Deep learning network optimization and hyperparameter tuning for seismic lithofacies classification

Deep learning is increasingly being applied in many aspects of seismic processing and interpretation. Here, we look at a deep convolutional neural network approach to multiclass seismic lithofacies characterization using well logs and seismic data. In particular, we focus on network performance and hyperparameter tuning. Several hyperparameter tuning approaches are compared, including true and directed random search methods such as very fast simulated annealing and Bayesian hyperparameter optimization. The results show that improvements in predictive capability are possible by using automatic optimization compared with manual parameter selection. In addition to evaluating the prediction accuracy's sensitivity to hyperparameters, we test various types of data representations. The choice of input seismic data can significantly impact the overall accuracy and computation speed of the optimized networks for the classification challenge under consideration. This is validated on a 3D synthetic seismic lithofacies example with acoustic and lithologic properties based on real well data and structure from an onshore oil field.

Optimizing machine learning models for granular NdFeB magnets by very fast simulated annealing

The macroscopic properties of permanent magnets and the resultant performance required for real implementations are determined by the magnets’ microscopic features. However, earlier micromagnetic simulations and experimental studies required relatively a lot of work to gain any complete and comprehensive understanding of the relationships between magnets’ macroscopic properties and their microstructures. Here, by means of supervised learning, we predict reliable values of coercivity (μ0Hc) and maximum magnetic energy product (BHmax) of granular NdFeB magnets according to their microstructural attributes (e.g. inter-grain decoupling, average grain size, and misalignment of easy axes) based on numerical datasets obtained from micromagnetic simulations. We conducted several tests of a variety of supervised machine learning (ML) models including kernel ridge regression (KRR), support vector regression (SVR), and artificial neural network (ANN) regression. The hyper-parameters of these models were optimized by a very fast simulated annealing (VFSA) algorithm with an adaptive cooling schedule. In our datasets of randomly generated 1,000 polycrystalline NdFeB cuboids with different microstructural attributes, all of the models yielded similar results in predicting both μ0Hc and BHmax. Furthermore, some outliers, which deteriorated the normality of residuals in the prediction of BHmax, were detected and further analyzed. Based on all of our results, we can conclude that our ML approach combined with micromagnetic simulations provides a robust framework for optimal design of microstructures for high-performance NdFeB magnets.

Simulated annealing velocity analysis: automating the picking process

We propose an automated method for velocity picking that allows to estimate appropriate velocity functions for the normal moveout (NMO) correction of common depth point (CDP) gathers, valid for either hyperbolic or nonhyperbolic trajectories. In the hyperbolic velocity analysis case the process involves the simultaneous search (picking) of a certain number of time-velocity pairs where the semblance, or any other coherence measure, is high. In the nonhyperbolic velocity analysis case, a third parameter, usually associated with the layering and/or the anisotropy, is added to the searching process. The proposed technique relies on a simple but effective search of a piecewise linear curve defined by a certain number of nodes in a 2D or 3D space that follows the semblance maxima. The search is carried out efficiently using a constrained very fast simulated annealing algorithm. The constraints consist of static and dynamic bounding restrictions, which are viewed as a means to incorporate prior information about the picking process. This allows to avoid those maxima that correspond to multiples, spurious, and other meaningless events. Results using synthetic and field data show that the proposed technique permits to automatically obtain accurate and consistent velocity picks that lead to flattened events, in agreement with the manual picks. As an algorithm, the method is very flexible to accommodate additional constraints (e.g. preselected events) and depends on a limited number of parameters. These parameters are easily tuned according to data requirements, available prior information, and the user's needs. The computational costs are relatively low, ranging from a fraction of a second to, at most, 1-2 seconds per CDP gather, using a standard PC with a single processor.

Simulated Annealing, Differential Evolution and Directed Search Methods for Generator Maintenance Scheduling

Generator maintenance scheduling presents many engineering issues that provide power system personnel with a variety of challenges, and one can hardly afford to neglect these engineering issues in the future. Additionally, there is vital need for further development of the repair planning task complexity in order to take into account the vast majority of power flow constraints. At present, the question still remains as to which approach is the simplest and most effective, as well as appropriate for further application in the power flow-oriented statement of the repair planning problem. This research compared directed search, differential evolution, and very fast simulated annealing methods based on a number of numerical calculations and made conclusions about their prospective utilization in terms of a more complicated mathematical formulation of the repair planning task. A comparison of results shows that the effectiveness of directed search methods should not be underestimated, and that the pure differential evolution and very fast simulated annealing approaches are not essentially reliable for repair planning. The experimental results demonstrate the perspectivity of unifying single-procedure methods in order to net out risk associated with specific features of these approaches.

Inversion of downhole electrical measurements for proppant mapping using very fast simulated annealing

Proppant mapping is critical for optimizing fracture treatment design and improving wells’ productivity. An electrode-based resistivity tool concept was developed earlier for proppant mapping in cased-hole wells. An array of insulating gaps is installed and cemented in place as a permanent part of the casing string. The electrical measurements are performed by imposing a voltage across each insulating gap, one at a time, before and after hydraulic fracture operations. The voltages across other insulating gaps near the transmitter gap are recorded. The method relies on direct excitation of the casing, which is expected to overcome the severe limitations of induction tools in cased-hole wells. A forward model based on a finite volume method has been developed to simulate the tool’s response to one or multiple fractures. To enable the implementation of such a practical system in multistage fractured horizontal wells, a fast and robust inversion approach is required. To that end, we have developed a divide-and-conquer approach based on a global optimization algorithm very fast simulated annealing (VFSA). Specifically, the original inverse problem is divided into subproblems and each subproblem can be solved separately using VFSA. The results indicate that our approach can invert the data and output widths and radii of multiple fractures without requiring a large number of forward simulations. The robustness of the inverse solver is also tested by adding Gaussian noise to the synthetic data. We tested example cases that demonstrate that when up to 5% noise is introduced, VFSA still provides very accurate inversion results with moderate uncertainties. Inversion results with some more realistic conditions, e.g., tilted fractures, complex fractures, and so on, are also presented.

Using Very Fast Simulated Annealing in the Weighted AB Semblance method

ABSTRACTThe conventional velocity analysis does not consider AVO effects in reflection seismic data. These conditions lead to obtaining of inadequate velocity fields, making it difficult to execute other steps in seismic processing. To overcome this problem, researchers developed the Weighted AB semblance method, a coherence measure which deals with AVO effects in velocity spectra. It is based on the application of two sigmoid weighting functions to AB semblance, which depend on four coefficients. The values of these coefficients directly influence the resolution of the resulting velocity spectrum. In this work, we apply the inversion algorithm Very Fast Simulated Annealing (VFSA) to obtain these values. Numerical experiments show that VFSA is a quite effective method, obtaining correct coefficient values and allowing the generation of the velocity spectrum with an excellent resolution for both synthetic and real data. Results also proved that Weighted AB semblance is an optimal coherence measure to be used in velocity spectrum, because it is insensitive to AVO effects and reversal polarity and presents considerably a better resolution than conventional semblance.Keywords: velocity analysis, AVO, high-resolution velocity spectra RESUMOA análise de velocidades convencional não considera efeitos de AVO em dados sísmicos de reflexão. Essas condições levam à obtenção de campos de velocidades inadequados, dificultando a execução de outras etapas do processamento sísmico. Para superar esse problema, pesquisadores desenvolveram o método AB semblance Ponderado, uma medida de coerência que lida com efeitos de AVO em espectros de velocidades. Ela ´e baseada na aplicação de duas funções sigmoides à AB semblance, que depende de quatro coeficientes. Os valores desses coeficientes influenciam diretamente a resolução do espectro de velocidade resultante. Nesse trabalho, n´os aplicamos o algoritmo de inversão Very Fast Simulated Annealing (VFSA) para obter esses valores. Experimentos numéricos mostram que VFSA é um método bastante eficaz, obtendo valores corretos dos coeficientes e permitindo a geração do espectro de velocidade com uma excelente resolução tanto para dados sintéticos quanto para dados reais. Resultados também provam que o AB semblance Ponderado ´e uma medida de coerência ótima para ser usada no espectro de velocidade, porque ela é insensível aos efeitos de AVO e apresenta resolução consideravelmente melhor do que a semblance convencional.Palavras-chave: análise de velocidades, AVO, espectro de velocidades de alta resolução.