Swiss Energy Company

Building an advanced forecasting model to eliminate manual forecasting inefficiencies, leading to improved electricity production planning and significant cost reductions in power plant operations

Challenge

• Current forecasting solution provided inaccurate predictions for district heating
• Parts of the solution were conducted manually
• Poor predictions caused imprecise scheduling of electricity production in power plants
• Costly balancing fees due to over- or underproduction of electricity

Solution

• Leveraged Darts’ forecasting framework – an open-source library for time series forecasting, manipulation and anomaly detection
• Simplified the Windows Executable generation with a PyInstaller specification file
• Built a private repo/package on client’s Azure DevOps to enable better scalability, code decoupling, testing, and reproduction on any machine

Business Impact

• 10pp reduction of forecasting error (MAPE from 15.8% to 5.7%)
• 3000x faster speed of predictions compared to previous model
• Leading to more accurate electricity production schedules, more efficient plant operations & cost reduction in balancing fees