SEISMIC EVENT IDENTIFICATION AND LOCATION USING NEURAL NETWORKS

HNC, INC. PROPOSES TO EVALUATE THE CLASSIFICATION CAPABILITIES OF NEURAL NETWORKS TO REAL SEISMIC DATA, SPECIFICALLY, TO DETERMINE WHETHER OR NOT NEURAL NETWORKS CAN IMPROVE UPON THE CURRENT CAPABILITY (USING STANDARD SEISMOLOGICAL AND STATISTICAL METHODS) TO IDENTIFY REGIONAL SEISMIC EVENTS. THE EMPHASIS IS ON AUTOMATED SIGNAL PROCESSING TO MAKE TRACTABLE THE ROUTINE CLASSIFICATION OF LARGE NUMBERS OF SEISMIC EVENTS. THE ULTIMATE PROJECT OBJECTIVE IS THE POSITIVE IDENTIFICATION AND LOCATION OF NUCLEAR TESTS FROM WITHIN A LARGE NUMBER OF EARTHQUAKES AND PEACEFUL (CHEMICAL) EXPLOSIONS. SCIENCE APPLICATIONS INTERNATIONAL CORPORATION (SAIC) WILL SUPPORT HNC BY PROVIDING REAL DATA AND ASSISTING WITH PROBLEM DEFINITION. AS A LARGE TRAINING SET OF SEISMIC RECORDINGS OF NUCLEAR TESTS ARE NOT AVAILABLE FOR THE GEOGRAPHIC REGIONS OF INTEREST, THE NETWORK WILL NOT BE ASKED TO IDENTIFY AN EVENT AS A NUCLEAR TEST BASED ON PAST EXPERIENCE. RATHER, THE NEURAL NETWORK WILL BE ASKED TO IDENTIFY EVENTS THAT ARE "DIFFERENT" FROM THOSE IT HAS EXPERIENCE WITH. THAT IS, IT SHOULD FIND EVENTS WITH A LOW PROBABILITY OF BEING IN ANY OF THE OUTPUT CLASSIFICATIONS OF THE NETWORK.