Search

Acquisition geometry design aims at finding the most affordable acquisition geometry that satisfies the objectives of the seismic survey. The parameters of an acquisition geometry can be specified in terms of the number of sources and detectors, their location, the blending parameters and the DSA (dispersed source array) parameters. In our acquisition geometry design, we include the effects of the (expected) subsurface, i.e., we assume the subsurface to be known. Consequently, the ideal data set – carpet shooting and carpet detection – can be modeled. A practical data set can be considered to be a subset of this ideal one. Acquisition design comes down to determining the optimum subset. Following compressive sensing, this subset is sparse and irregular. As a quality measure, we apply decompression (deblending and interpolation) to the subset, which leads to an estimate of the ideal data set, and then compare this estimate with the known ideal data set. The difference is the residue that should satisfy a predefined quality criterion. This procedure is the inner loop of a genetic algorithm. A CNN (convolutional neural network) is trained to improve the efficiency of the genetic algorithm by enhancing the effectiveness of each next generation. Furthermore, the solution space is limited to reduce the amount of computations needed. Finally, in this application it is acceptable to end up in a local minimum. The latter corresponds to an acquisition geometry that fully satisfies the quality and economic criteria (although some acquisition geometry may exist that provides even better results). Our design method leads to results that are better than those obtained with randomized acquisition geometries., Accepted Author Manuscript, Applied Geophysics and Petrophysics

Acquisition geometry design aims at finding the most affordable acquisition geometry that satisfies the objectives of the seismic survey. The parameters of an acquisition geometry can be specified in terms of the number of sources and detectors, their location, the blending parameters and the DSA (dispersed source array) parameters. In our acquisition geometry design, we include the effects of the (expected) subsurface, i.e., we assume the subsurface to be known. Consequently, the ideal data set – carpet shooting and carpet detection – can be modeled. A practical data set can be considered to be a subset of this ideal one. Acquisition design comes down to determining the optimum subset. Following compressive sensing, this subset is sparse and irregular. As a quality measure, we apply decompression (deblending and interpolation) to the subset, which leads to an estimate of the ideal data set, and then compare this estimate with the known ideal data set. The difference is the residue that should satisfy a predefined quality criterion. This procedure is the inner loop of a genetic algorithm. A CNN (convolutional neural network) is trained to improve the efficiency of the genetic algorithm by enhancing the effectiveness of each next generation. Furthermore, the solution space is limited to reduce the amount of computations needed. Finally, in this application it is acceptable to end up in a local minimum. The latter corresponds to an acquisition geometry that fully satisfies the quality and economic criteria (although some acquisition geometry may exist that provides even better results). Our design method leads to results that are better than those obtained with randomized acquisition geometries., Accepted Author Manuscript, Applied Geophysics and Petrophysics