Imagine you are the CEO of a major utility. You need to decide whether to invest in a new solar plant. What data do you need?

The most important factor influencing the decision is the future price of power. However, this is notoriously hard to predict. So, analysts create scenarios, where they project demand and generation up to 40 years into the future and intersect the curves to compute the price.

When we started the project, such computations were made in Excel, with many difficulties and known problems. There was no auditing, and no backup, the files were hard to understand and maintain, and could only simulate a few days each year.

We reviewed the whole design, creating a cloud-native application based on AirFlow, Kubernetes, and cloud functions.

The latter was adopted to parallelize the solution of a convex optimization problem meant to model the behaviour of energy storage facilities, such as batteries.

This talk presents the main modules of the system, discussing the architectural decisions and the trade-offs we had to face.

Delivered at BI Days, on 10th January 2023, University of Wroclaw, Poland.
More info at https://p.wz.pwr.edu.pl/~business.intelligence/BI-Days/BIDay23Jan