DESC0025165

Quantum Ventura Inc. will develop and implement cutting-edge machine learning tools to forecast emissions from carbon capture technologies, specifically targeting amine degradation product emissions. Current emissions forecasting methods leverage advanced climate models and data analytics to predict atmospheric pollutant levels from a variety of data sources, including historical emissions data and satellite observations. Training machine learning models to predict trends in amine and degradation product emissions is a difficult task due to the sparse amount of data available due to few reactors having carbon capture installations, nonetheless collecting and standardizing emissions data from the installation. We aim to maximize the potential of the available data gathered by the Technology Center Mongstad (TCM) by leveraging advanced machine learning techniques in combination with the domain expertise of our partner who is a leading research institution in the field of carbon capture solutions. Our proposed project seeks to leverage historical emissions data and operational parameters from the TCM and other relevant datasets to create predictive models capable of: Real-time emissions forecasting, causal impact analysis, emissions mitigation recommendations, and atomistic models of carbon capture solvent degradation. We will perform detailed statistical analyses of the time series dataset to provide a deep understanding of the features of the dataset and their interactions in order to provide recommendations to facility operators to minimize the amount of amine and hydrocarbon emissions. Through classical machine learning, deep learning, and cutting edge chemical reaction neural networks, we will provide a well rounded understanding of the dataset itself and a broad range of emissions forecasting models for plant operators to mitigate amine and hydrocarbon emissions. By leveraging our integrated explainable AI (XAI) methods which streamline the process of storing and preprocessing datasets, training deep learning and ML models, and deploying those models, we can provide detailed insights into the behavior of predictions made by our models to ensure they are fundamentally sound. In addition, we will integrate first-principles simulations to obtain a detailed understanding of the solvent degradation process by leveraging our work with the Navy where we are developing a workflow for generating molecular features for machine learning and running DFT simulations of molecules to predict polymer flammability characteristics. With a single SMILES string representation of a molecule, we generate thousands of features for machine learning and simulate the molecule using density functional theory (DFT) to obtain thermodynamic and kinetic properties. This would allow us to simulate the degradation process of the CESAR1 solvent, uncovering the intermediary reactions leading to the degradation of the solvent and resulting emissions. QVI's AI/ML tech will accelerate the development of transformational technologies to significantly improve the emissions reductions and environmental performance of coal and natural gas use. This will then be extended to manufacturing and industrial facilities. The Phase 2 technology technical and commercial viability demonstration serves as an opportunity for QVI to show how our AI/ML tech can help reduce carbon dioxide emissions released from coal electric generation facilities and natural gas electric generation facilities for commercial deployment. When working alongside other industry stakeholders, QVI's participation in the planned construction and operation of a proof of concept under the demonstration program will be essential. Our project will demonstrate the following capabilities: ? Real-time prediction of future emissions based on historical test data, including various operating parameters. ? Comprehensive causal impact analysis to establish emissions baselines and the effects of operational changes. ? Strategic emissions mitigation recommendations through "what if" scenario modeling.