Education
- B.S. in Physics, Zhejiang University, China, 2011-2015
- M.S. in Data Analytics, University of Warwick, UK, 2017-2018
- Ph.D in Computer Science, University of Warwick, UK, 2018-2022 (expected)
RESEARCH PROJECTS
- Data-driven Analysis of Users’ Abnormal Gas Consumption Behavior
- Duration: 2021.04-2021.12
- Funded by: Shenzhen Gas Corporation
- My role: major participant
- Achievement: We have proposed a data-driven unsupervised method to perform anomaly detection on gas consumption data of gas users. The method combines a rule-based model with a deep learning based model to detect various abnormal conditions including abnormal meter readings and abnormal gas usage patterns. The project is carried out on a large amount of real data, and the accuracy of our detection results has been confirmed through expert experience and on-site investigation. A visual anomaly detection system is also built and deployed in Shenzhen Gas Company for daily use.
- Water Contamination Detection and Identification Algorithm Based on Intelligent Sensor Network
- Duration: 2020.06-2022.06
- Funded by: Science and Technology Innovation Commission of Shenzhen
- My role: major participant
- Achievement: We have proposed a solution for cross-network contamination source identification (CSI) that can transfer the CSI knowledge learned from one water distribution network (WDN) to a different WDN. This greatly increases the usability of deep learning methods in water contamination detection and identification. Based on the structure of graph neural network (GCN) and the idea of transfer learning, we designed a model that can capture the spatial correlation of small areas in the network topology and the temporal correlation of water quality sensors, so as to realize the detection and identification of water contamination across WDNs
- Development and Application of Big Data-based Platform for Intelligent Prevention and Control of Urban Public Security Risks
- Duration: 2020.01-2021.12
- Funded by: National Key R&D Program of China
- My role: major participant
- Achievement: A large part of urban big data comes from various sensor data. Due to the problems of data collection, transmission and storage, missing values are a common problem in time series sensor data. We have proposed a method for time series data imputation to provide more accurate estimates for missing values in time series. It combines the time series decomposition method Seasonal Trend decomposition using Loess (STL) with recurrent neural networks (RNNs) to capture temporal dynamics from a long-term and short-term perspective. Our method has proved its effectiveness on multiple open datasets, including air quality data, water pressure data, and gas consumption data.
- Detection of Conflict Dissolution and Tolerant Intrusion for Information Security and Functional Safety in Industrial Cyber-Physical Systems
- Duration: 2019.01-2022.12
- Funded by: National Natural Science Foundation of China
- My role: major participant
- Achievement: In the data isolated distributed deep learning scenario, the distribution of training data used by each node could be highly imbalanced because of data isolation. This brings difficulty to the normalization process in neural network training, because the traditional batch normalization (BN) method will fail under this kind of data imbalanced scenario. Aiming at such data isolation scenarios, we have proposed a comprehensive data isolation deep learning scheme. Specifically, we use synchronous stochastic gradient descent algorithm for data exchange during training, and provides several normalization approaches to the problem of BN failure caused by data imbalance.
