I am a joint PhD student in Environmental Engineering at City University of Hong Kong and Tongji University, supervised by Prof. Zhiguo Yuan @CityUHK and Prof. Zhenliang Liao @Tongji. I’m currently a member of the Smart Water and EnviroBiotech Team (SWEB) at CityUHK. I received my B.Eng. in Environmental Engineering from Tongji University in 2020. My research interests lie on smart urban drainage networks, specifically in modelling and real-time control with learning-based methods.
Derives the gradients of the MPC objective function (output) to the control action (input) along the neural network for a L-BFGS-B optimizer.
Control-oriented real-time modelling of drainage flow routing with a spatio-temporal GNN model with physics-guided hydraulic regulations.
Using multi-agent reinforcement learning to control multiple drainage valves and pumps with real-time communication fail-safe performance.
Compare rule-based control, model predictive control, and reinforcement learning control in computing load and normalized performance.
Assess both risk and resilience with 1D2D modelling and threshold classifications.
Detect contamination timely and accurately from monitoring data considering water quality baseline and possible anomalies.
Avoid local optima problem and ensure effcient locatilization of contaminated sources in WDS combining SA and PSO.
Pretrain RL agent for RTC with knowledge-carrying data from engineering experiences and strategies.
Use the derivative of UDS resilience indicator to train a deep reinforcement learning agent.
Interpret the learned control policy of DRL with Sobol, tree-based logics, and conditional distributions.
Training a Koopman emulator of urban drainage dynamis for reinforcement learning control.
RL agents' decisions are action votes and further evaluated in real time to improve reliability and safety.
Safe learning control of urban drainage systems with a robust reward function.
Learning a 3D hydrodynamic lake model from monitoring data using Koopman operator and transfer learning.
Use spatial and temporal attention mechanisms to enhance CNN-LSTM models for predicting surface water quality.
Use transformer networks and loss with punishment to optimize management of urban water supply system.