Petronor – Soft Sensors
Petronor
Sector: Energy
Business Case
Determining process stream properties, such as pentanes in LPG streams or naphtha vapour pressure.
Objectives
Obtaining real-time values of the quality of intermediate and final products to: 1) optimise production and 2) minimise consumption associated with rework or corrective actions with little time margin.
Use case
Obtaining real-time values of the quality of intermediate and final products to: 1) optimise production and 2) minimise consumption associated with rework or corrective actions with little time margin.
Infrastructure
Hybrid
Technology
Machine learning and deep learning
Data
Time series of furnace temperatures, crude structure schedule values, reliability data and unit status. Private dataset.
Resources
IT personnel, production, programming, data analysts. Internal server infrastructure. In-house technology.
Difficulties and learning
Heavy reliance on expert knowledge to process data. Algorithm quality degradation resulting from changes in plant and dataset quality makes it necessary to re-train the algorithms, as well as clustering or algorithm ensemble techniques to improve the robustness of the algorithms.
KPIs (business impact and metrics of the model)
Availability and reliability of the algorithm. Error made in predictions with respect to laboratory analysis.
Funding
Project partially financed by Hazitek.
Collaborators, Partners
Vicomtech.
