Short-term Memory Deep Network Research Articles

Accurate prediction of machining cycle times and feedrates with deep neural networks using BiLSTM

This paper presents a deep neural network-based approach to predict machining cycle (run) times along complex 3-axis machining part programs. CAD/CAM systems are utilized to generate part programs to machine parts with complex geometries and sculptured surfaces. The part program is processed by the numerical control (NC) unit of the machine tool to plan a feed profile. For such complex part programs, the actual (observed) feed profile and commanded (desired) one are significantly different. This paper uses bi-directional long short-term memory deep networks (bi-LSTM) to model the interpolation behavior of the NC systems to accurately predict how the feedrate profile is planned for any given part program. The G-line length, commanded feed, change in the feed direction through consecutive G-lines are used as a primary set of inputs to the network. The neural network is then trained by running series of toolpaths on the NC system and recording the resultant feedrate profile. Simulation and experimental studies are conducted on NC kernels to validate effectiveness of the proposed strategy. It is shown that machining cycle times on complex toolpaths can be predicted with > 94 % accuracy.

Integrated adversarial long short-term memory deep networks for reheater tube temperature forecasting of ultra-supercritical turbo-generators

As an important equipment of the ultra-supercritical turbine systems, real-time and accurate control of reheater tube temperature is the key to the normal operation of the system. To improve the accuracy of reheater tube temperature prediction, an integrated adversarial long short-term memory deep network is proposed in this study. Considering the correlation and dynamics between reheater temperature features, different types of single long short-term memory deep networks are batch mixed and stacked in parallel to integrate adversarial training. The training process shares network layer weight parameters to synthesize the decision boundary of each network and complete the dichotomous integration. The proposed networks predict information for the next ten time steps, wirelessly approximate the mapping relationship between multi-source input and output data, and maximize the prediction accuracy and generalization. The proposed networks with single long short-term memory deep networks and 21 algorithms are compared for temperature prediction on the reheater tubes of thermal power plants in Guangdong Province, China, from March 2020 to May 2020. Compared with 90 networks without an integrated adversary, the mean square error obtained by the integrated networks have a 90.85% reduction. Furthermore, compared with the algorithm with the second smallest error among the 22 algorithms, the mean absolute percentage error obtained by the integrated networks is significantly reduced by 99.66%, and the mean absolute error is reduced by 98.36%.

A Sentiment Analysis Method Based on a Blockchain-Supported Long Short-Term Memory Deep Network.

Traditional sentiment analysis methods are based on text-, visual- or audio-processing using different machine learning and/or deep learning architecture, depending on the data type. This situation comes with technical processing diversity and cultural temperament effect on analysis of the results, which means the results can change according to the cultural diversities. This study integrates a blockchain layer with an LSTM architecture. This approach can be regarded as a machine learning application that enables the transfer of the metadata of the ledger to the learning database by establishing a cryptographic connection, which is created by adding the next sentiment with the same value to the ledger as a smart contract. Thus, a “Proof of Learning” consensus blockchain layer integrity framework, which constitutes the confirmation mechanism of the machine learning process and handles data management, is provided. The proposed method is applied to a Twitter dataset with the emotions of negative, neutral and positive. Previous sentiment analysis methods on the same data achieved accuracy rates of 14% in a specific culture and 63% in a the culture that has appealed to a wider audience in the past. This study puts forth a very promising improvement by increasing the accuracy to 92.85%.

Motor imagery EEG signal classification using long short-term memory deep network and neighbourhood component analysis

Motor imagery EEG signal classification using long short-term memory deep network and neighbourhood component analysis

Gait-based person identification using 3D LiDAR and long short-term memory deep networks

Gait recognition is one measure of biometrics, which also includes facial, fingerprint, and retina recognition. Although most biometric methods require direct contact between a device and a subject, gait recognition has unique characteristics whereby interaction with the subjects is not required and can be performed from a distance. Cameras are commonly used for gait recognition, and a number of researchers have used depth information obtained using an RGB-D camera, such as the Microsoft Kinect. Although depth-based gait recognition has advantages, such as robustness against light conditions or appearance variations, there are also limitations. For instance, the RGB-D camera cannot be used outdoors and the measurement distance is limited to approximately 10 meters. The present paper describes a long short-term memory-based method for gait recognition using a real-time multi-line LiDAR. Very few studies have dealt with LiDAR-based gait recognition, and the present study is the first attempt that combines LiDAR data and long short-term memory for gait recognition and focuses on dealing with different appearances. We collect the first gait recognition dataset that consists of time-series range data for 30 people with clothing variations and show the effectiveness of the proposed approach.