Search

This study introduces a novel approach for analyzing vital signals using continuous-wave (CW) radar, employing an integrated neural network model to overcome the limitations associated with traditional step-by-step signal processing methods. Conventional methods for vital signal monitoring, such as electrocardiograms (ECGs) and sphygmomanometers, require direct contact and impose constraints on specific scenarios. Conversely, our study primarily focused on non-contact measurement techniques, particularly those using CW radar, which is known for its simplicity but faces challenges such as noise interference and complex signal processing. To address these issues, we propose a temporal convolutional network (TCN)-based framework that seamlessly integrates noise removal, demodulation, and fast Fourier transform (FFT) processes into a single neural network. This integration minimizes cumulative errors and processing time, which are common drawbacks of conventional methods. The TCN was trained using a dataset comprising preprocessed in-phase and quadrature (I/Q) signals from the CW radar and corresponding heart rates measured via ECG. The performance of the proposed method was evaluated based on the L1 loss and accuracy against the moving average of the estimated heart rates. The results indicate that the proposed approach has the potential for efficient and accurate non-contact vital signal analysis, opening new avenues in health monitoring and medical research. Additionally, the integration of CW radar and neural networks in our framework offers a robust and scalable solution, enhancing the practicality of non-contact health monitoring systems in diverse environments. This technology can be leveraged in healthcare robots to provide continuous and unobtrusive monitoring of patients' vital signs, enabling timely interventions and improving overall patient care. [ABSTRACT FROM AUTHOR]

Copyright of Journal on Communication / Tongxin Xuebao is the property of Journal on Communications Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Airborne speech enhancement is always a major challenge for the security of airborne systems. Recently, multi-objective learning technology has become one of the mainstream methods of monaural speech enhancement. In this paper, we propose a novel multi-objective method for airborne speech enhancement, called the stacked multiscale densely connected temporal convolutional attention network (SMDTANet). More specifically, the core of SMDTANet includes three parts, namely a stacked multiscale feature extractor, a triple-attention-based temporal convolutional neural network (TA-TCNN), and a densely connected prediction module. The stacked multiscale feature extractor is leveraged to capture comprehensive feature information from noisy log-power spectra (LPS) inputs. Then, the TA-TCNN adopts a combination of these multiscale features and noisy amplitude modulation spectrogram (AMS) features as inputs to improve its powerful temporal modeling capability. In TA-TCNN, we integrate the advantages of channel attention, spatial attention, and T-F attention to design a novel triple-attention module, which can guide the network to suppress irrelevant information and emphasize informative features of different views. The densely connected prediction module is used to reliably control the flow of the information to provide an accurate estimation of clean LPS and the ideal ratio mask (IRM). Moreover, a new joint-weighted (JW) loss function is constructed to further improve the performance without adding to the model complexity. Extensive experiments on real-world airborne conditions show that our SMDTANet can obtain an on-par or better performance compared to other reference methods in terms of all the objective metrics of speech quality and intelligibility. [ABSTRACT FROM AUTHOR]

Deterministic wind power prediction can be used for long time-scale optimization of power dispatching systems, but the probability and fluctuation range of prediction results cannot be calculated. A Bayesian LSTM neural network (BNN-LSTM) is constructed based on Bayesian networks by placing a priori distributions on top of the LSTM network layer weight parameters. First, the temporal convolutional neural network (TCNN) is used to process the historical time-series data for wind power prediction, which is used to extract the correlation features of the time-series data and learn the trend changes of the time-series data. Then, the mutual information entropy method is used to analyze the meteorological dataset of wind power, which is used to eliminate the variables with small correlation and reduce the dimension of the meteorological dataset, so as to simplify the overall structure of the prediction model. At the same time, the Embedding structure is used to learn the temporal classification features of wind power. Finally, the time series data processed by TCNN, the meteorological data after dimensionality reduction, and the time classification feature data are fed into the BNN-LSTM prediction model together. Compared with a Bayesian neural network, continuous interval method, and Temporal Fusion Transformer (TFT), which is one of the most advanced time series prediction networks, the improved BNN-LSTM can respond more accurately to wind power fluctuations with better prediction results. The comprehensive index of probability prediction of pinball loss is smaller than those of the other three methods by 53.2%, 24.4%, and 11.3%, and the Winkler index is 3.5 %, 34.6 %, and 8.2 % smaller, respectively. [ABSTRACT FROM AUTHOR]

Out-of-stock prediction refers to the activity of forecasting the time when a product will not be available for purchase because of an inventory deficiency. Due to difficulties, out-of-stock forecasting models now face certain challenges. Incorrect demand forecasting may result in a lack or excess of goods in stock, a factor that affects client satisfaction and the profitability of companies. Accordingly, the new approach BCHO-TCN LightGBM, which is based on Buzzard Coney Hawk Optimization with a deep temporal convolutional neural network and the Light Gradient-Boosting Machine framework, is developed to deal with all challenges in the existing models in the field of out-of-stock prediction. The role that BCHO plays in the LightGBM-based deep temporal CNNis rooted in modifying the classifier to improve both accuracy and speed. Integrating BCHO into the model training process allows us to optimize and adjust the hyperparameters and the weights of the CNN linked with the temporal DNN, which, in turn, makes the model perform better in the extraction of temporal features from time-series data. This optimization strategy, which derives from the cooperative behaviors and evasion tactics of BCHO, is a powerful source of information for the computational optimization agent. This leads to a faster convergence of the model towards optimal solutions and therefore improves the overall accuracy and predictive abilities of the temporal CNN with the LightGBM algorithm. The results indicate that when using data from Amazon India’s product listings, the model shows a high degree of accuracy, as well as excellent net present value (NPV), present discounted value (PDV), and threat scores, with values reaching 94.52%, 95.16%, 94.81%, and 95.76%, respectively. Likewise, in a k-fold 10 scenario, the model achieves values of 94.81%, 95.60%, 96.28%, and 95.86% for the same metrics.

Accurate estimation of battery health is an effective means of improving the safety and reliability of electrical equipment. However, developing data-driven models to estimate battery state of health (SOH) is challenging when the amount of data is restricted. In this regard, this study proposes a method for estimating the SOH of lithium batteries based on sample data generation and a temporal convolutional neural network. First, we analyzed the charge/discharge curves of the batteries, from which we extracted features that were highly correlated with the SOH decay. Then, we used a Variational Auto-Encoder (VAE) to learn the features and distributions of the sample data to generate highly similar data and enrich the number of samples. Finally, a temporal convolutional neural network (TCN) was built to mine the nonlinear relationship between features and SOH by combining the source and extended domain data to realize SOH estimation. The experimental results show that the proposed method in this study has less than 2% error in SOH estimation, which improves the accuracy by 64.9% based on its baseline model. The feasibility of using data-driven models for battery health management in data-constrained application scenarios is demonstrated. [ABSTRACT FROM AUTHOR]

In the face of increasing irregular temperature patterns and climate shifts, the need for accurate power consumption prediction is becoming increasingly important to ensure a steady supply of electricity. Existing deep learning models have sought to improve prediction accuracy but commonly require greater computational demands. In this research, on the other hand, we introduce DelayNet, a lightweight deep learning model that maintains model efficiency while accommodating extended time sequences. Our DelayNet is designed based on the observation that electronic series data exhibit recurring irregular patterns over time. Furthermore, we present two substantial datasets of electricity consumption records from South Korean buildings spanning nearly two years. Empirical findings demonstrate the model's performance, achieving 21.23%, 43.60%, 17.05% and 21.71% improvement compared to recurrent neural networks, gated-recurrent units, temporal convolutional neural networks and ARIMA models, as well as greatly reducing model complexity and computational requirements. These findings indicate the potential for micro-level power consumption planning, as lightweight models can be implemented on edge devices. [ABSTRACT FROM AUTHOR]

There is recent interest in using model hubs – a collection of pre-trained models – in computer vision tasks. To employ a model hub, we first select a source model and then adapt the model for the target to compensate for differences. There still needs to be more research on model selection and adaption for renewable power forecasts. In particular, none of the related work examines different model selection and adaptation strategies for neural network architectures. Also, none of the current studies investigates the influence of available training samples and considers seasonality in the evaluation. We close these gaps by conducting the first thorough experiment for model selection and adaptation for transfer learning in renewable power forecast, adopting recent developments from the field of computer vision on 667 wind and photovoltaic parks from six datasets. We simulate different amounts of training samples for each season to calculate informative forecast errors. We examine the marginal likelihood and forecast error for model selection for those amounts. Furthermore, we study four adaption strategies. As an extension of the current state of the art, we utilize a Bayesian linear regression for forecasting the response based on features extracted from a neural network. This approach outperforms the baseline with only seven days of training data and shows that fine-tuning is not beneficial with less than three months of data. We further show how combining multiple models through ensembles can significantly improve the model selection and adaptation approach such that we have a similar mean error with only 30 days of training data which is otherwise only possible with an entire year of training data. We achieve a mean error of 9.8 and 14 percent for the most realistic dataset for PV and wind with only seven days of training data.

Copyright of Cyber Security & Data Governance is the property of Editorial Office of Information Technology & Network Security and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Airborne speech enhancement is always a major challenge for the security of airborne systems. Recently, multi-objective learning technology has become one of the mainstream methods of monaural speech enhancement. In this paper, we propose a novel multi-objective method for airborne speech enhancement, called the stacked multiscale densely connected temporal convolutional attention network (SMDTANet). More specifically, the core of SMDTANet includes three parts, namely a stacked multiscale feature extractor, a triple-attention-based temporal convolutional neural network (TA-TCNN), and a densely connected prediction module. The stacked multiscale feature extractor is leveraged to capture comprehensive feature information from noisy log-power spectra (LPS) inputs. Then, the TA-TCNN adopts a combination of these multiscale features and noisy amplitude modulation spectrogram (AMS) features as inputs to improve its powerful temporal modeling capability. In TA-TCNN, we integrate the advantages of channel attention, spatial attention, and T-F attention to design a novel triple-attention module, which can guide the network to suppress irrelevant information and emphasize informative features of different views. The densely connected prediction module is used to reliably control the flow of the information to provide an accurate estimation of clean LPS and the ideal ratio mask (IRM). Moreover, a new joint-weighted (JW) loss function is constructed to further improve the performance without adding to the model complexity. Extensive experiments on real-world airborne conditions show that our SMDTANet can obtain an on-par or better performance compared to other reference methods in terms of all the objective metrics of speech quality and intelligibility.

Accurate wind power prediction is crucial for the safe and stable operation of the power grid. However, wind power generation has large random volatility and intermittency, which increases the difficulty of prediction. In order to construct an effective prediction model based on wind power generation power and achieve stable grid dispatch after wind power is connected to the grid, a wind power generation prediction model based on WT-BiGRU-Attention-TCN is proposed. First, wavelet transform (WT) is used to reduce noises of the sample data. Then, the temporal attention mechanism is incorporated into the bi-directional gated recurrent unit (BiGRU) model to highlight the impact of key time steps on the prediction results while fully extracting the temporal features of the context. Finally, the model performance is enhanced by further extracting more high-level temporal features through a temporal convolutional neural network (TCN). The results show that our proposed model outperforms other baseline models, achieving a root mean square error of 0.066 MW, a mean absolute percentage error of 18.876%, and the coefficient of determination (R2) reaches 0.976. It indicates that the noise-reduction WT technique can significantly improve the model performance, and also shows that using the temporal attention mechanism and TCN can further improve the prediction accuracy.

Accurate estimation of battery health is an effective means of improving the safety and reliability of electrical equipment. However, developing data-driven models to estimate battery state of health (SOH) is challenging when the amount of data is restricted. In this regard, this study proposes a method for estimating the SOH of lithium batteries based on sample data generation and a temporal convolutional neural network. First, we analyzed the charge/discharge curves of the batteries, from which we extracted features that were highly correlated with the SOH decay. Then, we used a Variational Auto-Encoder (VAE) to learn the features and distributions of the sample data to generate highly similar data and enrich the number of samples. Finally, a temporal convolutional neural network (TCN) was built to mine the nonlinear relationship between features and SOH by combining the source and extended domain data to realize SOH estimation. The experimental results show that the proposed method in this study has less than 2% error in SOH estimation, which improves the accuracy by 64.9% based on its baseline model. The feasibility of using data-driven models for battery health management in data-constrained application scenarios is demonstrated.

In the face of increasing irregular temperature patterns and climate shifts, the need for accurate power consumption prediction is becoming increasingly important to ensure a steady supply of electricity. Existing deep learning models have sought to improve prediction accuracy but commonly require greater computational demands. In this research, on the other hand, we introduce DelayNet, a lightweight deep learning model that maintains model efficiency while accommodating extended time sequences. Our DelayNet is designed based on the observation that electronic series data exhibit recurring irregular patterns over time. Furthermore, we present two substantial datasets of electricity consumption records from South Korean buildings spanning nearly two years. Empirical findings demonstrate the model’s performance, achieving 21.23%, 43.60%, 17.05% and 21.71% improvement compared to recurrent neural networks, gated-recurrent units, temporal convolutional neural networks and ARIMA models, as well as greatly reducing model complexity and computational requirements. These findings indicate the potential for micro-level power consumption planning, as lightweight models can be implemented on edge devices.

The Indo-Malaysian region is a hot spot of rapid land-cover and land-use change (LCLUC) with little consensus about the rates and magnitudes of such change. Here we use temporal convolutional neural networks (TempCNNs) to generate a spatiotemporally consistent LCLUC data set for nearly thirty-five years (1982–2015), validated against two reference data sets with over 80 percent accuracy, better than other LCLUC products for the region. Our results both confirm and complicate estimates from earlier work that relied on decadal, rather than interannual, changes in regional land cover. We find forests decrease in mainland and maritime Southeast Asia and increase in South China and South Asia. Consistent with geographic theory about the drivers of land-use change, we find cropland expansion is a driving force for deforestation in mainland Southeast Asia with savannas playing a superior role, suggesting widespread forest degradation in this region. In contrast to earlier work and theory, we find that South China's increasing forest cover comes principally from savanna (rather than cropland) conversion. The explicit interannual LCLUC patterns, rates, and transitions identified in this study provide a valuable data source for studies of land-use theory, environmental and climate changes, and regional land-use policy evaluations. [ABSTRACT FROM AUTHOR]

In view of the issue that the runoff change was affected by many factors, characterized by randomness and nonlinearity and was difficult to predict accurately, based on the data of daily average runoff, daily precipitation and daily average sediment concentration of Huayuankou Hydrometrical Station on the Yellow River from 2008 to 2012, a multivariable input TCN Attention model combined with time convolution neural network (TCN) and attention mechanism was proposed to predict the daily runoff of the station, LSTM model and TCN model were selected for prediction and comparison experiments. The results show that the prediction performance of TCN model and TCN Attention model is better than that of LSTM model; The Attention mechanism can improve the prediction performance of TCN model by adjusting the feature vector weight. Compared with TCN model, the MAE, RMSE and MAPE values of TCN Attention model have been decreased by 20.25%, 24.90% and 24.39% respectively; TCN Attention model has better generalization performance, which can improve the prediction level of daily runoff. [ABSTRACT FROM AUTHOR]

Predicting remaining useful life (RUL) accurately is crucial for improving the reliability of the digitized maintenance and optimizing the operating cycle of devices. Based on the attention mechanism, this article proposes a temporal convolutional neural network (TCN)-bidirectional gate recurrent unit (BiGRU) model for the RUL prediction. The proposed model can accurately predict the RUL by fusing the time-series information recorded by multiple sensors. To suppress noise signals, the sensor’s raw signal filtered by the contribution rate is processed by variational mode decomposition (VMD) and decomposed into intrinsic mode components (IMFs) with different modes. Then, grid search and ${K}$ -fold cross validation are performed to determine the optimal hyperparameters of the proposed Attention TCN-BiGRU model to accurately capture the correlation of IMFs in the spatial dimension and the decay characteristics in the temporal dimension. The performance of the proposed model is validated on the commercial modular aero-propulsion system simulation (C-MAPSS) turbofan engine dataset and the sealing ring dataset. The experimental results indicate that the proposed model can achieve more accurate RUL predictions than other related models. [ABSTRACT FROM AUTHOR]

We employ and analyze various machine learning models for daily cryptocurrency market prediction and trading. We train the models to predict binary relative daily market movements of the 100 largest cryptocurrencies. Our results show that all employed models make statistically viable predictions, whereby the average accuracy values calculated on all cryptocurrencies range from 52.9% to 54.1%. These accuracy values increase to a range from 57.5% to 59.5% when calculated on the subset of predictions with the 10% highest model confidences per class and day. We find that a long-short portfolio strategy based on the predictions of the employed LSTM and GRU ensemble models yields an annualized out-of-sample Sharpe ratio after transaction costs of 3.23 and 3.12, respectively. In comparison, the buy-and-hold benchmark market portfolio strategy only yields a Sharpe ratio of 1.33. These results indicate a challenge to weak form cryptocurrency market efficiency, albeit the influence of certain limits to arbitrage cannot be entirely ruled out. [ABSTRACT FROM AUTHOR]

Wind energy reserves are large worldwide, but their randomness and volatility hinder wind power development. To promote the utilization of wind energy and improve the accuracy of wind power prediction, we comprehensively consider the influence of wind farm environmental factors and historical power on wind power generation. This paper presents a short-term wind power prediction model based on time convolution neural network (TCN) and variational mode decomposition (VMD). First, due to the non-smooth characteristics of the wind farm environmental data, this paper uses VMD to decompose the data of each environmental variable to reduce the influence of the random noise of the data on the prediction model. Then, the modal components with rich feature information are extracted according to the Pearson correlation coefficient and Maximal information coefficient (MIC) between each modal component and the power. Thirdly, a prediction model based on TCN is trained according to the preferred modal components and historical power data to achieve accurate short-term wind power prediction. In this paper, the model is trained and tested with a public wind power dataset provided by the Spanish Power Company. The simulation results show that the model has higher prediction accuracy, with MAPE and R2 are 2.79% and 0.9985, respectively. Compared with the conventional long short-term neural network (LSTM) model, the model in this paper has good prediction accuracy and robustness. [ABSTRACT FROM AUTHOR]

Radar high-resolution range profile (HRRP) sequences have received great attention in radar automatic target recognition (RATR) due to the advantages of rich target structure features, small storage space requirement, and easy acquisition and processing. The traditional HRRP sequence recognition methods cannot effectively extract the deep time-series features, resulting in poor recognition performance. In addition, the traditional methods are more complex, with weak adaptive capability to different length sequences and robustness to variant targets. To tackle these problems, we propose an efficient method for HRRP sequence recognition based on temporal convolutional networks with sequence length-adaptive algorithm and elastic net regularization (ERTCN-SLA). The proposed method extracts the deep temporal features of HRRP sequences by stacking multiple modified residual blocks of dilated causal convolution. The batch normalization (BN) is adopted to prevent the over-fitting issue in deep networks due to the offset in parameter distribution. Time-series features extracted by the temporal convolutional neural network are fused using adaptive average pooling, which can reduce the adverse effects of local noise, obtain the global pattern of the time-series features, and improve the robustness and generalizability of temporal convolutional networks (TCN). Further, the unstructured sparsity of the model is achieved by using the elastic net regularization method for feature selection. A sequence length-adaptive algorithm based on the Bayesian optimization algorithm is proposed to achieve adaptive end-to-end training and recognition for HRRP sequences of different lengths. Experimental results on moving target and stationary target acquisition and recognition (MSTAR) datasets show that the proposed method has superior recognition accuracy compared with other remarkable time-series methods. Furthermore, the proposed method has better robustness and sparsity. [ABSTRACT FROM AUTHOR]

Accurately predicting the burn-through point (BTP) is crucial for achieving stable control of the sintering process. However, accurately measuring the raw BTP is difficult due to the harsh production environment and poor thermocouple measurement accuracy of the temperature of exhaust gas in bellows. This paper proposes a prediction model of the BTP with data correction based on the feature matching of cross-section frame at discharge end. Firstly, a feature extraction method of cross-section frames at discharge end is designed. Next, the cross-section frame at discharge end features matching method is used to correct the raw BTP, and this method corrects anomalous data resulting from sensor failures. Finally, the temporal convolutional neural network and gated recurrent unit are used to predict the corrected BTP. The prediction model considers the cross-section frame feature at discharge end and state parameters as inputs, and it can achieve accurate prediction of the corrected BTP. A series of comparative experiments are conducted to verify the feasibility and effectiveness of the proposed model. At the same time, this paper also designs industrial implementation plan,and use actual operation data to verify the feasibility of the designed industrial implementation plan. • A BTP correction method based on cross-section frame feature matching is proposed. • A BTP prediction model with data correction using TCN-GRU is proposed. • The designed industrial implementation scheme realizes the application of the model. [ABSTRACT FROM AUTHOR]

The frequent incidents of password leakage have increased people's attention and research on password security. Password guessing is an essential part of password cracking and password security research. The progression of deep learning technology provides a promising way to improve the efficiency of password guessing. However, the mainstream models proposed for password guessing, such as RNN (or other variants, such as LSTM, GRU), GAN and VAE still face some problems, such as the low efficiency and high repetition rate of the generated passwords. In this paper, we propose a password-guessing model based on the temporal convolutional neural network (PassTCN). To further improve the performance of the generated passwords, we propose a novel password probability label-learning method, which reconstructs labels based on the password probability distribution of the training set and deduplicates the training set when training. Experiments on the RockYou dataset showed that, when generating 10 8 passwords, the coverage rate of PassTCN with password probability label learning (PassTCN-PPLL) reached 12.6%, which is 87.2%, 72.6% and 42.9% higher than PassGAN (a password-guessing model based on GAN), VAEPass (a password-guessing model based on VAE) and FLA (a password-guessing model based on LSTM), respectively. The repetition rate of our model is 25.9%, which is 45.1%, 31.7% and 17.4% lower than that of PassGAN, VAEPass and FLA, respectively. The results confirm that our approach not only improves the coverage rate but also reduces the repetition rate. [ABSTRACT FROM AUTHOR]

Keyword spotting (KWS) plays a crucial role in human–machine interactions involving smart devices. In recent years, temporal convolutional networks (TCNs) have performed outstandingly with less computational complexity, in comparison with classical convolutional neural network (CNN) methods. However, it remains challenging to achieve a trade-off between a small-footprint model and high accuracy for the edge deployment of the KWS system. In this article, we propose a small-footprint model based on a modified temporal efficient neural network (TENet) and a simplified mel-frequency cepstrum coefficient (MFCC) algorithm. With the batch-norm folding and int8 quantization of the network, our model achieves the accuracy of 95.36% on Google Speech Command Dataset (GSCD) with only 18 K parameters and 461 K multiplications. Furthermore, following a hardware/model co-design approach, we propose an optimized dataflow and a configurable hardware architecture for TENet inference. The proposed accelerator implemented on Xilinx zynq 7z020 achieves an energy efficiency of 25.6 GOPS/W and reduces the runtime by 3.1× compared with state-of-the-art work. [ABSTRACT FROM AUTHOR]

This research presents a new method based on a combined temporal convolutional neural network and long-short term memory neural network for the real-time assessment of short-term voltage stability to keep the electric grid in a secure state. The assessment includes both the voltage instability (stable state or unstable state) and the fault-induced delayed voltage recovery phenomenon subjected to disturbance. The trained model uses the time series post-disturbance bus voltage trajectories as the input in order to predict the stability state of the power system in a computationally efficient manner. The proposed method also utilizes a transfer learning approach that acclimates to the pre-trained model using only a few labeled samples, which assesses voltage instability under unseen network topology change conditions. Finally, the performance evaluated on the IEEE 9 Bus and New England 39 Bus test systems shows that the proposed method gives superior accuracy with higher efficacy and thus is suitable for online application. [ABSTRACT FROM AUTHOR]

In the recent research era, artificial intelligence techniques have been used for computer vision, big data analysis, and detection systems. The development of these advanced technologies has also increased security and privacy issues. One kind of this issue is Deepfakes which is the combined word of deep learning and fake. DeepFake refers to the formation of a fake image or video using artificial intelligence approaches which are created for political abuse, fake data transfer, and pornography. This paper has developed a Deepfake detection method by examining the computer vision features of the digital content. The computer vision features based on the frame change are extracted using a proposed deep learning model called the Cascaded Deep Sparse Auto Encoder (CDSAE) trained by temporal CNN. The detection process is performed using a Deep Neural Network (DNN) to classify the deep fake image/video from the real image/video. The proposed model is implemented using Face2Face, FaceSwap, and DFDC datasets which have secured an improved detection rate when compared to the traditional deep fake detection approaches.

Predicting change from multivariate time series has relevant applications ranging from the medical to engineering fields. Multisensory stimulation therapy in patients with dementia aims to change the patient's behavioral state. For example, patients who exhibit a baseline of agitation may be paced to change their behavioral state to relaxed. This study aimed to predict changes in one's behavioral state from the analysis of the physiological and neurovegetative parameters to support the therapist during the stimulation session. In order to extract valuable indicators for predicting changes, both handcrafted and learned features were evaluated and compared. The handcrafted features were defined starting from the CATCH22 feature collection, while the learned ones were extracted using a temporal convolutional network, and the behavioral state was predicted through bidirectional long short-term memory auto-encoder, operating jointly. From the comparison with the state of the art, the learned features-based approach exhibits superior performance with accuracy rates of up to 99.42% with a time window of 70 seconds and up to 98.44% with a time window of 10 seconds. [ABSTRACT FROM AUTHOR]

Recently, the end-to-end automatic speech recognition system based on Transformer has been widely popularized, but the multi-head self-attention mechanism in Transformer is not sensitive to the position information of the input sequence, and its flexible alignment has poor generalization performance in the face of noisy speech. To solve the above problems, firstly, this paper proposed to use TCN to enhance the capture of location information by neural network model. Secondly, on the basis of the above, it proposed the TCN-Transformer- CTC model by fusing CTC. Without using any language model, the experimental results on AISHELL-1,an open source speech database of Mandarin Chinese, show that the word error rate of TCN-Transformer-CTC is relatively reduced by 10.91% compared with Transformer, and the final word error rate of the model is reduced to 5.31% which verifies that the proposed model has a certain degree of advancement. [ABSTRACT FROM AUTHOR]

Smart district heating system (SDHS) is an important way to realize green energy saving and comfortable heating in the future, which is conducive to improving energy utilization efficiency and reducing pollution emissions. The accurate prediction algorithm of heating load plays an important role in on-demand heat supply, however, the heating load prediction is a complicated nonlinear optimization problem, and the prediction accuracy is limited due to the poor nonlinear expression ability of the traditional prediction algorithms. This paper proposes a heating load prediction model based on temporal convolutional neural network (TCN), which implements the rapid extraction of complex data features due to the integration of both the parallel feature processing of convolution neural network (CNN) and the time-domain modeling capability of recurrent neural network (RNN). The engineering data of four heat exchange stations located in Anyang, China in the 2018 heating season is used to evaluate and verify the performance of proposed prediction algorithm based on TCN, and the comprehensive comparisons with state-of-the-art algorithms, such as RFR, ETR, GBR, SVR, NuSVR, SGD, Bagging, Boosting, MLP, RNN, LSTM, etc., were analyzed carefully. The experimental results shown that the proposed heat load prediction algorithm based on TCN has performance superiority.

Wind energy reserves are large worldwide, but their randomness and volatility hinder wind power development. To promote the utilization of wind energy and improve the accuracy of wind power prediction, we comprehensively consider the influence of wind farm environmental factors and historical power on wind power generation. This paper presents a short-term wind power prediction model based on time convolution neural network (TCN) and variational mode decomposition (VMD). First, due to the non-smooth characteristics of the wind farm environmental data, this paper uses VMD to decompose the data of each environmental variable to reduce the influence of the random noise of the data on the prediction model. Then, the modal components with rich feature information are extracted according to the Pearson correlation coefficient and Maximal information coefficient (MIC) between each modal component and the power. Thirdly, a prediction model based on TCN is trained according to the preferred modal components and historical power data to achieve accurate short-term wind power prediction. In this paper, the model is trained and tested with a public wind power dataset provided by the Spanish Power Company. The simulation results show that the model has higher prediction accuracy, with MAPE and R2 are 2.79% and 0.9985, respectively. Compared with the conventional long short-term neural network (LSTM) model, the model in this paper has good prediction accuracy and robustness.

The frequent incidents of password leakage have increased people’s attention and research on password security. Password guessing is an essential part of password cracking and password security research. The progression of deep learning technology provides a promising way to improve the efficiency of password guessing. However, the mainstream models proposed for password guessing, such as RNN (or other variants, such as LSTM, GRU), GAN and VAE still face some problems, such as the low efficiency and high repetition rate of the generated passwords. In this paper, we propose a password-guessing model based on the temporal convolutional neural network (PassTCN). To further improve the performance of the generated passwords, we propose a novel password probability label-learning method, which reconstructs labels based on the password probability distribution of the training set and deduplicates the training set when training. Experiments on the RockYou dataset showed that, when generating 108 passwords, the coverage rate of PassTCN with password probability label learning (PassTCN-PPLL) reached 12.6%, which is 87.2%, 72.6% and 42.9% higher than PassGAN (a password-guessing model based on GAN), VAEPass (a password-guessing model based on VAE) and FLA (a password-guessing model based on LSTM), respectively. The repetition rate of our model is 25.9%, which is 45.1%, 31.7% and 17.4% lower than that of PassGAN, VAEPass and FLA, respectively. The results confirm that our approach not only improves the coverage rate but also reduces the repetition rate.

Multivariate time series (MTS) forecasting is a research field that is gaining more and more importance as time series data generators proliferate in the growing era of Internet of Things. Deep learning architecture for MTS data has been and still a very active research area as there is no comprehensive comparative study of the different architectures, let alone a perfect architecture that can solve all types of time series modeling problem. In this paper, we start by highlighting time series analysis requirements, to propose a new model for conditional multivariate time series forecasting based on the lately introduced WaveNet architecture. Our model is composed of stacked residual causal dilated convolutions that provide large scope in time series history and foster learning of long-term dependencies. Parameterized skip connections allow catching early trends and properties while conditioning enables modeling the associations that exist between multivariate data. We used group normalization for its stability and independence from the batch size. We present a deep comparison of the structure, flexibility, flow control, memory consumption, robustness and stability capacities of our new model against those of recurrent neural networks state-of-the-art approaches—long-short-term-memory (LSTM) and gated recurrent unit (GRU). To assess the performances of our model, we conduct extensive experimental testing on the task of urban air quality prediction in Marrakesh city using six real-world multi-sensor and multivariate time series datasets. We compare the results of our model against the results of LSTM and GRU. Our experiments revealed that our proposed WaveNet-temporal-CNN outperform recurrent models ability to learn long-term dependencies in a time-efficient way. [ABSTRACT FROM AUTHOR]

Photovoltaic (PV) power has attracted widespread attention from many countries around the world due to its clean and renewable characteristics. To ensure the stable operation of the power system, accurate PV power forecasting has become a mandatory and challenging task. Currently, deep learning methods have become a vital approach in the field of PV power forecasting. In this work, a multistage attention neural network based on neural ordinary differential equation (MANODE) is proposed to address the main limitations of previous deep learning methods applied to PV power forecasting. Based on the neural ordinary differential equation (NODE), MANODE optimizes the long short-term memory network (LSTM) and temporal convolutional neural network (TCN), and combines the attention mechanism to achieve fine-grained spatio-temporal information extraction of PV series. In addition, the proposed MANODE model is applied to three different PV series collected from the Alice Springs meteorological station. Compared to previous state-of-the-art methods, the proposed method reduces the PV power forecasting error by at least 12.05%, 13.15%, and 9.71% on three different PV datasets, in terms of mean absolute error metric. The average errors of the MANODE method in four-hour-ahead PV power forecasting on the three datasets are 0.321, 0.350, and 0.567. • A novel neural network method called MANODE is applied in PV power forecasting. • MANODE addresses drawbacks of previous PV forecasting methods using LSTMs or CNNs. • Three multivariate datasets located in Australia are adopted in this work. • MANODE outperforms previous state-of-the-art neural network method in PV forecasting. [ABSTRACT FROM AUTHOR]

Power generation and consumption predictions are fundamental for smart grid operations, addressing challenges posed by renewable energy variability and irregular consumer demand. This study introduces the Dual Sequence Predictive Model (DSPM) based on Spatiotemporal CNN (STCNN) architecture, offering a holistic solution for forecasting Electricity Generation (EG) and Electricity Consumption (EC) collectively. Leveraging STCNN, the DSPM efficiently extracts spatial and temporal information, improving prediction accuracy and processing speed. A 1D spatial attention module is added to capture crucial spatial dependencies in historical data. Incorporating shared historical weather information streamlines learning process and reduces model complexity. Extensive benchmark evaluations demonstrate the DSPM superior performance, achieving the lowest error rates compared to baseline models. The DSPM also employs the Kullback–Leibler Divergence (KLD) algorithm for power generation and consumption matching, ensuring efficient power distribution within smart and microgrids. Furthermore, the DSPM is evaluated alongside Single Sequence Prediction (SSP) models, providing a comprehensive analysis of its capabilities and comparisons with state-of-the-art models. The SSP model achieved an average RMSE reduction of 17.3% for solar EG data and 15.86% for IHEPC residential EC data, while DSPM reduced RMSE by 7.57% over PowerGrid dataset compared to baseline models. • Developed a DSPM for collective prediction of power generation and consumption. • Unified architecture for spatiotemporal feature learning. • Model evaluation for load and demand prediction individually. • Integration of the KLD algorithm for efficient power matching. • DSPM outperforms baseline models and reducing operational complexity by up to 20%. [ABSTRACT FROM AUTHOR]

This research presents a new method based on a combined temporal convolutional neural network and long-short term memory neural network for the real-time assessment of short-term voltage stability to keep the electric grid in a secure state. The assessment includes both the voltage instability (stable state or unstable state) and the fault-induced delayed voltage recovery phenomenon subjected to disturbance. The trained model uses the time series post-disturbance bus voltage trajectories as the input in order to predict the stability state of the power system in a computationally efficient manner. The proposed method also utilizes a transfer learning approach that acclimates to the pre-trained model using only a few labeled samples, which assesses voltage instability under unseen network topology change conditions. Finally, the performance evaluated on the IEEE 9 Bus and New England 39 Bus test systems shows that the proposed method gives superior accuracy with higher efficacy and thus is suitable for online application.

In this study, a novel temporal convolutional neural network (TCNN) model is developed for long-term streamflow projection in California within the Catchment Attributes for Large-Sample Studies (CAMELS) watershed regions. The TCNN model consists of several convolution blocks and causal convolution is used as physical constraint. The ensemble performance of the model is first compared with other machine learning models for streamflow prediction. The model is further assessed through comparison with reduced models and using different hyperparameters, with results suggesting that this model correctly ascertains the physical relationship between input variables and streamflow. The stability of the model and its behavior in the extrapolated regime is assessed through an idealized extreme test with quadruple precipitation and 5°C higher temperature. Future streamflow projections are then developed using daily high-resolution Localized Constructed Analogs dataset (LOCA). To understand the importance of the nonlinear machine learning approach, we estimate the degree of nonlinearity in the streamflow response among input variables. Our work shows the ability and potential for TCNNs to perform future hydrology projections.

The development of rapid and non-destructive prediction technology for fruit quality after harvest could enhance market competitiveness and profitability. The soluble solids content (SSC) is an essential quality index of fruit. This study aims to predict the SSC of crown pear using visible/near-infrared (Vis/NIR) spectroscopy (397–1187 nm) with deep learning model MLP-CNN-TCN. Firstly, the spectral information of crown pears was collected, and Savitzky-Golay (SG) smoothing method and Standard normal variate (SNV) were used to preprocess the spectral data. Secondly, the Multi-layer perceptron (MLP) method was used to reduce the dimension of the preprocessed data, and the reduced data was input into a one-dimensional convolutional neural network (1D-CNN) to extract spectral features. Finally, a Temporal convolutional neural network (TCN) was used to establish a regression model to predict pear SSC. This method was also compared with feedforward neural network (FNN), MLP, 1D-CNN, partial least squares regression (PLSR), and support vector regression (SVR). The MLP-CNN-TCN model obtained better prediction performance, with a prediction determination coefficient (R P 2) of 0.956 for SSC. This study demonstrated that the combination of Vis/NIR spectroscopy and MLP-CNN-TCN method could rapidly and non-destructively detect SSC of crown pear, and provide a new regression alternative for the prediction of fruit SSC. • The SSC of crown pears was predicted by visible/NIR spectroscopy. • A novel deep regression model MLP-CNN-TCN to predict pear SSC was proposed. • Data augmentation using all sampling points was investigated for SSC prediction. • Important wavelengths for SSC prediction of MLP-CNN-TCN was identified by Grad-CAM. [ABSTRACT FROM AUTHOR]

The prediction of chaotic time series, which constitutes many systems in the field of science and engineering, has recently become the focus of attention of researchers. Chaotic time series prediction is making future predictions about these systems using previously observed data for a nonlinear chaotic system with a known initial condition. Chaotic time series prediction can be applied in many fields such as weather forecasting, finance and stock markets. Many disciplines work on solving time series prediction problem, ranging from forecasting weather events days in advance to traders predicting the future of stocks. In recent studies, it has been observed that hybrid deep neural network methods give better performance in solving time series prediction problems and have gained popularity in order to benefit from the advantages of more than one method in solving such problems. In this study, a hybrid deep neural network architecture is proposed for chaotic time series prediction. The used hybrid approach includes both temporal convolutional network to extract low level features from input and recurrent neural network layers such as long short-term memory and gated recurrent units to capture temporal information. Simulations were carried out on nine different chaotic time series dataset which are obtained from Lorenz, Rössler and a Lorenz-like chaotic equation sets, and twenty-one electrocardiogram (ECG) recordings of patients with arrhythmias. In the benchmark study, in which twelve different methods, including classical machine learning, deep neural network and hybrid models were used, the proposed model achieved the best prediction performance with an average root-mean-square error (RMSE) value of 0.0022 for chaotic dataset and 0.0082 for ECG arrhythmia dataset. Performance evaluation metrics show that the proposed hybrid architecture can compete with the models in state-of-the-art studies in chaotic time series prediction. • Chaotic time series prediction using 30 different dataset. • Novel TCN with RNN architectures is used for chaotic time series prediction. • Benchmarking study with 12 different methods including TCN with RNNs, classical ML and DNNs. • Proposed method has obtained lowest average RMSE as 0.0022. [ABSTRACT FROM AUTHOR]

Heating, ventilation, and air-conditioning systems provide a comfortable indoor thermal environment, but high energy consumption is often necessary to achieve an adequate level of indoor thermal comfort. However, it is challenging to design an energy-efficient thermal comfort control strategy, mainly because the internal thermal environment is influenced by complicated factors and difficult to model accurately. To solve this problem, a control strategy incorporating the parallel temporal convolutional neural network (PTCN) and the improved chimp optimization algorithm (ICHOA) is proposed for thermal comfort control of buildings. Thermal comfort control is transformed into a cost-minimization problem by establishing an objective function for both the future thermal comfort of the occupants and energy consumption and optimizing multiple air-conditioning temperature set points for the coming day. First, to ensure the prediction performance, a PTCN model was developed to predict the energy consumption and thermal comfort under different factors. An opposition-learning-based adaptive chimp algorithm was then used to solve the objective function to output the optimal set temperature. Finally, the superiority of the PTCN-ICHOA optimization strategy was verified using an office building in Jinan as an example. In winter and summer experiments, the proposed PTCN model achieved the lowest prediction errors among the models compared in terms of energy and temperature prediction. Furthermore, the PTCN-ICHOA optimization model exhibited faster convergence than the other models for both experiments. Through the proposed optimization strategy, energy consumption savings of approximately 6.3%–8.1% can be achieved while maintaining indoor thermal comfort. • A predictive optimization model for building thermal comfort control. • Parallel temporal neural network extracts features of multiple factors separately. • Improved chimp algorithm using selected opposition learning and adaptive weight. • The predictive optimization model realized energy savings of 6.3%–8.1%. [ABSTRACT FROM AUTHOR]

In the recent research era, artificial intelligence techniques have been used for computer vision, big data analysis, and detection systems. The development of these advanced technologies has also increased security and privacy issues. One kind of this issue is Deepfakes which is the combined word of deep learning and fake. DeepFake refers to the formation of a fake image or video using artificial intelligence approaches which are created for political abuse, fake data transfer, and pornography. This paper has developed a Deepfake detection method by examining the computer vision features of the digital content. The computer vision features based on the frame change are extracted using a proposed deep learning model called the Cascaded Deep Sparse Auto Encoder (CDSAE) trained by temporal CNN. The detection process is performed using a Deep Neural Network (DNN) to classify the deep fake image/video from the real image/video. The proposed model is implemented using Face2Face, FaceSwap, and DFDC datasets which have secured an improved detection rate when compared to the traditional deep fake detection approaches., Competing Interests: The authors declare there are no competing interests., (©2022 Balasubramanian et al.)

Background: The dynamic tracking of tumors with radiation beams in radiation therapy requires the prediction of real-time target locations prior to beam delivery, as treatment involving radiation beams and gating tracking results in time latency.Objective: In this study, a deep learning model that was based on a temporal convolutional neural network was developed to predict internal target locations by using multiple external markers.Methods: Respiratory signals from 69 treatment fractions of 21 patients with cancer who were treated with the CyberKnife Synchrony device (Accuray Incorporated) were used to train and test the model. The reported model's performance was evaluated by comparing the model to a long short-term memory model in terms of the root mean square errors (RMSEs) of real and predicted respiratory signals. The effect of the number of external markers was also investigated.Results: The average RMSEs of predicted (ahead time=400 ms) respiratory motion in the superior-inferior, anterior-posterior, and left-right directions and in 3D space were 0.49 mm, 0.28 mm, 0.25 mm, and 0.67 mm, respectively.Conclusions: The experiment results demonstrated that the temporal convolutional neural network-based respiratory prediction model could predict respiratory signals with submillimeter accuracy. [ABSTRACT FROM AUTHOR]

• A new framework for automatic sleep stage classification from single-channel EEG. • Using a deep learning algorithm including TCNN and new data augmentation technique. • Evaluating our model using two public sleep datasets, Sleep-EDF-2013 and 2018. • The best total accuracy and kappa score compared to the state-of-the-art methods. Background and objective: This paper presents a new framework for automatic classification of sleep stages using a deep learning algorithm from single-channel EEG signals. Each segmented EEG signal appended with its label of stages is fed into a deep learning model to create an automatic sleep stage classification. This is one of the most important problems that is critical to the realization of monitoring patients with sleep disorder. Methods: In the present study, a neural network architecture is introduced utilizing Convolutional Neural Networks (CNNs) to extract features, followed by Temporal Convolutional Neural Network to extract the temporal features from the extracted features vector of CNN. Finally, the performance of our model is improved by a Conditional Random Field layer. We also employed a new data augmentation technique to enhance the CNNs training which has auxiliary effects. Results: We evaluated our model by two different single-channel EEG signals (i.e., Fpz-Cz and Pz-Oz EEG channels) from two public sleep datasets, named Sleep-EDF-2013 and Sleep-EDF-2018. The evaluation results on both datasets showed that our model obtains the best total accuracy and kappa score (EDF-2013: 85.39%- 0.80, EDF-2018: 82.46%- 0.76) compared to the state-of-the-art methods. Conclusions: This study will possibly allow us to have a wearable sleep monitoring system with a single-channel EEG. Also, unlike hand-crafted features methods, our model finds its own patterns through training epochs, and therefore, it may minimize engineering bias. [ABSTRACT FROM AUTHOR]

• Using deep temporal convolutional networks for building simulation surrogate models. • Deep network processes annual hourly weather time series data (≈ 150,000 inputs). • Accurate emulation of simulation outcomes for all locations in Canada. • 3% error in estimating annual heating demand for unseen locations and building designs. • Reasonable accuracy ( R 2 = 0.92) in estimating hourly demands given weather data. Surrogate models can emulate physics-based building energy simulation with a machine learning model trained on simulation input and output data. The trained model is extremely fast to run, allowing us to estimate simulation outcomes for thousands of different building designs in seconds. Recent studies have shown the diverse benefits for sustainable building design. Surrogates were applied to provide rapid feedback at the early design stage, to accelerate sensitivity analysis, uncertainty analysis and design optimization, or to improve building model calibration. However, the current process of surrogate modelling offers much room for improvement. In particular, a surrogate model is bound to the specific building design problem it has been trained for. This includes a specific site, requiring time-intensive retraining if the building performance at another location is to be analysed. In this paper, we develop a single surrogate model that spans arbitrarily many locations. For that purpose, we are among the first to use a deep temporal convolutional neural network to process annual multivariate weather data with hourly resolution (≈ 150,000 inputs). The network learns features relevant to estimate heating or cooling demand. We combine these location-specific weather features with building design parameters to serve as input to a single surrogate model (feed-forward neural network). In a case study with 569 weather files from locations in Canada, we show that the surrogate model deviates by less than 3% when predicting annual heating demand for new building designs at locations outside of the training data set. [ABSTRACT FROM AUTHOR]