Applied Intelligence最新文献

The micro- and nanopore throats in shale oil reservoirs are finer than those in conventional oil reservoirs and have a larger specific surface area, potentially resulting in a more pronounced crude oil boundary effect. The prediction of recoverable reserves in shale oil reservoirs is influenced by factors such as geological complexity, fracture characteristics, and multiphase flow characteristics. The application of conventional reservoir seepage theories and engineering methods is challenging because of the unique characteristics of shale formations. A novel computational framework is proposed for the prediction of recoverable reserves and optimization of fracturing parameters by combining machine learning algorithms with causal discovery. Based on the theory of causal inference, the framework discovers the underlying causal relationships of the data, mines the internal laws of the data, and evaluates the causal effects, aiming to build an interpretable machine learning model to better understand the properties of shale oil reservoirs. Compared to traditional methods, the interpretable machine learning model has an outstanding prediction ability, with R² of 0.94 and average error as low as 8.57%, which is 5.22% lower than that of traditional methods. Moreover, the maximum prediction error is only 21.84%, which is 25.2% smaller than the maximum error of traditional methods. The prediction robustness is good. An accurate prediction of recoverable reserves can be achieved. Furthermore, by integrating particle swarm optimization and TabNet, a fracturing parameter optimization model for shale oil reservoirs is developed. According to an on-site validation, this optimization results in an average increase of 13.45% in recoverable reserves. This study provides an accurate reference for reserve assessment and production design in the exploration and development of shale oil reservoirs.

Graphical Abstract

Short-term forecasting of the Limit Order Book (LOB) is challenging due to market noise. Traditionally, technical analysis using candlestick charts has been effective for market analysis and predictions. Inspired by this, we introduce a novel methodology. First, we preprocess the LOB data into long-term frame data resembling candlestick patterns to reduce noise interference. We then present the Long Short-Term Temporal Fusion Transformer (LSTFT), skillfully integrating both short-term and long-term information to capture complex dependencies and enhance prediction accuracy. Additionally, we propose a Temporal Attention Mechanism (TAM) that effectively distinguishes between long-term and short-term temporal relationships in LOB data. Our experimental results demonstrate the effectiveness of our approach in accurately forecasting the Limit Order Book in the short term.

An enhanced U-shaped network (EU-Net) based on deep semantic information fusion and edge information guidance is studied to improve the segmentation accuracy of road cracks under hazy conditions. The EU-Net comprises multimode feature fusion, side information fusion and edge extraction modules. The feature and side information fusion modules are applied to fuse deep semantic information with multiscale features. The edge extraction module uses the Canny edge detection algorithm to guide and constrain crack edge information from the neural network. The experimental results show that the method in this work is superior to the most widely used crack segmentation methods. Compared with that of the baseline U-Net, the mIoU of the EU-Net increases by 0.59% and 5.7% on the Crack500 and Masonry datasets, respectively.

Deep learning-based models have emerged as promising tools for multivariate long-term time series forecasting. These models are finely structured to perform feature extraction from time series, greatly improving the accuracy of multivariate long-term time series forecasting. However, to the best of our knowledge, few scholars have focused their research on preprocessing time series, such as analyzing their periodic distributions or analyzing their values and volatility at the global level. In fact, properly preprocessing time series can often significantly improve the accuracy of multivariate long-term time series forecasting. In this paper, using the cross-variable transformer as a basis, we introduce a statistical characteristics space fusion module to preprocess the time series, this module takes the mean and standard deviation values of the time series during different periods as part of the model’s inputs and greatly improves the model’s performance. The Statistical Characteristics Space Fusion Module consists of a statistical characteristics space, which represents the mean and standard deviation values of a time series under different periods, and a convolutional neural network, which is used to fuse the original time series with the corresponding mean and standard deviation values. Moreover, to extract the linear dependencies of the time series variables more efficiently, we introduce three different linear projection layers at different nodes of the model, which we call the Multi-level Linear Projection Module. This new methodology, called the SCSformer, includes three innovations. First, we propose a Statistical Characteristics Space Fusion Module, which is capable of calculating the statistical characteristics space of the time series and fusing the original time series with a specific element of the statistical characteristics space as inputs of the model. Second, we introduce a Multi-level Linear Projection Module to capture linear dependencies of time series from different stages of the model. Third, we combine the Statistical Characteristics Space Fusion Module, the Multi-level Linear Projection Module, the Reversible Instance Normalization and the Cross-variable Transformer proposed in Client in a certain order to generate the SCSformer. We test this combination on nine real-world time series datasets and achieve optimal results on eight of them. Our code is publicly available at https://github.com/qiuyueli123/SCSformer.

Siamese visual trackers based on segmentation have garnered considerable attention due to their high accuracy. However, these trackers rely solely on simple classification confidence to distinguish between positive and negative samples (foreground or background), lacking more precise discrimination capabilities for objects. Moreover, the backbone network excels at focusing on local information during feature extraction, failing to capture the long-distance contextual semantics crucial for classification. Consequently, these trackers are highly susceptible to interference during actual tracking, leading to erroneous object segmentation and subsequent tracking failures, thereby compromising robustness. For this purpose, we propose a Siamese visual segmentation and tracking network with classification-rank loss and classification-aware (Siam2C). We design a classification-rank loss (CRL) algorithm to enlarge the margin between positive and negative samples, ensuring that positive samples are ranked higher than negative ones. This optimization enhances the network’s ability to learn from positive and negative samples, allowing the tracker to accurately select the object for segmentation and tracking rather than being misled by interfering targets. Additionally, we design a classification-aware attention module (CAM), which employs spatial and channel self-attention mechanisms to capture long-distance dependencies between different positions in the feature map. The module enhances the feature representation capability of the backbone network, providing richer global contextual semantic information for the tracking network’s classification decisions. Extensive experiments on the VOT2016, VOT2018, VOT2019, OTB100, UAV123, GOT-10k, DAVIS2016, and DAVIS2017 datasets demonstrate the outstanding performance of Siam2C.

With the increase of communication devices and demands, the problems of high power consumption, tight spectrum resources, and low energy efficiency in the two-layer heterogeneous network are the popular topics, which need to be solved urgently. For the purpose of solving these problems in a two-layer heterogeneous network consisting of femtocell base stations in randomly distributed a macrocell base station, which can also be called the Macrocell/Femtocell two-layer heterogeneous network, the hierarchical clustering algorithm is firstly used to cluster femtocell base stations in accordance with a distance threshold, the spectrum partitioning mechanism and non-orthogonal multiple access technique are combined to obtain spectrum allocation schemes for different users. Then, the modified social network search algorithm is used to simulate the power allocation problem in the two-layer heterogeneous network with system energy efficiency as the objective function. By comparing with the previous algorithms, the proposed algorithm’s superior performance is verified on the test functions. The results show that the proposed method can effectively improve spectrum utilization and reduce interference. The modified social network search algorithm is more robust and widely applicable regarding energy and computational efficiency.

Evidence c-means(ECM) has certain advantages in dealing with uncertainty and imprecision, and it is widely applied to data clustering and image segmentation. However, ECM does not utilize spatial information and unable to recover noise, resulting in poor performance for noisy image segmentation. To address these problems, we propose a dynamic noise self-recovery ECM clustering algorithm with adaptive spatial constraints for image segmentation. The proposed algorithm has the following novelties. Firstly, the non-local spatial information is modified by initializing the noise probability to obtain more reliable spatial information. Secondly, the adaptive constraint factors are constructed by using the absolute difference between the original image and the modified non-local spatial information, which can reduce the sensitivity of the algorithm to noise. Finally, the self-recovery factors are constructed on the basis of the neighborhood belief degrees. And a dynamic anti-noise distance is proposed to replace the Euclidean distance. The dynamic anti-noise distance is more suitable for noise self-recover, enabling noise self-recovery during the iterative process. Extensive experiments on synthetic, natural, SAR and MR images show that the proposed algorithm has good performance for image segmentation.