Long Short-term Memory Network Model Research Articles

Application of recurrent neural networks for modeling and control of a quadruple‐tank system

AbstractThe quadruple tank (QT) system consists of four interacting tanks and can switch between the minimum and non‐minimum phase behavior with changes in the positions of pump valves and is considered a benchmark control problem. In the present study, long‐short term memory (LSTM), a type of recurrent neural networks (RNN) is designed for the benchmark QT system based on the model‐based control framework. Random input–output sequences are generated from the white box model of the QT system to train an LSTM network model. The LSTM network is tuned by adjusting its hyperparameters such as the number of hidden layers, hidden units, and epochs to minimize the prediction error on the test data. The trained model is cross validated both during and after training to avoid overfitting. Once a reasonably reliable model is obtained, another LSTM network is trained for use as a controller. The network architecture is constantly modified till the controller is able to track the test setpoints with minimum error. This procedure is repeated with a gated recurrent unit (GRU) network and the servo and regulatory response of both the network models and controller are evaluated in terms of standard performance measure namely root mean square error (RMSE), integral square error (ISE), and control effort (CE). It is observed that the controller designed based on RNN performs better than a conventional centralized controller.

Construction and Application of LSTM-Based Prediction Model for Tunnel Surrounding Rock Deformation

Tunnel surrounding rock deformation is a significant issue in tunnel construction and maintenance and has garnered attention from both domestic and international scholars. Traditional methods of predicting tunnel surrounding rock deformation involve fitting monitoring and measuring data, which is a laborious and resource-intensive process with low accuracy when predicting data with significant fluctuations. A deep learning approach can improve monitoring efficiency and accuracy while reducing labor costs. In this study, taking an actual tunnel project as an example, a long short-term memory (LSTM) network model was constructed based on the recurrent neural network algorithm with deep learning to model and analyze the tunnel monitoring and measurement data, and the model was used to analyze and predict the vault settlement of the tunnel. LSTM is a type of artificial neural network architecture that is commonly used in deep learning applications for sequence prediction tasks, such as natural language processing, speech recognition, and time-series forecasting. In predicting data with smaller fluctuations, the maximum error is 4.76 mm, the minimum error is 0.03 mm, the root mean square error is 2.64, and the coefficient of determination is 0.98. In predicting data with larger fluctuations, the maximum error is 8.32 mm, the minimum error is 0.13 mm, the root mean square error is 4.42, and the coefficient of determination is 0.88. The average error of the LSTM network model is 2.16 mm. With the growth of the prediction period, the prediction results become more and more stable and closer to the actual vault settlement, which provides a reliable reference for introducing the LSTM prediction method with deep learning to tunnel construction and promoting tunnel construction safety.

Ship Anomalous Behavior Detection Using Clustering and Deep Recurrent Neural Network

In this study, we propose a real-time ship anomaly detection method driven by Automatic Identification System (AIS) data. The method uses ship trajectory clustering classes as a normal model and a deep learning algorithm as an anomaly detection tool. The method is divided into three main steps: (1) quality maintenance of the original AIS data, (2) extraction of normal ship trajectory clusters using Density-Based Spatial Clustering of Applications with Noise (DBSCAN), in which a segmented improved Dynamic Time Warping (DTW) algorithm is used to measure the degree of trajectory similarity, (3) the clustering results are used as a normative model to train a Bi-directional Gated Recurrent Unit (BiGRU) recurrent neural network, which is used as a trajectory predictor to achieve real-time ship anomaly detection. Experiments were conducted using real AIS data from the port of Tianjin, China. The experimental results are manifold. Firstly, the data pre-processing process effectively improves the quality of raw AIS data. Secondly, the ship trajectory clustering model can accurately classify the traffic flow of different modes in the sea area. Moreover, the trajectory prediction result of the BiGRU model has the smallest error with the actual ship trajectory and has a better trajectory prediction performance compared with the Long Short-Term Memory Network model (LSTM) and Gated Recurrent Unit (GRU). In the final anomaly detection experiment, the detection accuracy and timeliness of the BiGRU model are also higher than LSTM and GRU. Therefore, the proposed method can achieve effective and timely detection of ship anomalous behaviors in terms of position, heading and speed during ship navigation, which provides insight to enhance the intelligence of marine traffic supervision and improve the safety of marine navigation.

End-to-end learning with interpretation on electrohysterography data to predict preterm birth

Prediction of preterm birth is a difficult task for clinicians. By examining an electrohysterogram, electrical activity of the uterus that can lead to preterm birth can be detected. Since signals associated with uterine activity are difficult to interpret for clinicians without a background in signal processing, machine learning may be a viable solution. We are the first to employ Deep Learning models, a long-short term memory and temporal convolutional network model, on electrohysterography data using the Term–Preterm Electrohysterogram database. We show that end-to-end learning achieves an AUC score of 0.58, which is comparable to machine learning models that use handcrafted features. Moreover, we evaluate the effect of adding clinical data to the model and conclude that adding the available clinical data to electrohysterography data does not result in a gain in performance. Also, we propose an interpretability framework for time series classification that is well-suited to use in case of limited data, as opposed to existing methods that require large amounts of data. Clinicians with extensive work experience as gynaecologist used our framework to provide insights on how to link our results to clinical practice and stress that in order to decrease the number of false positives, a dataset with patients at high risk of preterm birth should be collected. All code is made publicly available.

ResNext50 based convolution neural network-long short term memory model for plant disease classification

ResNext50 based convolution neural network-long short term memory model for plant disease classification

Research on migraine time-series features classification based on small-sample functional magnetic resonance imaging data

The extraction of neuroimaging features of migraine patients and the design of identification models are of great significance for the auxiliary diagnosis of related diseases. Compared with the commonly used image features, this study directly uses time-series signals to characterize the functional state of the brain in migraine patients and healthy controls, which can effectively utilize the temporal information and reduce the computational effort of classification model training. Firstly, Group Independent Component Analysis and Dictionary Learning were used to segment different brain areas for small-sample groups and then the regional average time-series signals were extracted. Next, the extracted time series were divided equally into multiple subseries to expand the model input sample. Finally, the time series were modeled using a bi-directional long-short term memory network to learn the pre-and-post temporal information within each time series to characterize the periodic brain state changes to improve the diagnostic accuracy of migraine. The results showed that the classification accuracy of migraine patients and healthy controls was 96.94%, the area under the curve was 0.98, and the computation time was relatively shorter. The experiments indicate that the method in this paper has strong applicability, and the combination of time-series feature extraction and bi-directional long-short term memory network model can be better used for the classification and diagnosis of migraine. This work provides a new idea for the lightweight diagnostic model based on small-sample neuroimaging data, and contributes to the exploration of the neural discrimination mechanism of related diseases.

Landmark tracking in liver US images using cascade convolutional neural networks with long short-term memory

Accurate tracking of anatomic landmarks is critical for motion management in liver radiation therapy. Ultrasound (US) is a safe, low-cost technology that is broadly available and offer real-time imaging capability. This study proposed a deep learning-based tracking method for the US image-guided radiation therapy. The proposed cascade deep learning model is composed of an attention network, a mask region-based convolutional neural network (mask R-CNN), and a long short-term memory (LSTM) network. The attention network learns a mapping from an US image to a suspected area of landmark motion in order to reduce the search region. The mask R-CNN then produces multiple region-of-interest proposals in the reduced region and identifies the proposed landmark via three network heads: bounding box regression, proposal classification, and landmark segmentation. The LSTM network models the temporal relationship among the successive image frames for bounding box regression and proposal classification. To consolidate the final proposal, a selection method is designed according to the similarities between sequential frames. The proposed method was tested on the liver US tracking datasets used in the medical image computing and computer assisted interventions 2015 challenges, where the landmarks were annotated by three experienced observers to obtain their mean positions. Five-fold cross validation on the 24 given US sequences with ground truths shows that the mean tracking error for all landmarks is 0.65 ± 0.56 mm, and the errors of all landmarks are within 2 mm. We further tested the proposed model on 69 landmarks from the testing dataset that have the similar image pattern with the training pattern, resulting in a mean tracking error of 0.94 ± 0.83 mm. The proposed deep-learning model was implemented on a graphics processing unit (GPU), tracking 47–81 frames s−1. Our experimental results have demonstrated the feasibility and accuracy of our proposed method in tracking liver anatomic landmarks using US images, providing a potential solution for real-time liver tracking for active motion management during radiation therapy.

Construction method of hybrid quantum long-short term memory neural network for image classification

Long-short term memory (LSTM) neural network solves the problems of long-term dependence, gradient disappearance and gradient explosion by introducing memory units, and is widely used in time series analysis and prediction. Combining quantum computing with LSTM neural network will help to improve its computational efficiency and reduce the number of model parameters, thus significantly improving the performance of traditional LSTM neural network. This paper proposes a hybrid quantum LSTM (hybrid quantum long-short term memory, HQLSTM) network model that can be used to realize the image classification. It uses variable quantum circuits to replace the nerve cells in the classical LSTM network to realize the memory function of the quantum network. At the same time, it introduces Choquet integral operator to enhance the degree of aggregation between data. The memory cells in the HQLSTM network are composed of multiple variation quantum circuits (VQC) that can realize different functions. Each VQC consists of three parts: the coding layer, which uses angle coding to reduce the complexity of network model design; the variation layer, which is designed with quantum natural gradient optimization algorithm, so that the gradient descent direction does not target specific parameters, thereby optimizing the parameter update process and improving the generalization and convergence speed of the network model; the measurement layer, which uses the Pauli Z gate to measure, and the expected value of the measurement result is input to the next layer to extract useful information from the quantum circuit. The experimental results on the MNIST, FASHION-MNIST and CIFAR datasets show that the HQLSTM model achieves higher image classification accuracy and lower loss value than the classical LSTM model and quantum LSTM model. At the same time, the network space complexity of HQLSTM and quantum LSTM are significantly reduced compared with the classical LSTM network.

Multi-Modal Point-of-Care Diagnostics for COVID-19 Based on Acoustics and Symptoms.

The COVID-19 pandemic has highlighted the need to invent alternative respiratory health diagnosis methodologies which provide improvement with respect to time, cost, physical distancing and detection performance. In this context, identifying acoustic bio-markers of respiratory diseases has received renewed interest. In this paper, we aim to design COVID-19 diagnostics based on analyzing the acoustics and symptoms data. Towards this, the data is composed of cough, breathing, and speech signals, and health symptoms record, collected using a web-application over a period of twenty months. We investigate the use of time-frequency features for acoustic signals and binary features for encoding different health symptoms. We experiment with use of classifiers like logistic regression, support vector machines and long-short term memory (LSTM) network models on the acoustic data, while decision tree models are proposed for the symptoms data. We show that a multi-modal integration of inference from different acoustic signal categories and symptoms achieves an area-under-curve (AUC) of 96.3%, a statistically significant improvement when compared against any individual modality ([Formula: see text]). Experimentation with different feature representations suggests that the mel-spectrogram acoustic features performs relatively better across the three kinds of acoustic signals. Further, a score analysis with data recorded from newer SARS-CoV-2 variants highlights the generalization ability of the proposed diagnostic approach for COVID-19 detection. The proposed method shows a promising direction for COVID-19 detection using a multi-modal dataset, while generalizing to new COVID variants.

Research on air quality prediction based on improved long short-term memory network algorithm.

Air quality is changing due to the influence of industry, agriculture, people's living activities and other factors. Traditional machine learning methods generally do not consider the time series of the data itself and cannot handle long-range dependencies, thus ignoring information relevant to the predicted items and affecting the accuracy of air quality predictions. Therefore, an attention mechanism is introduced based on the long short term memory network model (LSTM), which attenuates unimportant information by controlling the proportion of the weight distribution. Finally, an integrated lightGBM+LSTM-attention model was constructed based on the light gradient boosting machine (lightGBM), and the prediction results were compared with those of 11 models. The experimental results show that the integrated model constructed in this article performs better, with the coefficient of determination (R2) of prediction accuracy reaching 0.969 and the root mean square error (RMSE) improving by 5.09, 4.94, 4.85 and 4.0 respectively compared to other models, verifying the superiority of the model.

Instantaneous prediction of irregular ocean surface wave based on deep learning

Ocean surface wave environment due to its randomness has mostly been analyzed and understood from a statistical way. Accurate and reliable wave prediction is the basic guarantee for ocean transportation and offshore operation. However, the significant nonlinearity of the ocean surface wave in real sea will result in a simulation and prediction challenge. To solve the problem, this paper proposes a Deep Learning instantaneous prediction model based on a Convolutional Neural Network-Long Short Term Memory model (CNN-LSTM) and wavelet function proven as the activation function of the neural network model to strengthen the fitting of the nonlinear mapping relationship between waves appearing in two successive different time periods. And the results shows that the wavelet activation function has better generalization ability than the traditional activation function for irregular wave data. In the numerical verification test, the two-parameter spectrum of International Towing Tank Conference (ITTC) and the combined model of extreme wave and random wave are used to simulate the irregular wave. The standardized time series of free surface elevation is made into a data set for model training and verification. In order to improve the performance of the model as accurate as possible, main parameters of the model are investigated in this paper. The optimal parameters are obtained and the model tuning suggestion is developed for future studies.

A DEEP LEARNING PROGRAM FOR PREDICTING SAP FLOW OF LARIX OLGENSIS

Plant sap flow is crucial to understanding plant transpiration, plant hydraulic functioning and physiological properties. In this study, a method for predicting trunk sap flow of Larix olgensisusing deep learning was proposed. The method is based on the combined use of Long-short term memory network (LSTM) and transformer model, noted as LSTM-transformer model. The experimental results show that the proposed method provides more accurate prediction quality in terms of correlation coefficient (R2), root mean square error (RMSE) and mean absolute error (MAE), compared to the state of the art forecast methods such as BP, DNN, LSTM, and transformer models.

Study of SOC Estimation by the Ampere-Hour Integral Method with Capacity Correction Based on LSTM

The estimation of the state of charge (SOC) of a battery’s power is one of the key technologies in a battery management system (BMS). As a common SOC estimation method, the traditional ampere-hour integral method regards the actual capacity of the battery, which is constantly changed by the usage conditions and environment, as a constant for calculation, which may cause errors in the results of SOC estimation. Considering the above problems, this paper proposes an improved ampere-hour integral method based on the Long Short-Term Memory (LSTM) network model. The LSTM network model is used to obtain the actual battery capacity variation, replacing the fixed value of battery capacity in the traditional ampere-hour integral method and optimizing the traditional ampere-hour integral method to improve the accuracy of the SOC estimation method. The experimental results show that the errors of the results obtained by the improved ampere-hour integral method for the SOC estimation are all less than 10%, which proves that the proposed design method is feasible and effective.

Stacked machine learning algorithms and bidirectional long short-term memory networks for multi-step ahead streamflow forecasting: A comparative study

Prediction of river flow rates is an essential task for both flood protection and optimal water resource management. The high uncertainty associated with basin characteristics, hydrological processes, and climatic factors affecting river flows make streamflow prediction a very challenging problem. These reasons, together with the increasingly wide availability of data relating to flow rates and rainfall, frequently lead to a preference for data-driven models over physically based or conceptual forecasting models.This study shows the results of an in-depth comparison between two different daily streamflow prediction models: a novel simpler model based on the stacking of the Random Forest and Multilayer Perceptron algorithms, using the Elastic Net algorithm as meta-learner, and a more complex model based on bi-directional Long Short-Term Memory (LSTM) networks. Bayesian optimization was employed in selecting hyperparameters. The two prediction models were compared through the analysis of four different case studies: the Bacchiglione River, the Raccoon River, the Wilson River, and the Trent River.The two models showed comparable forecasting capabilities. In the most favourable cases, both models demonstrated very high accuracy (R2 greater than 0.93, Mean absolute percentage error approximately equal to 10%). The stacked model outperformed the bi-directional LSTM network model in several cases in predicting peak flow rates but was less accurate in forecasting low flow rates. In addition, its computation times are significantly shorter. The prediction accuracy of both models decreased as the forecast horizon increased. The length of the time series plays an essential role in developing models with satisfactory forecasting capabilities. In this study, the effectiveness of forecasting models was not influenced by the river regime. However, a high variance of the input dataset and a large number of outliers in the time series can reduce the accuracy of prediction models.

Forecasting of Total Column Ozone using Regression Analysis and LSTM-RNN Machine Learning Approach

Objectives: Atmospheric Ozone plays an important role in global climate change, human health and environmental conditions. Accurate and timely prediction of variation in Total Column Ozone (TCO) concentration is very important for both climatology and environment. In this context, the present study aims to utilize the advancement of Artificial Intelligence in Machine learning and a conventional method to develop models for the prediction of TCO concentration using historical time series data over a tropical region, India. Methods: Long Short-Term Memory (LSTM) deep learning networks are very useful in classifying, processing and making predictions based on time series data. In this work, Multiple Linear Regression (MLR) model, Long Short-Term Memory (LSTM) models are developed for forecasting TCO over a tropical region. The Statistical Parameters such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) were used to analyze and evaluate the performance of the proposed models. Findings: In the present study the TCO concentration varied between 217.4 DU and 288.6 DU. The performances of MLR and LSTM model were compared. From the results, it is evident that the values of RMSE, MAE and MAPE of the MLR model are 3.807924, 2.93766, 1.15 respectively whereas the values for LSTM model are 3.074492, 2.574, 1.01 respectively. LSTM was found to be the accurate model for the study data set and the predicted TCO concentration has a very good correlation with the actual observations and it does not rely on the meteorological parameters. Novelty: The LSTM models are usually employed in the field of machine translation, speech recognition and for the prediction of traffic levels, stock levels and air pollutants. This work utilizes the fast numerical machine learning computing library Tensorflow and a high level neural network library Keras that runs on top of Tensorflow to develop a LSTM network model for the prediction of the concentration levels of TCO. Keywords: Total Column Ozone; Regression; Forecasting; Climate Change; Long Short-Term Memory

A simulation and machine learning informed diagnosis of the severe accidents

We propose a simulation and machine learning informed model (SMLIM) for the diagnosis of severe accidents. A machine learning model which consisted of one hidden Long Short Term Memory (LSTM) layer and two dense layers with variations in the number of neurons and regularization parameters and an Adams optimizer was constructed for the multi-time step ahead forecasting analysis and the regression analysis. Using feature variables of lower plenum liquid level, core liquid level, reactor vessel pressure, and dry-well pressure, which were monitored during the Fukushima Daiichi Nuclear Power Plant (FDNPP) accident, the target variable of the dry-well pressure in time was predicted. Training data representing the reference scenario was produced by a MELCOR simulation of the FDNPP accident at unit 1. The first test set obtained from the MELCOR simulation was chosen to consider the aleatoric uncertainty, while the second test set from the MELCOR simulation addresses the epistemic uncertainty. It is shown that the proposed SMLIM model can forecast the time series data with reasonable accuracy for the test cases selected. Then another test sets were prepared using the measurement data taken from the FDNPP. While the forecasted prediction was quite different from the actual measurement data, which were corrupted, the LSTM network model showed better agreement between the prediction and actual data when the measurement data were corrected. This trend was similar in the case of regression analysis. We suggest that the proposed simulation and machine learning informed model can help the operator forecast the future trend of accident progression and diagnose corrupted measurement signals.

Research and Application of Deformation Prediction Model for Deep Foundation Pit Based on LSTM

Deep foundation pit is a door with a long history, but it has new disciplines; in this paper, firstly, the modeling method and process of LSTM (long short-term memory) network are discussed in detail, then the optimization algorithm used in the model is described in detail, and the parameter selection methods such as initial learning rate, activation function, and iteration number related to LSTM network training are introduced in detail. LSTM network is used to process the deformation data of deep foundation pit, and random gradient descent, momentum, Nesterov, RMSProp, AdaGmd, and Adam algorithms are selected in the same example for modeling prediction and comparison. Two examples of horizontal displacement prediction of pile and vertical displacement prediction of column in deep foundation pit show that the LSTM network model established by different optimization algorithms has different prediction accuracy, and the LSTM network model established by Adam optimization algorithm has the highest accuracy, which proves that the selection of optimization algorithm plays an important role in LSTM and also verifies the feasibility of LSTM network in the data processing and prediction of deep foundation pit deformation.

Research on muscle fatigue recognition model based on improved wavelet denoising and long short-term memory

The automatic recognition technology of muscle fatigue has widespread application in the field of kinesiology and rehabilitation medicine. In this paper, we used surface electromyography (sEMG) to study the recognition of leg muscle fatigue during circuit resistance training. The purpose of this study was to solve the problem that the sEMG signals have a lot of noise interference and the recognition accuracy of the existing muscle fatigue recognition model is not high enough. First, we proposed an improved wavelet threshold function denoising algorithm to denoise the sEMG signal. Then, we build a muscle fatigue state recognition model based on long short-term memory (LSTM), and used the Holdout method to evaluate the performance of the model. Finally, the denoising effect of the improved wavelet threshold function denoising method proposed in this paper was compared with the denoising effect of the traditional wavelet threshold denoising method. We compared the performance of the proposed muscle fatigue recognition model with that of particle swarm optimization support vector machine (PSO-SVM) and convolutional neural network (CNN). The results showed that the new wavelet threshold function had better denoising performance than hard and soft threshold functions. The accuracy of LSTM network model in identifying muscle fatigue was 4.89% and 2.47% higher than that of PSO-SVM and CNN, respectively. The sEMG signal denoising method and muscle fatigue recognition model proposed in this paper have important implications for monitoring muscle fatigue during rehabilitation training and exercise.

AI-federated novel delay-aware link-scheduling for Industry 4.0 applications in IoT networks

PurposeWith the advent of AI-federated technologies, it is feasible to perform complex tasks in industrial Internet of Things (IIoT) environment by enhancing throughput of the network and by reducing the latency of transmitted data. The communications in IIoT and Industry 4.0 requires handshaking of multiple technologies for supporting heterogeneous networks and diverse protocols. IIoT applications may gather and analyse sensor data, allowing operators to monitor and manage production systems, resulting in considerable performance gains in automated processes. All IIoT applications are responsible for generating a vast set of data based on diverse characteristics. To obtain an optimum throughput in an IIoT environment requires efficiently processing of IIoT applications over communication channels. Because computing resources in the IIoT are limited, equitable resource allocation with the least amount of delay is the need of the IIoT applications. Although some existing scheduling strategies address delay concerns, faster transmission of data and optimal throughput should also be addressed along with the handling of transmission delay. Hence, this study aims to focus on a fair mechanism to handle throughput, transmission delay and faster transmission of data. The proposed work provides a link-scheduling algorithm termed as delay-aware resource allocation that allocates computing resources to computational-sensitive tasks by reducing overall latency and by increasing the overall throughput of the network. First of all, a multi-hop delay model is developed with multistep delay prediction using AI-federated neural network long–short-term memory (LSTM), which serves as a foundation for future design. Then, link-scheduling algorithm is designed for data routing in an efficient manner. The extensive experimental results reveal that the average end-to-end delay by considering processing, propagation, queueing and transmission delays is minimized with the proposed strategy. Experiments show that advances in machine learning have led to developing a smart, collaborative link scheduling algorithm for fairness-driven resource allocation with minimal delay and optimal throughput. The prediction performance of AI-federated LSTM is compared with the existing approaches and it outperforms over other techniques by achieving 98.2% accuracy.Design/methodology/approachWith an increase of IoT devices, the demand for more IoT gateways has increased, which increases the cost of network infrastructure. As a result, the proposed system uses low-cost intermediate gateways in this study. Each gateway may use a different communication technology for data transmission within an IoT network. As a result, gateways are heterogeneous, with hardware support limited to the technologies associated with the wireless sensor networks. Data communication fairness at each gateway is achieved in an IoT network by considering dynamic IoT traffic and link-scheduling problems to achieve effective resource allocation in an IoT network. The two-phased solution is provided to solve these problems for improved data communication in heterogeneous networks achieving fairness. In the first phase, traffic is predicted using the LSTM network model to predict the dynamic traffic. In the second phase, efficient link selection per technology and link scheduling are achieved based on predicted load, the distance between gateways, link capacity and time required as per different technologies supported such as Bluetooth, Wi-Fi and Zigbee. It enhances data transmission fairness for all gateways, resulting in more data transmission achieving maximum throughput. Our proposed approach outperforms by achieving maximum network throughput, and less packet delay is demonstrated using simulation.FindingsOur proposed approach outperforms by achieving maximum network throughput, and less packet delay is demonstrated using simulation. It also shows that AI- and IoT-federated devices can communicate seamlessly over IoT networks in Industry 4.0.Originality/valueThe concept is a part of the original research work and can be adopted by Industry 4.0 for easy and seamless connectivity of AI and IoT-federated devices.

Research on remaining useful life of rolling bearings using EWT-DI-ALSTM

The bearing is one of the most essential parts of rotating machinery and equipment, and the accurate prediction of the remaining useful life (RUL) of the bearing is of great engineering significance. This paper proposes a novel hybrid prediction architecture called the long short-term memory (LSTM) based on an attention mechanism combined with empirical wavelet transform (EWT) to improve the prediction accuracy of the RUL of bearings. The prediction architecture consists of three parts: firstly, the vibration signal of the bearing is stationarily processed into several intrinsic mode functions (IMFs) by EWT decomposition. Then, a novel IMF degradation index based on weighted energy entropy and the variance contribution rate is proposed to select the representative IMFs with more degradation characteristics of the bearing. Finally, the selected IMFs are regarded as inputs for the network model, and the LSTM network model based on the attention mechanism is used to track the degradation state of the bearing and accurately used to predict its remaining life. The effectiveness of the proposed method is demonstrated by the experimental data.