Conference Agenda

Renewable energy and energy efficiency are crucial for creating new economic opportunities and reducing environmental impacts. Anaerobic digestion (AD) transforms organic materials into a clean, renewable energy source and is recognised as an important part of the UK's net-zero strategy. Co-digestion of various organic wastes and energy crops addresses the disadvantages of single substrate digestion, increasing production flexibility and enhancing gas yields, but adding complexity and sensitivity to the process. This study employs a model-predictive control strategy to optimize biogas production while simultaneously considering global warming potential to find optimal feeding schedules to meet dynamic gas demands via a nonlinear programming model for dynamic optimization of the overall system.

The NLP model incorporates a combined heat and power system to maximize production flexibility and capitalize on dynamic electricity, heat, and gas prices and considers various physical and economic parameters, including biomethane potential, chemical oxygen demand, and substrate density. A cardinal temperature and pH model is utilized to account for substrate degradation and gas production rates under varying conditions. The MPC strategy, implemented using the GEKKO optimization tool, provides fine-grained control over the digester temperature and feeding, accounting for real-world complexities such as time delays in heating/cooling systems, varying ambient conditions, and multiple feed components with different temperatures. We incorporate a hybrid model trained on real plant and experimental data to simulate system responses across varying realistic operating ranges.

Results demonstrate that the integrated model can simultaneously optimize the interaction between biogas generation and CHP operation for real-time profit maximization, while considering system environmental impact. The model provides detailed analyses of substrate utilization, total production volumes, and methane and carbon dioxide production, while offering insights into the dynamic behavior of the digester temperature control system. A case study validates the model's potential for guiding decision-making in biogas production facilities, emphasizing the necessity of strategic feedstock management and precise temperature control for optimizing biogas yield and CHP operations. This integrated approach represents a significant advancement in the modeling and control of anaerobic co-digestion systems, offering a powerful tool for enhancing the efficiency and profitability of biogas production facilities.

In bioprocess operations, maintaining consistent product quality and yield is critical, particularly given the inherent uncertainties in biological systems. To achieve effective control over these processes, system identification is typically performed to develop mathematical models that describe the dynamic behavior of the bioprocess. However, uncertainties in model parameters often persist due to biological variability, limitations in measurement accuracy, and simplifications made during model development. These uncertainties, which may span a range or follow a distribution, can lead to deviations in process performance if models rely solely on nominal parameter values. Therefore, it is crucial to develop control strategies that ensure key performance indicators (KPIs) (e.g., final product concentration and yield) are consistently met despite these uncertainties.

Real-time feedback control, commonly employed in industrial applications to manage such uncertainties, can be costly and impractical. This is due to the need for high-speed data processing, robust sensors, and rapid control actions, which can strain existing systems. To address these challenges, this paper proposes an approach that eliminates the dependence on real-time control while accounting for model uncertainties. Specifically, we aim to identify the largest possible operational space for the control variables that serves as a guideline for process operations. If the system operates within this defined space, the KPIs can be reliably achieved, regardless of the uncertainties in the system.

We reformulate the problem as an optimization task aimed at maximizing the operational space of relevant control variables, subject to path and terminal constraints imposed by process dynamics and performance specifications. We integrate symbolic frameworks using CasADi, a software tool for numerical optimization, and solve the formulated optimization problem using IPOPT, an interior-point optimizer. Also, a novel stage-wise optimization procedure is implemented to effectively reduce computational burden. Different from conventional surrogate-based methods, by avoiding surrogate models, our method preserves accuracy and fidelity to the original process dynamics, and can easily incorporate path constraints, while efficiently identifying the operational space.

The proposed method is validated through a case study on astaxanthin production, which involves a time-varying control variable (feed flowrate) and seven uncertain model parameters. Our approach successfully identifies an operational space for the control variable, ensuring that final product specifications are consistently met across a wide range of parameter variations. The operational space is further validated through extensive testing, demonstrating the effectiveness of the proposed control strategy.

Soybean is a highly valuable global commodity due to its versatility and numerous derivative products. During harvest, all non-seed materials become “straw,” for each ton of soybeans, 1.2 to 1.5 tons of straw are generated. Currently, this waste is primarily used for low-value purposes such as animal feed, landfilling, and incineration. To address this, the present work proposes a conceptual biorefinery aimed at converting soybean straw into higher-value products. The study began with data collection to identify potential conversion routes for this biomass. Based on this information, a superstructure was developed, comprising nine conversion processes: five thermochemical routes (pyrolysis, combustion, hydrothermal gasification, liquefaction, and deoxy-liquefaction), three biological routes (enzymatic hydrolysis, fermentation, and anaerobic fermentation), and one chemical route (alkaline extraction). Each process was evaluated based on product yields, conversion times, and associated costs from the literature. Using this data, a MILP (Mixed-Integer Linear Programming) optimization model was built in Pyomo with a CPLEX solver. the model comprises 23 possible products (biochar, bio-oil, syngas, ethanol, acetic acid, formic acid, hydroxymethylfurfural, furfural, biopolyols, fiber, methane, biohydrogen, propionic acid, iso-butyric acid, n-butyric acid, iso-valeric acid, n-valeric acid, biogas, xylose, glucose, energy, methanol and dimethyl ether). The variable in this problem was the amount of biomass processed by each conversion process. The objective function was to maximize the profit based on the product mix. The optimization considered a maximum raw material supply of 0.625 tons per year. As a result, fermentation was identified as the most profitable route, yielding annually $2.45 million in revenue. This route utilizes diluted pre-treated biomass and glucose as a supplement to produce five main products: ethanol (5.57 g L^-1), acetic acid (1.74 g L^-1), formic acid (1.03 g L^-1), furfural (0.02 g L^-1), and hydroxymethylfurfural (0.01 g L^-1). In conclusion, soybean straw offers significant potential for value-added biorefinery applications, with fermentation emerging as the most profitable conversion route. Future research will focus on optimizing other processes and exploring additional soybean biomasses while scaling up biorefinery technologies. This could foster sustainable industrial development to valorize wastes generated in agro-industrial sector.

Bioprocesses are seeing increased use for the production of renewable plastics, fuels, and other valuable bioproducts. Supporting bioprocess development during the ever prevalent fourth industrial revolution faces many challenges, including low yields in reactor scale-up, significant batch-to-batch variation and by-product accumulation. Modelling plays an important role in overcoming these challenges in its application to process optimisation and control, as well as process monitoring for fault detection.

However, modelling can be challenging as biosystems are dynamic with complicated catalytic and inhibitory relationships. Their continued exploration has uncovered deeper understanding of fundamental bioprocess mechanisms, allowing the development of kinetic models that rely on physical assumptions to explain the dynamic behaviour of the system. However, in many cases these assumptions only hold true within a narrow operational space and the biosystem will deviate from the assumed behaviour in a manner that is inexplainable through the existing physical assumptions and derivations. In such cases, data-driven modelling may be utilized to capture complicated nonlinear trends that a physical model cannot capture. However, purely data-driven modelling requires extensive data collection to identify the nonlinear model structure. In order to overcome the limitations of physical assumptions and reduce the data requirement to train the model, hybrid modelling can be employed. This uses fundamental physical understanding as the foundation upon which a data-driven model is applied in order to capture the simplified remaining nonlinearities, thus improving upon the accuracy of a purely kinetic model, while reducing the data requirement of a purely data-driven model.

In this study, an Artificial Neural Network (ANN) is employed as the data-driven component to improve the fundamental kinetic model accuracy. This is done by changing the most uncertain kinetic parameters from constants to time-varying outputs of the ANN. The ANN inputs are state variables to ensure the overall dynamic model is a function of state variables only, thus extending its application to the real-time modelling and prediction of fed-batch operation. Depending on the type of measurements, different ANNs are constructed using either offline data or online data to calculate time-varying parameters. To test the efficiency and accuracy of the modelling approach, a case study involving the production of lutein is used. Lutein is a valuable xanthophyll carotenoid utilized in several industries, including food, cosmetics, and pharmaceuticals. The microalga Chlorella sorokiniana has been previously identified for its potentially high lutein and microalgal biomass production, with a recent study exploring its growth and lutein production (Xie et al., 2022). This study also explores and compares the utilization of offline and online data to maximize the predictive capabilities of the model, meanwhile reliably estimating the hybrid model uncertainty. The novelty of this work lies in the application of hybrid modelling to the production of lutein from high-cell-density C. sorokiniana using offline and online data with the scope of conducting optimization under uncertainty for future fed-batch system design.

References:

Y. Xie et al, 2022, High-cell-density heterotrophic cultivation of microalga Chlorella sorokiniana FZU60 for achieving ultra-high lutein production efficiency, Bioresource Technology, Volume 365, https://doi.org/10.1016/j.biortech.2022.128130

Understanding the factors that determine the stability of microbial communities remains a central question in microbial ecology. While ecological principles are increasingly applied to microbiology to study microbial dynamics, the modelling approaches are limited to two extremes: generalized population models that lack spatial detail and genome-scale models that are too detailed for community-level analyses. We advocate for using individual-based modelling (IbM) to overcome these limitations. By explicitly modelling how microbes release toxins to modify their immediate environment, IbM provides a realistic framework to elucidate the principles governing microbial interactions. In our study, we employ IbM to investigate how different interaction types—self-inhibition, amensalism, and rock-paper-scissors (RPS) dynamics—affect community stability in response to varying interaction strengths and growth rates. By comparing the results from IbM with those from stochastic ODE and PDE models, we reveal that simple community-level features can dictate emergent behaviours. Specifically, we find that the type of interaction significantly influences community stability more than the underlying biological processes alone. By capturing these aspects, IbM offers deeper insights into the emergent stability of microbial communities, advancing our understanding of ecological dynamics and potentially informing the management of microbial systems.