AI Subsystem

ARTIFICIAL INTELLIGENCE (AI) SUBSYSTEM METHODOLOGY

parametrization and training of machine learning models based on TITAN historical data 🡪 prediction of Key Performance Indicators (KPIs) to be optimized as direct or composite function of (directly) manipulated and non-manipulated variables
- data preprocessing
  - non-operation intervals elimination
  - outliers elimination
  - normalization
  - automatic feature selection with Recursive Feature Elimination with Cross-Validation (RFECV) and Sequential Forward Selection (SFS) methods
  - automatic split with respect to time with clustering 🡪 separate model per cluster
- regression models (supervised learning) with various hyperparameters
  - Linear Regression (LR)
  - Multilayer Perceptrons (MLP)
  - Random Forest (RF)
  - K-nearest-neighbors (KNN)
  - LightGBM (LGBM)
  - Gradient Boosting Regressor (GBR)
  - Extreme Gradient Boosting Regressor (XGBoost)
  - CatBoost
  - Transformed Target Regressor (TTR)
application of trained models to real-time data
automatic retraining of models based on production conditions changes with reinforcement learning models

The project SIROKO was implemented in the context of the Action RESEARCH – CREATE – INNOVATE and co-financed by the European Union and national resources through the OP. Competitiveness, Entrepreneurship & Innovation (EPANEK).

The project started in June 2021 and is expected to be completed in December 2023.

Scientific co-ordinator is Dr. Dimitrios Tzovaras from CERTH/ITI.

AI Subsystem

Contact