Research Of Low Frequency Modeling Method And Its Application In AVO Inversion

It is well known that the frequency component of seismic data is limited from severalto dozens Hz, lack both low and high frequency component. The high frequency componentis correlated to the resolution of seismic data, and the low frequency component iscorrelation to the precision of velocity analysis and also correlation to the quality ofmigration and imaging. Especially, the low frequency component will influence the result ofpre-stack or post-stack inversion, influence the defined description of oil field. So, the lowfrequency component of seismic data is also very important. This paper makes a study ofhow to compensate the low frequency component and method of building the low frequencymodel of AVO inversion.Although, the low frequency composition for seismic data is a problem that exist forquite a long time, but it doesn’t has been solved till now, and there is not that much researchabout how to compensate the low frequency component. For the reliability of compensatedlow frequency, it should originate from seismic data itself. Here, I introduced a methodoriginated from compressive sensing and constrained sparse pulse inversion for lowfrequency compensation, and tested it by applying it both to model and real seismic data,the result is very well. In addition, I also introduced a method of stabilizing low frequencycomponent by using a space filter in frequency domain, the test result showed that the lowfrequency component can be much stabilized by applying this approach, and it can alsomake the inversion result be better.Quantitative analysis will be unavailable by pre-stack AVO inversion without lowfrequency model, and we just can do qualitative analysis at that situation. Normally, the lowfrequency model for inversion is build from well logs using space interpolation method constrained by geology structure. I analyzed the modeling error of co-kriging interpolationmethod and sequential gaussian simulation interpolation method, find that the interpolationresult is not stable when there is just one constrain condition. So, I test a modeling methodusing multi attributes constraints, the result is much stable.Finally, I did inversion using no low frequency trend, low frequency model build bysequential gaussian simulation method, and model build by multi attributes method. Afteranalyzed the inversion result of the three models, I found that the inversion result is not verywell when there is no low frequency trend or the low frequency model is not that accurate,but the low frequency model build by multi attributes method worked well, the result ismuch better than the two models introduced before.