- Type
- Industry
- Focus
- PyTorch · Spatiotemporal forecasting
- Context
- Algorithm Engineer · Oct 2024–May 2025
My contribution
Owned the end-to-end architecture, model design, optimization, and implementation of a highway traffic-balancing system, delivering most core models and code modules.
I worked as an Algorithm Engineer at JSTI Group Limited, Shanghai, from October 2024 to May 2025. Deliverables also included a project proposal and an initial technical manuscript.
Data and forecasting
I engineered preprocessing and feature pipelines for PB-scale raw multimodal data from ETC gantries, probe vehicles, video checkpoints, incidents, traffic controls, road topology, and historical records.
Using Python and PyTorch, I implemented incident-aware spatiotemporal forecasting with Transformer, Bi-LSTM, and STGCN architectures. Validation considered observed traffic curves, congestion duration, road speed, and throughput for real-time and short-term predictions.
Quantifying traffic imbalance
I developed a data-driven framework using dynamic thresholds, unsupervised clustering, information entropy, and PCA to estimate the location, length, severity, and duration of imbalance.
Completed components were iterated through a self-built simulator, expert review, and field-operation feedback.
Planned decision layer
I designed an integration plan for a future decision layer involving knowledge graphs, deep reinforcement learning, large language models, and multi-agent collaboration. This was a design proposal, separate from the forecasting and imbalance-analysis components already implemented.