Time Series Data Mining Techniques for Seismic Data Analysis

We applied advanced data mining techniques for seismic data processing, developing seismic signal discovery, classification and monitoring algorithms. In signal discovery we developed a semi-supervised motif discovery algorithm that forms a nearest neighbor graph to identify chains of nearest neighbors from the given events and demonstrated that the chains are likely to identify hidden patterns in the data. In signal classification we developed a machine-learned model that takes input from a signal detector and produces phase types as output for a signal associator. The model is a combination of convolutional and long short-term memory networks. It outperforms existing baselines and has consistently good performance for novel sources and stations. In signal monitoring we developed a technique that efficiently computes both filtering and correlation in a single step. We achieved an order of magnitude speed-up by maintaining frequency transforms over sliding windows. The method is exact, devoid of sensitive parameters, and easily parallelizable. We have provided a publicly available real-time system that employs the algorithm for monitoring a seismic network.