Short-term Memory Recurrent Neural Network Research Articles

Multi-Modality and Feature Fusion-Based COVID-19 Detection Through Long Short-Term Memory

The Coronavirus Disease 2019 (COVID-19) pandemic poses the worldwide challenges surpassing the boundaries of country, religion, race, and economy. The current benchmark method for the detection of COVID-19 is the reverse transcription polymerase chain reaction (RT-PCR) testing. Nevertheless, this testing method is accurate enough for the diagnosis of COVID-19. However, it is time-consuming, expensive, expert-dependent, and violates social distancing. In this paper, this research proposed an effective multi-modality-based and feature fusion-based (MMFF) COVID-19 detection technique through deep neural networks. In multi-modality, we have utilized the cough samples, breathe samples and sound samples of healthy as well as COVID-19 patients from publicly available COSWARA dataset. Extensive set of experimental analyses were performed to evaluate the performance of our proposed approach. Several useful features were extracted from the aforementioned modalities that were then fed as an input to long short-term memory recurrent neural network algorithms for the classification purpose. Extensive set of experimental analyses were performed to evaluate the performance of our proposed approach. The experimental results showed that our proposed approach outperformed compared to four baseline approaches published recently. We believe that our proposed technique will assists potential users to diagnose the COVID-19 without the intervention of any expert in minimum amount of time.

Identification of Oscillatory Modes in Power System Using Deep Learning Approach

The increase in electric power demand pushes the modern power system for more interconnected networks. It leads to a lack of inertia and creates more critical disturbances in the power system. When this oscillation isn’t damped out, it results in cascade tripping. Immediate detection of low-frequency oscillatory modes and their parameters will help the power system operator to act on a particular event without consuming much time. This research paper proposes novel strategies for identifying low-frequency modes using deep learning techniques, and the model can predict the LFO modes in different topologies. This work presents the Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) approach to predict the instantaneous mode oscillatory parameters in the power system. Once the LSTM-RNN model is trained for different power disturbance situations, it can be used for any events associated with the system. Simulation results are verified using two area Kundur systems at various disturbance conditions. The simulations are performed using MATLAB software and python tensor flow library. The results are validated using statistical methods, and it confirms the superior viability and adaptability of the proposed approach in predicting the instantaneous mode parameters.

Estimating Underlying Articulatory Targets of Thai Vowels by Using Deep Learning Based on Generating Synthetic Samples From a 3D Vocal Tract Model and Data Augmentation

Representation learning is one of the fundamental issues in modeling articulatory-based speech synthesis using target-driven models. This paper proposes a computational strategy for learning underlying articulatory targets from a 3D articulatory speech synthesis model using a bi-directional long short-term memory recurrent neural network based on a small set of representative seed samples. Using a seeding set from VocalTractLab, a larger training set was generated that provided richer contextual variations for the model to learn. The deep learning model for acoustic-to-target mapping was then trained to model the inverse relation of the articulation process. This method allows the trained model to map the given acoustic data onto the articulatory target parameters which can then be used to identify the distribution based on linguistic contexts. The model was evaluated based on its effectiveness in mapping acoustics to articulation, and the perceptual accuracy of speech reproduced from the articulation estimated from the recorded speech by native Thai speakers. The model achieved more than 80% phoneme classification accuracy in the listening test conducted with 25 native Thai speakers. The results indicate that the model can accurately imitate speech with a high degree of phonemic precision.

Long Short-Term Memory Recurrent Neural Network for Estimating State of Charge of Energy Storage System for Grid Services

Long Short-Term Memory Recurrent Neural Network for Estimating State of Charge of Energy Storage System for Grid Services

Confirming Customer Satisfaction With Tones of Speech

As customer satisfaction explicitly leads to repurchase behavior, the level of customer satisfaction affects both sales performance and enterprise growth. Traditionally, measuring satisfaction requires customers spending extra time to fill out a post-purchase questionnaire survey. Recently, ASR (Automatic Speech Recognition) is utilized to extract spoken words from conversation to measure customer satisfaction. However, as oriental people tend to use vague words to express emotion, the approach has its limitation. To solve the problem, this study strived to complete following tasks: devising a process to collect customer voice expressing satisfaction and corresponding verifiable ground truth; a data bank of 150 customer voices speaking in Mandarin was collected; the features of the voice data were extracted with MFCC; as the size of data bank was limited, Auto Encoder was utilized to further reduce the features of voices; models based on Long Short-Term Memory (LSTM) Recurrent Neural Networks (RNNs) and Support Vector Machine (SVM) were constructed to predict satisfaction. With nested cross validation, the average accuracy of LSTM and SVM could reach 71.95% and 73.97%, respectively.

Downhole quantitative evaluation of gas kick during deepwater drilling with deep learning using pilot-scale rig data

Gas kick occurs frequently during deep-water drilling operations caused by the lack of safe margin between pore pressure and leakage pressure. The existing research is limited to gas kick classification and cannot quantitatively evaluate the gas kick risk in the downhole very well. Thus, the objective of this work is to systematically use Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) models based on pilot-scale rig data for quantitative evaluation of gas kick risk. Furthermore, the quantitative evaluation is not surface but downhole. First, the gas kick simulation experiment is accomplished in the pilot-scale test well and produces the gas kick dataset, which is based on the multi-source data fusion through the surface monitoring technologies, riser monitoring technologies and downhole monitoring technologies. Second, the training features are selected and grouped as Sets1-5 to study the features' sensitivity. Third, the raw data is processed and prepared for the following machine learning framework. Fourth, there are five (5) LSTM models trained on Sets1-5. The results indicate that the models’ Loss decrease with the increase of feature number, which has fully demonstrated the effectiveness of PWD, EKD, and Doppler parameters. Finally, there are four representative case studies (artificial gas kick) that are used to test the above five models. The compressed air injected rate (AR) prediction error and detection time-delay decrease with the increase of feature number. The LSTM model trained with the combination of surface-riser-downhole comprehensive detection technologies performs the best in reducing both the prediction error and detection time delay, which could be used to quantitatively evaluate the downhole gas kick risk in the more accurate, faster, more stable, more reliable, and cost-effective manner, and it is effective and worthy of promotion. • The autonomous experiment setup in the pilot-scale test well is carried out to simulate deep-water gas kicks. • There are five novel LSTM models that are developed and proposed for downhole quantitative evaluation of gas kicks. • The proposed LSTM models could quantitatively evaluate the downhole gas kicks in the more accurate and faster manner. • The deployed LSTM models are applied in four representative case studies to further evaluate their performance.

Systematic Characterization of Lysine Post-translational Modification Sites Using MUscADEL.

Among various types of protein post-translational modifications (PTMs), lysine PTMs play an important role in regulating a wide range of functions and biological processes. Due to the generation and accumulation of enormous amount of protein sequence data by ongoing whole-genome sequencing projects, systematic identification of different types of lysine PTM substrates and their specific PTM sites in the entire proteome is increasingly important and has therefore received much attention. Accordingly, a variety of computational methods for lysine PTM identification have been developed based on the combination of various handcrafted sequence features and machine-learning techniques. In this chapter, we first briefly review existing computational methods for lysine PTM identification and then introduce a recently developed deep learning-based method, termed MUscADEL (Multiple Scalable Accurate Deep Learner for lysine PTMs). Specifically, MUscADEL employs bidirectional long short-term memory (BiLSTM) recurrent neural networks and is capable of predicting eight major types of lysine PTMs in both the human and mouse proteomes. The web server of MUscADEL is publicly available at http://muscadel.erc.monash.edu/ for the research community to use.

Comparative Study on Deformation Prediction Models of Wuqiangxi Concrete Gravity Dam Based on Monitoring Data

The deformation prediction models of Wuqiangxi concrete gravity dam are developed, including two statistical models and a deep learning model. In the statistical models, the reliable monitoring data are firstly determined with Lahitte criterion; then, the stepwise regression and partial least squares regression models for deformation prediction of concrete gravity dam are constructed in terms of the reliable monitoring data, and the factors of water pressure, temperature and time effect are considered in the models; finally, according to the monitoring data from 2006 to 2020 of five typical measuring points including J23 (on dam section ), J33 (on dam section ), J35 (on dam section ), J37 (on dam section ), and J39 (on dam section ) located on the crest of Wuqiangxi concrete gravity dam, the settlement curves of the measuring points are obtained with the stepwise regression and partial least squares regression models. A deep learning model is developed based on long short-term memory (LSTM) recurrent neural network. In the LSTM model, two LSTM layers are used, the rectified linear unit function is adopted as the activation function, the input sequence length is 20, and the random search is adopted. The monitoring data for the five typical measuring points from 2006 to 2017 are selected as the training set, and the monitoring data from 2018 to 2020 are taken as the test set. From the results of case study, we can find that (1) the good fitting results can be obtained with the two statistical models; (2) the partial least squares regression algorithm can solve the model with high correlation factors and reasonably explain the factors; (3) the prediction accuracy of the LSTM model increases with increasing the amount of training data. In the deformation prediction of concrete gravity dam, the LSTM model is suggested when there are sufficient training data, while the partial least squares regression method is suggested when the training data are insufficient.

A Novel Tunicate Swarm Algorithm With Hybrid Deep Learning Enabled Attack Detection for Secure IoT Environment

The Internet of Things (IoT) paradigm has matured and expanded rapidly across many disciplines. Despite these advancements, IoT networks continue to face an increasing security threat as a result of the constant and rapid changes in the network environment. In order to address these vulnerabilities, the Fog system is equipped with a robust environment that provides additional tools to beef up data security. However, numerous attacks are persistently evolving in IoT and fog environments as a result of the development of several breaches. To improve the efficiency of intrusion detection in the Internet of Things (IoT), this research introduced a novel tunicate swarm algorithm that combines a long-short-term memory-recurrent neural network. The presented model accomplishes this goal by first undergoing data pre-processing to transform the input data into a usable format. Additionally, attacks in the IoT ecosystem can be identified using a model built on long-short-term memory recurrent neural networks. There is a strong correlation between the number of parameters and the model’s capability and complexity in ANN models. It is critical to keep track of the number of parameters in each model layer to avoid over- or under-fitting. One way to prevent this from happening is to modify the number of layers in your data structure. The tunicate swarm algorithm is used to fine-tune the hyper-parameter values in the Long Short-Term Memory-Recurrent Neural Network model to improve how well it can find things. TSA was used to solve several problems that couldn’t be solved with traditional optimization methods. It also improved performance and shortened the time it took for the algorithm to converge. A series of tests were done on benchmark datasets. Compared to related models, the proposed TSA-LSTMRNN model achieved 92.67, 87.11, and 98.73 for accuracy, recall, and precision, respectively, which indicate the superiority of the proposed model.

Dynamic Space Allocation Based on Internal Demand for Optimizing Release of Shared Parking.

The size of cities has been continuously increasing because of urbanization. The number of public and private transportation vehicles is rapidly increasing, thus resulting in traffic congestion, traffic accidents, and environmental pollution. Although major cities have undergone considerable development in terms of transportation infrastructure, problems caused by a high number of moving vehicles cannot be completely resolved through the expansion of streets and facilities. This paper proposes a solution for the parking problem in cities that entails a shared parking system. The primary concept of the proposed shared parking system is to release parking lots that are open to specific groups for public usage without overriding personal usage. Open-to-specific-groups parking lots consist of parking spaces provided for particular people, such as parking buildings at universities for teachers, staff, and students. The proposed shared parking system comprises four primary steps: collecting and preprocessing data by using an Internet of Things system, predicting internal demand by using a recurrent neural network algorithm, releasing several unoccupied parking lots based on prediction results, and continuously updating the real-time data to improve future internal usage prediction. Data collection and data forecasting are performed to ensure that the system does not override personal usage. This study applied several forecasting algorithms, including seasonal ARIMA, support vector regression, multilayer perceptron, convolutional neural network, long short-term memory recurrent neural network with a many-to-one structure, and long short-term memory recurrent neural network with a many-to-many structure. The proposed system was evaluated using artificial and real datasets. Results show that the recurrent neural network with the many-to-many structure generates the most accurate prediction. Furthermore, the proposed shared parking system was evaluated for some scenarios in which different numbers of parking spaces were released. Simulation results show that the proposed shared parking system can provide parking spaces for public usage without overriding personal usage. Moreover, this system can generate new income for parking management and/or parking lot owners.

Prediction of Lower Extremity Multi-Joint Angles during Overground Walking by Using a Single IMU with a Low Frequency Based on an LSTM Recurrent Neural Network.

The joint angle during gait is an important indicator, such as injury risk index, rehabilitation status evaluation, etc. To analyze gait, inertial measurement unit (IMU) sensors have been used in studies and continuously developed; however, they are difficult to utilize in daily life because of the inconvenience of having to attach multiple sensors together and the difficulty of long-term use due to the battery consumption required for high data sampling rates. To overcome these problems, this study propose a multi-joint angle estimation method based on a long short-term memory (LSTM) recurrent neural network with a single low-frequency (23 Hz) IMU sensor. IMU sensor data attached to the lateral shank were measured during overground walking at a self-selected speed for 30 healthy young persons. The results show a comparatively good accuracy level, similar to previous studies using high-frequency IMU sensors. Compared to the reference results obtained from the motion capture system, the estimated angle coefficient of determination () is greater than 0.74, and the root mean square error and normalized root mean square error (NRMSE) are less than 7° and 9.87%, respectively. The knee joint showed the best estimation performance in terms of the NRMSE and among the hip, knee, and ankle joints.

Real-Time LSTM-RNN Classification of Floors With Different Friction Coefficients for a Walking Humanoid Robot Wearing a 3D Force System

In the study of biped humanoid robots it is crucial to achieve high precision and robustness in locomotion. Humanoid robots that operate in real world environments need to be able to physically recognize different grounds to best adapt their gait without losing their dynamic stability. This work proposes a technique to classify in real time the type of floor from a set of possibilities learnt off-line. Hence, the paper describes the collection and preparation of a dataset of contact forces, obtained with a wearable instrumented system, mixed with the information of the robot internal inertial sensor to classify the type of underlying surface of a walking humanoid robot. For this classification, the data are acquired for four different slippery floors at a rate of 100 Hz and it is used as input for a long short-term memory (LSTM) recurrent neural network (RNN). After testing different learning models architectures and tuning the models parameters, a good mapping between inputs and targets is achieved with a test classification accuracy greater than 92%. A real time experiment is presented to demonstrate the suitability of the proposed approach for the multi-classification problem addressed.

Transformer-Based Deep Neural Language Modeling for Construct-Specific Automatic Item Generation

Algorithmic automatic item generation can be used to obtain large quantities of cognitive items in the domains of knowledge and aptitude testing. However, conventional item models used by template-based automatic item generation techniques are not ideal for the creation of items for non-cognitive constructs. Progress in this area has been made recently by employing long short-term memory recurrent neural networks to produce word sequences that syntactically resemble items typically found in personality questionnaires. To date, such items have been produced unconditionally, without the possibility of selectively targeting personality domains. In this article, we offer a brief synopsis on past developments in natural language processing and explain why the automatic generation of construct-specific items has become attainable only due to recent technological progress. We propose that pre-trained causal transformer models can be fine-tuned to achieve this task using implicit parameterization in conjunction with conditional generation. We demonstrate this method in a tutorial-like fashion and finally compare aspects of validity in human- and machine-authored items using empirical data. Our study finds that approximately two-thirds of the automatically generated items show good psychometric properties (factor loadings above .40) and that one-third even have properties equivalent to established and highly curated human-authored items. Our work thus demonstrates the practical use of deep neural networks for non-cognitive automatic item generation.

A Deep Learning Model for Forecasting Velocity Structures of the Loop Current System in the Gulf of Mexico

Despite the large efforts made by the ocean modeling community, such as the GODAE (Global Ocean Data Assimilation Experiment), which started in 1997 and was renamed as OceanPredict in 2019, the prediction of ocean currents has remained a challenge until the present day—particularly in ocean regions that are characterized by rapid changes in their circulation due to changes in atmospheric forcing or due to the release of available potential energy through the development of instabilities. Ocean numerical models’ useful forecast window is no longer than two days over a given area with the best initialization possible. Predictions quickly diverge from the observational field throughout the water and become unreliable, despite the fact that they can simulate the observed dynamics through other variables such as temperature, salinity and sea surface height. Numerical methods such as harmonic analysis are used to predict both short- and long-term tidal currents with significant accuracy. However, they are limited to the areas where the tide was measured. In this study, a new approach to ocean current prediction based on deep learning is proposed. This method is evaluated on the measured energetic currents of the Gulf of Mexico circulation dominated by the Loop Current (LC) at multiple spatial and temporal scales. The approach taken herein consists of dividing the velocity tensor into planes perpendicular to each of the three Cartesian coordinate system directions. A Long Short-Term Memory Recurrent Neural Network, which is best suited to handling long-term dependencies in the data, was thus used to predict the evolution of the velocity field in each plane, along each of the three directions. The predicted tensors, made of the planes perpendicular to each Cartesian direction, revealed that the model’s prediction skills were best for the flow field in the planes perpendicular to the direction of prediction. Furthermore, the fusion of all three predicted tensors significantly increased the overall skills of the flow prediction over the individual model’s predictions. The useful forecast period of this new model was greater than 4 days with a root mean square error less than 0.05 cm·s−1 and a correlation coefficient of 0.6.

An Evaluation of Speech-Based Recognition of Emotional and Physiological Markers of Stress

Life in modern societies is fast-paced and full of stress-inducing demands. The development of stress monitoring methods is a growing area of research due to the personal and economic advantages that timely detection provides. Studies have shown that speech-based features can be utilised to robustly predict several physiological markers of stress, including emotional state, continuous heart rate, and the stress hormone, cortisol. In this contribution, we extend previous works by the authors, utilising three German language corpora including more than 100 subjects undergoing a Trier Social Stress Test protocol. We present cross-corpus and transfer learning results which explore the efficacy of the speech signal to predict three physiological markers of stress—sequentially measured saliva-based cortisol, continuous heart rate as beats per minute (BPM), and continuous respiration. For this, we extract several features from audio as well as video and apply various machine learning architectures, including a temporal context-based Long Short-Term Memory Recurrent Neural Network (LSTM-RNN). For the task of predicting cortisol levels from speech, deep learning improves on results obtained by conventional support vector regression—yielding a Spearman correlation coefficient (ρ) of 0.770 and 0.698 for cortisol measurements taken 10 and 20 min after the stress period for the two corpora applicable—showing that audio features alone are sufficient for predicting cortisol, with audiovisual fusion to an extent improving such results. We also obtain a Root Mean Square Error (RMSE) of 38 and 22 BPM for continuous heart rate prediction on the two corpora where this information is available, and a normalised RMSE (NRMSE) of 0.120 for respiration prediction (−10: 10). Both of these continuous physiological signals show to be highly effective markers of stress (based on cortisol grouping analysis), both when available as ground truth and when predicted using speech. This contribution opens up new avenues for future exploration of these signals as proxies for stress in naturalistic settings.

Center of Inertia Frequency Estimation Using Deep Learning Algorithm

Abstract Increasing the number of generation units connected to the grid via power electronic devices potentially implies negative impacts on the power system frequency stability and, depending on the power system inertia value, implies the necessary contribution of wind power plants to inertial response of the system. An alternative approach to the active power control of wind power plants, without the impact of local frequency deviation on the output power, is the application of a control strategies based on the center of inertia frequency. Since control schemes based on the input variable of the center of inertia frequency require a satisfactory level of signal transmission capacity in real time and the advanced telecommunication infrastructure of the power system, the paper considers an alternative approach to estimate the input signal value. According to the developed long short-term memory recurrent neural network, paper presents the idea of center of inertia frequency estimation by monitoring the speed of several generators in the system and passing the sequence of input data for a certain time interval, after the occurrence of imbalance, to the artificial intelligence module.

Recognition of Unknown Radar Emitters With Machine Learning

Classifiers based on machine learning are usually trained to distinguish between several known classes. For an electronic intelligence application, however, it is of great importance to recognize if an intercepted signal belongs to an unknown radar emitter. In the machine learning literature, this task is called open-set recognition. This article investigates six approaches in several configurations to recognize unknown emitters. It is based on a hierarchical emission model that understands emissions as a language with an inherent hierarchical structure. We consider two general approaches, which are the “memoryless” Markov chain and the Long Short-Term Memory recurrent neural network, which is especially designed to “remember” the past. The performance is demonstrated with two evaluation metrics in ten scenarios that contain different combinations of known and unknown emitters. An evaluation with corrupted data provides an estimate on the methods’ accuracies under challenging conditions. The results show that unknown emitters that do not use known waveforms are reliably recognized even with corrupted data, while unknown emitters that are more similar to known ones are harder to detect.

Point-of-use sensors and machine learning enable low-cost determination of soil nitrogen.

Overfertilization with nitrogen fertilizers has damaged the environment and health of soil, but standard laboratory testing of soil to determine the levels of nitrogen (mainly NH4+ and NO3-) is not performed regularly. Here we demonstrate that point-of-use measurements of NH4+, combined with soil conductivity, pH, easily accessible weather and timing data, allow instantaneous prediction of levels of NO3- in soil (R2 = 0.70) using a machine learning model. A long short-term memory recurrent neural network model can also be used to predict levels of NH4+ and NO3- up to 12 days into the future from a single measurement at day one, with [Formula: see text] and [Formula: see text], for unseen weather conditions. Our machine-learning-based approach eliminates the need for dedicated instruments to determine the levels of NO3- in soil. Nitrogenous soil nutrients can be determined and predicted with enough accuracy to forecast the impact of climate on fertilization planning and to tune timing for crop requirements, reducing overfertilization while improving crop yields.

Cyber-Resilient Smart Cities: Detection of Malicious Attacks in Smart Grids

• Introduces a new type of attack against distributed state estimation of power systems, which operates on inter-area boundary buses. • The proposed attack can bypass existing robust state estimators and the convergence-based detection methods. • A deep learning-based cyber-attack detection mechanism to detect such attacks. • The results show that the proposed detector outperforms existing machine learning-based cyber-attack detection methods. A massive challenge for future cities is being environmentally sustainable by incorporating renewable energy resources (RES). At the same time, future smart cities need to support resilient environments against cyber-threats on their supported information and communication technologies (ICT). Therefore, the cybersecurity of future smart cities and their smart grids is of paramount importance, especially on how to detect cyber-attacks with growing uncertainties, such as frequent topological changes and RES of intermittent nature. Such raised uncertainties can cause a significant change in the underlying distribution of measurements and system states. In such an environment, historical measured data will not accurately exhibit the current network’s operating point. Hence, future power grids’ dynamic behaviors within smart cities are much more complicated than the conventional ones, leading to incorrect classification of the new instances by the current attack detectors. In this paper, to address this problem, a long short-term memory (LSTM) recurrent neural network (RNN) is carefully designed by embedding the dynamically time-evolving power system’s characteristics to accurately model the dynamic behaviors of modern power grids that are influenced by RES or system reconfiguration to distinguish natural smart grid changes and real-time attacks. The proposed framework’s performance is evaluated using the IEEE 14-bus system using real-world load data with multiple attack cases such as attacks to the network after a line outage and combination of RES. Results confirm that the developed LSTM-based attack detection model has a generalization ability to catch modern power grids’ dynamic behaviors, excelling current traditional approaches in the designed case studies and achieves accuracy higher than 90% in all experiments.

A Hybrid Prognostic Method for PEMFC With Aging Parameter Prediction

Prognostic of the proton-exchange membrane fuel cell can effectively extend the fuel cell lifespan, which can contribute to its large-scale commercialization. In this article, a hybrid prognostic approach is proposed to predict the fuel cell output voltage and other aging parameters that can reflect the stack’s internal degradation. During the training stage, the prognostic parameters are obtained by using the extended Kalman filter (EKF). Besides, the fuel cell output voltage is used to train the long short-term memory (LSTM) recurrent neural network. During the prediction stage, the hybrid EKF and LSTM method will predict the output voltage and aging parameters, and the degradation can be predicted under dynamic conditions. The proposed method is validated by experimental tests under static, quasi-dynamic, and dynamic conditions. Results indicate that the hybrid method can accurately predict the degradation trend of fuel cell voltage and aging parameters. The RMSE of the method is less than 0.0110, 0.0262, and 0.0317 under static, quasi-dynamic, and dynamic conditions, respectively, which are smaller than the conventional model-based methods or data-driven methods. Furthermore, the hybrid method can provide more detailed information for prognostic decision-making and better prolong the fuel cell lifespan.