About Research Project
As a critical node in any supply chain systems, warehousing is a common yet complex industrial practice that involves a variety of planning and operation problems, such as layout design, equipment configuration, inventory allocations and picking policies. Different strategies and operation rules need to be developed according to the various nature of goods stored in the warehouses, and usually need to consider future technologies, what-if scenarios, market fluctuation and variation in demand. Although simulation based optimization is shown to be a practical and effective tool for the decision making, the main drawback comes with the time complexity for simulating and acquiring the optimal results, for a large-scale system or the problem associated with high uncertainty.
This research is targeting to fill the gap by strengthening and re-engineering the simulation based decision-making process to fit the “smart warehousing”, in which real-time decision are provided automatically based on the instant operational state. A dual-track architecture is to be proposed, with 1) online decision making and 2) offline simulation optimization to be carried out concurrently, aiming to prevent time latency due to the coupling of the two processes. The two processes are synchronized by statistical learning or deep neural networks.
The learning procedure is to be designed for effective analytics on both historical and simulated data, and mapping the relationship between environmental state and corresponding best action. New algorithms and methodologies should be developed under the corresponding architecture, and to be engineered in an efficient way to release the power of an HPC infrastructure.
