· Sofia Helsinki, Sofiankatu 4 C, 00170 Helsinki
An evening of talks at Sofia Helsinki, sponsored by Cactos: data-driven battery modelling, stochastic models for energy storage, and lightning talks on satellite imagery, ArviZ 1.0, and Python visualisation; plus snacks, drinks, networking, and the all-new PyData Helsinki quiz.
Sponsored by Cactos.
Jussi Pyörre, CTO at Cactos, introduces their work in battery energy storage systems and the data science challenges involved in operating them at scale.
Olli Ruokojoki from Cactos presents his Master's thesis work on creating battery models directly from operational data, skipping expensive laboratory testing entirely.
Learn how Cactos uses data science to solve a critical challenge in battery energy storage: accurately estimating state of charge (SoC) when you can't measure it directly. Olli demonstrates an innovative iterative approach that extracts battery parameters from messy real-world data, achieving remarkable accuracy (7.65 mV RMSE) that outperforms traditional laboratory-based models.
Tuomas Sivula from Cactos presents how stochastic optimization improves battery energy storage system (BESS) operations. Learn how probabilistic programming frameworks like Stan help model uncertainty in ancillary service markets, specifically AFR (Automatic Frequency Restoration Reserve) activation patterns.
Quentin Salomé presents a project using Earth observation data to track changes in Mexico's Lake Pátzcuaro over a 10-year period. Using satellite imagery from the European Union's Copernicus program, he demonstrates how to process cloud coverage, apply spectral indices to identify water bodies, and calculate surface area changes over time. The analysis reveals a 6.2% reduction in lake area, potentially linked to both meteorological factors and human activities such as avocado farming.
Osvaldo Martin presents the upcoming ArviZ 1.0 release, a complete refactor of this essential Python library for Bayesian modelling. Learn about the split into three modular libraries (arviz-base, arviz-stats, arviz-plots), improved dependency management, and a new flexible plotting syntax inspired by grammar of graphics. The redesign enables minimal dependencies for statistical computing while maintaining batteries-included functionality for users who want the full package.
A lightning talk surveying the Python visualization landscape, from Matplotlib's origins as a MATLAB-compatible plotting library to modern alternatives like Seaborn, UltraPlot, Plotnine, and Bokeh.
This event has taken place. The original event listing is on Meetup.