Industrial time series forecasting enables forecasting of future factory production scenarios, thereby enhancing production efficiency and reducing costs. Existing time series forecasting methods often suffer from boundary information loss, incomplete feature learning due to unimodal extraction, and insufficient analysis of interchannel correlations. Therefore, a novel time-frequency feature fusion method with triple-attention (FreMixer) is proposed for multivariate time series forecasting of industrial processes, which consists of the frequency-domain enhanced long-sequence extensible patching (FLEP) and the residual multidimensional attention network (RMDANet). The FLEP is designed to capture more comprehensive sequence information and mitigate boundary value loss. Then, the RMDANet is designed to model intrapatch dependencies, interpatch relationships, and correlations between global variables. Meanwhile, the contrastive chunk consistency loss is applied to enforce semantic coherence between patches. Finally, the FreMixer is validated on six benchmarks and a proprietary Polyethylene Terephthalate industrial dataset. Compared to other baselines, experimental results demonstrate at least 7.1% and 4.6% reductions in the average mean square error and the average mean absolute error of the proposed FreMixer, indicating state-of-the-art performance with strong industrial application potential.

Novel View Synthesis (NVS) seeks to generate realistic novel views from limited source images, offering an effective solution for 3D reconstruction in complex or unknown environments. Achieving high generalization under occlusion, varying illumination, and sparse observations remains challenging, largely hinging on the effective extraction, optimization, and fusion of image features and spatial geometry. In this work, we propose JOIG — a Joint Optimization Model of Image Features and Constraint Geometry Fusion for generalizable 3D Gaussian splatting. JOIG introduces three key components: Multiscale Dimension Rotation Fusion (MDRF) to capture intrinsic dependencies across feature dimensions for enhanced image encoding, Geometry Self-Correcting Aggregation (GSCA) to refine multi-view geometry with depth-guided reweighting, and Geometry-Image Feature Aggregation (GIFA) to achieve pixel-aligned fusion of spatial and image information. Extensive experiments on DTU, LLFF, NeRF Synthetic, and Tanks and Temples datasets demonstrate that JOIG achieves state-of-the-art generalization performance, significantly improving both quantitative metrics and visual fidelity in novel view synthesis.

Wind energy, characterized by its clean and replenishable nature, is increasingly used worldwide due to its environmental friendliness and wide distribution of resources. However, ice accretion on turbine blades in cold regions, often resulting from cold weather conditions, significantly impacts both the operational performance and security of wind energy production, which significantly increases maintenance costs, resulting in a significant reduction in the energy output performance of wind turbines. Specifically, blade icing alters the aerodynamic characteristics of the blade surface, increases wind resistance, and reduces wind energy conversion efficiency. Furthermore, ice accretion may result in a non-uniform mass allocation across the blade surfaces. This imbalance can induce vibrations within the turbine system, thereby compromising its operational stability and structural integrity. The key challenges are complex sensor parameter variations, high labeling costs, and data imbalance, making accurate icing prediction difficult. In order to tackle these difficulties, this study introduces a technique based on dual-path contrastive learning. This method balances the dataset using a sliding window technique and utilizes both icing loss features and expert features for dual-path processing to fully exploit the feature sets. Furthermore, ice accretion may result in a non-uniform mass allocation across the blade surfaces. This imbalance can induce vibrations within the turbine system, thereby compromising its operational stability and structural integrity, particularly exhibiting excellent performance in handling data imbalance.

Blind image quality assessment (BIQA) is crucial for user satisfaction and the performance of various image processing applications. Most BIQA methods directly use the pre-trained model to extract features and then perform feature fusion. However, the features extracted by pre-trained models may contain irrelevant information to BIQA. Although some methodspre-train the feature extraction network from scratch, these approaches raise computational costs and resource demands. In this letter, a Feature-selected Pyramid Network(FsPN) is proposed to address this issue from a different perspective. First, a spatial selection module selects useful information from the features extracted by the pre-trained model. Additionally, a pyramid network based on skip connections is utilized to fuse the selected multi-scale features. The proposed method is verified in six public datasets, where it consistently outperformed existing state-of-the-art methods, affirming its effectiveness and adaptability.

Various uncertain disturbances in industrial processes bring noise to industrial process data, which brings great challenges to industrial soft sensor modeling. Traditional soft sensor models focused on removing noise in the process data, but it is almost impossible to remove all noise in actual engineering. Therefore, a novel noise adaptive filtering method integrating the multiscale neural network (NAF-MSNN) is proposed for the soft sensor, which can incorporate a noise processing mechanism that adaptively removes noise at different scales during feature extraction. The MSNN extracts overall trend and local trend features through the multiscale convolution. Then, the NAF converts multiscale features into frequency domain features, and constrains the noise filter matrix through proposed piecewise regularization to select important frequency domain components at different scales. Moreover, the multiscale fusion module controls denoised multiscale features exchange fusion between different scale based on the important measurement of each corresponding scale. Finally, the gated recurrent unit (GRU) establishes the dynamic relationships between the fused multiscale features and the key indicator. The proposed NAF-MSNN is compared with state-of-the-art soft sensor models in three datasets. In terms of R² metrics, the accuracy improvement of NAF-MSNN reaches 4%, 8% and 3% in the public dataset and two industrial datasets.

In order to ensure the safe operation of industrial systems, the timely diagnosis of incipient faults is gradually gaining attention. The kernel entropy component analysis (KECA) has been widely used in the fault diagnosis of nonlinear industrial processes. However, the KECA often performs unsatisfactorily in the case of incipient faults. Therefore, a novel incipient fault detection and diagnosis method based on the statistical feature KECA integrating the twofold weighted (TWSFKECA) is proposed. The residual function in the local approach is combined with the KECA to construct statistical features of the data. Then, in order to highlight the influence of incipient faults of statistical features, the statistical feature sample weighting strategy is established based on the dissimilarity analysis between the test and training samples. Furthermore, the statistical feature component weighting strategy is developed for the sensitive components, which are judged by applying the Durbin-Watson (DW) criterion to calculate the extent to which the sample-weighted statistical feature components contain significant information. Moreover, based on the statistical features of twofold weights, two statistics indexes are created for incipient fault detection. In addition, the strategy for process fault diagnosis using a variable contribution plot method is proposed to isolate faulty variables. Finally, the continuous stirred tank reactor control system and the Tennessee Eastman process illustrate the superiority of the proposed method for incipient fault detection and diagnosis.Note to Practitioners—Effective detection of incipient faults prevents the evolution of accidents and ensures the smooth operation of the production process. In nonlinear industrial processes, the statistical feature KECA integrating the twofold weighted is proposed for incipient fault detection and diagnosis. The residual function is introduced in the kernel entropy component analysis to construct the statistical features of the data, which helps extract the incipient fault information. Then, the twofold weighting strategy weights the statistical features in terms of samples and components, highlighting the influence of the main samples and sensitive components in the incipient faults, respectively. In addition, a variable contribution plot method is developed to solve the problem of not being able to find out the cause of faults through control plots. The experimental results further verify the applicability of the proposed method for monitoring the occurrence of incipient faults.

Long-term series forecasting (LTSF) plays a crucial role in energy efficiency analysis and optimization in industrial production processes. However, due to the complexity and nonstationarity of industrial production data, the fluctuations and transformations between data are deeply mixed, and the variables interact with each other, making prediction extremely challenging. Therefore, this article proposes a novel multiscale hybrid decomposition feature extraction network (MHDN) for industrial long-term series prediction. The MHDN consists of the multiscale feature extraction (MFE) and the time-mixing predictor (TMP) to fully extract multiscale temporal features. The MHDN breaks the traditional operation of using decomposition as a preprocessing block and uses time series decomposition as a basic internal block of deep models. Specifically, the MFE extracts features from seasonal and trend components separately, achieving full extraction of time patterns and decoupling of data fluctuations from complex process data. The TMP integrates multiple predictors in the time domain and the feature domain and uses residual connections to avoid information loss, effectively utilizing multiscale information for complementary prediction. Finally, the MHDN is validated on six benchmarks and an actual industrial production dataset. The experimental results show that compared with the current baseline, the MHDN method achieves state-of-the-art results, reducing the average mean square error and average absolute error by at least 7.8% and 5.6% which can effectively guide industrial production.

Time-series anomaly detection plays an important role in ensuring industrial safety. Currently, many anomaly detection methods mainly target complete time series and ignore the widespread problem of data missing in the real world. Therefore, this article proposes a novel anomaly detection method for time series with sparse observations based on imputation consistency using a mixture of patch information inference network (MoPIN). Due to the robustness of the imputation method modeling to the random mask, different imputed series of the same normal time series with different random masks should have consistency. Then, a novel imputation consistency is used to detect anomalies in sparse observation series. Moreover, the MoPIN imputes series by a two-step imputation and multiscale modeling of patch information. Meanwhile, the similarity of imputed series under different masks is used to measure imputation consistency, which well constructs the relationship between sparse observation series and anomaly scores. Finally, the MoPIN can accurately detect anomalies while imputing series. Extensive experiments on four real-world benchmarks in different domains of imputation and anomaly detection tasks and a real fluid catalytic cracking (FCC) process case demonstrate the effectiveness of the proposed method. Specifically, the MoPIN achieved at least 8.05% mean absolute error (MAE) relative improvement in imputation and 3.74% $F1$ relative improvement in anomaly detection.

Due to the lack of fault sample of the acoustic signal and the susceptibility of the acoustic signal to interference from reverberation and background noise, constructing an acoustic diagnostic model is difficult. Therefore, in this article, a novel migration diagnosis of different devices from vibration to acoustic (MD3VA) of the rotating machinery is proposed. First, the improved recursive least squares and the resonance-based sparse signal decomposition are used to remove the reverberation and background noise of the collected sound signal. Then, the fault component distribution model reflecting fault categories is learned by combining multisource-domain fault vibration data with the deep convolutional generative adversarial network (GAN). And the amplitude variation distribution model is learned based on the GAN. Combining the normal state acoustic signal and common characteristics distribution models of the vibration signal, the acoustic fault sample is generated by using the mechanism character generative model (MCGM). Finally, the acoustic diagnosis model is constructed based on the convolutional neural network utilizing virtual acoustic fault samples. The accuracy of the MD3VA method is validated using the experiment and the industrial data. The outcomes demonstrate that the average diagnostic accuracy of the MD3VA reaches 76.3% in various tasks, which is more than 7.13% higher than the accuracy of the comparative method.

The self-attention mechanism has advantages in analyzing the internal characteristics of data and capturing local information. Therefore, it is applied to fault detection. Since complex industrial processes usually contain a mixture of properties, such as nonlinearity, dynamics, and non-Gaussianity, this places higher demands on fault detection methods. To further improve the performance of the self-attention mechanism for fault detection in complex industrial processes, an innovative self-attention mechanism integrating the adaptive double subspace (ISA-ADS) method is proposed for fault detection. First, the query matrix, the key matrix, and the value matrix of the self-attention mechanism are designed so that they can better focus on local information through the adaptive sample weight allocation strategy. At the same time, this design reduces the interference of nonlinear and noisy information and amplifies the impact of important fault information in the sample. Second, considering the problem of Gaussian and non-Gaussian distributions of the data and the different degrees of influence of different subspaces on the final process monitoring results, an adaptive Gaussian and non-Gaussian double subspace fault detection model based on the self-attention output matrix is established, which adaptively assigns different weights to different subspaces by determining their importance. The statistics of different subspaces are fused using the Bayesian inference and a new adaptive monitoring statistic is constructed to obtain a more accurate monitoring of the process state. Finally, the ISA-ADS is validated using the Tennessee Eastman (TE) process and the actual polyethylene production process. The results show that ISA-ADS has a low false alarm rate and a high fault detection rate, verifying the effectiveness of its fault detection performance.

Multivariate time series anomaly detection plays an important role for the safe operation of industrial devices and systems. At present, many effective methods have the major limitation that the changes in information propagation between variables are not considered when anomalies occur. Therefore, this article proposes a novel graph structure change-based anomaly detection on multivariate time series (GSC-MAD). First, a stable graph structure under normal conditions is obtained and a single-step prediction for all variables is achieved from a high-dimensional time-series embedding representation learned from the normal data. Then, anomaly detection is achieved by combining the variable behavior deviation reflected by prediction errors and the information propagation deviation between variables reflected by GSC. Extensive experiments on five real-world benchmarks are conducted to demonstrate the effectiveness of the proposed method and compared with current state-of-the-art (SOTA) baselines, a relative improvement of 6.64% on the average F1 is achieved. Moreover, an actual chemical industrial case is provided to verify the effect of the GSC-MAD and a relative improvement of 4.03% is achieved on the F1 metric compared with SOTA baselines. Comparison experiment results show that the proposed method achieves the SOTA results in terms of current baselines. Further experiment analysis shows the good interpretability of the proposed method for detected anomalies.

With the development of urban water supply systems, the leakage detection of water supply pipe networks is of great significance for the safe operation of urban water supply systems. In practice, due to the short of the important data, traditional detection models often fail to achieve good detection results. Therefore, this article proposes a novel pipeline attention integrating the long short-term memory (LSTM) to detect the leakage. The density-based spatial clustering of applications with noise (DBSCAN) method divides the water supply network into several regions according to the leakage characteristics of pipelines. Then, the LSTM extracts dynamic time-varying hydraulic features of pipelines, and the pipeline attention extracts the dynamically changing features between pipelines. Finally, the Attention-LSTM model is used to detect the leak region of urban water supply systems. Compared with multilayer perceptron classifier,$K$neighbors classifier, decision tree classifier, support vector machine classifier, multiscale fully convolutional network, one-dimensional multichannel convolution neural network, and variational autoencoder, the F1-Score of the Attention-LSTM is greatly improved, which are 116%, 120%, 254%, 180%, 21%, 26%, and 27%, respectively. Therefore, the proposed model can accurately detect the leakage of water supply network and effectively reduce the waste of resources.

Thermal power generation is one of the main forms of electricity generation in the world, and the share of thermal power generation in total electricity generation has long been maintained at over 80% in 2018. However, power plants are often shut down due to boiler accidents, which are mostly caused by water wall damage. At present, the detection method for water wall defects is still in the stage of manual detection, which has a high risk coefficient, long time-frame, and low efficiency. In this article, a deep learning method integrating deep convolutional generating adversarial networks (DCGAN) and a seam carving algorithm to solve the problem of small sample defect detection is proposed. The proposed method uses the seam carving algorithm to solve the overfitting of the DCGAN, for which the DCGAN generates high-quality images. Then, the intelligent small sample defect detection model is built by convolutional neural networks. Finally, the proposed method is used in the defect detection of water walls in the actual thermal power generation plant. To evaluate the performance of our proposed method, we conduct comparison experiments among different GANs and different detection networks integrating different processes used and not used the proposed data expansion method. The experimental results demonstrate that the proposed method can achieve a detection accuracy of 98.43%, which is higher than other methods, and has the best generalization ability.

In order to overcome the low robustness and weak generalization in existing deep autoencoder (AE) for soft sensor modeling, a novel feature-disentangled AE (FDAE) integrating residual network (Resnet) (FDAE-Resnet) is proposed in this article. Different from the traditional deep AE that only can learn entangled features, the FDAE can obtain disentangled multisource features including trend features, periodic features and spatial features by a new trend-periodic long short-term memory (TPLSTM) and a novel dynamic self-attention convolutional neural network (DSACNN). The trend and periodic signals decomposed from input variables are fed into the TPLSTM to learn trend and periodic features in time and frequency domain, respectively. Then, the DSACNN is utilized to capture dynamic spatial features in spatial domain by adding a new attention mechanism. Moreover, disentangled multisource features are obtained by concatenating trend features, periodic features and spatial features together. Finally, the Resnet is utilized to build the soft sensor model by establishing the relationship between disentangled multisources features and outputs. To illustrate the effectiveness and superiority of the proposed method, the FDAE-Resnet is applied in the actual polypropylene process industry for melt index modeling. The experiment results show that compared with other state-of-the-art methods, the FDAE-Resnet can reduce the root mean square error by 26.2% and the mean absolute percentage error by 38.2% on average in the changed working conditions, respectively.

The existing fault diagnosis methods can achieve good results when various status fault data are available. However, the construction of the diagnosis model is often unachievable in the actual application because only normal data are available, which is actually a few-shot fault diagnosis problem. Therefore, a novel intelligent few-shot fault diagnosis method of rotating machinery based on the convolutional neural network (CNN) using virtual samples generated by the mechanism character generative model (MCGM) integrating the generative adversarial network (GAN) is proposed. The distribution pattern of common parameters that reflect the fault category is learned using the GAN and source domain fault data. Then, the normal state data of the target domain is combined with the distribution common parameters to generate virtual samples in target domain based on the MCGM. Moreover, the fault diagnosis model is trained by virtual samples based on the CNN. Finally, the proposed fault diagnosis method is validated using the laboratory bearing data, the industrial data and the public data of the rotating machinery, respectively. The results show that the proposed method achieves an average accuracy of 93.38% in the diagnostic task, exhibiting at least 4.56% better performance than other comparison methods.

The power system measurement data has high-dimensional features and strong noise, which is difficult to be directly used for intrusion detection. Traditional machine learning methods used in the power system intrusion detection take the feature processing as a preprocessing step and perform separately from the training, which makes the features not well adapted to the training. Therefore, this article proposes a novel binary particle swarm-wrapped feature selection optimization framework (BPSWO), which can improve the intrusion detection accuracy of machine learning methods by strengthening the coupling between the feature selection and the training. First, the improved transfer function is used to make the method converge to the global optimal particle. Second, the chaotic transformation and the Hamming distance are used to solve the premature problem of the traditional particle swarm optimization. Then, the different classifiers can be embedded in the particle swarm for training. The BPSWO trains the classifier while selecting features and the final training classifier is used for intrusion detection. The proposed method is examined on the public power system from Oak Ridge National Laboratory, USA and the IEEE 57-bus system. Compared with the existing power system intrusion detection methods based on machine learning, the experimental results show that the BPSWO can achieve the state-of-the-art in the detection accuracy, which proves the effectiveness and stability of the proposed method.

The modular multilevel converter (MMC) is the main part of MMC-based high-voltage direct current (HVDC) system. The MMC bridge arm inductance fault and the submodule IGBT fault have the greatest influence on the transmission quality of transmission systems. Therefore, this article proposes a novel fault diagnosis method based on short-time wavelet entropy integrating the long short-term memory network (LSTM) and the support vector machine (SVM). The proposed short-time wavelet entropy calculation method is used to extract the fault information. First, the optimal short-term wavelet packet calculation period is determined. Moreover, the improved LSTM topology can process the wavelet entropy fault information in the time dimension. Then, the output of the LSTM is set as the input of the SVM to obtain the fault diagnosis result based on the adaptive classification. Finally, through the MMC fault diagnosis experiment of the double-ended MMC-HVDC transmission system, the effectiveness of the proposed method is verified. Compared with the traditional fault diagnosis method, the proposed method has better robustness, adaptability, and accuracy, which can greatly reduce the number of electrical signal samples and realize the fault diagnosis of multiple fault types by collecting a single signal.

Industrial process data are usually time-series data collected by sensors, which have the characteristics of high nonlinearity, dynamics, and noises. Many existing soft sensor modeling methods usually focus on dominant variables and auxiliary variables at a single time point while ignoring the timing characteristics of industrial process data. Meanwhile, the soft-sensing methods considering timing characteristics based on the deep learning are usually faced with gradient vanishing and the difficulty in parallel computing. Therefore, a novel Gated Convolutional neural network-based Transformer (GCT) is proposed for dynamic soft sensor modeling of industrial processes. The GCT encodes short-term patterns of the time series data and filters important features adaptively through an improved gated convolutional neural network (CNN). Then, the multihead attention mechanism is applied to modeling the correlation between any two moments. Finally, the prediction results are obtained through a linear neural network layer with the highway connection. In this article, the experiments in the dynamic soft sensor modeling of polypropylene and purified terephthalic acid industrial processes show that the proposed method achieves state-of-the-art comparing with the back propagation neural network, the extreme learning machine, the long short-term memory (LSTM) and the LSTM based on the CNN.

Industrial process data is usually time series data collected by sensors with high non-linearity, dynamics and high noise. Many existing soft sensor models usually focus on the temporal features of industrial process data, while ignoring the spatial interaction between auxiliary variables. Therefore, a novel spatio-temporal attention neural network (STAN) is proposed for dynamic soft sensor modeling of industrial processes. The temporal attention module and the spatial attention module extract the temporal and spatial interaction features of sequence data, respectively. Then the spatio-temporal fusion module adaptively controls the fusion of the temporal and spatial interactive features to obtain the spatio-temporal interactive features. Finally, the spatio-temporal interaction features are input into the fully connected layer and the highway layer of the STAN to output the final prediction result. The STAN is applied in the dynamic soft sensor modeling of polypropylene melt index. Compared with backpropagation neural network (BPNN), extreme learning machine (ELM), long short-term memory (LSTM) and convolutional LSTM ($CNN+LSTM$), the STAN achieves the state-of-the-art results.

Deep convolutional neural network can effectively extract hidden patterns in images and learn realistic image priors from the training set. And fully convolutional networks (FCNs) have achieved state‐of‐the‐art performance in the image segmentation. However, these methods have the disadvantages of noise, boundary roughness and no prior shape. Therefore, this study proposes a level set with the deep prior method for the image segmentation based on the priors learned by FCNs. The FCNs can learn high‐level semantic patterns from the training set. Also, the output of the FCNs represents the high‐level semantic information as a probability map and the global affine transformation can obtain the optimal affine transformation of the intrinsic prior shape. Moreover, the improved level set method integrates the information of the original image, the probability map and the corrected prior shape to achieve the image segmentation. Compared with the traditional level set method of simple scenes, the proposed method solves the disadvantage of FCNs by using the high‐level semantic information to segment images of complex scenes. Finally, Portrait data set are used to verify the effectiveness of the proposed method. The experimental results show that the proposed method can obtain more accurate segmentation results than the traditional FCNs.

In order to establish an effective water flooded layer recognition model to deal with complex chromatogram data and correctly identify the water flooded layer in the oil and gas reservoirs, this paper proposes a modeling approach based on ensemble classifier. First, the proposed approach utilizes the function fitting method to obtain the effective chromatogram characteristic information (CCIs). Moreover, in order to transform the sparse classification problem into a general classification problem, the synthetic minority over-sampling technique (SMOTE) algorithm is used to process the unbalanced training sample as a general training sample. Compared with the traditional classification approach, the robustness and effectiveness of the ensemble classifier model composed of the model-free classification (MFBC) algorithm, the k-nearest neighbor (KNN) algorithm and the support vector machine (SVM) algorithm were validated through the standard data source from the UCI (University of California at Irvine) repository. Finally, the proposed model is validated through an application in a complex oil and gas recognition system of China petroleum industry. The CCIs and the prediction results are obtained to provide more reliable water flooded layer information, guide the process of reservoir exploration and development and improve the oil development efficiency.

Building a robust and accurate energy analysis model is considered as an important issue in the field of petrochemical industries. Under the circumstance of small samples, the accuracy of the energy analysis model is unacceptable. In order to solve this problem, a novel noise injection integrated with extreme learning machine based nonlinear virtual sample generation method is proposed. Through injecting noise in the output matrix of the hidden layer of ELM, a virtual information matrix that is different the original one generated using the original small dataset can be obtained. Then the newly generated information matrix is adopted to produce good-quality virtual samples for supplement knowledge for small samples. To authenticate the effectiveness of the proposed method, a standard trigonometric function is first selected; and then the proposed method is developed as an energy analysis model for an ethylene production process. Simulation results indicate that good virtual samples can be generated using the proposed method, and the accuracy of the energy analysis model is much improved with the aid of the newly generated virtual samples. The proposed method will effectively help production departments of petrochemical industries set more suitable targets of energy consumption and make better use of available resources.

Comprehensive energy modeling and optimization play a key role in sustainable development of complex petrochemical industries. However, it is difficult to make effective energy modeling and optimization due to the characteristics of uncertainty, high nonlinearity, and with noise of modeling data from the practical production. To deal with this problem, a novel energy modeling and efficiency optimization method using a novel extreme learning fuzzy logic network (ELFLN) is proposed. In the proposed method, Mamdani type fuzzy inference system (FIS) and multi-layer feedforward artificial neural network (MLFANN) are adopted. First, the fuzzy inference replaces the hidden layers of artificial neural network (ANN). Then the proposed framework takes fuzzy membership degrees instead of precise values as the output. Meanwhile, an extreme learning algorithm based on Moore-Penrose Inverse is utilized to train the network efficiently. Three levels of energy efficiency of “low efficiency, median efficiency and high efficiency” can be effectively achieved using the proposed method. For inefficiency samples, valid slack variables are predicted for finding the direction of improving the efficiency. The energy efficiency optimization performance and the practicality of the proposed method is confirmed through an application of China ethylene industry. Finally, the energy saving potential is indicted as 8.82% and practical ethylene production can be guided by the result of the demonstration analysis.

Modeling knowledge graph completion by encoding each entity and relation into a continuous tensor space becomes very hot. Meanwhile, many models including TransE, TransH, TransR, CTransR, TransD, TranSpare, TransDR, STransE, DT, FT and OrbitE are proposed for knowledge graph completion. However, all these previous works take less attention to the asymmetrical and the imbalance of many relations (some relations link a subject and many objects, and other relations link many subjects and many objects). Therefore, this paper proposes a novel asymmetrical embedding model(AEM) for knowledge graph completion. Because of the different properties of the head and tail entities in the triplets of the same relationship, every head entity vector and every tail entity vector are weighted by the corresponding head relation vector and the corresponding tail relation vector, respectively. And then new entity vector representations are obtained and the new entity vectors in the same triple are similar. Because the AEM weights each dimension of the entity vectors, it can accurately represent the latent attributes of entities and relationships. Moreover, the number of parameters of the AEM is so small that it is easier to train. Finally, compared with previous embedding models, the AEM obtains a better link prediction performance through two benchmark datasets FB15K and WN18.

Fully convolutional neural network (FCN) has achieved state-of-the-art performance in the task of time series classification without any heavy preprocessing. However, the FCN cannot effectively capture features of different frequencies. Therefore, this paper proposed a novel FCN structure based on the multi-frequency decomposition (MFD) method. In order to extract more features of different frequencies, the MFD based on real fast Fourier transform (RFFT) is set as a layer of the FCN to decompose the original signal into n sub-signals of different frequency bands. And then the improved FCN fuse those features of different frequencies together to obtain time series classification. Finally, compared with the existing state-of-the-art methods, the proposed method is effectively verified through some datasets in UCR Time Series Classification archive.

Energy conservation and emission reduction in the ethylene industry is the main way to attain sustainable development, which can be achieved if the energy efficiency of petrochemical industries can be accurately analyzed and predicted. This paper proposes an improved radial basis function neural network based on fuzzy C-means (FCM) algorithm integrated with principal component analysis (PCA) technology (PCA-FRBF). The PCA is used to denoise and reduce dimensions of data to decrease the training time and errors of the modeling process. The FCM is used to separate every fuzzy class in input space and decide the number of neurons in hidden layer to overcome the shortcoming of setting them by experience subjectively. Meanwhile, the robustness and effectiveness of the PCA-FRBF model are validated through the standard data set from the University of California Irvine repository. Moreover, to predict the energy efficiency of ethylene plants, a multi-inputs and single-output model of energy efficiency is established based on the PCA-FRBF for monthly data of ethylene production process. We obtain a rational allocation of crude oil, fuel, steam, water, and electricity, and the greatest benefit of ethylene plants under different technologies. Finally, the empirical results show the effectiveness and practicability of the PCA-FRBF model applied to predict and guide the ethylene production in the petrochemical industry.