Uncertainty Quantification of Reservoir Performance Using Artificial Neural Network and Structured Expert Judgment

Uncertainty quantification of reservoir performance using machine learning algorithms and structured expert judgment

When physical and data sciences are not sufficient to support models and/or decisions, expert judgment is a recognized approach to quantify the uncertainties around specific issues. Among different expert judgment methods, Structured Expert Judgment (SEJ) employs a formalized, documented procedure for obtaining probabilistic belief statements from a group of experts about unknown quantities or parameters. This provides an attractive approach for performing assessments at volcanoes characterized by large knowledge gaps by integrating diverse kinds of information. Uncertainties are likely to be large in these cases, and SEJ can quantify these uncertainties to provide scientists and decision-makers with indications of the reliability of the assessments. The final goal of this approach is to obtain the group’s synthesized uncertainty distribution (representing a new “virtual expert” often called Decision Maker - DM) around specific items (or “target questions”), that result from combining elicited judgments of all the experts.More specifically, performance-based expert elicitation relies on validating expert probability assessments through an impartial empirical trial. Thus, in a performance-based elicitation, experts are tasked with providing their estimations of probability distributions for a set of known quantities, often referred to as “seed items,” which, jointly, serve as calibration benchmarks for expert performance. Experts’ responses provide the basis for performance scoring using the Classical Model algorithm (Cooke, 1991), determined empirically on the individual expert’s attainment, jointly in terms of two separate metrics: “statistical accuracy” and “informativeness”, which are evaluated on the set of seed items overall.Traditionally, the management of a performance-based expert elicitation is time consuming and involves the collection of tens of questionnaires, often manually copied from hard-copies of the responses or email attachments. The performance-calibration algorithms and the production of standard outputs, including statistical samples of the DM responses, are necessary steps every time an elicitation is conducted, but relied on different scripts and independent pieces of software.In this study we present the latest version of the recently released software ELICIPY (de’Michieli Vitturi et al. 2024), which allows organizing and managing performance-based expert elicitation sessions in a partially automated way, resulting in a significantly enhanced efficiency. This new version includes, among other improvements, an “agreement index” to quantify the level of agreement among experts, a continuous version of the Classical Model to compute expert weights, the possibility to import weights from an external file, new plot options, an interactive dashboard for results exploration.ELICIPY enables the smooth conduction of elicitations with a relatively large number of questions and/or experts. Moreover, the management of multiple elicitation sessions, with questionnaire modifications and response updates are made easier. Some of these improvements are demonstrated on the case study of a hazard and risk assessment for the active Kolumbo submarine Volcano (Aegean Sea, Greece), which was made of 20 Seed Items and 64 Target Items in total, the latter structured into 17 subject matter groups (Bevilacqua et al., 2025).

Workshop report

This paper presents an investigation of an artificial neural network (ANN) to predict the flutter characteristics of a simple wing that under went multiple weight variation changes. The weight variations were used to simulate the addition of non-structural weight or structural repair weight. The test case wing had 24 designated repair areas where repair weights were added as lumped masses. Weight and location of the repairs were randomly generated to increase the sample size and help the generalization of the ANN. The ANN was evaluated for both the flutter speed and flutter frequency for multiple levels of structural damping. The number of test cases required to train the neural network was also evaluated for both the error between the input and output results as well as the generalization of the neural networks.

The Brain Emotional Learning (BEL) is a novel bio-inspired machine learning approach mentioned as a new class of artificial neural network (ANN). The artificial emotional neural network (AENN) is one of the BEL methods which used the genetic algorithm (GA) to compute proper weights, weights of the amygdala (AMYG), orbitofrontal cortex (OFC) weights and a bias value. AENN trained by GA has been reported that it could produce low error rates. However, AENN still has more rooms to enhance its prediction of performance, especially generalization and the prediction of performance. Therefore, this paper aims to propose a new training method for AENN. The JAYA optimization algorithm optimized the weights and the biases of AENN. Two new proposed models are named as AENN-Max-JAYA and AENN-Mean-JAYA. Their names are according to the way of selecting the additional expanded feature which obtained through either the max or the average of input patterns, respectively. From the experimental results for wind power forecasting dataset, the proposed methods proved that the results are better in generalization performance and give lower error rates which compared to the comparative AENN models and traditional ANNs.

For the past few years Artificial Neural Networks (ANN) have made a strong comeback to the scientific community. They are used in a variety of tasks where adaptive computing can enhance process performance. There has been a handful of papers suggesting the use of artificial neural networks in the petroleum industry1-3. These papers can be classified into two major categories. First category includes papers that recommend the use of ANN in classification of lithologies from well logs. Second category includes papers that employ ANN to pick the proper reservoir model for well testing purposes. This paper introduces a new implementation of the neuro-computing technology in petroleum engineering. It is shown in this study that artificial neural networks possess numerous capabilities, and can be much more useful to petroleum engineers than previously thought. An implementation of artificial neural networks in characterization of reservoir heterogeneity is presented in this paper. A methodology is introduced through which different rock properties in highly heterogeneous reservoirs can be predicted with good accuracy using information deduced from geophysical well logs. Examples of such networks are presented using field data for verification. The underlying reasons (theories) that make achievement of such complex tasks possible are discussed. The notion that artificial intelligence and neural networks in particular have immense potentials in solving complex engineering and scientific problems are addressed. The innovation now lies on the creativity of the researchers to recognize and define petroleum engineering problems that can be addressed by artificial intelligence technology.

Brain research up to date has revealed that structure and function are highly related. Thus, for example, studies have repeatedly shown that the brains of patients suffering from schizophrenia or other diseases have a different connectome compared to healthy people. Apart from stochastic processes, however, an inherent logic describing how neurons connect to each other has not yet been identified. We revisited this structural dilemma by comparing and analyzing artificial and biological-based neural networks. Namely, we used feed-forward and recurrent artificial neural networks as well as networks based on the structure of the micro-connectome of C. elegans and of the human macro-connectome. We trained these diverse networks, which markedly differ in their architecture, initialization and pruning technique, and we found remarkable parallels between biological-based and artificial neural networks, as we were additionally able to show that the dilemma is also present in artificial neural networks. Our findings show that structure contains all the information, but that this structure is not exclusive. Indeed, the same structure was able to solve completely different problems with only minimal adjustments. We particularly put interest on the influence of weights and the neuron offset value, as they show a different adaption behaviour. Our findings open up new questions in the fields of artificial and biological information processing research.

The information processing capability of artificial neural networks (ANNs) is closely related to its architecture and weights. The paper describes a new evolutionary system for evolving artificial feedforward neural networks, which is based on the particle swarm optimisation (PSO) algorithm. Both the architecture and the weights of ANNs are adaptively adjusted according to the quality of the neural network. This process is repeated until the best ANN is accepted or the maximum number of generations has been reached. A strategy of evolving added nodes and a partial training algorithm are used to maintain a close behavioural link between the parents and their offspring. This system has been tested on two real problems in the medical domain. The results show that ANNs evolved by PSONN have good accuracy and generalisation ability.

Objective: The main objective of this paper is to improve the intrusion detection accuracy of neural networks by hybridized with Jordan network, applying novel Lyapunov function and changing input values as fuzzy values using fuzzy logic. Methods: In this research, the intrusion detection of NSL KDD dataset is carry out by neural network first . The performance of neural networks mainly depends on the parameters like number of hidden layer, no of node in the hidden layer and the no of epoch. Selecting the proper parameters values and weight initialization are the main difficulty in neural network. To overcome this issue a hybrid network by combining neural network with Jordan network is proposed which is highly sensitive to weight convergence. In hybrid network Lyapunov function is used to achieve the stability of equilibrium between two networks. But in this work a novel Lyapunov function is proposed to achieve global robust stability in hybrid network. A novel Lyapunov function uses a class of general activation functions which are not to be differentiable, bounded or monotonically nondecreasing. A set of criteria are derived to guarantee the existence, uniqueness and global robust stability of the equilibrium of hybrid networks with time delays. Then the learning ability of hybrid network is improved by using fuzzy logic. The hybrid network improved by a novel Lyapunov and fuzzy logic is called as Improved Fuzzy Hybrid Jordan network and Artificial Neural Network (IHFJANN). Findings: Artificial Neural Network (ANN), Hybrid Jordan network and Artificial Neural Network (HJANN) and Improved Hybrid Fuzzy Jordan network and Artificial Neural Network (IHFJ ANN) are applied on NSL KDD training dataset to lean the type of available attacks. NSL training dataset contains 2500 instances and 38 attributes where as testing dataset contains 995 instances and 38 attributes. The learned model of three classifiers is used to predict the classes in test dataset. The performance measures are evaluated in terms of accuracy, precision, recall and f-measure values. Improvement: The classification accuracy of Artificial Neural Network (ANN), Hybrid Jordan network and Artificial Neural Network (HJANN) and Improved Hybrid Fuzzy Jordan network and Artificial Neural Network (IHFJANN) are 72%, 80% and 84% respectively. Accuracy is increased by 12% in IHFJANN than the ANN and 4% than the HJANN. The precision value of ANN, HJANN and IHFJANN are 0.46, 0.54 and 0.57 respectively. Precision value is increased by 0.9 in IHFJANN than the ANN and 0.3 than the HJANN. The recall value of ANN, HJANN and IHFJANN are 0.73, 0.81 and 0.84 respectively. Recall value is increased by 0.9 in IHFJANN than the ANN and 0.3 than the HJANN. The F-measure value of ANN, HJANN and IHFJANN are 0.58, 0.65 and 0.68 respectively. F-measure value is increased by 0.10 in IHFJANN than the ANN and 0.3 than the HJANN. The results proved that the proposed IHFJANN provides better performance than ANN and HJANN.

Benchmark exercise on expert judgment techniques in PSA Level 2

این مطالعه به منظور بررسی پیش بینی شاخص های عملکرد تولیدی در مرغان تخم گذار با استفاده از شبکه های عصبی مصنوعی و رگرسیون غیرخطی چندگانه انجام شد. بررسی بر روی اطلاعات چهار دوره متوالی پرورش در یک واحد پرورش مرغ تخم گذار صورت گرفت. روش های داده-کاوی شامل رگرسیون خطی و غیر خطی، شبکه عصبی پرسپترون سه لایه، شبکه عصبی پرسپترون چهار لایه و شبکه عصبی با تابع پایه ای شعاعی بود. در این مدل ها از متغیرهای سن گله، میزان خوراک مصرفی و فصل تولید به عنوان متغیر پیشگو و شاخص های عملکرد تولیدی شامل درصد تخم-گذاری، وزن توده ای تخم مرغ تولیدی و ضریب تبدیل غذایی به عنوان متغیر پاسخ استفاده شد. نتایج نهایی رگرسیون های خطی نشان داد که برای تمامی متغیرهای وابسته مورد مطالعه متغیر مستقل سن گله معنی دار می باشد. بنابراین رگرسیون غیر خطی شاخص های عملکرد تولیدی در مقابل سن برای مقایسه با شبکه های عصبی مختلف مورد بررسی قرار گرفت و برای مقایسه کلیه مدل ها از ضریب تعیین (R2) و میانگین قدر مطلق خطا (MAE) استفاده شد. نتایج نشان داد بین شبکه های عصبی مصنوعی مختلف مورد مطالعه، شبکه با تابع پایه ای شعاعی بهتر از سایر مدل های در پیش‌بینی شاخص های عملکرد تولیدی مرغان تخم گذار عمل می کند.

This thesis evaluates the utility of Artificial Neural Networks (ANNs) applied to financial market and macroeconomic forecasting. In application, ANNs are evaluated in comparison to traditional forecasting models to evaluate if their nonlinear and adaptive properties yield superior forecasting performance in terms of robustness and accuracy. Furthermore, as ANNs are data-driven models, an emphasis is placed on the data collection stage by compiling extensive candidate input variable pools, a task frequently underperformed by prior research. In evaluating their performance, ANNs are applied to the domains of: exchange rate forecasting, volatility forecasting, and macroeconomic forecasting. Regarding exchange rate forecasting, ANNs are applied to forecast the daily logarithmic returns of the EUR/USD over a short-term forecast horizon of one period. Initially, the analytic method of Technical Analysis (TA) and its sub-section of technical indicators are utilized to compile an extensive candidate input variable pool featuring standard and advanced technical indicators measuring all technical aspects of the EUR/USD time series. The candidate input variable pool is then subjected to a two-stage Input Variable Selection (IVS) process, producing an informative subset of technical indicators to serve as inputs to the ANNs. A collection of ANNs is then trained and tested on the EUR/USD time series data with their performance evaluated over a 5-year sample period (2012 to 2016), reserving the last two years for out of sample testing. A Moving Average Convergence Divergence (MACD) model serves as a benchmark with the in-sample and out-of-sample empirical results demonstrating the MACD is a superior forecasting model across most forecast evaluation metrics. For volatility forecasting, ANNs are applied to forecast the volatility of the Nikkei 225 Index over a short-term forecast horizon of one period. Initially, an extensive candidate input variable pool is compiled consisting of implied volatility models and historical volatility models. The candidate input variable pool is then subjected to a two-stage IVS process. A collection of ANNs is then trained and tested on the Nikkei 225 Index time series data with their performance evaluated over a 4-year sample period (2014 to 2017), reserving the last year for out-of-sample testing. A GARCH (1,1) model serves as a benchmark with the out-of-sample empirical results finding the GARCH (1,1) model to be the superior volatility forecasting model. The research concludes with ANNs applied to macroeconomic forecasting, where ANNs are applied to forecast the monthly per cent-change in U.S. civilian unemployment and the quarterly per cent-change in U.S. Gross Domestic Product (GDP). For both studies, an extensive candidate input variable pool is compiled using relevant macroeconomic indicator data sourced from the Federal Bank of St Louis. The candidate input variable pools are then subjected to a two-stage IVS process. A collection of ANNs is trained and tested on the U.S. unemployment time series data (UNEMPLOY) and U.S. GDP time series data. The sample periods are (1972 to 2017) and (1960 to 2016) respectively, reserving the last 20% of data for out of sample testing. In both studies, the performance of the ANNs is benchmarked against a Support Vector Regression (SVR) model and a Naïve forecast. In both studies, the ANNs outperform the SVR benchmark model. The empirical results demonstrate that ANNs are superior forecasting models in the domain of macroeconomic forecasting, with the Modular Neural Network performing notably well. However, the empirical results question the utility of ANNs in the domains of exchange rate forecasting and volatility forecasting. A MACD model outperforms ANNs in exchange rate forecasting both in-sample and out-of-sample, and a GARCH (1,1) model outperforms ANNs in volatility forecasting.

Joint A-SNN: Joint training of artificial and spiking neural networks via self-Distillation and weight factorization

The article explores modern neural network architectures for the automatic detection and recognition of marine surface objects and obstacles of given classes throughout the full image area, applicable for execution in real or near real time on an optoelectronic vision system to au-tomate and improve the safety of civil marine navigation. A formal statement of the problem of automatic detection of objects on images is given. The state-of-the-art algorithms for detecting objects in images based on use of artificial convolutional neural networks were reviewed, their comparison was made and a reasonable choice was made in favor of the most efficient neuralnetwork architecture in terms of computational complexity to recognition accuracy. The subject area is studied, as well as publicly available databases of surface objects suitable for use in the training of algorithms using artificial neural networks. The article concluded that there is insuffi-cient labeled data for training neural network algorithms, as a result of which the authors inde-pendently collected research images and video sequences, prepared and labeled the collected data containing surface marine objects and other obstacles that represent a navigation hazard for ships. Based on the selected neural network architecture, a new neural network algorithm for automatic full-frame detection and recognition of surface objects was developed, and an artificial neural network was trained using the prepared database of images of typical objects. The resulting algorithm was tested by the authors on a validation data set, the quality of its work was estimated using various metrics, and the algorithm’s performance was measured. Conclusions are made about the necessity to expand the collected database of images of typical marine objects, further steps are proposed to improve the accuracy of the developed software and algorithmic complex and its implementation to be used in a marine optoelectronic machine vision system for automa-tion and improving the safety of civil navigation.

Some features of cognitive modeling are presented, including the prerequisites for a cognitive approach to solving complex problems. Cognitive modeling involves the use of various artificial neural networks, including convolutional neural networks. The classification of artificial neural networks according to various characteristics is given. The features of self-organizing neural networks and networks using deep learning methods are considered. The artificial neural network, which is a three-layer unidirectional direct propagation network, the interface of a computer program used to approximate functions using the specified neural network, as well as the solution of the image recognition problem using an artificial convolutional neural network, in which the neural network parameters are adjusted for each recognizable image fragment in order to adaptively filter the image, are considered in detail. The analysis of images in video surveillance systems in order to detect fires allows them to be detected at an early stage and, thus, prevent the fire propogation.