A comparative experiment was carried out to discuss the detectability and influence factors on heterogeneous distributions of the stress field for underground panels in North China. The panels are situated in the center and the eastern margin of the North China plate, having similar layouts and stress states. Seismic surveys were performed before coal extractions, and travel-time topographies for refracted P-waves from stiff roofs were computed with a ray-based inversion. The tomographic distributions of P-wave velocity were interpreted by a K-means clustering to categorize the stress fields. The estimated high-stress anomalies show strong correlations with the monitored dynamic roof ruptures and high-energy microseisms. Besides, the relationship among high-stress areas, geological factors, and mined-out voids is also discussed through feature comparisons, Monte–Carlo simulations, and point process statistics. Our results show that the curvature variations and the mined-out voids are dominant factors that affect the heterogeneous distribution of the stress field in the panels. However, the leading causes of the curvature variations in the small-scale panels are different, but they do follow the large-scale regional patterns of the geological regime.

Inverse mapping from geophysics to geology is a difficult problem due to the inherent uncertainty of geophysical data and the spatially heterogeneous patterns (structure) in geology. We improve conditional facies simulation based on GANs (GANSim), a new Earth model simulation method based on generative adversarial networks (GANs), to bridge the gap between remotely sensed geophysical information and geology, by introducing a specially designed loss function and an input architecture for probability maps representing geophysical interpretation. After training, the GANSim is then used to produce multiple geological facies models conditioned to the input geophysics-interpreted probability maps alone or together with input well observations and global features. By evaluation, the generated facies models are realistic, diversified, and consistent with all input conditions. We demonstrate that the GAN learns the implicit geological pattern knowledge from the training data set and the knowledge of conditioning to inputs from human-defined explicit functions. Given the commonality of remotely sensed geophysical information, sparse measurements, and global features, GANSim should be applicable to many problems of geosciences.

Online solvers for partially observable Markov decision processes have difficulty scaling to problems with large action spaces. This paper proposes a method called PA-POMCPOW to sample a subset of the action space that provides varying mixtures of exploitation and exploration for inclusion in a search tree. The proposed method first evaluates the action space according to a score function that is a linear combination of expected reward and expected information gain. The actions with the highest score are then added to the search tree during tree expansion. Experiments show that PA-POMCPOW is able to outperform existing state-of-the-art solvers on problems with large discrete action spaces.

Online solvers for partially observable Markov decision processes have difficulty scaling to problems with large action spaces. Monte Carlo tree search with progressive widening attempts to improve scaling by sampling from the action space to construct a policy search tree. The performance of progressive widening search is dependent upon the action sampling policy, often requiring problem-specific samplers. In this work, we present a general method for efficient action sampling based on Bayesian optimization. The proposed method uses a Gaussian process to model a belief over the action-value function and selects the action that will maximize the expected improvement in the optimal action value. We implement the proposed approach in a new online tree search algorithm called Bayesian Optimized Monte Carlo Planning (BOMCP). Several experiments show that BOMCP is better able to scale to large action space POMDPs than existing state-of-the-art tree search solvers.

We present an integrated study in which we develop a range of techniques for deriving images of rock properties, such as porosity and shaliness, from cross-well seismic tomograms. One of the keys is to incorporate rock physics knowledge of the relations between velocity, porosity, and clay content which were developed in the laboratory and calibrated in the field. Geostatistical techniques, such as kriging and cokriging, are used as a means to combine the heterogeneous data set, consisting of well logs, the tomogram, and laboratory results.