Research
A Learn-Based Approach for Order-Pooling and Order-Dispatching in a Large-Scale Ride-Sharing System
Columbia University, Jun. 2021 ~ Dec. 2021
Advisor: Prof. Sharon Di
- Proposed a learning-based approach that embedded the residual gated graph convolutional neural network model into a local search algorithm to learn the matching policy in the ride-sharing system
- Designed a novel training method that ensembles imitation learning and evolutionary strategy; Trained the model to imitate the Blossom algorithm first and self-evolute by interacting with the environment
- Developed a ride-sharing simulation environment leveraging the historical taxi data in New York for testing different management strategies
- Conducted a series of numerical experiments and showed that the proposed method outperformed the traditional heuristic algorithm regarding customer satisfaction and computation efficiency
(source)
Optimized Energy-Spong Electric Vehicle Sharing System with Integration of Renewable Energy
Remote, Jul. 2020 ~ Mar. 2021
Advisor: Prof. Xin Wang
Collaborator: Yikang Hua
- Proposed a robust and stochastic optimization model for a spatially distributed electric vehicle fleet with renewable energy integrated to serve as a backup reservation interfacing with transportation and power grid system
- Developed a data-driven approach for constructing uncertainty set in robust optimization to deal with the temporal-spatial correlation in uncertain renewable energy generation and avoid over-conservative
- Proved that the above robust and stochastic optimization model under correlated uncertainty could be linearized by adding auxiliary variables and extra constraints
Energy-Sponge Electric Vehicle Sharing System Design
University of Wisconsin-Madison, May 2019 ~ Jan. 2020
Advisor: Prof. Xin Wang
Collaborator: Yikang Hua
- Established a profit-driven planning framework for electric vehicle sharing system to optimize its strategies in energy bidding, serving customers, charging, and relocation
- Implemented a two-stage stochastic model for electric vehicle sharing management incorporating the uncertainty of customer demand in spatiality, temporality, and quantity
- Built and solved the stochastic model via Sample Average Approximation method using Python and Gurobi; Conducted a case study in Austin to demonstrate the managerial insights
Design Optimization of Composite Wind Turbine Blades
Zhejiang University, Sept. 2018 ~ Apr. 2019
Advisor: Prof. Weifei Hu
- Modeled both the lightning strike dielectric breakdown failure and multi-axial fatigue failure mechanisms for the structural design of composite wind turbine blades
- Proposed a design optimization framework that integrates realistic lightning strike electrostatic and fatigue analyses for designing reliable and economical composite wind turbine blades
- Conducted a case study of the structural design optimization of a 5 MW composite wind turbine blade using the above optimization framework written by MATLAB
