Surrogate models have been widely applied to correlate design variables and performance parameters in turbomachinery optimization applications. With more design variables and uncertain factors taken into account in an optimization design problem, the mathematical relations between the design variables and the performance parameters might present linear, low-order nonlinear or even high-order nonlinear characteristics, and are usually analytically unknown. Therefore, it is required that surrogate models have high adaptability and prediction accuracy for both the linear and nonlinear characteristics. The paper mainly investigates the effectiveness of an adaptive region segmentation combining surrogate model based on support vector regression and kriging model applied to a transonic axial compressor to approximate the complicated relationships between geometrical variables and objective performance outputs with different sampling methods and sizes. The purpose is to explore the prediction accuracy and computational efficiency of this adaptive surrogate model in real turbomachinery applications. Three different sampling techniques are studied: (1) uniform design; (2) Latin hypercube sampling method; (3) Sobol quasi-random design. For the low dimensional case with five variables, the adaptive region segmentation combining surrogate model performs better (not worse) than the single component surrogate in terms of prediction accuracy and computational efficiency. In the meanwhile, it is also noted that the uniform design applied to the adaptive surrogate model has more advantages over the Latin hypercube sampling method especially for the small sample size cases, both performing better than the Sobol quasi-random design. Moreover, a high dimensional case with 12 variables is also utilized to further validate the prediction advantage of the adaptive region segmentation combining surrogate model over the single component surrogate, and the computational results favor it. Overall, the adaptive region segmentation combining surrogate model has produced acceptable to high prediction accuracy in presenting complex relationships between the geometrical variables and the objective performance outputs and performed robustly for a transonic axial compressor problem.
Generative adversarial networks (GANs) are paid more attention to dealing with the end-to-end speech enhancement in recent years. Various GAN-based enhancement methods are presented to improve the quality of reconstructed speech. However, the performance of these GAN-based methods is worse than those of masking-based methods. To tackle this problem, we propose speech enhancement method with a residual dense generative adversarial network (RDGAN) contributing to map the log-power spectrum (LPS) of degraded speech to the clean one. In detail, a residual dense block (RDB) architecture is designed to better estimate the LPS of clean speech, which can extract rich local features of LPS through densely connected convolution layers. Meanwhile, sequential RDB connections are incorporated on various scales of LPS. It significantly increases the feature learning flexibility and robustness in the time-frequency domain. Simulations show that the proposed method achieves attractive speech enhancement performance in various acoustic environments. Specifically, in the untrained acoustic test with limited priors, e.g., unmatched signal-to-noise ratio (SNR) and unmatched noise category, RDGAN can still outperform the existing GAN-based methods and masking-based method in the measures of PESQ and other evaluation indexes. It indicates that our method is more generalized in untrained conditions. 相似文献
The rapidly increasing popularity of mobile devices has changed the methods with which people access various network services and increased network traffic markedly. Over the past few decades, network traffic identification has been a research hotspot in the field of network management and security monitoring. However, as more network services use encryption technology, network traffic identification faces many challenges. Although classic machine learning methods can solve many problems that cannot be solved by port- and payload-based methods, manually extract features that are frequently updated is time-consuming and labor-intensive. Deep learning has good automatic feature learning capabilities and is an ideal method for network traffic identification, particularly encrypted traffic identification; Existing recognition methods based on deep learning primarily use supervised learning methods and rely on many labeled samples. However, in real scenarios, labeled samples are often difficult to obtain. This paper adjusts the structure of the auxiliary classification generation adversarial network (ACGAN) so that it can use unlabeled samples for training, and use the wasserstein distance instead of the original cross entropy as the loss function to achieve semisupervised learning. Experimental results show that the identification accuracy of ISCX and USTC data sets using the proposed method yields markedly better performance when the number of labeled samples is small compared to that of convolutional neural network (CNN) based classifier. 相似文献
Neural Computing and Applications - In this paper, an image cipher is presented based on DNA sequence operations, image filtering and memrisitve chaotic system. Firstly, plain image is preprocessed... 相似文献
Neural Computing and Applications - Deep convolutional neural networks have achieved great success for image denoising recently. However, increasing the depth of the neural network cannot... 相似文献
Neural Computing and Applications - Most traditional genetic programming methods that handle symbolic regression are random algorithms without memory and direction. They repeatedly search for the... 相似文献
Journal of Intelligent Manufacturing - With the advance in Industry 4.0, smart industrial monitoring has been proposed to timely discover faults and defects in industrial processes. Steel is widely... 相似文献
The explosive growth of Chinese electronic market has made it possible for companies to better understand consumers?? opinion towards their products in a timely fashion through their online reviews. This study proposes a framework for extracting knowledge from online reviews through text mining and econometric analysis. Specifically, we extract product features, detect topics, and identify determinants of customer satisfaction. An experiment on the online reviews from a Chinese leading B2C (Business-to-Customer) website demonstrated the feasibility of the proposed method. We also present some findings about the characteristics of Chinese reviewers. 相似文献
Abstract— A psychophysical experiment was carried out to assess the perceptual contrast on a large‐sized liquid‐crystal display (LCD) under nine phases of viewing conditions. Based on the results, six contrast models employing different color attributes or the index of just‐noticeable difference (JND) were developed. Their performances were also tested by the visual data of the nine phases, which indicated that the model CQ employing CIECAM02 brightness gives the best performance in predicting visual data under different viewing conditions among those models. A preprocessing step was also proposed to utilize the contrast model CQ conveniently in the practical contrast evaluation of LCDs. 相似文献
We study on-line scheduling on parallel batch machines. Jobs arrive over time. A batch processing machine can handle up to
B jobs simultaneously. The jobs that are processed together form a batch and all jobs in a batch start and are completed at
the same time. The processing time of a batch is given by the processing time of the longest job in the batch. The objective
is to minimize the makespan. We deal with the unbounded model, where B is sufficiently large. We first show that no deterministic on-line algorithm can have a competitive ratio of less than
1+(?{m2+4}-m)/21+(\sqrt{m^{2}+4}-m)/2
, where m is the number of parallel batch machines. We then present an on-line algorithm which is the one best possible for any specific
values of m. 相似文献