Uncertainty Quantification Method for Trend Prediction of Oil Well Time Series Data Based on SDMI Loss Function DOI Open Access
Yun Shen, Xiang Wang, Yixin Xie

et al.

Processes, Journal Year: 2024, Volume and Issue: 12(12), P. 2642 - 2642

Published: Nov. 23, 2024

IoT sensors in oilfields gather real-time data sequences from oil wells. Accurate trend predictions of these are crucial for production optimization and failure forecasting. However, well time series exhibit strong nonlinearity, requiring not only precise prediction but also the estimation uncertainty intervals. This paper first proposed a denoising method based on Variational Mode Decomposition (VMD) Long Short-Term Memory (LSTM) to reduce noise present data. Subsequently, an SDMI loss function was introduced, combining respective advantages Soft Dynamic Time Warping Mean Squared Error (MSE). The additionally accepts upper lower bounds interval as input is optimized with sequence. By predicting next 48 points, results using existing three common functions compared multiple sets. before after shown. experimental demonstrate that average coverage rate predicted intervals across seven wells 81.4%, accurately reflect trends real

Language: Английский

Dynamic behavior of multi-dimensional chaotic systems based on state variables and unknown parameters with applications in image encryption DOI
Jingfeng Jie,

Ping Zhang,

Yang Yang

et al.

Physica Scripta, Journal Year: 2025, Volume and Issue: 100(2), P. 025222 - 025222

Published: Jan. 14, 2025

Abstract To explore the impact of unknown terms and parameters on chaotic characteristics in systems, this paper examines effects state variables parameters. The study focuses different combinations linear, nonlinear, constant It primarily investigates role multi-order their application to system models varying dimensions. Firstly, by simulating a three-dimensional system, analyzes how nonlinear initial conditions affect system's behavior. Secondly, it evaluates four-dimensional combining with parameters, using tools such as Lyapunov index diagrams, sample entropy, dynamic trajectory plots. Finally, integrates constructed mapping develop two-level key image encryption thoroughly assessing its security resistance interference.

Language: Английский

Citations

0
A comparative analysis of LSTM, GRU, and Transformer models for construction cost prediction with multidimensional feature integration DOI Creative Commons
Shi Tang, Kazuya SHIDE

Journal of Asian Architecture and Building Engineering, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 16

Published: Jan. 18, 2025

Construction cost prediction remains a complex challenge due to the multidimensional nature of construction data and external factors. The objective this study is identify most effective deep learning model for accurately predicting costs by comparing performance LSTM, GRU, Transformer models. Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), are advanced machine regression models widely utilized tasks. This investigates these models' using feature framework. Through comprehensive evaluation comparison, demonstrated superior performance, particularly excelling in handling interactions long-sequence data. LSTM model, while capturing temporal dependencies, shows reliable but lags behind accuracy. GRU although faster training, proved less accurate outliers. Key features such as Total Area (TA), Site (SA), Number Floors (NF) were identified significant predictors across all models, with proving adept at interactions. By integrating features, contributes improved management, thereby enhancing accuracy reliability.

Language: Английский

Citations

0
GHENet: Attention-based Hurst exponents for the forecasting of stock market indexes DOI
João B. Florindo,

Reneé Rodrigues Lima,

Francisco Alves dos Santos

et al.

Physica A Statistical Mechanics and its Applications, Journal Year: 2025, Volume and Issue: unknown, P. 130540 - 130540

Published: March 1, 2025

Language: Английский

Citations

0
Machine Learning Models Informed by Connected Mixture Components for Short- and Medium-Term Time Series Forecasting DOI Creative Commons
Andrey Gorshenin,

Anton L. Vilyaev

AI, Journal Year: 2024, Volume and Issue: 5(4), P. 1955 - 1976

Published: Oct. 22, 2024

This paper presents a new approach in the field of probability-informed machine learning (ML). It implies improving results ML algorithms and neural networks (NNs) by using probability models as source additional features situations where it is impossible to increase training datasets for various reasons. We introduce connected mixture components information that can be extracted from mathematical model. These are formed special algorithm merging parameters sliding window mode. has been proven effective when applied real-world time series data short- medium-term forecasting. In all cases, informed showed better than those did not use them, although different may datasets. The fundamental novelty research lies both informing demonstrated forecasting accuracy applications. For geophysical spatiotemporal data, decrease Root Mean Square Error (RMSE) was up 27.7%, reduction Absolute Percentage (MAPE) 45.7% compared with without informing. best metrics values were obtained an ensemble architecture fuses Long Short-Term Memory (LSTM) network transformer. Squared (MSE) electricity transformer oil temperature ETDataset had improved 10.0% vanilla methods. MSE value random forest. introduced allows us outperform NN architectures classical statistical

Language: Английский

Citations

2
The Time Series Classification of Discrete-Time Chaotic Systems Using Deep Learning Approaches DOI Creative Commons
Ömer Faruk Akmeşe, Berkay Emi̇n, Yusuf Alaca

et al.

Mathematics, Journal Year: 2024, Volume and Issue: 12(19), P. 3052 - 3052

Published: Sept. 29, 2024

Discrete-time chaotic systems exhibit nonlinear and unpredictable dynamic behavior, making them very difficult to classify. They have properties such as the stability of equilibrium points, symmetric behaviors, a transition chaos. This study aims classify time series images discrete-time by integrating deep learning methods classification algorithms. The most important innovation this is use unique dataset created using systems. In context, large representing various behaviors was for nine different initial conditions, control parameters, iteration numbers. based on existing system solutions in literature, but structures these much more complex than ordinary image datasets due their nature. Although there are studies literature continuous-time systems, no been found obtained were classified with models DenseNet121, VGG16, VGG19, InceptionV3, MobileNetV2, Xception. addition, integrated algorithms XGBOOST, k-NN, SVM, RF, providing methodological innovation. As best result, 95.76% accuracy rate DenseNet121 model XGBOOST algorithm. takes graphical representations an advanced level provides powerful tool respect, classifying offers adapting datasets. findings thought provide new perspectives future research further advance studies.

Language: Английский

Citations

1
KRC-APM: Key region cutting and artificial prior model for breast cancer recognition in ultrasound images DOI
Lin Yi,

Haosen Wang,

Jingchi Jiang

et al.

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 257, P. 125092 - 125092

Published: Aug. 13, 2024

Language: Английский

Citations

0
Uncertainty Quantification Method for Trend Prediction of Oil Well Time Series Data Based on SDMI Loss Function DOI Open Access
Yun Shen, Xiang Wang, Yixin Xie

et al.

Processes, Journal Year: 2024, Volume and Issue: 12(12), P. 2642 - 2642

Published: Nov. 23, 2024

IoT sensors in oilfields gather real-time data sequences from oil wells. Accurate trend predictions of these are crucial for production optimization and failure forecasting. However, well time series exhibit strong nonlinearity, requiring not only precise prediction but also the estimation uncertainty intervals. This paper first proposed a denoising method based on Variational Mode Decomposition (VMD) Long Short-Term Memory (LSTM) to reduce noise present data. Subsequently, an SDMI loss function was introduced, combining respective advantages Soft Dynamic Time Warping Mean Squared Error (MSE). The additionally accepts upper lower bounds interval as input is optimized with sequence. By predicting next 48 points, results using existing three common functions compared multiple sets. before after shown. experimental demonstrate that average coverage rate predicted intervals across seven wells 81.4%, accurately reflect trends real

Language: Английский

Citations

0