recently, the team of Zhang Kai has made new progress in the field of history matching, and the related research has been published in the Journal of China University of Petroleum(Edition of Natural Sciences). The paper is titled " History matching method based on ensemble and neural architecture search ".
Innovation: In this paper, a novel PSO algorithm is proposed to automatically discover the optimal network architecture of the deep autoencoder for image classification problems without manual intervention. Meanwhile, in order to reduce the time spent in the optimization process, the strategy that the neural network is only trained in the early stage of optimization is adopted. The autoencoder of the optimal network architecture can reduce more dimensions of data and occupy less computing resources while ensuring the accuracy of dimensionality reduction and reconstruction. DAE, an optimal architecture, is used to parameterize geological models, and ES-MDA is used for automatic reservoir history fitting. In terms of geological model reconstruction, the DAE model optimized by the proposed method is obviously better than the parameterized results of the DAE model before optimization. The inversion results obtained by automatic historical fitting are significantly improved, and the permeability field obtained by inversion is closer to the real permeability field.
Abstract: Due to the limitations of manual experience selection, it is difficult to obtain the optimal network parameters that determine the accuracy of model reconstruction is currently one of the difficulties when using deep learning methods to reduce the dimensionality of reservoir geological models in automatic history matching. In response to this, the automatic search of the best network architecture was realized by combining the deep autoencoder and the particle swarm optimization algorithm, and based on this, an automatic reservoir history matching method based on aggregate data assimilation and neural network architecture automatic search was constructed. A two-dimensional fluvial reservoir permeability field distribution model and a SPE-10 single-layer reservoir numerical model were used to verify the comparison experiment with a single automatic history matching method and the proposed method. The comparison experiment results show that the automatic history matching method which automatically searches optimal neural network framework after optimization could extract the geological characteristics of the reservoir numerical model more accurately than the single automatic history matching method and same method before optimization.
The Journal of China University of Petroleum(Edition of Natural Sciences), founded in 1959, is a comprehensive academic journal sponsored by China University of Petroleum and supervised by the Ministry of Education of the People's Republic of China. It mainly publishes innovative achievements and high-level papers in the fields of oil and gas geology and exploration engineering, oil and gas drilling and production engineering, oil and gas storage and transportation and mechanical engineering, oil and gas chemical engineering, and basic and applied research of interdisciplinary disciplines.
Citation:
Zhang L M, Chen X S, Li G X, et al. History matching method based on ensemble and neural architecture search[J]. Journal of China University of Petroleum(Edition of Natural Sciences), 2022.
