I have graduated from DTU with a MSc in Engineering. Here are some points that worth mentioning.

Subject, Tool and Experiment

The most important thing I realised is that there is a huge difference between tools and subjects. Most of the courses I took during bachelor should be classified as subjects. For example, “power electronics in renewable energy” and “distributed energy system and theory”. Only some math courses in the freshman year provide means to handle calculation in subject courses. Though I was exposed to all kinds of “technologies”, only one subject will be the subject during work or PhD.

I started thinking about this distinction when choosing the topic of my Bachelor thesis. The idea was to focus on a tool that can utilise my “knowledge” about different energy “technologies”. I did find a tool, which I think should be called physical modelling. The subjects are thee kinds of thermal engines and heat transfers. The thesis got me an award from the university, and it is hosted in edxu96/StaticDESim.

At that time, the MSc program focusing on energy system analysis at DTU caught my eyes. I had read a book for energy system analysis [introduction2021blok] that time and loved the idea. However, during the first year at DTU, I realised those subjects are too broad. For example, International Energy Agency (IEA) is the top institution in this field, and publishes lots of data and analysis about country-level energy systems. There is a very important concept called “scenario” that I never encountered before. For example, there is a tool called TIMES, in which DTU participated. It seems to me that most of the energy system analysis models are based on parameters from such “scenarios”, because the spatial and temporal scope is so broad that statistical models are not applicable. Such models provide a convenient way for the public, politicians and “researchers” to communicate, and contribute a lot, for example, to Danish climate policies. There are multiple courses on such modelling, but statistics never play an important role. The building of “scenarios” is too subjective. Besides, I don’t think I can find a job or build something new in such field, largely because I am an outsider. Gradually, I lost interest. (There are other topics in energy system analysis like life cycle analysis [burkhardt2011life] that I don’t want to talk about.)

Luckily, another focus of that major is optimisation. I took many courses, though never go beyond the limit of MILP. In my mind, they focused too much on algorithms, which are important, indeed, but boring. How to apply MILP to practical problems [williams2013model] was never mentioned. Besides, the area, decision making under uncertainty [conejo2010decision], interests me and it requires statistics. So I had to learn statistics from scratch. I did take course in probability in Taiwan, but it was mainly about combinatorial analysis (see, for example, chapter 1 of first2014ross) and a bit about Bayes’ theorem. Now, MILP and basic statistical models have been in my toolbox.

In summer 2019, I took my favourite course, stochastic simulation, (for which I worked as TA in summer 2020). I didn’t really understand the course until the teacher summarised it, in the last lecture, as “computer experiments with mathematical models” [santner2018design]. For example, discrete event simulation. [law2015simulation] I begun to understand the importance of scientific experiments. Also, to make friends with the teacher, I audited the stochastic processes course [pinsky2010introduction] in autumn 2019, which is the hardest course I ever take. The benefit is that I has become familiar with other topics in operations research. Overall, every aspects for mathematical and physical simulation become no stranger to me.

In that autumn, I also became addicted to the graph theory course, and audited every lecture. Furthermore, it played a dominant role in my master thesis and my part-time job at Utiligize company. Such experiences contribute to edxu96/mgrid, and this subject, utility networks, has become my focus since.

Collaboration

The most important lesson I learnt in Denmark is collaboration. Every single course more or less emphasises collaboration, even when individual assignments are the only requirement. There were some collaborative courses during my bachelor, but the first thing we did was to spit the task equally and finished it on our own. Surprisingly, in Denmark, the usual way in terms of collaboration is to do exercises/assignments together (real-time and face-to-face), aka group work (see a post in DTU International Student Blog), which I never see in China.

There are advantages and disadvantages. I really hate this group work stuff and tried to avoid it during the first two years, because it is not how software engineers work together and my oral English was too poor at that time. But this way of study is deeply integrated, since it usually takes half of the course time. Teachers and TAs will be present, answering questions. I regret not taking the opportunity now, because to receive help from teachers and TAs can largely improve the final grade. It’s not sufficient to know everything from lectures. Instead, after the lecture, I went back to my dorm, where I read books and wrote code alone.

Another example, I used to love writing LaTeX using overleaf.com, and in one course, my teammates decided to rely on its real-time collaboration feature. I never agreed with the idea and spent my time writing Julia code, in order to obtain results our final assignment of that course. The writing was a disaster, because this issue, What if a collaborator makes changes I don’t want to keep?, was never resolved. The worst scenario was that we could not compile the file until main contents were pasted in another template. Then it occurred to me that version control can be apply to writing as well, so I stopped using overleaf.com afterwards.

A more suitable approach I can think of, which is also adopted by some teams, is to have multiple short meetings followed by individual work. The biggest two problems of group work are irregular interruptions from teammates and the mix of formal and casual conversation. So at the time of group work, the classroom is very noisy. Scheduled short meetings can be focused on formal conversation and afford students multiple long periods of uninterrupted time.

In terms of code, I was never able to collaborate with others, because I never saw anyone capable of basic code refactoring. In summer 2020, I managed to follow the practice of continuous integration and master GitHub issues. It always surprises me how much effort is required to reach this milestone, and how much it helps after. I stuck to this practice when coding and writing for my master thesis.

Schedule

To schedule day-to-day life and work is not trivial. I have never seen a teacher or a working colleague who doesn’t schedule their daily activities. I started to reply heavily on timetables in the second semester, when the courses I took were very hard. My plan was to pass all the courses, so I formulated & adjusted my timetable accordingly everyday. Moreover, during the pandemic, online meetings can be fully integrated in the timetable software.

In my experience, a good starting point is to separate any day into three 4-hour periods: morning, afternoon and night. A very, if not most, important thing is to leave enough time for sleep and meals, because the daily schedule must be sustainable. As discussed above, GitHub issues can facilitate collaboration. It can also log to-do list and the study progress. What’s next can be easily found and prioritised.

Self-Study

Self-study skills are always worth mentioning. To read is the most important one, and it is the cornerstone of all the other skills. For example, I always believe only a small portion of information is delivered in lectures or slides, so I always read all the materials before lectures. Also, one can never write until read some materials in the same domain. The last example has been discussed before in work as TA. The only way to understand some algorithm is to read its pseudo-code or description.

To admit one’s reading limit is not easy. There are many milestones in terms of levels in my mind. To read and understand textbooks is just the first step. It’s harder to read journal papers because of the inadequate editing and different writing styles. The materials in the next level are news and magazines, and I admit consulting dictionaries a lot. There are many words with which I am not familiar. The highest level, I think, is to read novels. The only novel in English I have ever dared to read is “The Little Prince”.

Again, one can formulate reading lists using GitHub issues. It also serves as an outline in case some contents are needed in the future.

Before the pandemic, I always print books and gather printed papers to books and it helps a lot. It takes a while to adjust text sizes sometimes, but there are many advantages. To write directly on pages really enhance the reader’s concentration. Also, to transcribe formulas strengthen the reader’s visualisation memory. Gradually, I got used to reading long texts and became able to read papers on screen, when the pandemic hit.

Another important skill is test (validation), which has been discussed in work as TA as well. The last one I want to mention is to pose questions. For example, I posted a question about my application of KKT condition and received lots of valuable feedbacks. To write a well-defined question requires effort, but reward with high-quality answers and an improvement in writing.

Reference

• Blok, K., & Nieuwlaar, E. (2021). Introduction to energy analysis. Routledge.
• Burkhardt III, J. J., Heath, G. A., & Turchi, C. S. (2011). Life cycle assessment of a parabolic trough concentrating solar power plant and the impacts of key design alternatives. Environmental science & technology, 45(6), 2457-2464.
• Conejo, A. J., Carrión, M., & Morales, J. M. (2010). Decision making under uncertainty in electricity markets (Vol. 1). New York: Springer.
• Law, A. M. (2015). Simulation modeling and analysis. New York: McGraw-Hill.
• Ross, S. (2014). A first course in probability. Pearson.
• Ross, S. (2012). Simulation. Academic Press.
• Pinsky, M., & Karlin, S. (2010). An introduction to stochastic modeling. Academic press.
• Santner, T. J., Williams, B. J., Notz, W. I., & Williams, B. J. (2018). The design and analysis of computer experiments. New York: Springer.
• Williams, H. P. (2013). Model building in mathematical programming. John Wiley & Sons.