Time-FLM: Universal large model for fed-batch process time series forecasting

doi:10.1016/j.knosys.2026.115580

Time-FLM: Universal large model for fed-batch process time series forecasting

2026-02-17

0

PRE

AI

DOI：10.1016/j.knosys.2026.115580

Original

Original request for help

Save

Abstract

En 中文

• Time-FLM: a large pretraining framework tailored for fed-batch industrial processes. • FLM-Data: a 4M-point real-world dataset covering four fed-batch systems. • Time Stage Embedding and MoE routing enable strong cross-domain generalization. • Time-FLM outperforms SOTA models in both zero-shot and full-shot forecasting tasks.

Keywords:

Time-FLM

fed-batch process

time series forecasting

large pretraining framework

cross-domain generalization

Time-FLM: Universal large model for fed-batch process time series forecasting

Abstract

Journal

Organization