Research On Earthquakes And Explosions Identification

This thesis reviews the studying history of earthquake and explosion identification, and presents a systematic summary about the criterion and methods for earthquake and explosion identification.In identification criterion study, a method is put forward to quantify the difference in instantaneous frequency between explosion and earthquake in time-frequency domain, and instantaneous frequency complexity is also presented as an indentificaion criterion. In comprehensive identification method study, a statistical pattern recognition algorithm-Stepwise Accumulating Minimal Cost-is put forward, and self-organizing competitive neural network is modified from non-supervision to under supervision.18earthquakes and26explosions in Beijing and Huailai county, Hebei are selected as the samples for examination. The major studying content is summarized as follows:1. Total5criteria were extracted from the collected waveform data to make earthquake and explosion identification, i.e., first motion direction, the ratio of P-wave first motion amplitude to S-wave maximum amplitude, the maximum amplitude ratio of P-wave to S-wave P-wave, instantaneous frequency complexity and P-wave spectral ratio. The difference of P-wave instantaneous frequency between earthquake and explosion is quantified, and a new criterion about instantaneous frequency complexity-IFC is obtained, which reflects the difference in P-wave instantaneous frequency variation between explosion and earthquake.2. A statistical pattern recognition algorithm-Step Accumulating Minimal Cost, i.e. SAMC-is put forward. By defining a certain cost function to achieve path optimization, this algorithm can obtain full cost function to distinguish earthquake and explosion. C test for SAMC on samples with the above-mentioned5extracted features shows that the recognition rate is95%. It shows that not only the cost function may well reflect the characterized trend, but also its absolute value can be used to judge the reliability of recognition. Furthermore, the cost function can be easily calculated, and can be applied in real time monitoring of small-size explositions..3. The self-organizing competitive neural network is developed from non-supervision pattern to under supervision, and a new algorithm SSOM is put forward on the base of SOM. Using samples with the extracted5features mentioned above, C test for self-organizing competitive neural network under supervision shows that the recognition rate is98%. Compare the test results between SOM and SSOM algorithm, it shows that SSOM can effectively remove "dead neurons" and improve the efficiency of classification.4. By comparing the recognition results from the new2algorithms, i.e., SAMC and SSOM with those from other3extensively used pattern recognition algorithms, i.e., Fisher, minimum distance and ICHAM, it shows that the recognition rate is100%for Fisher method,98%for the supervised self-organizing competitive neural network,95%for the recognition rate of accumulating minimal cost, which demonstrates that the2new algorithm, i..e., SAMC and SSOM, can be applied in the recognition of small explosion and small earthquake. The recognition rate is lower than82%for other2method, i.e., minimum distance method and ICHAM in this region.Earthquake and explosion recognition research is of great significance for seismic verification emergency, earthquake catalog editing and other relevant seismological research. A new identification criteria, instantaneous frequency complexity-IFC and two new comprehensive identification algorithms, i.e., Step Accumulating Minimal Cost-SAMC and self-organizing competitive neural network under supervision-SSOM are put forward, which can effectively distinguish explosions from earthquakes, and can be applied in earthquake and explosion identification.