Privacy-Preserving Federated Learning for Chemical Process Optimization
A new framework enables collaborative model training across distributed chemical plants without sharing raw data. Each plant trains a neural-network-based process model locally using time-series sensor data, transmitting only model parameters to a central aggregation server via secure mechanisms. This preserves data locality and industrial confidentiality. Experiments used datasets from three independent plants.
Key facts
- arXiv:2604.26073
- Federated learning enables collaborative model training without sharing raw operational data
- Each plant trains a neural-network-based process model locally
- Only model parameters are transmitted to a central aggregation server
- Secure aggregation mechanisms are used
- Experimental evaluation used process datasets from three independent chemical plants
- Plants operate under heterogeneous conditions
- Framework addresses data confidentiality constraints in industrial chemical plants
Entities
—