A hybrid mechanistic machine learning approach to model industrial network dynamics for sustainable design of emerging carbon capture and utilization technologies
A hybrid mechanistic machine learning approach to model industrial network dynamics for sustainable design of emerging carbon capture and utilization technologies
复制标题
一种混合机械机器学习方法，用于模拟工业网络动态，以实现新兴碳捕获和利用技术的可持续设计
DOI：
10.1039/d3se01032e
复制
发表时间：
2023
影响因子：
5.6
通讯作者：
Singh, Shweta
中科院分区：
文献类型：
--
作者：
Shekhar, Abhimanyu Raj;Moar, Raghav R.;Singh, Shweta
工业网络由多个工业节点组成，这些节点通过支持网络整体生产目标的物质交换相互作用。这些产业网络表现出复杂的非线性动力学，由于多尺度的性质，行业之间的相互作用和每个产业节点的固有动态。此外，这些整体动态对这些网络的可持续设计具有重大影响，沿着的是整个网络的资源消耗和排放动态。然而，理解工业网络的整体动态是具有挑战性的，因为对于整个网络动态不存在数字模型，特别是对于新兴的工业系统，并且对其进行模拟分析可能在计算上是昂贵的。为了克服这一局限性，我们提出了一种基于数据驱动系统识别的混合机制机器学习方法，以建立工业节点的代理动态模型，这些模型可以耦合到评估整个工业网络动态。此外，我们建议利用整体网络动态量化动态碳足迹和设计的工业网络的最大碳汇。我们应用我们的方法来评估藻类生物柴油工业网络的动态碳足迹，包括5个独立的动态工业系统。根据整个网络的动态变化，采用修改后的技术参数对网络进行重新设计，可使CO2封存率提高约2%，达到29 750.34 kg h−1，根据网络的非线性模型，50小时运行的净CO2足迹准确计算为−1485069.47 kg。 动态模型还用于分析在特定年份使用特定藻类生物柴油网络在特定区域完全消除与能源相关的CO2排放所需的净中和时间，从而深入了解该技术在实现气候减缓目标方面的潜力。因此，所提出的方法建立了一个路径，以评估工业网络动态的任何新兴系统，依靠机械模型和数据驱动的系统识别和通知未来工业网络的可持续设计。
Industrial networks consist of multiple industrial nodes interacting with each other through material exchanges that support the overall production goal of the network. These industrial networks exhibit complex nonlinear dynamics arising due to the multiscale nature of interactions among industries and the inherent dynamics of each industrial node. Furthermore, these overall dynamics have a significant impact on the sustainable design of these networks, along with the resource consumption and emission dynamics of the overall network. However, understanding the overall dynamics of industrial networks is challenging as digital models do not exist for the whole network dynamics, especially for emerging industrial systems, and simulative analyses of the same can be computationally expensive. To overcome this limitation, we propose a hybrid mechanistic machine learning approach based on data-driven system identification to build surrogate dynamic models of industrial nodes, which can be coupled to evaluate the overall industrial network dynamics. Furthermore, we propose utilizing the overall network dynamics to quantify the dynamic carbon footprint and design of industrial networks for a maximum carbon sink. We apply our methodology to evaluate the dynamic carbon footprint of an algal-biodiesel industrial network comprising 5 separate dynamic industrial systems. The redesign of the network with the modified technological parameters informed by overall network dynamics results in an approximately 2% enhanced CO2 sequestration rate of 29 750.34 kg h−1, with the net CO2 footprint being accurately calculated as −1485069.47 kg for 50 hours of operation based on the nonlinear model obtained for the network. The dynamic models were also used to analyze the net neutralization time required to completely remove the energy-related CO2 emissions using this specific algal biodiesel network for a specific region in a particular year, providing insights into the potential of this technology to meet the climate mitigation goals. Hence, the proposed approach establishes a pathway to evaluate industrial network dynamics for any emerging system by relying on mechanistic models and data-driven system identification and informing the sustainable design of future industrial networks.