scMARK an 'MNIST' like benchmark to evaluate and optimize models for unifying scRNA data

Today's single-cell RNA analysis tools provide enormous value in enabling researchers to make sense of large single-cell RNA (scRNA) studies, yet their ability to integrate different studies at scale remains untested. Here we present a novel benchmark dataset (scMARK), that consists of 100,000 cells over 10 studies and can test how well models unify data from different scRNA studies. We also introduce a two-step framework that uses supervised models, to evaluate how well unsupervised models integrate scRNA data from the 10 studies. Using this framework, we show that the Variational Autoencoder, scVI, represents the only tool tested that can integrate scRNA studies at scale. Overall, this work paves the way to creating large scRNA atlases and 'off-the-shelf' analysis tools.

Intelligent Operating Envelope Integrated with Automated Well Models Improves Asset Wide PCP Surveillance and Optimization

Petroleum Development Oman (PDO) has been a pioneer in improving Well management processes utilizing its valuable human resources, continuous improvement and digitalization. Managing several PCP wells through Exception Based Surveillance (EBS) methodology had already improved PCP surveillance and optimization across assets. The key to trigger EBS was to keep Operating Envelope (OE), Design Limits updated in Well Management Visualization System (WMVS) after every change in operating speed (RPM), workover and new completion. The sustainable solution was required for automatic update of OEs, having well inflow potential and oil gain opportunities available for quicker optimization decisions for further improvements. PDO has completed a project automating PCP well modeling process where models are built and sustained automatically in Well Management System (WMS) for all active PCP wells, with huge impact on day-to-day operational activities. The paper discusses utilization of physics based well models from WMS to automatically update OE, identify oil gain potential daily and enable real time PCP performance visualization in WMVS. The integration of WMS and WMVS was completed to share data between two systems and automatically update well's OE daily. A tuned well model from WMS was utilized to provide well performance data and sensitivity analysis results for various RPMs. Among the various data obtained from WMS, live OE of torque and fluid above pump (FAP) for various speeds, operating limits, design limits, locked in potential (LIP) for optimization and pump upsize were utilized to process PCP well EBS and create live OE visualization. The visualization is created on a torque-speed chart where a live OE and FAP can be observed in provided picture with current RPM and torque with optimum operating condition. The project is completed after conducting successful change management across PDO assets and after thorough analysis of implementation following benefits were observed: 5% net gain of total PCP production is being executed with zero CAPEX using LIP reports. 50% of engineer's time was saved by updating OEs in WMVS automatically, reduction of false EBS and EBS rationalization. 200% improvement in PCP well performance diagnostics capabilities of Engineers. 15% CAPEX free optimization and pump upsize cases were identified based on well inflow potential. 100% visibility to PCP well's performance was achieved using well model. The visualization has supported engineers monitoring well performance in real time and easily identifying ongoing changes in well and pump performance. PCP well models have supported engineers in new PCP well design and pump upsize. The current efforts in utilizing real time well models, inferred production, automating processes to update OE is one more step toward Digitalization of PCP Surveillance and optimization and to achieve self well optimization for further improving operational efficiency.

Mental Workload Detection Based on EEG Analysis

The study of mental workload becomes essential for human work efficiency, health conditions and to avoid accidents, since workload compromises both performance and awareness. Although workload has been widely studied using several physiological measures, minimising the sensor network as much as possible remains both a challenge and a requirement. Electroencephalogram (EEG) signals have shown a high correlation to specific cognitive and mental states like workload. However, there is not enough evidence in the literature to validate how well models generalize in case of new subjects performing tasks of a workload similar to the ones included during model’s training. In this paper we propose a binary neural network to classify EEG features across different mental workloads. Two workloads, low and medium, are induced using two variants of the N-Back Test. The proposed model was validated in a dataset collected from 16 subjects and shown a high level of generalization capability: model reported an average recall of 81.81% in a leave-one-out subject evaluation.

Full Integrated Model as a Tool for Strategy Evaluation of the Greenfield

The main goal of this work is to evaluate and select the best strategy for the development of the field in the first stage of development. To solve this problem, a full-scale integrated model was created that takes into account the physics of the reservoir, wells and surface infrastructure, as well as their mutual influence. The integrated model was calculated for the full development of the asset. The integrated model of the A. Zhagrin field is based on three simulation models, well models and surface infrastructure, which are linked through an integrator program. All constituent parts of the model are configured to accurately reproduce their actual operation. Greenfield is characterized as active drilling, so the planned well count is modeled by replacing wells with "typical" well models, which are selected by taking into account the expected input flow rate, well design, well completion and well trajectory. Fields of the first stage of development are characterized by limitations related to oil transportation and treatment. These constraints are also specified and taken into account in the model The concept of surface infrastructure is formed depending on the potential production capabilities of the reservoir and has considerable variability. The total number of actual and planned wells in the field is more than 1,300, including more than 700 production wells and about 600 injection wells. All wells are ESP lifted. Considering infrastructure capacity constraints and requirements for optimal pipeline utilization, the use of different numbers of drilling rigs directly affecting the utilization of oil treatment and delivery facilities was evaluated. 29 main variations of the field development strategy until 2060 were formed and calculated, based on the integrated model. The main parameters of variation were the capacity of preparation facilities, the degree of oil separation, the scheme of product transportation, gas utilization capabilities, drilling rigs and subsurface equipment. All scenarios in the integrated model took into account constraints - on target bottomhole, wellhead and line pressures, in order to operate real facilities in accident-free mode. In the course of calculations, an optimal scenario was selected, which made it possible to increase oil production in 2021 by optimizing the transportation of produced products to the treatment facilities. This scenario formed the basis of the asset development strategy.

Understanding Collections of Related Datasets Using Dependent MMD Coresets

Understanding how two datasets differ can help us determine whether one dataset under-represents certain sub-populations, and provides insights into how well models will generalize across datasets. Representative points selected by a maximum mean discrepancy (MMD) coreset can provide interpretable summaries of a single dataset, but are not easily compared across datasets. In this paper, we introduce dependent MMD coresets, a data summarization method for collections of datasets that facilitates comparison of distributions. We show that dependent MMD coresets are useful for understanding multiple related datasets and understanding model generalization between such datasets.

Closed Looped Production Rate Evaluation by Means of Virtual Metering and Pressure Transient Analysis

The Integration of real-time high frequency data in well models allows to infer useful information regarding well and field performance. Virtual Metering (VM) algorithms aim at providing real time well rates solving an inverse problem based on flow equation in the wellbore. Although VM methodologies are based on Pressure/Temperature measurements, they rely on availability of calibration measurements. Pressure Transient Analysis (PTA) can provide useful insight for VM calibration. An innovative closed-loop workflow combining VM and PTA has been developed to face unreliable or absent rate measurements. VM requires periodical separator tests for model calibration. PTA played an important role in estimating well production rates, using it as a virtual well test to compensate the lack of field tests. VM rates are used as first guess for the PTA interpretation of build-up where production rates are unreliable. PTA log-log derivative plot is compared with the reference one which was interpreted to calibrate the formation K•H. The loop is iterated correcting VM calibration parameters until the match is acceptable. An implementation of the closed loop rate estimation workflow on an offshore oil asset is presented as an application of the methodology. The asset comprises 15 production wells, most of them with high Gas-Oil Ratio. Virtual Metering has been applied on wells fully equipped with wellhead and bottom-hole sensors. The joint application of PTA with an iterative closed loop philosophy was fundamental to compensate the lack of separator tests and of the sometimes unreliable choke opening data. The accuracy of the production profiles simulated by the VM is confirmed by the comparison with the reference asset fiscal production and by the final pressure history matching obtained with the PTA. The application of the iterative closed-loop workflow plays a fundamental role in the improvement of backallocation, in real time production monitoring and in the implementation of production optimization. Well models based on VM algorithm have been included in production optimization workflow to improve the well line-up and identify production optimization opportunities. Virtual Metering allowed to monitor results of optimization actions by estimating the actual wells production increment. This paper contains a novel approach, consisting in a reliable and robust closed loop virtual metering workflow, which integrates different tools with the common objective of assessing the actual well production rates for maximising the asset performance. The real-time data and model sharing allowed to set-up a collaborative environment optimizing effective problem solving and field production performance.

Study of factors determining the amplitude-frequency characteristics of acoustic noise in well logging

The article considers the factors that affect the spectral characteristics of acoustic noise recorded in active horizontal wells during logging. Based on the results of physical experiments performed on well models, the presence of resonance phenomena in the acoustic volume of the wellbore, the influence of the location of sensor and the tool body on the measurement result under identical conditions, the need to calibrate downhole tools for bringing their characteristics to a single scale are shown.

Maximising Asset Value through Implementation of Dynamic Well Operating Envelop

Dynamic Well Operating Envelop (WOE) allows to ensure that well is maintained and operated within design limits and operated in the safe, stable and profitable way. WOE covers the Well Integrity, Reservoir constraints and Facility limitations and visualizes them on well performance chart (Hamzat et al., 2013). Design and operating limits (such as upper and lower completion/facilities design pressures, sand failure, erosion limitations, reservoir management related limitations etc) are identified and translated into two-dimensional WOE (pressure vs. flowrate) to ensure maximum range of operating conditions that represents safe and reliable operation are covered. VLP/IPR performance curves were incorporated based on latest Validated Well Model. Optimum well operating window represents the maximum range of operating conditions within the Reservoir constraints assessed. By introducing actual Well Performance data the optimisation opportunities such as production/injection enhancement identified. During generating the Well Operating Envelops tremendous work being done to rectify challenges such as: most static data (i.e. design and reservoir limitations) are not digitized, unreliable real-time/dynamic data flow (i.e. FTHP, Oil/Gas rates etc), disintegrated and unreliable well Models and no solid workflows for Flow assurance. As a pre-requisite the workflows being developed to make data tidy i.e.ready and right, and Well Model inputs being integrated to build updated Well Models. Successful WOE prototype is generated for natural and artificially lifted Oil and Gas wells. Optimisation opportunities being identified (i.e. flowline pressure reduction, reservoir stimulation and bean-up) Proactive maintenance is made possible through dynamic WOE as a real time exceptional based surveillance (EBS) tool which is allowing Asset engineers to conduct the well performance monitoring, and maintain it within safe, stable and profitable window. Additionally, it allows to track all Production Enhancement jobs and seamless forecasting for new opportunities.

Information Theory: A Swiss Army Knife for system characterization, learning and prediction

<p>In this contribution, I will – with examples from hydrology - make the case for information theory as a general language and framework for i) characterizing systems, ii) quantifying the information content in data, iii) evaluating how well models can learn from data, and iv) measuring how well models do in prediction. In particular, I will discuss how information measures can be used to characterize systems by the state space volume they occupy, their dynamical complexity, and their distance from equilibrium. Likewise, I will discuss how we can measure the information content of data through systematic perturbations, and how much information a model absorbs (or ignores) from data during learning. This can help building hybrid models that optimally combine information in data and general knowledge from physical and other laws, which is currently among the key challenges in machine learning applied to earth science problems.</p><p>While I will try my best to convince everybody of taking an information perspective henceforth, I will also name the related challenges: Data demands, binning choices, estimation of probability distributions from limited data, and issues with excessive data dimensionality.</p>