Chapter 5
Case studies
Selected projects, anonymised. Each one is described in the same order: problem, solution, result. Where the result has been measured, the number is given.
Automated PV farm control at negative prices
Problem: farms switched off by hand at negative prices. During downtime nobody knew whether it was a fault or a DSO curtailment.
Solution: a system that fetches prices from the Polish TSO every 15 minutes, controls the inverters and keeps operators informed in a messenger group. Developed and maintained since 2022.
Production plans sent with no one at the keyboard
Problem: production plans for the grid operator prepared from fixed profiles.
Solution: a production model for each farm driven by weather forecasts, and automatic XML submission to the operator’s API.
Result: daily error about 30–40% lower than with naive methods, plans submitted every day about four weeks ahead.
Day-ahead price and negative price risk forecast
Problem: decisions about running installations made without knowing the price over the next hours.
Solution: a probabilistic forecast for today and tomorrow, stored in a database and available in a dashboard.
Result: about 40% lower error than “same as yesterday”, ROC-AUC 0.93 for negative prices.
Automating the design process at an architecture office
Problem: repetitive work in Revit, manual reporting, manual project start-up.
Solution: a set of Revit add-ins, automated project start-up with spatial data (3D site plan generator), reporting and CI/CD in the design process.
Result: 40 hours of manual work per week replaced by 10 minutes of scripts and 2 hours of review.
Automated data validation at a general contractor
Problem: structural model data extracted and checked by hand, project data scattered across many formats.
Solution: a Python pipeline (IfcOpenShell, pandas) processing about 200 IFC models, models, drawings and specifications unified in a central database, a local vision model reading documents.
Result: process cut from 3 days of manual work to 3 hours of review, data from many projects available for management analytics.
Profitability analyses for renewable projects
Problem: assessing locations and profitability of PV farms, wind turbines and linear infrastructure.
Solution: parametric project geometry from GIS data and monthly forecasts from irradiance, weather and energy price data.
Open source
Simple Asset Manager (a Blender add-on), meshroom2blender and fixes to IfcOpenShell. Leysa builds on open source and gives some of the work back.