Hyperight

Time Series Forecasting at Scale – Lessons learnt from more than 10000 training jobs – Edgar Bahilo Rodríguez & Mohamed Ahmed, Siemens Energy

Bookmark
play-rounded-fill
[

Time series forecasting production workloads are challenging by nature. More than often individual simple statistical models like ARIMA or Prophet tend to give sufficient performance to not consider more sophisticated approaches. However, maintaining and retraining so many individual models can become a challenge. In this session you will learn what Siemens Energy is doing to solve this challenge and its future plans to optimize their current solutions.

Key Takeaways

  • Forecasting strategies (recursive, direct, mimo and hybrid)
  • Metadata oriented machine learning
  • Complex cloud machine learning pipelines
  • Serverless active model monitoring

Add comment