Research

Research agenda of Shunyu Wu: optimization, forecasting, quantitative modeling, and reliable implementation for large-scale operational systems.

Research Vision

I study how forecasting, optimization, and engineering constraints can be combined into reliable quantitative workflows. The common thread is data-informed decision making for large-scale operational systems, with an emphasis on forecasting, system modeling, and practical implementation.

Research Agenda

1) Optimization and Quantitative Decision Methods

• Train predictive models against downstream operational objectives rather than proxy accuracy alone.
• Diagnose and stabilize training when decision maps are nonsmooth or surrogate gradients are inexact.
• Develop regret-aware objectives that better align learned representations with deployment outcomes.

2) Forecasting, Modeling, and Quantitative Analysis

• Build forecasting and system-modeling modules that respect domain structure, process understanding, and operator knowledge.
• Couple learned surrogates with operational data streams, engineering constraints, and practical rules.
• Use predictive models inside scheduling workflows, not only as standalone estimators.

3) Reliable Implementation in Practice

• Design constrained optimization methods with explicit feasibility handling.
• Integrate reliability mechanisms such as feasible-set tightening, staged validation, replay checks, and operator review.
• Build reusable decision stacks for long-running operation in power and water systems.

Current Focus (2025-2026)

• Quantitative methods for linking forecasting quality with downstream decisions.
• Constrained optimization under unknown or learned constraints.
• Operational planning workflows with deployable forecasting and optimization modules.
• Measurement design, scheduling, and efficiency improvement with deployment feedback.

Domain Applications

• Energy-System Applications: efficiency optimization under operational constraints.
• Urban Service Applications: scheduling, planning, and resilient operations.

Method and Tool Stack

• Modeling: Python-based forecasting and quantitative modeling tools, uncertainty-aware estimation, and domain-informed modeling.
• Optimization: LP/MILP, stochastic optimization, constrained optimization, and mathematical programming workflows.
• Implementation: config-driven services, operational data integration, staged release pipelines, monitoring, documentation, and operator-facing review workflows.