Generative Adversarial Networks (GANs) have shown impressive results for image generation. However, GANs face challenges in generating contents with certain types of constraints, such as game levels. Specifically, it is difficult to generate levels that have aesthetic appeal and are playable at the same time. Additionally, because training data usually is limited, it is challenging to generate unique levels with current GANs. In this paper, we propose a new GAN architecture named Conditional Embedding Self-Attention Generative Adversarial Network (CESAGAN) and a new bootstrapping training procedure. The CESAGAN is a modification of the self-attention GAN that incorporates an embedding feature vector input to condition the training of the discriminator and generator. This allows the network to model non-local dependency between game objects, and to count objects. Additionally, to reduce the number of levels necessary to train the GAN, we propose a bootstrapping mechanism in which playable generated levels are added to the training set. The results demonstrate that the new approach does not only generate a larger number of levels that are playable but also generates fewer duplicate levels compared to a standard GAN.
Hanabi is a cooperative card game with hidden information that has won important awards in the industry and received some recent academic attention. A two-track competition of agents for the game will take place in the 2018 CIG conference. In this paper, we develop a genetic algorithm that builds rulebased agents by determining the best sequence of rules from a fixed rule set to use as strategy. In three separate experiments, we remove human assumptions regarding the ordering of rules, add new, more expressive rules to the rule set and independently evolve agents specialized at specific game sizes. As result, we achieve scores superior to previously published research for the mirror and mixed evaluation of agents.
We develop a proxy model based on deep learning methods to accelerate the simulations of oil reservoirs-by three orders of magnitudecompared to industry-strength physics-based PDE solvers. This paper describes a new architectural approach to this task, accompanied by a thorough experimental evaluation on a publicly available reservoir model. We demonstrate that in a practical setting a speedup of more than 2000X can be achieved with an average sequence error of about 10% relative to the oil-field simulator. The proxy model is contrasted with a high-quality physics-based acceleration baseline and is shown to outperform it by several orders of magnitude. We believe the outcomes presented here are extremely promising and offer a valuable benchmark for continuing research in oil field development optimization. Due to its domain-agnostic architecture, the presented approach can be extended to many applications beyond the field of oil and gas exploration.
The economical performance of an oilfield operation is uncertain and highly influenced by strategic and operational decisions variables such as well placement, scheduling and control. Based on numerically intensive reservoir simula- tors, the evaluation of an extensive list of possible decisions across all possible realizations becomes computationally intractable and additional mathematical techniques are required. A common approach to dealing with this problem is the Response Surface Methodology (RSM) coupled with Design of Experiments (DoE) and sampling techniques. Existing approaches to construct surrogates depend on specific statistical/risk measures such as expected value and standard deviation. For example, in order to construct a surrogate for the standard deviation of NPV, one would compute the standard deviation associated with the simulation results over the selected geological realizations for each candidate production strategy and then fit a mathematical model to it. In this case, the idiosyncratic response of each geological realization with respect to the production strategy is lost, which may lead to a bad risk assessment and, consequently an inappropriate decision making process. In this paper, we propose Stochastic Response Surface Methodology (SRSM) to enhance the decision-making process over the determination of oil & gas production strate- gies while properly taking into consideration geological uncertainty. The SRSM does not depend on any pre-defined risk measure providing the necessary flexibility to evaluate the intrinsic risk-return trade-off associated with the economical performance of the reservoir. Our approach is based on the construction of surrogates for each geolog- ical realization selected by sampling procedure. We argue that constructing a different surrogate for each selected realization captures the idiosyncratic behavior of each representative geological setting and provides the flexibility of choosing any set of risk measures after the surrogate construction has been done. Based on the Brugge field, an SPE benchmark case study, we provide a numerical example to illustrate our methodology.
Decision making under uncertainty can be quite challenging, especially when complex numerical simulations are considered in the work flow and the decision has to be made relatively fast (e.g., in hours). This is the case when one needs to rank a given field portfolio within a limited budget and with acquisition constraints. If the ranking measure associated with each field is properly and rapidly evaluated, new prospect opportunities, which may lead to a favorable strategic position, can be readily identified.In this paper, we propose an efficient methodology for computing a "production-potential" measure that can be used to rank greenfield portfolios in the presence of geological uncertainty, quantifying both uncertainty and risk propagation. Next, we briefly describe the basics of the method proposed. First, uncertainty in sedimentary variability and flow behavior has to be characterized by a number of representative geological realizations. Sampling techniques are used to significantly reduce the number of realizations while preserving accuracy in the description and uncertainty propagation. Thereafter, multiple and varied field-development plans, based on primary/secondary-recovery mechanisms, are automatically generated while accounting for key parameters related to the number, drilling locations, and drilling sequence of wells. In these plans the reservoir is clustered by areas with similar production/injection potential, and the well locations and drilling schedules are obtained accordingly. The well controls are determined through estimations of the fieldrecovery factor. By means of experimental-design techniques a relatively small number of field-development plans are selected to capture the most significant production profiles. Each of these development plans is simulated for the realizations sampled previously, and the production-potential measure [e.g., average net present value (NPV) over all sampled realizations] is computed for all the plans. The highest of these measures (i.e., the best development plan) can be used for ranking the greenfield in the portfolio. Response-surface procedures are considered to perform additional analysis computations within iterative optimization procedures. It is important to note that other statistics related to the exploitation potential (e.g., standard deviation of the NPV) can also be used to complement the ranking, thereby mitigating the decision makers' risk tolerance. The methodology has been tested on the Brugge Field benchmark, which presents 104 realizations of multiple geological parameters. The benchmark has been modified to simulate a greenfield scenario. The ranking measure is the (discounted) NPV averaged over the 104 realizations. The proposed work flow yields a ranking measure of USD 5.43 billion, and the computational cost is approximately 1,900 simulations (performed in a parallel-computing environment). This NPV is somewhat higher than those found for the Brugge benchmark (with similar modified settings) by other researchers. To validate the result...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.