Search

Text classification is an important task in machine learning. Specifically, deep neural network has been shown strong capability to improve performance in different fields, for example speech recognition, objects recognition and natural language processing. However, in most previous work, the extracted feature models do not achieve the relative text tasks well. To address this issue, we introduce a novel multi-task learning approach called a hybrid representation-learning network for text classification tasks. Our method consists of two network components: a bidirectional gated recurrent unit with attention network module and a convolutional neural network module. In particular, the attention module allows for the task learning private feature representation in local dependence from training texts and that the convolutional neural network module can learn the global representation on sharing. Experiments on 16 subsets of Amazon review data show that our method outperforms several baselines and also proves the effectiveness of joint learning multi-relative tasks.

In this paper, the distribution of 210Po after high temperature processes in six units of coal-fired power plants (CFPs) were evaluated. The coal, bottom ashes, fly ashes from electrostatic precipitators (ESP), and flue gases from stacks were sampled from four CFPs and analyzed for 210Po contents. The results showed that 210Po was mainly captured by the ESP, with little left in the bottom ash, and a small fraction of 210Po was directly discharged into the environment through the stacks, accounting for 0.06%-0.6%, which was consistent with the reported data. It was also found that part of the 210Po could not be accounted for in the mass balance analysis for the whole combustion process in CFPs, which was also in line with the reported data. The results obtained in this study provided essential basic data for environmental radiological risk analysis for CFPs.

In cases of high radiation emergencies, we propose a surface contamination monitor (SCM) that can quickly measure and pinpoint the contamination distribution in the affected population. Thick gaseous electron multiplier (THGEM) has several advantages, including fast response time and good spatial resolution. Based on new THGEMs, a two-dimensional imaging detector was developed for alpha detection, with a position resolution greater than 3 mm. The detector design and test results are described in this paper. Fast radiation imaging SCMs, with a 40 mm × 40 mm sensitive area, are currently under development.

With the future development of substation, the research of power fault detection algorithm has very important theoretical significance and wide application prospects. In order to improve the recognition of power line fault detection, one modeling method based on sparse self-encoding neural network is proposed. The dB3 wavelet is used to decompose the fault signal, and then the sub-band energy is calculated as parameters for the deep learning neural network. By the pre-training analysis and modeling for the characteristic of fault signal, the deep learning neural network is used as the fault recognition classifier. The simulation experiment based on IEEE 34 shows that the fault recognition rate exceeds 99%.

In order to improve the efficiency of electric system and reduce the works of maintenance, scholars from different countries have studied the substation automation. Nowadays, IEC61850 standard is the main rule to regulate the communication between substations. The paper studies the way to use substation configuration description language to describe models of IEC61850, compares the differences of SCL (substation configuration description language) file and discusses the rules of substation configuration description language (SCL) files in actual substation project.