Microseismic source location based on full waveform inversion driven neural network

Accurate localization of microseismic sources is essential in fields such as oil and gas extraction and underground energy storage. Current seismic source localization methods based on full waveform inversion exhibit a high degree of nonlinearity and involve complex gradient calculations for the objective function. However, data-driven neural network microseismic source localization methods lack physical constraints, which can compromise geological validity. To address these challenges, this paper proposes a microseismic source localization method that integrates full waveform inversion with a recurrent neural network. First, the seismic wavefield propagation operator is designed using convolutional kernels to achieve networked microseismic forward modeling. Next, chain differentiation of the neural network is employed to calculate the gradient for full waveform inversion in reverse, improving computational efficiency. Finally, by minimizing the error between the observed and forward-modeled data, the spatial components of the seismic source are optimized, and non-maximum suppression is applied to obtain the spatial location of the seismic source. The experimental results reveal that the proposed method achieves high localization accuracy, high computational efficiency, and resistance to noise.