Development of Strategic Data-Driven Approaches To Sustainable Bioprocess Modelling and Optimisation

为平台和高价值化学品（如生物燃料、食品补充剂、医疗保健相关材料）开发可持续的生物制造路线是建立低碳经济和实现能源/粮食安全的高度优先领域。生物过程（如发酵和光生产）利用不同类型微生物（如细菌、酵母和微藻）的代谢反应途径，将原料转化为目标产品。由于代谢活动和培养流体动力学（工业生产系统）之间复杂的相互作用，生物过程对操作条件（例如pH值、营养供应、培养混合）的变化高度敏感。因此，为了最大限度地提高工艺盈利能力和材料/能量转换效率，保持适宜的微生物生长和生物制品合成的操作条件至关重要。然而，对于通用的大规模生物制造系统，由于当前在线监测设备的限制，可用信息（例如可测量数据）的数量往往是稀缺的，并且高度嘈杂，阻碍了实时过程控制和优化的决策。这个博士项目旨在开发最先进的混合建模工具来模拟和优化复杂的生化过程。这将通过结合机器学习技术和物理驱动的方法来表示和重建工业生物生产场景来实现。然后，这些模型将与前沿的在线优化策略相结合，构建实时动态优化框架，以预测和加强现有生物过程在不同尺度上的表现。该项目将特别关注通过采用和探索各种机器学习方法（如不同类型的神经网络、高斯过程、集成学习和强化学习）来开发先进的数据驱动工具。这些策略将通过与英国、中国和墨西哥的研究小组和工业伙伴的广泛合作，在细菌发酵过程和微藻光生产系统中进行测试。

The development of sustainable bio-manufacturing routes for both platform and high-value chemicals (e.g. biofuels, food supplement, healthcare relevant materials) is a high priority area to establish a low carbon economy and achieve energy/food security. Bioprocesses (e.g. fermentation and photo-production) exploit the metabolic reaction pathways of different types of microorganisms, such as bacteria, yeast, and microalgae, to convert raw materials into target products. Due to the complex interaction between metabolic activities and culture fluid dynamics (industrial production systems), bioprocesses are highly sensitive to changes in operating conditions (e.g. pH, nutrient supply, culture mixing). Therefore, to maximise the process profitability and material/energy conversion efficiency, it is essential to maintain suitable operating conditions for microorganism growth and bioproduct synthesis. However, for generic large-scale bio-manufacturing systems, the amount of available information (e.g. measureable data) is often scarce and highly noisy due to the limitation on current online monitoring equipment, impeding the decision making for real-time process control and optimisation. This PhD project aims to develop state-of-the-art hybrid modelling tools to simulate and optimise complex biochemical processes. This will be achieved by coupling machine learning technologies and physically-driven methodologies to represent and recreate industrial bio-production scenarios. The models will then be combined with cutting-edge online optimisation strategies to construct real-time dynamic optimisation frameworks to predict and intensify the performance of existing bioprocesses at different scales. Particular attention of this project will be placed on the development of advanced data-driven tools through adoption and exploration of a variety of machine learning methodologies such as different types of neural networks, Gaussian Processes, ensemble learning, and reinforcement learning. The strategies will be tested for both bacterial fermentation processes and microalgal photo-production systems through extensive collaborations with research groups and industrial partners in the UK, China, and Mexico.