v0.5.5 (13th August 2025)

New Features:

• Realtime simulation: A simulation which runs in real-time instead of simulation time can be created and used for hardware-in-the-loop with real assets. Manual configuration of the agents is required for this.

Improvements:

• Changed Logging for DRL metrics: TensorBoard logging was restructured to separate metrics collected per gradient step and per simulation time step. This avoids unnecessary padding, ensures consistency, and prevents data loss across different logging frequencies.
• Improve checking for available solvers: Defines the list of available solvers only once
• More notebooks in CI: add notebook 10 to CI for functional validation of Demand-Side-Units (DSU).
• Correctly suppress highs output for linopy: this fixes an error when running the redispatch market clearing on Google collab
• Add JOSS paper
• Pin pip-tools to fix the docker build
• Additional learning strategy for renewables: Introduced a new learning strategy specifically designed for renewable energy sources. Most of the functionalities can just be inherited, we chose to add the availability of the unit into the individual observations and calculate the opportunity costs based on the available generation as well.

Full Changelog: v0.5.4...v0.5.5