A dynamic system identification problem is considered. It is an inverse modelling problem, where one needs to find the model in an analytical form and a dynamic system is represented with the observation data. In this study the identification problem was reduced to an optimization problem, and in such a way every solution of the extremum problem determines a linear differential equation and coordinates of the initial value. The proposed approaches do not require any assumptions of the system order and the initial value coordinates and estimates the model in the form of a linear differential equation. These variables are estimated automatically and simultaneously with differential equation coefficients. Problem-oriented evolution-based optimization techniques were designed and applied. Techniques are based on the evolutionary strategies algorithm and have been improved to achieve efficient solving of the reduced problem for every proposed determination scheme. Experimental results confirm the reliability of the given approach and the usefulness of the reduced problem solving tool.
1 November 2019 in Journal of Physics: Conference Series
The waveguide path systems consisting mainly of aluminum alloys with improved weight, size, and electrical parameters are used to ensure the functioning of the spacecraft and its onboard equipment in accordance with a given program. When assembling the elements of the waveguide paths, in particular a thin-walled pipe of rectangular cross section, coupling and flange, induction brazing is used. Element-by-element assembly of lightweight waveguide paths using the method of induction soldering is connected with a number of technological difficulties such as the uneven distribution of temperature fields, the thickness variation of the soldered products and the design features of the inductors. This paper presents a model describing the process of heating a pipe-inductor system during induction brazing. The presented model describes the process of induction heating of the pipe as a separate structural element of the waveguide path. In the process of simulation in the Comsol software, on the basis of numerical simulation methods with the connection of the necessary media involved in induction heating, the pipe heating graphs are obtained.
Актуальность работы. Оценка трещиноватости карбонатных коллекторов является одной из наиболее сложных задач нефтегазовой геофизики, поскольку наличие трещин существенно влияет на фильтрационно-емкостные свойства пород. Традиционные методы интерпретации геофизического каротажа не всегда обеспечивают необходимую достоверность выделения трещиноватых интервалов в условиях сложного геологического строения. Цель исследования. Разработка и оценка эффективности интеллектуальных методов обработки данных геофизического каротажа для прогнозирования трещиноватости карбонатных коллекторов. Методы исследования. Использованы данные акустического, нейтронного, плотностного и электрического каротажа, FMI-сканирования и керновых исследований. Для анализа применялись методы машинного обучения Random Forest, XGBoost, SVM и сверточные нейронные сети. Выполнены предварительная обработка данных, формирование признакового пространства и обучение моделей на размеченных выборках. Результаты. Установлено, что наибольшую эффективность при прогнозировании трещиноватости обеспечивают ансамблевые алгоритмы и комплексная обработка данных ГИС и FMI. Максимальная точность классификации достигла 92 % при использовании XGBoost. Выявлены устойчивые связи между локальной изменчивостью каротажных параметров и развитием трещинных систем. Полученные результаты подтверждают перспективность применения интеллектуальных технологий для повышения достоверности интерпретации карбонатных коллекторов Relevance.Fracture characterization of carbonate reservoirs remains one of the most challenging problems in petroleum geophysics because fracture systems significantly affect reservoir storage and flow properties. Conventional well-log interpretation techniques often provide insufficient reliability in identifying fractured intervals under complex geological conditions. Aim.To develop and evaluate intelligent data-processing approaches for predicting fracture distribution in carbonate reservoirs using well logging information. Methods.Acoustic, neutron, density, and electrical logging data, FMI images, and core analysis results were used. Machine learning methods including Random Forest, XGBoost, Support Vector Machine, and convolutional neural networks were applied. Data preprocessing, feature engineering, and model training were performed using labeled datasets. Results. Ensemble algorithms and integrated interpretation of well logging and FMI data demonstrated the highest prediction performance. The XGBoost model achieved a maximum classification accuracy of 92 %. Stable relationships between local variations of logging parameters and fracture development were identified. The obtained results confirm the effectiveness of intelligent data-processing methods for improving fracture detection reliability and enhancing geological characterization of carbonate reservoirs
30 November 2018 in IOP Conference Series: Materials Science and Engineering
Image recognition is an independent field of the computer science nowadays. Image classification is one of its main domains, in which investigated objects can be represented by an image or a video stream. The objective of the image classification is correct assigning of objects to corresponding classes, and there exist many effective approaches for solving this problem. One of the most popular approaches is artificial neural networks, which are a method from the field of machine learning. Despite the fact that neural networks cover a wide range of machine learning problems, they are also able to solve the problem of the image classification. However, there is one more specific approach for neural networks-based images classification that applies the deep learning conception. The best-known deep learning algorithm is called the convolutional neural network (CNN). The CNN uses a principle of using the same parts of a neural network to manipulate with different local parts of an input image. As well as the standard neural network architecture, the convolutional neural network should be fine-tuned for solving a certain problem. Because of the CNN's depth and complexity, the tuning process usually is complex and needs huge computational efforts. In this study, we have proposed an approach for creating ensembles of previously trained convolutional neural networks. The approach allows to increase the performance of the image classification. The results of experiments for image classification problems are presented and discussed. The experiments show that the proposed approach is able to outperform the standard perceptron and single convolutional neural network.
1 April 2020 in Journal of Physics: Conference Series
Large-scale global optimization (LSGO) problems are emergent in many domains of applied sciences. LSGO is a hard challenge for the majority of state-of-the-art optimization methods. Evolution algorithms (EAs) combined with Cooperative Coevolution (CC) are able to perform well when solving many real-world LSGO problems. The grouping of variables at the problem decomposition stage has a significant impact on the performance of the CC approach. This study proposes an improvement of the previously developed random adaptive grouping (RAG) approach for CC. The new method is titled as RAG2, and the whole optimization algorithm, based on RAG2, is called CC-SHADE-RAG2. The influence of the choice of the population size and the number of subcomponents on the algorithm performance have been investigated using the IEEE LSGO CEC’2013 benchmark. The set of test problems in the benchmark contains fifteen functions with dimensionality equal to one thousand. We have also compared the performance of the novel algorithm with some EAs, which are applied for solving LSGO problems.