I worked on this project together with two fellow students for an upcoming player in the green hydrogen market.

Scope

The company is planning on buying green energy from two sources, the day ahead market and through a power purchase agreement (PPA). Our project was to figure out how we can optimally buy (and sell) this power while also producing hydrogen to satisfy customers. The most challenging part of the problem was integrating monthly matching. This is a concept where you are allowed to buy grey energy and sell your hydrogen as green, as long as there is a surplus of green energy at the end of the month. It is obviously difficult to predict the yield of the PPA and the market prices later in each month. A deficit of green energy is no option, but missing out on profitable opportunities also is not efficient.

Data

The data consisted of the expected yields of the PPAs (wind and solar) and of market prices per MW/h. We had data for various previous and future years. There were some missing and wrong values but no real issues with the data.

Approach

We initially built a Java model that simulated a year of production with a single hydrogen plant. The model went through the data per day and made decisions about production, storage and buying additional power to produce more. This is phase I of the project.

In phase II, we focused on monthly matching and setting up an environment where the model could be run by anyone without a need to install additional software. To solve the monthly matching problem, we used a MIP optimization that took all constraints into account. To make the model easier to use the decision was made to write it in Python, such that it could be run from a Notebook on (practically) any computer, but additions could also easily be made. I mainly focused on building a clean GUI where only a handful of parameters needed to be filled in (together with a data file). After a successful run, the model gives an Excel file with relevant summary statistics and the overview of the simulation as output.

Results and learning experience

The model we developed did a much better job at visualizing the simulation, it is much more versatile than the initial Excel file the company worked with. Furthermore, the monthly matching approach was completely new. It gave good insight into an optimal production approach and showed that utilizing monthly matching could result in a much more efficient use of the plant, being almost always above 90%.

Overall, this was an exciting project, working with real data and with the challenges in the professional world were a good experience. The green hydrogen market is still innovating and there are many things that are just unknown or where there is not a general good practice for. This typically is not the case when working on an academic case or a personal project. But thinking about and solving those challenges with colleagues was one of the most fun things that I hope to do in future projects as well.

Unfortunately, I cannot post the repository of the project, or a more elaborate view of the model because it is internal.