GENETIC ALGORITHM (GA) MACHINE LEARNING OF SEISMIC WAVEFORM CHARACTERISTICS

THE SBIR PHASE I PROJECT PROPOSED CONSISTS OF TESTING THE APPLICATION OF GENETICS-BASED MACHINE LEARNING METHODS (GBML) TO TWO PROBLEMS; SEISMIC WAVEFORM CHARACTERISTICS IN THE CONTEXT OF SEISMIC DISCRIMINATION AND IN IDENTIFICATION OF SEISMIC ARRIVALS. OTHER APPLICATIONS OF THESE METHODS MAY ALSO BE EXPLORED IN THE PROPOSED WORK. IN THE APPLICATION TO DISCRIMINATION, THE FOLLOWING DATA SETS WILL BE USED: REGIONAL ARRAY DATA FOR EARTHQUAKES AND QUARRY BLASTS IN THE SCANDINAVIAN REGIONS, IRIS DATA RECORDED IN THE USSR AND SELECTED DATA SETS USED PREVIOUSLY TO DEDUCE DISCRIMINATES. THE PERFORMANCE OF THESE TECHNIQUES WILL BE EVALUATED BY VARIOUS DATA PARTITIONING AND RESAMPLING METHODS. THE OBJECTIVE OF THIS RESEARCH IS TO MAKE USE OF THE EXTENSIVE PARAMETRIC DATA GENERATED BY THE IMS. BY GENERATING NEW RULES FROM THE DATA BY THE MACHINE LEARNING ALGORITHMS AND INCLUDING THEM IN THE SYSTEM THE COGNITIVE CAPABILITIES OF THE IMS CAN BE CONTINUOUSLY UPGRADED. ANTICIPATED BENEFITS/POTENTIAL APPLICATIONS - THE AUTOMATIC DISCRIMINATION SCHEMES TO BE TESTED AND DEVELOPED UNDER THIS PROJECT CAN BE INCORPORATED INTO THE IMS AND UPGRADED CONTINUOUSLY AS NEW DATA BECOMES AVAILABLE. THIS WOULD LEAD TO THE ANTICIPATED BENEFIT OF CONTINUOUS IMPROVEMENT OF DISCRIMINATION CAPABILITY OF IMS. THE ALGORITHMS CAN ALSO BE USED IN NUMEROUS OTHER APPLICATIONS IN SEISMOLOGY AND OTHER FIELDS.