Abstract：A nonlinear seisimc convolution model is constructed by performing the multistage decomposition and multistage nonlinear transform of seismic wavelets.Combining this model with the neural network that is based on neuron F-P function model forms a ANNLOG technique for high-resolution wave impedance inversion under the constraints of logging and stratigraphic data.This technique involves following essential points: ·Multistage nonlinear transform causes fast iterative convergence and very high resolution in vertical direction. ·According to the lateral dynamic characteristic variation of seismic data,the neural networks storing multistage seismic wavelets(F-P model) enable reliable adaptive extrapolation inversion,and make continuous vertical resolution be stable in azimuthal direction. ·Stratigraphic restrained inversion makes ANNLOG technique suitable to complex geologic structures such as big throw faults,pinchouts and so on.
