Artificial intelligence techniques in the analysis of digital seismograms

Abstract

Abstract The Seismic Network Analyzer project aims at improving by Artificial Intelligence techniques' the performance of automatic systems for seismogram analysis. It closely models the expert seismologist's knowledge and interpretative behavior. The system architecture is based on the blackboard model : several modules (the knowledge sources ) cooperate in the analysis opportunistically, with an activation sequence adapted at run-time to the specific characteristics of data. Key features are the system's ability to focus the attention on relevant parts of signals and to drive the analysis of weak signals by expectations based on clear onsets. The current prototype SNA2 is able to give correct interpretations in about 90% of situations, thanks to the cooperation of knowledge from independent domains, as are time-series analysis and seismology. SNA2 provides preliminary analysis of events recorded by microearthquake networks in all distance ranges, from local to teleseismic. Tests made with data collected by two networks show that the encoded knowledge is general enough to handle data from different settings.