Hierarchical Reservoir Research Articles

DHESN: A deep hierarchical echo state network approach for algal bloom prediction

A deep hierarchical echo state network (DHESN) is designed for rectifying the shortcomings of the shallow coupled structure with less reservoir dynamics. This design is with reference to algal bloom which is a complex ecological phenomenon. Accurate prediction of algal bloom can reduce the environmental impact and economic loss. Since the formation of algal bloom has chaotic characteristics, the ESN has been employed to realize its prediction function. First, the candidate variables with strong causal relationship have been screened by transfer entropy, and the redundant variables is eliminated. Then, a hierarchical reservoir structure is established that is inspired by the hierarchical characteristics from the brain. The hierarchical reservoir has realized the connection between the representative nodes of each subreservoir, and improved the information processing ability of the reservoir. Finally, the pruning and compression of the output weights have been realized by the elastic regularization method, which improves the robustness of the prediction model. The simulation results demonstrate that the DHESN has appreciable prediction accuracy in both the chaotic and the public algal bloom datasets. The DHESN contains richer dynamic characteristics, and can realize the self-organization of the network structure. It provides a novel idea to realize the prediction model of algal bloom with a high accuracy and low complexity.

Hierarchical-Task Reservoir for Online Semantic Analysis From Continuous Speech.

In this article, we propose a novel architecture called hierarchical-task reservoir (HTR) suitable for real-time applications for which different levels of abstraction are available. We apply it to semantic role labeling (SRL) based on continuous speech recognition. Taking inspiration from the brain, this demonstrates the hierarchies of representations from perceptive to integrative areas, and we consider a hierarchy of four subtasks with increasing levels of abstraction (phone, word, part-of-speech (POS), and semantic role tags). These tasks are progressively learned by the layers of the HTR architecture. Interestingly, quantitative and qualitative results show that the hierarchical-task approach provides an advantage to improve the prediction. In particular, the qualitative results show that a shallow or a hierarchical reservoir, considered as baselines, does not produce estimations as good as the HTR model would. Moreover, we show that it is possible to further improve the accuracy of the model by designing skip connections and by considering word embedding (WE) in the internal representations. Overall, the HTR outperformed the other state-of-the-art reservoir-based approaches and it resulted in extremely efficient with respect to typical recurrent neural networks (RNNs) in deep learning (DL) [e.g., long short term memory (LSTMs)]. The HTR architecture is proposed as a step toward the modeling of online and hierarchical processes at work in the brain during language comprehension.

Functional connectivity analysis on hierarchical reservoir computing model

Functional connectivity analysis on hierarchical reservoir computing model

Hierarchical reservoir lithofacies and acoustic impedance simulation: Application to an oil field in SW of Iran

In this study, a hierarchical simulation algorithm was implemented to produce 3D models of lithofacies and acoustic impedance (AI) in an oil field in southwestern Iran. Hierarchical simulation is a powerful method that takes into account the spatial distribution and hard contact relationships of continuous variables within the categorical variables in reservoir characterization. To implement the hierarchical simulation algorithm in this study, pluri-Gaussian simulation was used to construct the layouts of one categorical variable (lithofacies domains) and turning bands simulation was used to produce one continuous variable (AI models) within each simulated domain. The dataset that is available to test the proposed method includes the lithofacies and AI logs of seven wells. For validation of the proposed model, one of wells was excluded from the initial dataset to check the accuracy of results. The algorithm was implemented on a gridded model to produce 50 realizations of 3D lithofacies and AI models. The statistical analyses of the produced models verified the capabilities of hierarchical simulation in reproducing the lithofacies domains and AI values related to each domain. The simulated AI models were compared with the seismic inversion model to corroborate the reliability of the results. In addition, to show the superiority of the hierarchical approach, the results were compared to the results of direct AI modeling in which the AI models were generated without considering any lithofacies data. The comparisons in the blind well indicated a high correlation between the modeled values and the true AI. The correlation coefficient between average AI models of the hierarchical approach and the true AI was 0.85. Whereas, this value was 0.75 for the average of directly simulated AI models and 0.74 for the seismic inversion, respectively. Histogram analysis also verified the capability of the hierarchical modeling in reproducing the statistical properties of the original data. The resulting AI model obtained using the method proposed in this study also showed a high compatibility with the deterministic seismic inversion cube.

Hierarchical architectures in reservoir computing systems

Reservoir computing (RC) offers efficient temporal data processing with a low training cost by separating recurrent neural networks into a fixed network with recurrent connections and a trainable linear network. The quality of the fixed network, called reservoir, is the most important factor that determines the performance of the RC system. In this paper, we investigate the influence of the hierarchical reservoir structure on the properties of the reservoir and the performance of the RC system. Analogous to deep neural networks, stacking sub-reservoirs in series is an efficient way to enhance the nonlinearity of data transformation to high-dimensional space and expand the diversity of temporal information captured by the reservoir. These deep reservoir systems offer better performance when compared to simply increasing the size of the reservoir or the number of sub-reservoirs. Low frequency components are mainly captured by the sub-reservoirs in later stage of the deep reservoir structure, similar to observations that more abstract information can be extracted by layers in the late stage of deep neural networks. When the total size of the reservoir is fixed, tradeoff between the number of sub-reservoirs and the size of each sub-reservoir needs to be carefully considered, due to the degraded ability of individual sub-reservoirs at small sizes. Improved performance of the deep reservoir structure alleviates the difficulty of implementing the RC system on hardware systems.

DeePr-ESN: A deep projection-encoding echo-state network

As a recurrent neural network that requires no training, the reservoir computing (RC) model has attracted widespread attention in the last decade, especially in the context of time series prediction. However, most time series have a multiscale structure, which a single-hidden-layer RC model may have difficulty capturing. In this paper, we propose a novel multiple projection-encoding hierarchical reservoir computing framework called Deep Projection-encoding Echo State Network (DeePr-ESN). The most distinctive feature of our model is its ability to learn multiscale dynamics through stacked ESNs, connected via subspace projections. Specifically, when an input time series is projected into the high-dimensional echo-state space of a reservoir, a subsequent encoding layer (e.g., an autoencoder or PCA) projects the echo-state representations into a lower-dimensional feature space. These representations are the principal components of the echo-state representations, which removes the high frequency components of the representations. These can then be processed by another ESN through random connections. By using projection layers and encoding layers alternately, our DeePr-ESN can provide much more robust generalization performance than previous methods, and also fully takes advantage of the temporal kernel property of ESNs to encode the multiscale dynamics of time series. In our experiments, the DeePr-ESNs outperform both standard ESNs and existing hierarchical reservoir computing models on some artificial and real-world time series prediction tasks.

Reservoir characteristics and three-dimensional architectural structure of a complex fault-block reservoir, beach area, China

The objective area in this paper is a complex fault-block reservoir, which is provided by China Petroleum Engineering Design Competition. Reservoir characteristics which comprise of stratigraphic features, vertical changes, and section properties (thickness, percentage sand, and percentage amalgamation) are documented. Through comprehensive analysis on structural architecture and reservoir characteristics, a three-dimensional quantitative reservoir modeling is taken at a regional scale of 69 × 97 × 37 m3 with the application of geostatistics as theoretical guidance. Then, a high-resolution hierarchical reservoir model of this field has been generated with a combination of hierarchical, structural, physical, and well trajectory data. The established three-dimensional geological model integrates all well and structural information, which can provide a basic model for subsequent sedimentary microfacies modeling and property modeling as well. Finally, a series of profiles are built successively, they are three-dimensional fence diagrams, connecting well sections and well group profiles, which can be a valuable tool for reflecting geologic body with great reality and thus achieve reservoir characterization of the complex fault-block reservoir.

Reservoir Characteristics and Three-dimensional Architectural Structure of a Complex Fault-block Reservoir, Beach Area, China

The objective area is a complex fault-block reservoir, and is provided by China Petroleum Engineering Design Competition(CPEDC). Reservoir characteristics, including stratigraphic features, vertical changes and section properties (thickness, percentage sand and percentage amalgamation) are documented. Based on comprehensive analysis on structural architecture and reservoir characteristics, a three-dimensional quantitative reservoir modeling is taken at a regional-scale of 69×97×37 m3 with the application of geostatistics as theoretical guidance, and thus high-resolution hierarchical reservoir model is generated with a combination of hierarchical data, structural data, physical data and well trajectory data. Therefore, the established three-dimensional geological model integrates all well-point information and structural information, and provides a basic model for subsequent sedimentary microfacies modeling and property modeling as well. Finally, three-dimensional fence diagrams, connecting-well sections and well group profiles are built successively.

The reservoir characterization of the Sea Lion Field

The reservoirs that form the Sea Lion Field comprise a series of canyon-fed fans deposited into a deep, anoxic lake, on the hanging wall of the basin boundary fault. The fans primarily form stratigraphic traps with an element of fault seal on the west flank. These fans are dominated by mass flow and high-density turbidite sands, with subordinate low-density turbidites set within lacustrine shales. The fans mapped on seismic profiles are partitioned into lobes, based on sand bodies penetrated into the wells or the conceptual model where correlation between wells is not possible. Ten facies are identified from core but attempts to discriminate facies met with mixed success. Facies are therefore combined into four associations (‘Rock Types’) for use in dynamic flow modelling. Rock Types are distributed within the fans using a hierarchical reservoir architecture (‘fan-lobe-surge’). This paper describes the reservoir and fluid characteristics, and outlines the challenges associated with converting the detailed geological model into a form suitable for reservoir simulation, while preserving the main reservoir features that will influence fluid movement within the reservoir

Sedimentology and reservoir modelling of the Ormen Lange field, mid Norway

Reservoir in the Ormen Lange field, offshore Norway, is provided by Upper Cretaceous and Lower Paleocene turbidite sandstones deposited in a sand-rich submarine fan. Systematic vertical and down-fan changes in facies and section properties (thickness, percentage sand and percentage amalgamation) are documented. Bulk sediment accumulation rates are very low suggesting that the system was not directly delta-fed. Paucity of mud in the supply mix prevented the development of channel-overbank systems with strong segregation of sands in channel-fills and muds in overbank areas. The resultant architecture is externally tabular with internally varying degrees of erosional amalgamation between beds. A novel approach to 3D quantitative reservoir modelling has been taken. At a regional-scale (475×180×3 km 3), 3D models integrate all well and structural information and provide a context for field-scale models. Nested within the regional models, high-resolution hierarchical reservoir models of the field have been generated. Models combine detailed sedimentological observations in core with petrophysical, pressure, seismic amplitude and structural data. Modelling rules are derived from partial analogues and a forward geometric model that addresses the nature of distal and lateral sandstone body terminations.