1. Student and Project Details
| Name |
Vishesh Mittal |
| College |
Indian Institute of Information Technology, Allahabad, India |
| Programme |
NTU-India Connect Research Programme 2022 (Summer Term) |
| Department |
School of Physical and Mathematical Sciences (SPMS), NTU |
| Research Title |
Time Series Analysis With Deep Learning Techniques and Beyond |
| Supervisor |
Asst Prof Tong Ping |
| Assistant Supervisors |
Li Tianjue, Wu Shucheng |
2. Study Background
Time series analysis has wide applications in many fields such as economics, finance, computer sciences, engineering, and earth sciences, where understanding patterns and trends is important to decision making and predicting future behaviours. For example, the ground motion caused by an earthquake is recorded as a course of time. Time-series data are always noisy and high-dimensional. There features increase the difficulty of deep and precise analysis. Recent development in deep learning provides some new techniques for efficient time-series analysis. In this project, we aim to create new models of unsupervised learning of features for time series analysis and prediction. Applications into seismic data for earthquake identification & prediction and for petroleum exploration will be explored.
3. Study Region
- For this project, we considered South-East Asia as our study region, primarily the area enclosed by the coordinate pairs (30 N, 90 E), (30 N, 132 E), (-12 N, 90 E) and (-12 N, 132 E).
- This study region was further sub-divided into 9 study regions, out of which the top-left region was allotted to me.
- My area of coverage can be defined using the following 4 coordinates, which can also be seen in the below image:
- 16.00 N, 90.00 E (Bay of Bengal)
- 30.00 N, 90.00 E (Nagqu, Tibet, China)
- 16.00 N, 104.00 E (Roi Et, Thailand)
- 30.00 N, 104.00 E (Sichuan, China)
Geographical Coverage of my Data Analysis
4. Data Types
- The project considered 2 different data-types, NSFE (Near Station Far Event) and NEFS (Near Event Far Station). For both of the data-types, data was collected with the aid of ObsPy package, via IRIS and GFZ data providers.
- NSFE - In NSFE, the stations are within the study region, and the events are 25° to 95° away from the study site. For NSFE, the data was downloaded for the period (01/01/2000 - 31/12/2021).
- NEFS - • In NEFS, the events are within the study region, and the stations are 25° to 95° away from the study site. For NSFE, the data was downloaded for the period (01/01/2000 - 31/12/2018).
Note - For keeping the report concise, in all the further sections, I have presented the findings for only a single year from both the data-types, NSFE and NEFS. For NSFE, I have considered the period (01/01/2021 - 31/12/2021) and for NEFS, I have considered the period (01/01/2018 - 31/12/2018). Furthermore, since each of the years, contain data for numerous events, I have only considered a small number of events from both of the years.
5. Distribution of the Downloaded Data
- The below plot illustrates the data distribution for the NSFE data for the period (01/01/2021 - 31/12/2021). The plot shows the data collected from both the data providers, i.e., IRIS and GFZ.
- As can be seen in the below plot, the stations are within the boundaries for my study region (outside plot), and the events are spread out in a much wider region (inside plot), 25° to 95° away from the study site, to be more specific.