological Systems Using Adaptive Nonlinear Model Predictive Control

Engineering aims to design systems that are predictable and highly controllable or homeostatic in nature. Process control has played a critical role in the development and operation of such systems. One way for process control to contribute to sustainability is by imposing the requirement that processes are operated in a manner that respects the capacity of ecosystems to supply goods and services that the process needs. These may include regulation of climate, air and water quality, and provisioning of clean water, minerals and other resources. Natural ecosystems tend to exhibit homeorhesis (I.e., variable and intermittent capacity without any set-point) such as varying river flowrate or difference in solar irradiance throughout the year. Traditionally, engineering has either ignored ecosystems or attempted to enforce homeostasis on them. Both can contribute to ecological degradation. To overcome such shortcomings, the techno-ecological synergy (TES) framework aims to design and operate engineered systems in harmony with nature’s cycles (1). Our previous work has shown that at steady-state (2–4), TES designs can be economically and environmentally superior to conventional (techno-centric) designs. However, operation of TES systems poses many new challenges due to phenomena across many scales, complex and nonlinear dynamics of ecosystems. In this work, we take initial steps toward addressing these challenges with the long-term goal of operating TES systems that meet human needs while respecting nature’s limits and restoring its services.

Traditionally, real-time control focuses on set-point tracking in the presence of uncertainty to maintain homeostasis. On the other hand, sustainability requires the integration of ecological and societal dynamics with technological systems, thus introducing safety and ecological constraints. The constraints need to be satisfied under varying ecological conditions while minimizing the cost of operation. Model predictive control (MPC), which is a popular optimization-based control strategy, is a well-suited framework for handling general economic and/or societal objectives in the presence of physical, ecological, and safety constraints (5). However, the vast majority of industrial MPC implementations are time-invariant and the underlying model, objective, and/or constraints are only modified after a significant perturbation or performance degradation has occurred. Furthermore, while the dynamics of technological systems are often relatively fast (on the order of minutes), ecosystems involve a mixture of fast and slow dynamics (on the order of days-seasons-years) that are difficult to account for in a time-invariant version of MPC. In this work, we propose a multi-scale MPC approach that adapts the objective and/or constraints in a time-varying fashion to explicitly account for variability in the capacity of ecological systems. Since ecological capacity is difficult to measure in real-time, we also discuss how time series-based forecasting models can be developed to predict this time-varying capacity using a combination of ecological models and correlated measurements. Using a ground-level ozone regulation case study, we show how this type of an adaptive, multi-scale MPC strategy can lead to economically and environmentally win-win solutions.

The case study is based on an existing chloralkali manufacturing facility located in Freeport, TX, which is part of Houston-Galveston-Brazoria (HGB) a ground-level ozone (O₃) non-attainment zone. A coal-fired electricity generator is available to meet its energy demands and it emits nitrogen dioxide, a precursor to O₃. The air pollutant emissions would lead to environmental deterioration and necessitates investment in air pollution mitigation technologies. Here, we consider a technological option of selective catalytic reduction (SCR) and an ecological alternative of reforestation to mitigate the impact of air emissions. Conventional, air quality policy is designed on multi-year statistic and ignores the capacity of ecosystems to regulate air pollutants. These air quality policies ignore short-term impact of poor air quality and is retrospective in nature. Based on EPA’s NowCast Air Quality Index (AQI) (6), we introduce short-term air quality constraints to mitigate impact of emissions. The AQI constraints are proactive in nature and require forecast of ambient concentration of air pollutant. Past meteorological data is used to forecast hourly capacity of ecosystems. While the AQI constraints are met on an hourly basis, an annualized social health impact cost of air pollutants is also evaluated. Thus, the problem is formulated as multi-scale MPC (spanning from hourly to annual time scales) that minimizes operational cost subject to constraints on AQI. By accounting for the capacity of the reforested trees, we observe that the proposed adaptive MPC approach substantially reduces operational costs compared to the traditional time-invariant form of MPC, while still meeting demand and AQI constraints. When the instantaneous demand is relaxed to a yearly average constraint, the multi-scale MPC further reduces operational costs by varying the production rate based on the currently available (predicted) pollution absorption capacity of the reforested trees. Lastly, we observed that, for sufficiently large reforestation areas, yearly demand and hourly AQI constraints could be satisfied without having to turn on the SCR.