A Heuristic-Learning Optimizer for Elastodynamic Waveform Inversion in Passive Seismics

• Published in: IEEE transactions on geoscience and remote sensing Vol. 57; no. 4; pp. 2234 - 2248
• Main Author: Vera Rodriguez, Ismael A.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01-04-2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Calibration; Design optimization; Earthquakes; Elastodynamics; Indexes; Inversion; Inversions; Iterative methods; Joint inversion; Locations (working); Mathematical model; Mathematical models; microseismic monitoring; Monitoring; Parameter estimation; Parameters; Particle swarm optimization; passive seismic; Receivers; Seismic activity; Seismic stability; Tensile stress; Tensors; Velocity; Viscoelasticity; Wave propagation; waveform inversion; Wavelet analysis; Workflow
• ISSN: 0196-2892; 1558-0644
• DOI: 10.1109/TGRS.2018.2872329

This paper explores (full-) waveform inversion in passive seismics to simultaneously optimize the source parameters of seismic events together with the properties of the medium of wave propagation. A heuristic optimization algorithm inspired in iterative design is proposed, which incorporates ideas from particle swarm optimization and variable projection. The algorithm is designed to accelerate convergence, improve stability and robustness, and minimize the number of user-defined parameters. The performance of the algorithm is illustrated with a real data example of hydraulic stimulation monitoring using a surface array of seismic receivers. In this example, only the locations of the receivers are known. The inversion is targeted to jointly optimize the parameters of a viscoelastic velocity model, associated models of receiver residual statics, wavelet signatures for direct compressional and shear arrivals, spatiotemporal locations of the input seismic events, and their corresponding moment tensors. Velocity model and locations are compared to estimations of the same parameters previously obtained through a well-established workflow using decoupled inversions. Results from both approaches show consistency. The main advantage of the joint approach is, therefore, the possibility of incorporating additional unknowns into the optimization to obtain self-consistent solutions using a single inversion process.