Data Assimilation for Discrete-Event Simulations

on YouTube ↗29m

" summary

A major automaker container fleet problem motivates a concrete goal: estimate the probability a factory runs out of containers two days ahead given only barcode hits at transshipment points (no per-container tags) and billions of dollars of inventory at stake. Exact filtering for the high‑dimensional GSMP is intractable (singular transition structure, no transition densities), so the practical procedure is to run many forward discrete‑event simulations, select trajectories matching the current observable via splitting or weight them by local smoothing using a fluid‑limit approximation, then simulate forward to estimate shortage risk; parameter learning and diurnal nonstationarity remain open.

AI-written summary · about

" chapters (8)

1. 0:00Introduction to the topic
2. 2:04Discrete event simulation
3. 4:43Markovian representations
4. 6:42Historical and theoretical
5. 9:14Container management case study
6. 13:34Filtering and state estimation
7. 18:57Practical filtering approaches
8. 25:54Conclusion and Q&A