To ensure the operation safety of industrial processes, process monitoring techniques have been receiving considerably increasing attention from both academia and industry. The advent of deep learning has revolutioniz...
详细信息
To ensure the operation safety of industrial processes, process monitoring techniques have been receiving considerably increasing attention from both academia and industry. The advent of deep learning has revolutionized data-driven process monitoring by effectively dealing with the inherent nonlinearity of industrial processes. In this work, a gated recurrent unit-based stacked sparse autoencoder with attention mechanism (GRU-SSAE-AM) model is developed for process monitoring. This model leverages the power of recurrent neural networks in conjunction with the sparsestackedautoencoder to tackle the temporal and nonlinear features present in process data. To enhance the model's ability to distill critical information indicative of process status, an attention mechanism is naturally integrated within the encoder-decoder framework. Experiments on an industrial benchmark of Tennessee Eastman process (TEP) and a real blast furnace ironmaking process (BFIP) are carried out to verify the capability and effectiveness of the proposed GRU-SSAE-AM based process monitoring method by comparison with other related methods.
Early identification of wheel defects can prevent serious damage to railways, considerably lowering maintenance costs for both railway administrations and rolling stock operators. Within this context, an unsupervised ...
详细信息
Early identification of wheel defects can prevent serious damage to railways, considerably lowering maintenance costs for both railway administrations and rolling stock operators. Within this context, an unsupervised methodology based on artificial intelligence techniques is presented, which allows the detection and classification of out-of-roundness damage wheels, such as wheel flats and polygonal wheels, based on dynamic responses induced on the track by crossing freight railway vehicles. The methodology involves the following steps: (i) data collection and pre-processing, (ii) feature extraction (iii) data fusion and (iv) feature discrimination. In the first phase, an FFT algorithm is applied to the acceleration track responses. Then, features are extracted after training a stacked sparse autoencoder, in which the main features of the responses are obtained after a compression stage using an encoder network. This lower dimensional layer forces the model to learn a compression of the input data. Then, these extracted features are merged using the Mahalanobis distance, which enhances the sensitivity to the damage recognition. Posteriorly, an outlier analysis is performed to distinguish a healthy wheel from a defective one and a cluster analysis to discriminate the two types of out-of-roundness (OOR) damage and classify the severity of each type of damage.
Combining deep learning (DL) and near-infrared (NIR) spectroscopy provides a new direction for revolutionizing agricultural quality assessment methods. Typical DL frameworks often require a large number of samples, an...
详细信息
Combining deep learning (DL) and near-infrared (NIR) spectroscopy provides a new direction for revolutionizing agricultural quality assessment methods. Typical DL frameworks often require a large number of samples, and the acquisition of NIR spectral samples is usually constrained by various factors. For this reason, this paper combines NIR spectral feature selection with improved stacked sparse autoencoder (ISSAE) DL to build a classification model suitable for small NIR samples and validate the model's performance with maize seed variety identification. An improved backward-spaced partial least squares method (IBIPLS) is used to select the feature wavelengths. IBIPLS keeps the number of spectral segments traveled within tolerable bounds. The correlation coefficient to the root mean square error of the cross-validation ratio was used to evaluate the merit of the extracted bands to achieve effective feature wavelength selection. ISSAE blends unsupervised feature learning with refined network parameter tuning, using a cross-entropy function as a supervised penalty term. A total of 276 NIR spectral samples from 13 maize varieties were collected, and IBIPLS selected 789 spectral features among 1845 spectral variables. The sparsity parameter and the number of hidden layer nodes were adjusted to obtain the best performance during the training of ISSAE. The model performs best when the sparsity parameter is 0.1 and the number of hidden layer nodes is 130, with a test accuracy of 98.91%. Comparing this model with models such as k-nearest neighbor (KNN), support vector machine (SVM), back propagation (BP) neural network, and 1-D CNN, it performs the best. The above test results show that the method proposed in this paper can achieve accurate and fast classification of maize varieties with good reliability and generalization performance, which provides a new idea for NIR small sample classification using DL.
Network security situation awareness enables networks to actively and effectively defend against network attacks, relying on the extraction of network situation elements as an initial and decisive step. In existing st...
详细信息
Network security situation awareness enables networks to actively and effectively defend against network attacks, relying on the extraction of network situation elements as an initial and decisive step. In existing studies, the stacked sparse autoencoder (SSAE) has been employed to extract features from unlabeled network flows. However, obtaining the optimal hyperparameter combination is challenging due to its numerous hyperparameters. To address this issue, we propose a novel approach named DBO-SSAE that leverages dung beetle optimization (DBO) to select the optimal hyperparameters for SSAE automatically. Applied to the well-known UNSW-NB15 dataset, our model yields an optimal feature subset, which is evaluated across various binary classifiers with different metrics. Experimental results demonstrate that our approach improvesaccuracyandF1-measureby 0.2% to 1.5% while reducing the false negative rate(FNR)and false positive rate(FPR) by 0.06% to 7%, surpassing other feature extraction methods on the same classifier for the UNSW-NB15 dataset. Particularly, in conjunction with a lightweight bidirectional longs hort-term memory (BiLSTM), our model achieves metrics of 98.84%accuracy, 98.96%F1-measure, 1.86%FNR, and 0.6%FPR. This study could provide novel insights into the effective representation of network situation elements and lay the groundwork for a high-efficiency intrusion detection system
Cost management plays a vital role in ensuring the successful execution of different engineering projects, with precise costing serving as the cornerstone of effective cost management strategies. Recently, the machine...
详细信息
Cost management plays a vital role in ensuring the successful execution of different engineering projects, with precise costing serving as the cornerstone of effective cost management strategies. Recently, the machine learning technique offers an accurate and efficient method for forecasting construction expenses, introducing a novel approach to cost accounting other than conventional calculation techniques. This paper provides an overview of the current research landscape in the realm of cost prediction utilizing machine learning, also addresses some new research focuses and limitations. By utilizing highway tunnel engineering as a case study, this study employs an Interpretative Structural Model (ISM) to analyze the primary factors influencing construction costs. Subsequently, a construction cost prediction model is developed, leveraging a stacked sparse autoencoder (SSAE) network within a deep learning framework. Last, the proposed model is trained using real construction projects as samples. Results show that there are some good prediction outputs with a remarkably low mean absolute percentage error of 0.71%. Thereby, they verified the identifying precision of key influencing factors and the reliability of the cost prediction model.
Six-phase transmission lines have the capability to address the continually evolving power demand. It allows upgrading the power transfer capability of the prevailing three-phase double circuit line without major chan...
详细信息
Six-phase transmission lines have the capability to address the continually evolving power demand. It allows upgrading the power transfer capability of the prevailing three-phase double circuit line without major changes in the transmission corridors. However, the operational performance of any six-phase system is highly dependent on its protection scheme. The possibility of larger number of faults in six-phase system complicates the protection task. Furthermore, the harmonics intrusion arising because of nonlinear loading compromises the reliability of the conventional threshold-based protection schemes. In this regard, this article addresses the above-mentioned challenges by developing a protection scheme based on the hybrid frameworks of bat algorithm and stacked sparse autoencoder-deep neural network (SAE-DNN). To overcome the limitation of SAE-DNN regarding optimal selection of architecture and tuning parameters, the selection task has been formulated as an optimization problem and solved using Bat algorithm. The use of raw voltage and current signals as input to the SAE-DNN reduces the overall complexity of the protection scheme. The efficacy of the proposed scheme has been validated for all 120 types of faults under varying operating, loading and fault scenarios. Furthermore, the proposed scheme has been validated for practical settings by performing real-time simulations on OPAL-RT digital simulator.
Web robots are automated computer programs that can be exploited for benign and malicious activities such as website indexing, monitoring, or unauthorized content scraping and scalping. Several methods are available t...
详细信息
Web robots are automated computer programs that can be exploited for benign and malicious activities such as website indexing, monitoring, or unauthorized content scraping and scalping. Several methods are available to detect automated web robots through their footprints and behaviors. Although the accuracy and efficiency of existing methods depend highly on the labeled web log data, countless web requests are generated daily with the help of web robots. Exhaustive and accurate manual labeling of reconstructed sessions is time-consuming and challenging. Further, effective detection of web robots is more challenging with unlabeled or partially labeled data. To address the aforementioned issues, we reformulated web robot detection as a semi-supervised learning problem. In this paper, we propose a deep learning-based Semi-Supervised stacked sparse autoencoder (Web-S4AE) for web robot detection. The proposed model uses content-based features and features extracted from web access log data to effectively classify web robots. The experiments were conducted on publicly available web log data from a library and information portal to assess the performance of Web-S4AE. The Web-S4AE model was trained in two phases. The first phase;comprises training the model with unlabeled data to extract the hidden information, and in the second phase, the model is fine-tuned using labeled data. The results suggest that incorporating more unlabeled data can significantly improve the classifier's performance. The Web-S4AE model's performance was also compared with other models such as the Decision Tree (DT), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Multi-Layer Perceptron (MLP).
Connected healthcare systems face more and more cyber attacks recently. With the development of technology, people use intrusion detection systems (IDS) to detect network attacks and achieve effective results. The exi...
详细信息
Connected healthcare systems face more and more cyber attacks recently. With the development of technology, people use intrusion detection systems (IDS) to detect network attacks and achieve effective results. The existing methods do not take into account the limited storage and computing power of wireless devices on connected healthcare systems. IDSs in the connected healthcare systems need to be real-time and lightweight. This paper proposes an IDS method based on stacked sparse autoencoder (sSAE) and sliced gated recurrent unit (SGRU). The sSAE reduces the dimensionality of the original traffic data and the memory required to calculate the covariance matrix. We slice the original data and input the processed data into the SGRU networks which are paralleled. Therefore, SGRU networks achieve real-time response. The method uses the AWID dataset. The experimental results show that our scheme is better than deep neural network (DNN), recurrent neural network (RNN), long short-term memory (LSTM) and other methods. Especially, the F1-score of the method is 2-5% higher than existing schemes, the detection time is 5-13 times shorter than other solutions, and the model size is smaller than the size of other models by at least 4 times.
In recent years, abnormal condition monitoring has provoked vast amount o attention and research from multiple disciplines. Especially in the microservice architecture based on the cloud platform, abnormal condition m...
详细信息
In recent years, abnormal condition monitoring has provoked vast amount o attention and research from multiple disciplines. Especially in the microservice architecture based on the cloud platform, abnormal condition monitoring has emerged as a powerful instrument for improving the stability of service. However, along with the development of geometric growth in the number of microservice, due to the lack of systematic methods, it is difficult to effectively analyze and extract valuable information from the large amount of data collected, the abnormal monitoring in cloud platform O&M is facing new difficulties and challenges. Traditional operation and maintenance work mainly relies on human experience to analyze a large number of indicators to determine whether there is a failure, which is very inefficient and relies heavily on expert opinions. To address these problems, in this paper, we propose a abnormal condition monitoring method based on stacked sparse autoencoder for cloud platform. Our method aims to model the operating state based on the historical experience with deep autoencoder, then the constructed model is used to analyze the current status of the service and predict the possible abnormal condition. Experimental results on real-world datasets demonstrate the effectiveness of our method compared to the traditional methods.
The rate of development of technology and its associated security issues are increasing in integrated circuit industry. This provides a space for implanting dormant nature of threats, named Hardware Trojan (HT) in var...
详细信息
The rate of development of technology and its associated security issues are increasing in integrated circuit industry. This provides a space for implanting dormant nature of threats, named Hardware Trojan (HT) in various process of integrated circuits (IC) design. The impact of HT's emanates the encrypted signals, privacy disruption, performance degradation or denial of service. The threat models are unimaginable and it can be intruded at any stage which complicates the HT detection process. Therefore, a deep learning-based malicious module identification method is proposed in this work by implementing stackedautoencoder and stacked sparse autoencoder model. The simulation results shows that the proposed stacked sparse autoencoder outperforms the best in detecting the malicious modifications with an average accuracy of 97.53%, true positive rate of 93% and moreover the true negative rate achieved is 98.14% which proves the effectiveness of sparsity nature in extracting suitable features in the proposed schemes.
暂无评论