This work develops and tests microseism event-classification techniques. Research was performed in collaboration with CREWES using microseismic data from Cold Lake, Alberta that was provided by Imperial Oil Ltd. The objective was to develop passiveseismic signal classification algorithms capable of precisely and automatically distinguishing between microseismic events warranting further investigation from noise events that are generally not of interest. Novel methods involving frequency-filtering, event-length detection, and statistical analysis were developed.
After extensive testing, it was found that developed statistical analysis algorithms performed best. Principal components analysis was applied to statistical analysis algorithm outputs to optimize classification.
A MATLAB implementation scheme was created that yielded classification accuracies between 90% and 99.5% when tested on a wide range of datasets. Given that up to tens of thousands of microseismic events are detected daily at Cold Lake, this work could have significant future impact.
View full article as PDF (0.49 Mb)