Short-term Memory Method Research Articles

Neural Machine Translation of Spanish-English Food Recipes Using LSTM

Nowadays, food is one of the things that has been globalized, and everyone from different parts of the world has been able to cook food from other countries through existing online recipes. Based on that, this study developed a translation formula using a neural machine translation (NMT). NMT is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder–decoders. Our experiment led to novel insights and practical advice for building and extending NMT with the applied long short-term memory (LSTM) method to 47 bilingual food recipes between Spanish-English and English-Spanish. LSTM is one of the best machine learning methods for translating languages because it can retain memories for an extended period concurrently, grasp complicated connections between data, and provides highly useful information in deciding translation outcomes. The evaluation for this neural machine translation is to use BLEU. The comparing results show that the translation of recipes from Spanish-English has a better BLEU value of 0.998426 than English-Spanish with a data-sharing of 70%:30% during epoch 1000. Researchers can convert the country's popular cuisine recipes into another language for further research, allowing it to become more widely recognized abroad.

THE APPLICATION OF ARTIFICIAL INTELINGENCE IN DJIA STOCKS TO IMPROVE THE INVESTMENT PROFITABILITY USING PHYTON

Technological developments, competitive economics climate and demanding competition have led the investment industry to experienced rapid and continuous development in the last few decades.
 Some of the rapid and continuous key developments are transformation in financial microstructures, development of investment strategies, the progression in computing capacity and the new trend of the investment performance of pioneers in algorithmic traders surpassing that of the human, discretionary investors (Jansen, Stefan 2017)
 These four key factors have driven the investment company and hedge fund to develop algorithmic trading methods even further to achieve a more stable and reliable profit over time.
 Therefore, to manifest aforementioned concerns, this research will conduct the process of building hybrid machine learning in Dow Jones Industrial Average stocks by using Long Short Term Memory (LSTM) Method to improve the investment profitability using phyton programming language.
 The Result of this research shows that the prediction made by the software has acceptable rate of errors. The several measurements of errors used are namely, Median Absolute Error, Mean Absolute Percentage Error and Median Absolute Percentage Error.
 
 Keywords: Artificial Intelligence, Stock Market, Data Science , LSTM, Phyton

Backhand-Approach-Based American Sign Language Words Recognition Using Spatial-Temporal Body Parts and Hand Relationship Patterns.

Most of the existing methods focus mainly on the extraction of shape-based, rotation-based, and motion-based features, usually neglecting the relationship between hands and body parts, which can provide significant information to address the problem of similar sign words based on the backhand approach. Therefore, this paper proposes four feature-based models. The spatial–temporal body parts and hand relationship patterns are the main feature. The second model consists of the spatial–temporal finger joint angle patterns. The third model consists of the spatial–temporal 3D hand motion trajectory patterns. The fourth model consists of the spatial–temporal double-hand relationship patterns. Then, a two-layer bidirectional long short-term memory method is used to deal with time-independent data as a classifier. The performance of the method was evaluated and compared with the existing works using 26 ASL letters, with an accuracy and F1-score of 97.34% and 97.36%, respectively. The method was further evaluated using 40 double-hand ASL words and achieved an accuracy and F1-score of 98.52% and 98.54%, respectively. The results demonstrated that the proposed method outperformed the existing works under consideration. However, in the analysis of 72 new ASL words, including single- and double-hand words from 10 participants, the accuracy and F1-score were approximately 96.99% and 97.00%, respectively.

Sparse Data-Extended Fusion Method for Sea Surface Temperature Prediction on the East China Sea

The accurate temperature background field plays a vital role in the numerical prediction of sea surface temperature (SST). At present, the SST background field is mainly derived from multi-source data fusion, including satellite SST data and in situ data from marine stations, buoys, and voluntary observing ships. The characteristics of satellite SST data are wide coverage but low accuracy, whereas the in situ data have high accuracy but sparse distribution. For obtaining a more accurate temperature background field and realizing the fusion of measured data with satellite data as much as possible, we propose a sparse data-extended fusion method to predict SST in this paper. By using this method, the actual observed sites and buoys data in the East China Sea area are fused with Advanced Very High Resolution Radiometer (AVHRR) Pathfinder Version 5.0 SST data. Furthermore, the temperature field in the study area were predicted by using Long Short-Term Memory (LSTM) and Gate Recurrent Unit (GRU) deep learning methods, respectively. Finally, we obtained the results by traditional prediction methods to verify them. The experimental results show that the method we proposed in this paper can obtain more accurate prediction results, and effectively compensate for the uncertainty caused by the parameterization of ocean dynamic process, the discrete method, and the error of initial conditions.

Jakarta Pandemic to Endemic Transition: Forecasting COVID-19 Using NNAR and LSTM

In December 2021, the latest COVID-19 variant, Omicron, was confirmed in Indonesia. Unlike the Delta variant, the number of deaths in the Omicron type did not increase significantly and remained constant, even though the cases increased significantly. It is hoped that Indonesia will declare COVID-19 endemic. Jakarta is the capital of Indonesia and the first city where the new COVID-19 virus emerged. Therefore, we are trying to model COVID-19 cases in Jakarta and predict future cases to see if endemic conditions are identified. We applied Neural Network Auto-Regressive (NNAR) and Long Short-Term Memory (LSTM) methods. It is found that the NNAR forecast better for positive confirmed cases with an R-squared 0.939 and the LSTM forecast better for cases of death with an R-squared 0.9337. The forecasting results for the next 7 days reveal that positive confirmed cases of COVID-19 in Jakarta will increase slightly. In addition, the death cases experienced a very small increase, only one new case. According to the results of this study, it can be concluded that COVID-19 in Jakarta will enter an endemic phase in Jakarta, with no substantial increase in cases and a low mortality rate.

LONG-SHORT TERM MEMORY (LSTM) FOR PREDICTING VELOCITY AND DIRECTION SEA SURFACE CURRENT ON BALI STRAIT

The strategic role of the Bali Strait as a connection between the islands of Java and Bali is growing in line with the increase in the economy and tourism of the two islands. Therefore, it is necessary to have a further understanding of the condition of the waters in the Bali strait, one of which is ocean currents. This study aims to predict future ocean currents based on 30-minute data in the Bali Strait in the range of 16 May 2021 to 9 June 2021 obtained from the Perak II Surabaya Maritime Meteorological Station. In this study, the Long Short Term Memory method was used. The parameters used are hidden layer, batch size, and learn rate drop. Based on the parameters used, the results showed that the smallest MAPE value was 18.64% for U ocean current velocity data and 5.29% for V ocean current velocity data.

Research on Prediction of Solar Power Considering the Methods of Statistical and Machine Learning – Based on the Data of Australian Solar Power Market

In this paper, we use the methods of machine learning and traditional time series to predict solar power generation, which is based on the Australia Market Data. In the paper, we analyze Ausgrid's Solar by using long and short-term memory (LSTM) methods and time series models (multiple regression models with related errors) to accurately estimate the parameters of photovoltaic (PV) array models, which is using the data of household electricity consumption from July 1, 2010, to June 30, 2013. The results show that the regression model with correlated errors is better than the machine learning-based LSTM algorithm, which is based on the differential MSE performance. The final prediction accuracy rate is as high as 98%, so the regression model can accurately predict solar power generation.

Price Prediction Model for Red and Curly Red Chilies using Long Short Term Memory Method

The price data of the Strategic Food Price Information Center from May 2018 to May 2021 in 34 provinces show a fluctuated trend. Our study aimed to build predictive modeling of red chili and curly chili prices in West Java province using the Long Short Term Memory method. The red chili and curly chili prices prediction model in our study was successfully constructed and is considered very representative of predicting prices in traditional and modern markets in West Java Province. The best parameter model for red chili in the traditional market is a neuron value of 64 and a learning rate of 0.0005, and in the modern market, there are neuron values of 48 and a learning rate of 0,005. For curly chili, the best parameter model in traditional markets is a neuron value of 48 and a learning rate of 0.00075, and in the modern market, there are neuron values of 32 and a learning rate of 0,001. All models use the number of the epoch 100. The best prediction model for the price of red chili and curly red chili in traditional markets obtained the smallest root mean square error values on the test data of 2.57% and 2.07%, respectively. Meanwhile, the best price prediction model in the modern market obtained the smallest root mean square error values on the test data of 2.11% and 2.17%, respectively. Based on the root mean square error value obtained, the model is better than the other research method and shows that the variation in the value produced by a model is close to the variation in the actual value.

Deep fake detection using a sparse auto encoder with a graph capsule dual graph CNN.

Deepfake (DF) is a kind of forged image or video that is developed to spread misinformation and facilitate vulnerabilities to privacy hacking and truth masking with advanced technologies, including deep learning and artificial intelligence with trained algorithms. This kind of multimedia manipulation, such as changing facial expressions or speech, can be used for a variety of purposes to spread misinformation or exploitation. This kind of multimedia manipulation, such as changing facial expressions or speech, can be used for a variety of purposes to spread misinformation or exploitation. With the recent advancement of generative adversarial networks (GANs) in deep learning models, DF has become an essential part of social media. To detect forged video and images, numerous methods have been developed, and those methods are focused on a particular domain and obsolete in the case of new attacks/threats. Hence, a novel method needs to be developed to tackle new attacks. The method introduced in this article can detect various types of spoofs of images and videos that are computationally generated using deep learning models, such as variants of long short-term memory and convolutional neural networks. The first phase of this proposed work extracts the feature frames from the forged video/image using a sparse autoencoder with a graph long short-term memory (SAE-GLSTM) method at training time. The first phase of this proposed work extracts the feature frames from the forged video/image using a sparse autoencoder with a graph long short-term memory (SAE-GLSTM) method at training time. The proposed DF detection model is tested using the FFHQ database, 100K-Faces, Celeb-DF (V2) and WildDeepfake. The evaluated results show the effectiveness of the proposed method.

Traffic prediction enabled dynamic access points switching for energy saving in dense networks

To meet the ever-increasing traffic demand and enhance the coverage of cellular networks, network densification is one of the crucial paradigms of 5G and beyond mobile networks, which can improve system capacity by deploying a large number of Access Points (APs) in the service area. However, since the energy consumption of APs generally accounts for a substantial part of the communication system, how to deal with the consequent energy issue is a challenging task for a mobile network with densely deployed APs. In this paper, we propose an intelligent AP switching on/off scheme to reduce the system energy consumption with the prerequisite of guaranteeing the quality of service, where the signaling overhead is also taken into consideration to ensure the stability of the network. First, based on historical traffic data, a long short-term memory method is introduced to predict the future traffic distribution, by which we can roughly determine when the AP switching operation should be triggered; second, we present an efficient three-step AP selection strategy to determine which of the APs would be switched on or off; third, an AP switching scheme with a threshold is proposed to adjust the switching frequency so as to improve the stability of the system. Experiment results indicate that our proposed traffic forecasting method performs well in practical scenarios, where the normalized root mean square error is within 10%. Furthermore, the achieved energy-saving is more than 28% on average with a reasonable outage probability and switching frequency for an area served by 40 APs in a commercial mobile network.

Petite term traffic flow prediction using deep learning for augmented flow of vehicles

An Intelligent Transport System (ITS) model that is contingent on the compulsion and expertise of the Traffic Prediction System in the contemporary urban context is proposed in this paper. Deep Learning (DL) is computationally becoming comfortable to train and set as many hyperparameters automatically as possible. The researchers and practitioners crave to set as many hyperparameters inevitably as possible in the DL. To be a great enabler, ITS has to find suitable solutions to issues like—alert on live time traffic information to interested parties along with facility to retrieve on demand the long-term statistical data, reduce the middling waiting time for commuters, offer protected, consistent, value-added services, control with vitality the signal timing based on the traffic flow etc., All these limitations call for instant attention. Among all the listed issues the problems like the sharp nonlinearities due to changeovers amid free flow, breakdown, retrieval and congestion. The contributions in this paper are as follows: (i) Adopt an ascendable approach to kindle the scarce information formed; (ii) Exploit the attention mechanism to exterminate the disadvantages of Long Short-Term Memory (LSTM) methods for traffic prediction; (iii) Suggest a new fusion smoothing model; (iv) Investigating, developing, and utilizing the Bayesian contextual bandits; (v) Recommend a Linear model based on LSTM, in combo with Bayesian contextual bandits. The travel speed prediction is done by LSTM. The results authenticate that the proposed model can adeptly achieve the goal of developing a system. The proposed model is definitely the best solution to overcome the issues.

Comparative Analysis of Machine Learning and Numerical Modeling for Combined Heat Transfer in Polymethylmethacrylate.

This study has compared different methods to predict the simultaneous effects of conductive and radiative heat transfer in a polymethylmethacrylate (PMMA) sample. PMMA is a type of polymer utilized in various sensors and actuator devices. One-dimensional combined heat transfer is considered in numerical analysis. Computer implementation was obtained for the numerical solution of the governing equation with the implicit finite difference method in the case of discretization. Kirchhoff transformation was used to obtain data from a non-linear equation of conductive heat transfer by considering monochromatic radiation intensity and temperature conditions applied to the PMMA sample boundaries. For the deep neural network (DNN) method, the novel long short-term memory (LSTM) method was introduced to find accurate results in the least processing time compared to the numerical method. A recent study derived the combined heat transfer and temperature profiles for the PMMA sample. Furthermore, the transient temperature profile was validated by another study. A comparison proves the perfect agreement. It shows the temperature gradient in the primary positions, which provides a spectral amount of conductive heat transfer from the PMMA sample. It is more straightforward when they are compared with the novel DNN method. Results demonstrate that this artificial intelligence method is accurate and fast in predicting problems. By analyzing the results from the numerical solution, it can be understood that the conductive and radiative heat flux are similar in the case of gradient behavior, but the amount is also twice as high approximately. Hence, total heat flux has a constant value in an approximated steady-state condition. In addition to analyzing their composition, the receiver operating characteristic (ROC) curve and confusion matrix were implemented to evaluate the algorithm’s performance.

Modeling and Fault Detection of Brushless Direct Current Motor by Deep Learning Sensor Data Fusion.

Only with new sensor concepts in a network, which go far beyond what the current state-of-the-art can offer, can current and future requirements for flexibility, safety, and security be met. The combination of data from many sensors allows a richer representation of the observed phenomenon, e.g., system degradation, which can facilitate analysis and decision-making processes. This work addresses the topic of predictive maintenance by exploiting sensor data fusion and artificial intelligence-based analysis. With a dataset such as vibration and sound from sensors, we focus on studying paradigms that orchestrate the most optimal combination of sensors with deep learning sensor fusion algorithms to enable predictive maintenance. In our experimental setup, we used raw data obtained from two sensors, a microphone, and an accelerometer installed on a brushless direct current (BLDC) motor. The data from each sensor were processed individually and, in a second step, merged to create a solid base for analysis. To diagnose BLDC motor faults, this work proposes to use data-level sensor fusion with deep learning methods such as deep convolutional neural networks (DCNNs) for their ability to automatically extract relevant information from the input data, the long short-term memory method (LSTM), and convolutional long short-term memory (CNN-LSTM), a combination of the two previous methods. The results show that in our setup, sound signals outperform vibrations when used individually for training. However, without any feature selection/extraction step, the accuracy of the models improves with data fusion and reaches 98.8%, 93.5%, and 73.6% for the DCNN, CNN-LSTM, and LSTM methods, respectively, 98.8% being a performance that, according to our reading, has never been reached in the analysis of the faults of a BLDC motor without first going through the extraction of the characteristics and their fusion by traditional methods. These results show that it is possible to work with raw data from multiple sensors and achieve good results using deep learning methods without spending time and resources on selecting appropriate features to extract and methods to use for feature extraction and data fusion.

Prediction and Diagnosis of Respiratory Disease by Combining Convolutional Neural Network and Bi-directional Long Short-Term Memory Methods.

ObjectiveBased on the respiratory disease big data platform in southern Xinjiang, we established a model that predicted and diagnosed chronic obstructive pulmonary disease, bronchiectasis, pulmonary embolism and pulmonary tuberculosis, and provided assistance for primary physicians.MethodsThe method combined convolutional neural network (CNN) and long-short-term memory network (LSTM) for prediction and diagnosis of respiratory diseases. We collected the medical records of inpatients in the respiratory department, including: chief complaint, history of present illness, and chest computed tomography. Pre-processing of clinical records with “jieba” word segmentation module, and the Bidirectional Encoder Representation from Transformers (BERT) model was used to perform word vectorization on the text. The partial and total information of the fused feature set was encoded by convolutional layers, while LSTM layers decoded the encoded information.ResultsThe precisions of traditional machine-learning, deep-learning methods and our proposed method were 0.6, 0.81, 0.89, and F1 scores were 0.6, 0.81, 0.88, respectively.ConclusionCompared with traditional machine learning and deep-learning methods that our proposed method had a significantly higher performance, and provided precise identification of respiratory disease.

Forecasting Shanghai Container Freight Index: A Deep-Learning-Based Model Experiment

With the increasing availability of large datasets and improvements in prediction algorithms, machine-learning-based techniques, particularly deep learning algorithms, are becoming increasingly popular. However, deep-learning algorithms have not been widely applied to predict container freight rates. In this paper, we compare a long short-term memory (LSTM) method and a seasonal autoregressive integrated moving average (SARIMA) method for forecasting the comprehensive and route-based Shanghai Containerized Freight Index (SCFI). The research findings indicate that the LSTM deep learning models outperformed SARIMA models in most of the datasets. For South America and the east coast of the U.S. routes, LSTM could reduce forecasting errors by as much as 85% compared to SARIMA. The SARIMA models performed better than LSTM in predicting freight movements on the west and east Japan routes. The study contributes to the literature in four ways. First, it presents insights for improving forecasting accuracy. Second, it helps relevant parties understand the trends of container freight markets for wiser decision-making. Third, it helps relevant stakeholders understand overall container shipping market trends. Lastly, it can help hedge against the volatility of freight rates.

Press-fit process fault diagnosis using 1DCNN-LSTM method

PurposePress-fit with force and displacement monitoring is commonly adopted in automotive mechatronic system assembling. However, suitable methods for the press-fit study are still at initial investigation phase. The sequential data physical meaning, small data sets from different resources and computing efficiency should be considered. Therefore, this paper aims to better identify press-fit fault types.Design/methodology/approachThis paper proposed one-dimensional convolutional neural network (1DCNN)–long short-term memory (LSTM) method to perform press-fit fault diagnosis into automotive assembling practice which is in accordance with current product development procedure. Specialized data augmentation method is proposed to merge different data resources and increase the sample size. Referring one-way sequential data characteristics, LSTM and batch normalization layers are integrated in 1DCNN to improve the performance.FindingsThe proposed 1DCNN-LSTM method is feasible with small data sets from different sources. Using data augmentation to make data unified and sample size increased, the accuracy could reach more than 99%. Training time has reduced from 90 s/Epoch to 4 s/Epoch compare to pure LSTM method.Originality/valueThe proposed method shows better performance with less training time compared to LSTM. Therefore, the method has practical value and is worthy of industrial application.

BUĞDAY VERİM TAHMİNİ İÇİN YENİLEMELİ SİNİR AĞI TABANLI MODEL GELİŞTİRME

In the study carried out in line with the stated purposes, monthly rain, humidity and temperature data, wheat production amount, and wheat productivity data of Konya province between 1980-2020 were used. Using these data, wheat productivity estimation was performed with (Gated Recurrent Units) GRU and Long Short Term Memory (LSTM) methods, which are Recurrent Neural Network (RNN) based algorithms. When wheat productivity estimation performance was examined with the implemented GRU-based model, 0.9550, 0.0059, 0.0280, 0.0623, 7.45 values were obtained for the R2 score, MSE, RMSE, MAE and MAPE values, respectively. In the performance results obtained with the LSTM method, which is another RNN-based method, 0.9667, 0.0054, 0.0280, 0.0614, 7.33 values were obtained for the R2 score, MSE, RMSE, MAE and MAPE values, respectively. Although the LSTM method gave better results than the GRU method, the training modelling time of the LSTM method took longer than that of the GRU method.

Bayesian CNN-BiLSTM and Vine-GMCM Based Probabilistic Forecasting of Hour-Ahead Wind Farm Power Outputs

The importance of the accurate forecasting of power outputsof wind-based generation systems is increasing, as their contributions to the total system generation are rising. However, wind energy resource exhibits strong and stochastic spatio-temporal variations, which further combine with the uncertainties in WF operating regimes, i.e., numbers of wind turbines in normal operation, under curtailment, or that are faulty/disconnected. This paper presents a novel approach for efficient dealing with uncertainties in hour-ahead forecasted WF power outputs. It first applies Bayesian convolutional neural network-bidirectional long short-term memory (Bayesian CNN-BiLSTM) method, which allows for a more accurate probabilistic forecasting of wind speed, air density and wind direction, using the nearby WFs as additional input data. The WF operating regimes are also predicted using the same Bayesian CNN-BiLSTM structure. Afterwards, a high-dimensional Vine-Gaussian mixture Copula model is combined with Bayesian CNN-BiLSTM model to evaluate uncertainties in the WF outputs based on a cross-correlational conditioning of the forecasted weather variables and operating regimes. The proposed combined model is applied and validated using the actual field measurements from two WF clusters in close locations in Croatia and is also benchmarked against several other models.

Forecasting cryptocurrency prices using Recurrent Neural Network and Long Short-term Memory

The rapid development of cryptocurrencies over the past decade is one of the most controversial and ambiguous innovations in the modern global economy. Numerous and unpredictable fluctuations in cryptocurrencies rates, as well as the lack of intelligent and proper management of transactions of this type of currency in most developing countries and users of this type of currency, has led to increased risk and distrust of these roses in investors. Capitalists and investors prefer to invest in programs which have the least risk, the most profit and the least time to achieve the main profit. Therefore, the issue of developing appropriate methods and models for predicting the price of cryptographic products is essential both for the scientific community and for financial analysts, investors and traders. In this research, a new deep learning model is used to predict the price of cryptocurrencies. The proposed model uses a Recurrent Neural Networks (RNN) algorithm based on Long Short-Term Memory (LSTM) method to predict the price. In the presented results of the simulation of the proposed method, factors such as the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), R-Squared (R2) were compared with other similar methods. Finally, the superiority of the proposed method over other methods was proven.

Prediction of Cumulative Installed Power of Geothermal Power Plants in Turkey by Using Artificial Neural Network and Bidirectional Long Short-Term Memory

Turkey has a great potential for renewable energies. The number of power plants (PP) producing electricity from renewable energy sources and accordingly the installed power has risen over the years. As of the end of December 2021, the cumulative installed power of Turkey reached 99819.6 MW and the share of the total installed power of the PPs generating electricity from renewable energy sources was 53.72%. Although the installed power has increased, the percentage of PPs using renewable energy sources in total electricity generation is not yet at the desired level. However, geothermal energy is being used more and more in electricity generation alongside the other most well-known types of renewable energy. It can be observed that the installed power of geothermal power plants (GPP) in Turkey started to increase gradually after 2007, and as of the end of December 2021, the cumulative installed power reached 1676.2 MW. In this study, with the data for the cumulative installed power of GPPs in Turkey in the 2007-2021 period, the cumulative installed power of GPPs in Turkey for 2022 was predicted by using Artificial Neural Network (ANN) and Bidirectional Long Short-Term Memory (BLSTM) methods, and the results were compared and interpreted.