Deep Belief Network Research Articles

Power system transient stability preventive control optimization method driven by Stacking Ensemble Learning

The dynamic characteristics of the power system are becoming more and more complex, and the difficulty of operation control is increasing. Preventive control is the main means of power system transient stability control. This paper proposes a stacking ensemble learning-driven power system transient stability preventive control optimization method. Firstly, a transient stability assessment model based on Stacking Ensemble Deep Belief Nets (SEDBN) network is established in this research. The performance of weak classifiers is improved by SEDBN’s multi-layer ensemble structure, and the created transient stability estimator can extract diverse features and has better robustness and generalization abilities. Secondly, the trained transient stability estimator is integrated into the Aptenodytes Forsteri Optimization (AFO) algorithm as a “transient stability constraint discriminator”. Finally, with the goal of minimizing the cost of preventive control, an optimization algorithm for the preventive control of power system transient stability driven by SEDBN is established. Simulation results on IEEE 39-bus systems show that the proposed method can achieve highly efficient control solutions.

Dynamic Keystroke Technique for a Secure Authentication System based on Deep Belief Nets

The rapid growth of electronic assessment in various fields has led to the emergence of issues such as user identity fraud and cheating. One potential solution to these problems is to use a complementary authentication method, such as a behavioral biometric characteristic that is unique to each individual. One promising approach is keystroke dynamics, which involves analyzing the typing patterns of users. In this research, the Deep Belief Nets (DBN) model is used to implement a dynamic keystroke technique for secure e-assessment. The proposed system extracts various features from the pressure-time measurements, digraphs (dwell time and flight time), trigraphs, and n-graphs, and uses these features to classify the user's identity by applying the DBN algorithm to a dataset collected from participants who typed free text using a standard QWERTY keyboard in a neutral state without inducing specific emotions. The DBN model is designed to detect cheating attempts and is tested on a dataset collected from the proposed e-assessment system using free text. The implementation of the DBN results in an error rate of 5% and an accuracy of 95%, indicating that the system is effective in identifying users' identities and cheating, providing a secure e-assessment approach.

Adaptive Particle Grey Wolf Optimizer with Deep Learning-based Sentiment Analysis on Online Product Reviews

The increasing use of e-commerce websites and social networks is continually generating an immense amount of data in various forms, such as text, images or sounds, videos, etc. Sentiment analysis (SA) in online product reviews is a method of identifying the overall sentiment of customers about a specific product or service. This study used Natural Language Processing (NLP) and Machine Learning (ML) algorithms to identify and extract opinions and emotions expressed in text. Online reviews are often written in informal language, slang, and dialects, making it difficult for ML models to accurately classify sentiments. In addition, the use of misspelled words or incorrect grammar can further complicate the analysis. The recent developments of Deep Learning (DL) models can be used for the accurate classification of sentiments. This paper presents an Adaptive Particle Grey Wolf Optimizer with Deep Learning Based Sentiment Analysis (APGWO-DLSA) method to accurately classify sentiments in product reviews. Initially, data pre-processing was performed to improve the quality of the product reviews using the word2vec embedding process. For sentiment classification, the proposed method used a Deep Belief Network (DBN) model. Finally, the hyperparameter tuning of the DBN was performed using the APGWO algorithm. An extensive experimental analysis demonstrated the improved results of APGWO-DLSA over other methods, showing a maximum accuracy of 94.77% and 85.31% on the Cell Phones And Accessories (CPAA) and Amazon Products (AP) datasets.

Evaluation of Soil Rocky Desertification in Karst Region Based on Deep Belief Network

Abstract Dynamic features from remote sensing photos may be successfully extracted using deep learning and symmetric network structure, which can then be used to direct them to carry out accurate classification. The DBN model can more effectively extract features from photos since it uses unsupervised learning. It can be reduced to the many symmetric Restricted Boltmann Machines (RBM) training problem. In this paper, a soil rocky desertification (RD) assessment model based on a deep belief network (DBN) is created in light of the complicated influencing aspects of Karst RD risk assessment encompassing several geographical elements. The model builds upon the conventional RBM framework and incorporates the influence layer of related elements as an auxiliary requirement for retrieving Geographic Information System (GIS) score data. Then, in order to forecast the level of soil rocky desertification, it learns the features of many elements. The experimental results show that the proposed model proposed in this paper has better prediction performance and faster convergence speed, and its classification results for different degrees of RD are more consistent with the actual risk assessment results.

Defects Detection of Onboard Cable Termination in EMUs Using Partial Discharge Measurement and SDP-DL Framework

Termination is the most important part of on-board cable, which plays a vital role in ensuring the continuous and reliable power supply to EMUs (electrical multiple units). The maintenance time of EMUs is limited, so the time left for PD (partial discharge) measurement of on-board cable termination is very short, leading to the obvious decreasing of detection accuracy. For addressing this issue, this paper proposed an SDP (symmetrized dot pattern)–DL (deep learning) framework to detect the insulation defects using time series PD pulse signal. First, a PD measurement platform and experimental samples were prepared in laboratory to obtain the time series PD pulse signals of four typical insulation defects. Then, the time series PD signals were converted into SDP images using the proposed parameter optimization method. Finally, the SDP-DL framework was proposed, and three specific and typical methods were utilized, i.e., CNN (convolutional neural network), SAE (stacked auto encoder) and DBN (deep belief network). The results show that the performance of SDP-DBN method is the best, and the insulation defects can be detected with accuracy of 96.1%. In addition, the visualization ability of data increases after SDP transformation of the original PD time series signal.

Automated Video Classification System Driven by Characteristics of Emotional Human Brainwaves Caused by Audiovisual Stimuli

This paper proposes a method of automatically classifying videos into human emotional categories by imitating human neural processes, in which emotion-specific electroencephalography (EEG) characteristics are generated by audiovisual stimulation. In the proposed method, sophisticated emotional features are first extracted from EEG signals that are generated while a subject watches a video, using a sampleattention-based deep neural network encoder. Next, the direct mapping relationship between the extracted emotional EEG features and audiovisual features extracted from the contents of the corresponding videos are learned by a deep belief network. For practical application, the EEG features corresponding to the input video are automatically generated based on the machine’s learned ability without measuring human EEG signals and subsequently applied to a segment-attention-based deep neural network decoder for emotion classification of the video. The experimental results demonstrated that the proposed method significantly outperforms existing methods with an average accuracy of about 95% for classifying four emotional classes of the video. As for automated emotional video classification, our artificial emotional EEG features-based approach obtains competitive performance, comparable to models that directly measure EEG, and can be generalized to various audiovisual datasets.

Torque Ripple Reduction of PMSM Based on Modified DBN-DNN Surrogate Model

Due to the nonlinearity and dead time of the inverter, a great number of harmonic components is introduced into the motor input current, causing the torque ripple of the motor. The paper aims to suppress the torque ripple of a low-speed and high-torque permanent magnet synchronous motor (PMSM) with a two-step stator voltage optimization method, including a high precision harmonic voltage injection (HPHVC) method and a deep belief network (DBN)-deep neural network (DNN) voltage prediction model. Firstly, a closed-loop harmonic current detection system with Type-II Chebyshev filter is adopted to enhance the harmonic current extraction accuracy. Secondly, a harmonic voltage algorithm is designed considering the dq-axis current coupling relationship, which improves the accuracy of harmonic voltage calculation. Thirdly, a dq-axis voltage prediction model is constructed using DBN-DNN algorithm as the surrogate model, which further reduces the current harmonic components caused by the harmonic injection structure and the PI control algorithm, reduces the torque pulsation of the motor, and boosts the dynamic performance of the motor. Finally, the experimental platform is constructed by embedding the proposed strategy into the field oriented control system for reducing the torque pulsation. The experimental results verify the proposed strategy.

Hyperspectral image-aided LiDAR point cloud labeling via spatio-spectral feature representation learning

Urban scene-level 3D point cloud labeling is a very laborious and expensive task compared to images. Conversely however, image processing techniques, deep learning or otherwise are more established and mature. Thus, in a multi-source data environment, the labeling of a point cloud scene via an automated image process as an initial step, followed by a manual human verification process is an effective way to save man hours and cost. With the above as the goal, this study presents a simple but robust spatio-spectral feature representation approach. In this approach, a class-aware band selection and reduction (CBSR) technique is developed for optimal hyperspectral feature representation. A double-branched convolutional Gaussian Bernoulli deep belief network (CGBDBN) is then used for hierarchical spatial feature extraction from LiDAR-derived data and the CBSR data. Using stacked ensemble learning, spatio-spectral features are generated from the two feature streams via a fusion rule and then classified — the results of which are used in labeling a raw 3D LiDAR point cloud through projection. To evaluate this study, extensive experiments were conducted on the IEEE 2018 Houston dataset — the only publicly available dataset with both hyperspectral image (HSI) and 3D point cloud covering the same area — for urban scene classification. The results indicate that the developed CBSR attained comparatively competitive results with state-of-the-art approaches, thus making it a robust spectral feature representation technique. Also, the weight-sharing property, probabilistic modeling, and hierarchical nature of CGBDBN gives our approach the ability to capture high-level contextual features. Furthermore, compared to the spatial- or spectral-only features, the generated spatio-spectral features are more discriminative and significantly aided in improving the proposed model’s efficacy. Overall, the proposed approach, based on the evaluation metrics, is a robust and effective approach for both coarse- and fine-grained raw LiDAR point cloud labeling tasks.

Emotion classification using EEG signal for women safety application based on deep learning

An emotion is a conscious logical response that varies for different situations in women’s life. These mental responses are caused by physiological, cognitive, and behavioral changes. Gender-based violence undermines the participation of women in decision-making, resulting in a decline in their quality of life. More accurate and automatic classification of women’s emotions can enhance human-computer interfaces and security in real time. There are some wearable technologies and mobile applications that claim to ensure the safety of women. However, they rely on limited social action and are ineffective at ensuring women’s safety when and where it is needed. In this work, a novel CDB-LSTM network has been proposed to accurately classify the emotions of women in seven different classes. The electroencephalogram (EEG) offers non-radioactive methods of identifying emotions. Initially, the EEG signals are preprocessed and they are converted into images via Time-Frequency Representation (TPR). A smoothed pseudo-Wigner-Ville distribution (SPWVD) is employed to convert the EEG time-domain signals into input images. Consequently, these converted images are given as input to the Convolutional Deep Belief Network (CDBN) for extracting the most relevant features. Finally, Bi-directional LSTM is used for classifying the emotions of women into seven classes namely: happy, relax, sad, fear, anxiety, anger, and stress. The proposed CDB-LSTM network preserves the high accuracy range of 97.27% in the validation phase. The proposed CDB-LSTM network improves the overall accuracy by 6.20% 32.98% 6.85% and 3.30% better than CNN-LSTM, Multi-domain feature fusion model, GCNN-LSTM and CNN with SVM and DT respectively.

Intelligent Diagnosis Using Continuous Wavelet Transform and Gauss Convolutional Deep Belief Network

Bearing fault diagnosis is of significance to ensure the safe and reliable operation of a motor. Deep learning provides a powerful ability to extract the features of raw data automatically. A convolutional deep belief network (CDBN) is an effective deep learning method. In this article, a novel vibration amplitude spectrum imaging feature extraction method using continuous wavelet transform and image conversion is proposed, which can extract the image features with two-dimensional and eliminate the effect of handcrafted features under low signal-to-noise ratio conditions, different operating conditions, and data segmentation. Then, a novel CDBN with Gaussian distribution is constructed to learn the representative features for bearing fault classification. The proposed method is tested on motor bearing dataset with four and ten classifications. The results have been compared with other methods. The experiment results show that the proposed method has achieved significant improvements and is more effective than the traditional methods.

Stock Market Predictions through Deep Belief Neural Network

Abstract: Owing to the huge potential returns, the proportionate degree of exposure, and the adaptability of the transaction, a big number of people choose to invest their money in stocks. This is due to the fact that stocks come with both of these factors. If investors are able to successfully foresee changes in stock values, they have the opportunity to generate considerable returns from their financial investments. The value of a stock may be affected by a wide variety of variables, including the current state of the market and the macroeconomic environment as a whole, significant events in the economic and social spheres, investor attitude, and management actions made by the firm. Predicting the value of a company's shares has always been considered as the most challenging and essential component of academic research. This view continues to hold true today. Utilizing mathematical and statistical models is a common component of the traditional approaches used for predicting stock values. These methods have been around for quite some time. Despite this, these strategies are not enough for dealing with the intricate dynamics of the stock market, which are always evolving. The fast development of computer technology over the last several decades has led to an increase in the prevalence of the use of machine learning among academics. Therefore, this paper outlines an effective strategy for stock market prediction that utilizes K Nearest Neighbors, Linear Regression, along with Deep Belief Network and Decision making.

IOT Based Transformer Monitoring and Controlling

Abstract: Transformers are critical components of electric powersystems, yet precise fault identificationremains difficult. The study presents a novel transformer defect diagnostic approach based on an Internet of Things (IoT) monitoring sysem and ensemble machine learning (EML). The IoTbased monitoring system is divided into two parts: data measuring subsystemand a data reception subsystem. To begin, the data measuring subsystem measures transformer vibration signals, which are then relayed to the remote server via the data receipt subsystem. Then an EML is proposed that is made up of deep belief networks (DBNs), stacked denoising auto encoders (SDAs) with distinct activation functions, and relevance vector machines (RVMs). DBN sand SDAs are respectively used to extract features from the signals, and RVMs are respectively employed as classifier. In order to ensure efficient of the EML, a novel combination strategy is proposed. A transformer fault diagnosis experiment is performed

Patterns of influence of different landslide boundaries and their spatial shapes on the uncertainty of landslide susceptibility prediction

Some landslide susceptibility modeling uses idealized landslide points or buffer circles as landslide boundaries, adding uncertainty to the susceptibility modeling. However, landslide boundaries and their spatial shapes are typically presented as irregular polygonal surfaces, such as semicircles and bumps. To study the influence of different landslide boundaries on modeling uncertainty, 370 landslides and 11 environmental factors in Ruijin were chosen in order to establish landslide boundaries and their frequency ratio correlations with environmental factors. Then, these borders were formed, utilizing, respectively, landslide points, buffer circles, and precisely encoded and drawn polygons. Then, models like Point, Circle, and Polygon-based DBN and RF were built using deep belief network (DBN) and random forest (RF). Finally, the distribution pattern of the susceptibility index and its variability were used, along with the receiver operating characteristic (ROC) accuracy, to analyze the modeling uncertainty. The results indicate that: (1) while correct landslide polygon borders are more successful in ensuring modeling accuracy and dependability, using landslide points or buffer circles as boundaries can increase modeling uncertainty. (2) but the. (3) in the absence of precise landslide borders, the landslide susceptibility results derived by employing points and buffer circles as landslide barriers can reflect the spatial distribution pattern of landslide likelihood in the studied area as a whole.

A robust deep neural network framework for the detection of diabetes

Significant developments occurred in numerous industries and fields during the digital age (1997–2006). One industry that has seen similar changes is the healthcare sector. Big data has primarily come from the healthcare sector since the 1990s. The outcomes of data analysis and dissemination have boosted healthcare and awareness. Perspectives and insights are the outcomes. The most important public health issue has been identified as diabetes and its repercussions. Based on patient medical imaging and records gathered, various techniques have been used to forecast diabetic complications. The technology of data mining has not been applied with much effort. This technique requires unstructured medical records, data entry, and output. Numerous methods have been employed to foresee diabetes problems. This study employs a deep learning technique to create a healthcare system to categorize and forecast the development of diabetes mellitus (Type 2). The Deep Belief Network, which includes the data collecting, pre-training, and classification processes of forecasting diabetes, is used to predict the complications of diabetes mellitus. The diabetic data set was subjected to the proposed DBN approach, which had an accuracy of 81.25%. Compared to other machine learning techniques, the suggested method produces results with higher accuracy.

Review of Research on Fault Diagnosis of Rolling Bearings Based on Deep Learning

Deep learning has powerful capabilities in deep feature extraction and expression, and has been successfully applied in equipment fault diagnosis, overcoming the shortcomings of traditional diagnostic methods that rely on expert experience. It can save costs while improving diagnostic accuracy. This article briefly introduces three commonly used neural networks: Deep Belief Networks (DBN), Convolutional Neural Networks (CNN), and Long Short-Term Memory Networks (LSTM), and points out the problems in rolling bearing diagnosis and analyzes future development directions.

Multi‐disease classification model using deep neural network and Strassen's rectilinear fine‐tune bouncing training algorithm

AbstractA deep neural network (DNN) is being used in the healthcare industry to improve care delivery at a lower cost and in less time. A DNN is well‐known for its diagnostic applications. However, it is also increasingly being utilized to guide healthcare management decisions. DNNs have been applied to real‐life disease classification problems. The performances of DNN are improved by reducing the training time, performing fast classification, and improving the parameters such as accuracy and error rate, and so forth. We have proposed a hybrid DNN named Strassen's rectilinear fine‐tune bouncing training (SRFBT) Algorithm by combining Strassen's theorem and bouncing training algorithm. The simulation result shows that the SRFBT algorithm outperforms both the DNN, support vector machine, radial basis function network and deep belief network algorithms in terms of network training, testing time and accuracy on various UCI machine learning healthcare datasets. The SRFBT has improved performance with a minimum average training time of 43.6249 and a maximum accuracy of 96.86.

Adaptive-CSSA: adaptive-chicken squirrel search algorithm driven deep belief network for student stress-level and drop out prediction with MapReduce framework

Adaptive-CSSA: adaptive-chicken squirrel search algorithm driven deep belief network for student stress-level and drop out prediction with MapReduce framework

Temperature modeling of wave rotor refrigeration process based on elastic net variable selection and deep belief network

Optimal operation modeling plays an important role in wave rotor refrigeration process; however, considering covariance among multiple variables and high nonlinearity in wave rotor refrigeration process, it becomes more and more difficult to establish an accurate operation modeling using first-principles methods. This study proposed a novel modeling algorithm for the temperature parameter of the wave rotor refrigeration process based on elastic net and dingo optimization deep belief network (Enet-DOA-DBN). Firstly, to determine the correlation between the input variables and reduce the dimension of the input variables, the elastic net (Enet) algorithm is used to select the input variables that are irrelevant to the temperature parameter of the wave rotor refrigeration process. In this way, the covariance between multiple variables is eliminated and the model structure is simplified. Secondly, in order to improve the generalization of the temperature parameter model, a deep belief network (DBN) deep learning is proposed for modeling the temperature parameter of the wave rotor refrigeration process. Considering that the numerous hyperparameters of DBN algorithm have a great impact on the training and prediction results, the hyperparameters are optimized by the dingo optimization algorithm (DOA). The proposed Enet-DOA-DBN algorithm is validated by simulation using the benchmark data sets and the wave rotor refrigeration industrial process data sets. The simulation results show that the proposed Enet-DOA-DBN algorithm has good generalization ability, meanwhile it can effectively implement variable selection and simplify the model structure.

Decision-Refillable-Based Shared Feature-Guided Fuzzy Classification for Personal Thermal Comfort

Different types of buildings in different climate zones have their own design specifications and specific user populations. Generally speaking, these populations have similar sensory feedbacks in their perception of environmental thermal comfort. Existing thermal comfort models do not incorporate personal thermal comfort models for specific populations. In terms of an algorithm, the existing work constructs machine learning models based on an established human thermal comfort database with variables such as indoor temperature, clothing insulation, et al., and has achieved satisfactory classification results. More importantly, such thermal comfort models often lack scientific interpretability. Therefore, this study selected a specific population as the research object, adopted the 0-order Takagi–Sugeno–Kang (TSK) fuzzy classifier as the base training unit, and constructed a shared feature-guided new TSK fuzzy classification algorithm with extra feature compensation (SFG-TFC) to explore the perception features of the population in the thermal environment of buildings and to improve the classification performance and interpretability of the model. First, the shared features of subdatasets collected in different time periods were extracted. Second, the extra features of each subdataset were independently trained, and the rule outputs corresponding to the key shared features were reprojected into the corresponding fuzzy classifiers. This strategy not only highlights the guiding role of shared features but also considers the important compensation effect of extra features; thereby, improving the classification performance of the entire classification model. Finally, the least learning machine (LLM) was used to solve the parameters of the “then” part of each basic training unit, and these output weights were integrated to enhance the generalization performance of the model. The experimental results demonstrate that SFG-TFC has better classification performance and interpretability than the classic nonfuzzy algorithms support vector machine (SVM) and deep belief network (DBN), the 0-order TSK, and the multilevel optimization and fuzzy approximation algorithm QI-TSK.

Composite Fault Diagnosis of Aviation Generator Based on EnFWA-DBN

Because of the existence of composite faults, which consist of both short-out and eccentricity faults, the characteristics of the output voltage and internal magnetic field of aviation generators are less different than those of single short-out faults; this causes the eccentricity fault to be difficult to identify. In order to solve this problem, this paper proposes a fault diagnosis method using an enhanced fireworks algorithm (EnFWA) to optimize a deep belief network (DBN). The aviation generator model is built according to the finite element method (FEM), whereas the output of different combinations of composite faults are obtained using simulations. The EnFWA algorithm is used to train and optimize the DBN network to obtain the best structure. Meanwhile, an extreme learning machine (ELM) classifier performs fault diagnosis and classification on the test data. The diagnosis results show that a pinpoint accuracy can be achieved using the proposed method in the diagnosis of composite faults in aviation generators.