(3fd) Data-Driven Energy Systems Design Under Uncertainty

During my PhD, I have worked in the area of Process Systems Engineering (PSE) under the guidance of Prof. Ignacio Grossmann at Carnegie Mellon University. The focus of my thesis is developing algorithms and software for optimization under uncertainty. The techniques that I have developed have been applied to power systems infrastructure planning with high penetration of renewables and shale gas development under price/production uncertainty. My future research interests are outlined as follows.

Sustainable Energy Systems Design and Operation

The design of energy systems is increasingly focusing on systems that involve renewable energies. As of 2019, 64% of the electric generation capacity additions come from solar and wind. I am interested in designing energy systems with high penetration of renewables, such as wind, solar, biomass, and energy storage facilities. The operation of renewable generation like wind and solar is mostly restricted to power systems. I plan to investigate the interplay of renewable energy generation for power systems with chemical processes that require electricity, and natural gas systems, which could in turn motivate novel designs and operational modes of traditional chemical processes. Renewable generation level is usually volatile due to the dependence on weather conditions. Therefore, short-term operating decisions have to be integrated into the long-term planning model to guarantee systems feasibility, which is related to my second research interest, hybrid multiscale modelling and optimization. The uncertainties in renewable generation is one of the motivations for my third research interest, data-driven optimization under uncertainty.

Hybrid Multiscale Modeling and Optimization of Supply Chains

Decision-making in PSE involves large temporal and spatial scales. Strategic decisions are made on a yearly basis, e.g., supply chain design of a large geographical region. Tactical decisions, such as production target, are made on a monthly or weekly basis. Scheduling and real-time optimization decisions are made on a daily or hourly basis. Process control of a reactor or flowsheet are made every few minutes or seconds. One could spend his/her career in one of the five levels of the decision-making processes. However, making decisions at the strategic level while neglecting the decisions at lower levels could lead to suboptimal or even infeasible supply chain designs. On the other hand, integrating models at various scales prevents information loss and ensures that models are consistent across scales. It also allows uncertainty to be propagated across scales to ensure that both the true uncertainty and the source of that uncertainty are known. I am interested in developing a hybrid multi-scale modelling approach that combines data-driven reduced order models with rigorous optimization algorithms. I plan to develop reduced order models that can characterize the optimal decisions from one level of the decision-making process, so that it can be incorporated into the higher-level model. I plan to develop theory and algorithms for the reduced order models to meet constraints of the physical systems. I also plan to use my strength in solving large-scale optimization problems to improve decomposition algorithms for solving multi-scale models. Moreover, I am eager to investigate the aid of parallel computing, GPU computing, and quantum computing, in solving these multi-scale models.

Data-Driven Optimization under Uncertainty

Based on my PhD research, I will continue working in the area of optimization under uncertainty. I am especially interested in problems under the uncertainty of rare events, such as supply chain network design under disruptions, electricity infrastructure planning under power outages, and medical resource management under a pandemic. The challenge in solving rare-event uncertainty problem is that the dimensionality of the uncertain parameters is extremely large. To address this computational challenge, I plan to develop efficient data-driven sampling methods for optimization under uncertainty by exploiting knowledge from high dimensional statistics. I also intend to investigate using distributionally robust optimization for solving this type of problem.

There are different mathematical frameworks that can deal with decision-making under uncertainty, including stochastic programming, robust optimization, reinforcement learning, and dynamic programming. Each method has its advantages and limitations. I plan to explore the possibility of combining the advantages of different approaches in unified modeling approach and solution strategy. For example, reinforcement learning has the advantage of quickly computing online policies, with the recent success in the Alpha-Go project. Stochastic programming is too slow to be deployed in real-time optimization, but can provide rigorous optimality guarantees. It would be ideal to develop a framework that can provide both quick online computations and theoretical guarantees.

Application of PSE tools to Systems Biology

PSE tools, such as mathematical optimization, control theory, statistical data science, have been applied a number of process systems. However, the application of PSE tools to biological systems are sparse; examples include computational protein design, metabolic networks. The increasing abundance of data in systems biology and the breakthroughs in PSE tools for solving large-scale problems, have made the integration of the two fields look promising. I am interested in collaborating with colleagues to develop predictive and optimization models for systems biology applications. I am especially interested in building new genome-scale predictive models and simulation tools of metabolism and protein expression. These models can be used in optimizing desired products in biosynthesis. I am also looking forward to other collaboration opportunities with colleagues in the biological sciences.

Teaching Interests

At the undergraduate level, I am especially interested in teaching senior courses like Chemical Process Design, Chemical Process Control and Optimization. I will emphasize recently developed techniques for process design and process control, such as systematic synthesis strategies, mathematical optimization, flowsheet simulation, and model predictive control. The courses should prepare the students for pursuing graduate research in systems engineering, or for working as process engineers in industry. I am also comfortable with teaching any introductory chemical engineering courses, such as Introduction to Chemical Engineering, Transport Phenomena, and Thermodynamics. At the graduate level, I am interested in teaching courses with a focus on quantitative methodologies, such as an advanced course on Process Systems Engineering, and Numerical Methods.

I am also interested in developing new courses. In particular, I would like to design a course focused on machine learning and mathematical optimization. With the current boom of artificial intelligence and machine learning, more and more companies in the chemical industry are looking for students with strong data science and analytical skills. I have taken all the classes required to obtain a master degree in machine learning at CMU’s School of Computer Science, and all the PhD-level mathematical optimization-related classes at CMU’s Tepper Business School. I believe I have a strong background to develop this new course. Such a course will not only include the basic machine learning and mathematical tools, but also include case studies from chemical engineering-related industries, such as the petrochemical and pharmaceutical industries. Students will apply the data science and analytical tools to solve these case studies by writing their code in Matlab or Python. The class can be offered at both undergraduate and graduate levels. The course offered at the graduate level will have stronger mathematical content.