One-dimensional Convolutional Neural Network Research Articles

A Novel Approach for Diagnosing Transformer Internal Defects and Inrush Current Based on 1DCNN and LSTM Deep Learning

In power systems, power transformer (Pt) protection plays a vital role in ensuring that customers have a reliable power supply. Correctly recognizing inrush currents from internal defects and preventing differential relay malfunctions are two of biggest challenges facing the differential protection of power transformers. Although previous approaches suggested to overcome these issues have promising outcomes, increasing the accuracy and reducing the execution time and complexity of transformer differential relays are still interesting topics for researchers. Accordingly, a new fault diagnostic method based on wavelet transform (WT) and deep learning is introduced in paper. In the proposed approach, Discrete WT is used to extract the features of differential currents, and combined one-dimensional convolutional neural networks and long short-term memory (1DCNN-LSTM)) is applied for classify internal faults from other abnormal events. High accuracy, no need for any thresholds or transformer parameters and fast fault detection are the main advantages of the proposed approach. The simulation results for a 132/11 kV, 63 MVA power transformer approved the proposed method for its ability to accurately differentiate between inrush currents and internal defects after 5 ms, as well as its accuracy for abnormal event classification of about 99.4%.

Generalized method for distributed detection and quantification of cracks in bridges

The development of a generalized machine learning approach based on distributed detection and quantification of cracks by optical fibers is described in this article. A Brillouin scattering optical fiber sensor system was employed to develop, test, and verify the method. The main components of the approach described herein consist of an unsupervised crack identification module based on the iForest algorithm and a crack quantification component by the one-dimensional convolutional neural network method. The main attribute of this model is the versatility for application in various types of structures. The proposed method does not require further application-dependent training or calibration as long as the structural applications employ the same optical fiber type and installation adhesives. The effectiveness of the proposed method was verified by two experiments involving a 15-m steel beam in the laboratory and monitoring a twin set of 332-m-long, five-span continuous box girder concrete bridges. Regarding crack detection capabilities, it was possible to detect 107 out of 112 cracks in the laboratory beam and 20 out of the 21 in the bridges. The resolution of crack opening displacements for the steel beam and concrete bridges were 20.6 and 21.7 µm, respectively. The verification experiments further indicated the generality of the approach in applications to various types of structures and materials.

Fusion model based RUL prediction method of lithium-ion battery under working conditions

Under working conditions, since the remaining useful life (RUL) prediction of lithium-ion battery is subject to uncertainties of random charging and discharging, and infeasibility of battery capacity test, a fusion model based RUL prediction method was proposed. First, the feature learning method of lithium-ion batteries was developed by synthesizing manual extraction and one-dimensional convolutional neural network (1DCNN) extraction. Then, a fused method was proposed to estimate the historical available capacity through exploring the spatial and temporal relationship of features, and the long short-term memory (LSTM) network model was adopted for predicting the RUL of lithium-ion battery. The proposed method was verified through the comparison of different methods, and the results show that it can realize highly precise and stable capacity estimation and RUL prediction under working conditions.

Fusing multimodal data of nature-economy-society for large-scale urban building height estimation

The building height holds significant importance for comprehensively understanding urban morphology, enhancing urban planning, and fostering sustainable development. Although many methods using optical and SAR images have been presented for building height estimation, these methods fall short in capturing the influences of economic and social attributes on building height. In this study, we introduced a Nature-Economy-Society (NES) feature model to comprehensively represent building height information, and established a multi-scale One-Dimensional (1-D) Convolutional Neural Network for predicting building heights, referred to as NES-CNN. First, we derived the natural attributes of urban buildings from time-series Sentinel-1 SAR images and Sentinel-2 multispectral images, as well as World Settlement Footprint (WSF) data and Digital Elevation Model (DEM), economic attributes from nighttime light and Gross Domestic Product (GDP) data, and social function attributes from Points of Interest (POI) data. Second, an autoencoder is employed to reduce the dimensionality of the high-dimensional natural attribute features, minimizing data redundancy. Finally, the multi-scale 1-D CNN model is presented to explore the correlations between the multi-source and heterogeneous NES features and building height information, facilitating the prediction of building height. In experiments, we applied the proposed method to estimate building heights in Beijing and Shanghai at a spatial resolution of 10 m. The results indicated that for Beijing, the RMSE, MAE, and R values are 6.93 m, 4.41 m, and 0.84, respectively, while for Shanghai, these values are 7.57 m, 5.38 m, and 0.80, respectively. The addition of social and economic attribute information decreases the RMSE by 6 % in both Beijing and Shanghai compared with using only natural attributes. In comparison to existing studies at the same mapping resolution, RMSE decreases by 39 % for Beijing and 51 % for Shanghai. The innovative and inspiring nature of this study lies in its application to large-scale building height estimation.

Combining data augmentation and deep learning for improved epilepsy detection.

In recent years, the use of EEG signals for seizure detection has gained widespread academic attention. Aiming at the problem of overfitting deep learning models due to the small number of EEG signal data during epilepsy detection, this paper proposes an epilepsy detection method that combines data augmentation and deep learning. First, the Adversarial and Mixup Data Augmentation (AMDA) method is used to realize the data augmentation, which effectively enriches the number of training samples. To further improve the classification accuracy and robustness of epilepsy detection, this paper proposes a one-dimensional convolutional neural network and gated recurrent unit (AM-1D CNN-GRU) network model based on attention mechanism for epilepsy detection. The experimental results show that the performance of epilepsy detection achieved by using augmented data is significantly improved, and the accuracy, sensitivity, and area under the subject's working characteristic curve are up to 96.06, 95.48%, and 0.9637, respectively. Compared with the non-augmented data, all indicators are increased by more than 6.2%. Meanwhile, the detection performance was significantly improved compared with other epilepsy detection methods. The results of this research can provide a reference for the clinical application of epilepsy detection.

Integrating machine learning and feature analysis for predicting and managing thermal deformation in machine tools

This study develops a method to predict and compensate for thermal deformation in machine tools, focusing on selecting temperature-sensitive points, establishing a predictive model, and detecting temperature anomalies. Optimal sensing points were identified using feature ranking algorithms and Partial Dependence Plots, improving Z-axis deformation identification. The research also evaluated various machine learning models for Z-axis deformation prediction, notably optimizing Gaussian Process Regression for superior accuracy. Additionally, anomaly detection in temperature sensors was addressed using One-Dimensional Convolutional Neural Networks and Long Short-Term Memory Networks, enhancing system reliability and robustness. This approach offers a comprehensive solution for thermal deformation in machine tools, contributing to the field's precision and efficiency.

A hybrid fault diagnosis method for rolling bearings based on GGRU-1DCNN with AdaBN algorithm under multiple load conditions

The fault diagnosis of rolling bearings is a critical aspect of rotating machinery, as it significantly contributes to the overall operational safety of the mechanical equipment. In the practical engineering environment, the complex and variable working conditions, along with the presence of overlapping noise, contribute to intricate frequency information in the acquired signals and their highly time-dependent characteristics, which makes it difficult to extract the available fault features hidden in the signal. Based on this, a hybrid fault diagnosis method named GGRU-1DCNN-AdaBN is introduced, which combines improved gap-gated recurrent unit network (GGRU), one-dimensional convolutional neural network (1DCNN), and adaptive batch normalization (AdaBN). The proposed approach involves several parts to enhance fault diagnosis accuracy in vibration signals under constant load conditions and variable load conditions. Firstly, the end-layer structure of the traditional GRU is replaced with a one-dimensional global average pooling layer to aggregate the influence components of defects and reduce model training parameters. Secondly, the fusion of different types of frequency and sequence features is achieved by combining 1DCNN, addressing the limitation of a single network’s feature extraction capability and the loss of temporal features in a cascaded hybrid model. Subsequently, the fused features are input into a softmax multi-classifier to obtain fault type identification results. Lastly, the GGRU-1DCNN method is further improved by incorporating the AdaBN algorithm, enhancing the model’s domain adaptive capability under variable load conditions and noisy environments. The method is validated using datasets obtained from Case Western Reserve University, aero-engine bearings, Xi’an Jiaotong University, and the Changxing Sumyoung Technology. The findings suggest that the proposed method demonstrates superior accuracy and robustness in fault diagnosis, as well as excellent generalization capability and universal applicability.

Two Proposed Models for Face Recognition: Achieving High Accuracy and Speed with Artificial Intelligence

In light of the development in computer science and modern technologies, the impersonation crime rate has increased. Consequently, face recognition technology and biometric systems have been employed for security purposes in a variety of applications including human-computer interaction, surveillance systems, etc. Building an advanced sophisticated model to tackle impersonation-related crimes is essential. This study proposes classification Machine Learning (ML) and Deep Learning (DL) models, utilizing Viola-Jones, Linear Discriminant Analysis (LDA), Mutual Information (MI), and Analysis of Variance (ANOVA) techniques. The two proposed facial classification systems are J48 with LDA feature extraction method as input, and a one-dimensional Convolutional Neural Network Hybrid Model (1D-CNNHM). The MUCT database was considered for training and evaluation. The performance, in terms of classification, of the J48 model reached 96.01% accuracy whereas the DL model that merged LDA with MI and ANOVA reached 100% accuracy. Comparing the proposed models with other works reflects that they are performing very well, with high accuracy and low processing time.

Most influential feature form for supervised learning in voltage sag source localization

The paper investigates the application of machine learning (ML) for voltage sag source localization (VSSL) in electrical power systems. To overcome feature-selection challenges for traditional ML methods and provide more meaningful sequential features for deep learning methods, the paper proposes three time-sample-based feature forms, and evaluates an existing feature form. The effectiveness of these feature forms is assessed using k-means clustering with k = 2 referred to as downstream and upstream classes, according to the direction of voltage sag origins. Through extensive voltage sag simulations, including noises in a regional electrical power network, k-means identifies a sequence involving the multiplication of positive-sequence current magnitude with the sine of its angle as the most prominent feature form. The study develops further traditional ML methods such as decision trees (DT), support vector machine (SVM), random forest (RF), k-nearest neighbor (KNN), an ensemble learning (EL), and a designed one-dimensional convolutional neural network (1D-CNN). The results found that the combination of 1D-CNN or SVM with the most prominent feature achieved the highest accuracies of 99.37% and 99.13%, respectively, with acceptable/fast prediction times, enhancing VSSL. The exceptional performance of the CNN was also approved by field measurements in a real power network. However, selecting the best ML methods for deployment requires a trade-off between accuracy and real-time implementation requirements. The research findings benefit network operators, large factory owners, and renewable energy park producers. They enable preventive maintenance, reduce equipment downtime/damage in industry and electrical power systems, mitigate financial losses, and facilitate the assignment of power-quality penalties to responsible parties.

A Lithium-Ion Battery Degradation Prediction Model With Uncertainty Quantification for Its Predictive Maintenance

Battery degradation modeling in the presence of uncertainty is a key but challenging issue in the applications of battery predictive maintenance. This paper develops a capacity prediction model with uncertainty quantification for lithium-ion batteries and proposes a dynamic maintenance strategy which can help to make an optimized decision at each battery cycle stage. To be specific, after using the one-dimensional convolution neural network (1dCNN), deep representative features hidden in original measured signals are extracted. Then, the bi-directional long short-term memory (Bi-LSTM) is applied to estimate the battery capacities, while the quantile regression (QR) layer is embedded into the construction of the Bi-LSTM network to obtain the capacities for different quantiles. Next, the kernel density estimation (KDE) is utilized to derive the probability density of the predicted points at each battery cycle stage. Thus, the combination of 1dCNN, Bi-LSTM, QR and KDE, named 1dCNN-BiLSTMQR-KDE, forms an efficacious capacity prediction model with reliable uncertainty management. Finally, the costs of different decisions at each battery cycle stage are evaluated, and the decision with the lower cost will be chosen. The whole proposition is verified on battery degradation datasets from NASA, and the comparison with other methods show that the proposed method is competitive.

A 1D CNN-Based Emulator of CMAQ: Predicting NO2 Concentration over the Most Populated Urban Regions in Texas

Abstract In this study, we developed an emulator of the Community Multiscale Air Quality (CMAQ) model by employing a one-dimensional (1D) convolutional neural network (CNN) algorithm to predict hourly surface nitrogen dioxide (NO2) concentrations over the most densely populated urban regions in Texas. The inputs for the emulator were the same as those for the CMAQ model, which includes emission, meteorological, and land-use/land-cover data. We trained the model over June, July, and August (JJA) of 2011 and 2014 and then tested it on JJA of 2017, achieving an index of agreement (IOA) of 0.95 and a correlation of 0.90. We also employed temporal threefold cross validation to evaluate the model’s performance, ensuring the robustness and generalizability of the results. To gain deeper insights and understand the factors influencing the model’s surface NO2 predictions, we conducted a Shapley additive explanations analysis. The results revealed solar radiation reaching the surface, planetary boundary layer height, and NOx (NO + NO2) emissions are key variables driving the model’s predictions. These findings highlight the emulator’s ability to capture the individual impact of each variable on the model’s NO2 predictions. Furthermore, our emulator outperformed the CMAQ model in terms of computational efficiency, being more than 900 times as fast in predicting NO2 concentrations, enabling the rapid assessment of various pollution management scenarios. This work offers a valuable resource for air pollution mitigation efforts, not just in Texas, but with appropriate regional data training, its utility could be extended to other regions and pollutants as well. Significance Statement This work develops an emulator of the Community Multiscale Air Quality model, using a one-dimensional convolutional neural network to predict hourly surface NO2 concentrations across densely populated regions in Texas. Our emulator is capable of providing rapid and highly accurate NO2 estimates, enabling it to model diverse scenarios and facilitating informed decision-making to improve public health outcomes. Notably, this model outperforms traditional methods in computational efficiency, making it a robust, time-efficient tool for air pollution mitigation efforts. The findings suggest that key variables like solar radiation, planetary boundary layer height, and NOx (NO + NO2) emissions significantly influence the model’s NO2 predictions. By adding appropriate training data, this work can be extended to other regions and other pollutants such as O3, PM2.5, and PM10, offering a powerful tool for pollution mitigation and public health improvement efforts worldwide.

Application of one-dimensional hierarchical network assisted screening for cervical cancer based on Raman spectroscopy combined with attention mechanism

Cervical cancer is one of the most common malignant tumors among women, and its pathological change is a relatively slow process. If it can be detected in time and treated properly, it can effectively reduce the incidence rate and mortality rate of cervical cancer, so the early screening of cervical cancer is particularly critical and significant. In this paper, we used Raman spectroscopy technology to collect the tissue sample data of patients with cervicitis, Low-grade Squamous Intraepithelial Lesion, High-grade Squamous Intraepithelial Lesion, Well differentiated squamous cell carcinoma, Moderately differentiated squamous cell carcinoma, Poorly differentiated squamous cell carcinoma and cervical adenocarcinoma. A one-dimensional hierarchical convolutional neural network based on attention mechanism was constructed to classify and identify seven types of tissue samples. The attention mechanism Efficient Channel Attention Networks module and Squeeze-and-Excitation Networks module were combined with the established one-dimensional convolutional hierarchical network model, and the results showed that the combined model had better diagnostic performance. The average accuracy, F1, and AUC of the Principal Component Analysis-Squeeze and Excitation-hierarchical network model after 5-fold cross validations could reach 96.49%±2.12%, 0.97±0.03, and 0.98±0.02, respectively, which were 1.58%, 0.0140, and 0.008 higher than those of hierarchical network. The recall rate of the Principal Component Analysis-Efficient Channel Attention-hierarchical network model was as high as 96.78%±2.85%, which is 1.47% higher than hierarchical network. Compared with the classification results of traditional CNN and ResNet for seven types of cervical cancer staging, the accuracy of the Principal Component Analysis-Squeeze and Excitation-hierarchical network model is 3.33% and 11.05% higher, respectively. The experimental results indicate that the model established in this study is easy to operate and has high accuracy. It has good reference value for rapid screening of cervical cancer, laying a foundation for further research on Raman spectroscopy as a clinical diagnostic method for cervical cancer.

Simplified detection of genetic background admixture using artificial intelligence.

Admixture refers to the mixing of genetic ancestry from different populations. Admixture is important for genomic medicine because it can affect how an individual responds to certain medications, how they metabolize drugs, and susceptibility to certain diseases. For example, some genetic variants associated with drug metabolism and response may be more common in certain populations, and individuals with admixed ancestry may have a different frequency of these variants than individuals from the ancestral populations. Understanding the patterns of admixture in a population can also help researchers identify new genetic variants associated with diseases or traits and develop more personalized and targeted treatments. In this study, we compared and classified the known and self-reported genetic backgrounds from 1000 Genomes Project and admixed samples from GTEx projects using supervised, unsupervised and statistical classification methodologies. We developed a novel tool called Admix-AI that uses a one-dimensional convolutional neural network to understand and classify admixed genetic backgrounds using 213 DNA-marker based genetic background labels. Admix-AI can be used to discover admixed proportions in samples and ultimately aid personalized genomic medicine by identifying specific biomarker systems. We compared Admix-AI to the existing admixture categorization software and found our tool to be computationally faster with 2× speedup and streamlined usage. Admix-AI is available as open-source code under GPL version 3.0 license at https://github.com/rpauly/Admix-AI.

Research on fault diagnosis of rolling bearing based on improved convolutional neural network with sparrow search algorithm.

Traditional approaches to the intelligent fault diagnosis of rolling bearings have predominantly relied on manual expertise for feature extraction, a practice that compromises robustness. In addition, the existing convolutional neural network (CNN) is characterized by an overabundance of parameters and a substantial requirement for training samples. To address these limitations, this study introduces a novel fault diagnosis algorithm for rolling bearings, integrating a one-dimensional convolutional neural network (1DCNN) with a support vector machine (SVM) to form an enhanced 1DCNN-SVM model. This model is further refined using the sparrow search algorithm (SSA) for the optimal adjustment of the parameters of 1DCNN-SVM. Specifically, by substituting the CNN's final softmax layer with an SVM, the model becomes better suited for processing limited data volumes. In addition, the incorporation of batch normalization and dropout layers within the CNN framework significantly augments its fault classification accuracy for rolling bearings, concurrently mitigating the risk of overfitting. The SSA is subsequently applied to refine three principal hyper-parameters: batch size, initial learning rate, and the L2 regularization coefficient, thereby overcoming the challenges associated with manually adjusting parameters, such as extended processing times and unpredictable outcomes. Empirical tests on Case Western Reserve University (CWRU) datasets revealed the model's superior performance, with the SSA-optimized 1DCNN-SVM showcasing diagnostic accuracies over 98%, marked improvements over conventional models, and a significant reduction in processing times. This method not only marks a significant advancement in intelligent fault diagnosis for rolling bearings but also demonstrates the potential of integrating machine learning for more precise and efficient diagnostics. The SSA-1DCNN-SVM model, optimized for accuracy and minimal data use, sets a new standard in fault diagnosis, relevant for machinery health monitoring and maintenance strategies across various industries.

The deep convolutional networks for the classification of multi-class arrhythmia

An arrhythmia is an irregular heartbeat. Many researchers in the AI field have carried out the automatic classification of arrhythmias, and the issue that has been widely discussed is imbalanced data. A popular technique for overcoming this problem is the synthetic minority oversampling technique (SMOTE) technique. In this paper, the author adds some sampling of data obtained from other datasets into the primary dataset. In this case, the main dataset is the Massachusetts Institute of Technology–Beth Israel Hospital (MIT-BIH) arrhythmia database and an additional dataset from the MIT-BIH supraventricular arrhythmia database. The classification process is carried out with one-dimensional convolutional neural network model (1D-CNN) to perform multiclass and subject-class advancement of medical instrumentation (AAMII) classifications. The results obtained from this study are an accuracy of 99.10% for multiclass and 99.25% for subject-class.

Bolt Looseness Detection in Brake Disc of High-Speed Rail Using Wavelet Packet Decomposition and 1D Convolutional Neural Network

The effective and in-time detection of bolt looseness in the brake disc of high-speed rail is of great significance to ensure the safe operation of the train. The hammer tapping method is one of the most widely used methods for bolt looseness detection. Due to the complex structure of the brake disc, the vibration characteristics associated with bolt looseness are coupled with the transmission path from the head of the bolt through the brake disc to the sensors, which make it difficult to quantitively identify the bolt looseness. In this work, a method based on wavelet packet decomposition and a one-dimensional convolutional neural network (1D CNN) is proposed to quantitively detect the bolt looseness of brake disc. Firstly, the vibration signals collected from the two sensors mounted on the nut side are fused by autocorrelation summation to reduce the influence of transmission path. Secondly, the fused signal is decomposed to sub signals in low frequency and high frequency components by wavelet packet decomposition. The relative difference of wavelet packet energy among sub signals is extracted as the features to enhance the difference among different degrees of looseness. Finally, the 1D CNN model is established and trained by the features of energy relative difference to quantitively identify the bolt looseness. To validate the effectiveness of the proposed method, an experimental platform for bolt looseness detection in brake disc is constructed. Compared with the single-channel 1D CNN and fused signal-1D CNN models, the accuracy of the proposed method is approximately 97%, which confirms the superiority of the proposed method.

Ternary Precursor Centrifuge Rolling Bearing Fault Diagnosis Based on Adaptive Sample Length Adjustment of 1DCNN-SeNet

To address the issues of uneven sample lengths in the centrifuge machine bearings of the ternary precursor, inaccurate fault feature extraction, and insensitivity of important feature channels in rolling bearings, a rolling bearing fault diagnosis method based on adaptive sample length adjustment of one-dimensional convolutional neural network (1DCNN) and squeeze-and-excitation network (SeNet) is proposed. Firstly, by controlling the cumulative variance contribution rate in the principal component analysis algorithm, adaptive adjustment of sample length is achieved, reducing data with uneven sample lengths to the same dimensionality for various classes. Then, the 1DCNN extracts local features from bearing signals through one-dimensional convolution-pooling operations, while the SeNet network introduces a channel attention mechanism which can adaptively adjust the importance between different channels. Finally, the 1DCNN-SeNet model is compared with four classic models through experimental analysis on the CWRU bearing dataset. The experimental results indicate that the proposed method exhibits high diagnostic accuracy in rolling bearings, demonstrating good adaptability and generalization capabilities.

Prediction of impurity spectrum function by deep learning algorithm

By using the numerical renormalization group (NRG) method, we construct a large dataset with about one million spectral functions of the Anderson quantum impurity model. The dataset contains the density of states (DOS) of the host material, the strength of Coulomb interaction between on-site electrons (U), and the hybridization between the host material and the impurity site (Γ). The continued DOS and spectral functions are stored with Chebyshev coefficients and wavelet functions, respectively. From this dataset, we build seven different machine learning networks to predict the spectral function from the input data, DOS, U, and Γ. Three different evaluation indexes, mean absolute error (MAE), relative error (RE) and root mean square error (RMSE), are used to analyze the prediction abilities of different network models. Detailed analysis shows that, for the two kinds of widely used recurrent neural networks (RNNs), gate recurrent unit (GRU) has better performance than the long short term memory (LSTM) network. A combination of bidirectional GRU (BiGRU) and GRU has the best performance among GRU, BiGRU, LSTM, and BiLSTM. The MAE peak of BiGRU+GRU reaches 0.00037. We have also tested a one-dimensional convolutional neural network (1DCNN) with 20 hidden layers and a residual neural network (ResNet), we find that the 1DCNN has almost the same performance of the BiGRU+GRU network for the original dataset, while the robustness testing seems to be a little weak than BiGRU+GRU when we test all these models on two other independent datasets. The ResNet has the worst performance among all the seven network models. The datasets presented in this paper, including the large data set of the spectral function of Anderson quantum impurity model, are openly available at https://doi.org/10.57760/sciencedb.j00113.00192.

AC-CfC: An attention-based convolutional closed-form continuous-time neural network for raw multi-channel EEG-based emotion recognition

Emotion recognition based on electroencephalogram (EEG) is a critical task in the field of affective brain-computer interfaces. However, due to the non-stationarity and individual variability of EEG, hand-designed features cannot adequately capture the nonlinear and high-dimensional properties of raw EEG. Currently, spatial–temporal models have been verified to effectively capture the spatial–temporal information for EEG. However, these models are characterized by complex structures, large parameter counts, and the need for extensive training data. To overcome these disadvantages, this paper propose an end-to-end spatial–temporal model for emotion recognition based on raw EEG, called attention-based convolutional closed-form continuous-time neural network (AC-CfC). This model employs a channel attention mechanism to weight and encode EEG, capturing channel dependencies. Subsequently, one-dimensional convolutional neural networks and closed-form continuous-time neural networks are used to extract deep spatial–temporal features. Additionally, an adaptive loss-controlling mechanism is designed to enhance the model's decision-making ability between classes that are easily confused. To verify the effectiveness of the proposed model, experiments are conducted on the DEAP and DREAMER datasets. The average accuracies of the proposed model reach to 94.76% and 93.01% for valence and arousal in subject-independent experiments on the DEAP dataset, with an improvement of 11.8% and 8.73% respectively compared with the state-of-art model ACRNN. On the DREAMER dataset, the average accuracies reach to 81.83%, 81.22%, and 80.63% for valence, arousal and dominance, with an improvement of 2.55%, 6.64%, and 6.98% respectively over ACRNN. These results show that our proposed model exhibits better performance than some state-of-art models in the same category.

Neural network-based prediction of auto-ignition temperature of ternary mixed liquids

Auto-ignition temperature (AIT) is one of the crucial exponents in the design of fire and explosion safety measures. Therefore, in this study, quantitative structure-property relationship approach was used to predict the AIT of ternary hybrid liquids based on molecular structure information. The optimal molecular descriptors were calculated and filtered using Mordred software. Twelve mixing rules were proposed for calculating molecular descriptors of mixtures. A prediction model for the AIT value of binary liquid mixtures was developed, validated and evaluated using a back propagation neural network (BPNN) and a one-dimensional convolutional neural network (1DCNN). The relative contribution and positive and negative correlations between individual molecular descriptors and AIT in the model were interpreted using the shapley additive explanations method. The results show that BPNN and 1DCNN models using mixing rule 1 have the best fitting ability, stability and prediction ability. The determination coefficient of the BPNN and 1DCNN models in the training set were 0.996 and 0.992, the root mean square errors were 3.613 °C and 5.284 °C, the mean absolute errors were 2.483 °C and 4.144 °C, the nash efficiency coefficient was 0.996 and 0.992, respectively, the willmott index was 0.999 and 0.998. and the values of the top three molecular descriptors of relative contribution, SssCH2, SsOH and SsCH3, were negatively correlated with the AIT values. The BPNN and 1DCNN models provide an accurate and reliable method for predicting ternary mixing liquid AIT.