Search

Insulation plays an important role in power system. During the production and operation of power equipment, there will inevitably be insulation defects occurring. Realizing the detection and localization of these defects is of great significance to ensure the safe and stable operation of power system. In this paper, different defect localization algorithms are proposed for transformer, cable and whole substation equipment, respectively. Experimental results show that these algorithms can reach a high localization accuracy.

Ultrasonic method is a non-invasive partial discharge (PD) detection method for power equipment, which plays a more and more important role in partial discharge location. The propagation characteristics of ultrasonic in converter transformer can be effectively analyzed by using numerical method. In this paper, according to the actual size of a 110kV converter transformer, the 3-D finite element simulation model of converter transformer is established to simulate the propagation of ultrasonic signals in the tank. A 2-D model is then established to simulate the influence of windings on the propagation of ultrasonic signals. The simulation results show that when the partial discharge is generated, it will spread around in the form of a spherical wave, which spreads over the entire converter transformer space within 2000μs. The sound pressure amplitude is exponentially decreased with the increase of distance between PD sources and prbes, and the drastic attenuation is caused by the windings and core. When the discharge source is located between the winding and the core or between the windings, the wave arrival time from the partial discharge source to the ultrasonic sensor will be increased. The analysis can provide the basis for the detection and localization of partial discharge in converter transformer.

Deterioration of insulation materials caused by partial discharge is one of the important reasons for the failure of power equipment. The noise caused by the interference of the external electromagnetic environment seriously affects the accuracy of partial discharge detection. Aiming at the problem of partial discharge noise suppression, this paper proposed a partial discharge signal noise suppression method based on adaptive window singular value decomposition. This method first randomly generated a series of windows with different positions and lengths on the noisy partial discharge signal; Then singular value decomposition method was employed to denoise the signals in each window; Finally, the denoising index was chosen as the fitness function, through the particle swarm algorithm to obtain the optimal solution of the window position and length. The noise signals obtained by simulation and experiment were verified, and the method was compared with adaptive singular value decomposition, wavelet denoising, and other methods. The results showed that this method could achieve a better denoising effect.

Epoxy resin and Al 2 O 3 /epoxy resin composite materials are widely used in electrical equipment insulation. In order to evaluate the effectiveness of frequency domain spectroscopy (FDS) for epoxy resin and alumina/epoxy resin composite insulation damp defects, in this paper, two kinds of insulation samples are prepared to study the influence of voltage and moisture on FDS curves of two insulation materials by using dielectric response test platform. The results show that the applied voltage has little effect on the high frequency range (1Hz-5kHz for pure epoxy resin , 20mHz-5kHz for Al 2 O 3 /epoxy resin) of FDS curves, while the tanδ of low frequency band decreases with the increase of voltage. With the decrease of water content, the low frequency band of FDS curve decreases gradually, while the Al 2 O 3 /epoxy resin curve decreases over the entire frequency band (1mHz-5kHz). Since moisture has the greatest influence on tanδ of the two insulations at 0.01 Hz, this frequency point can be used as the characteristic frequency for diagnosing damping defects of the two insulations.

High-voltage bushing is an important part of the operation of transformer equipment. In order to accurately and effectively evaluate the moisture condition of the bushing, this paper uses the polarization and depolarization current (PDC) non-destructive measurement method to do it. In the current research, the characteristics of loss peaks and curve fluctuations of tanδ-f curve can judge whether the bushing is evenly damp, but it lacks the judgment of the high moisture position of the internal damp. This paper uses oil-paper samples of different moisture content to make a preliminary qualitative exploration of PDC measurement methods. Then the polarity reversal method is used to judge the bushings with different moisture conditions, Combined with the migration characteristics of hydrolyzed ions in the oil-paper insulation, it is analyzed to verify the effectiveness of the method.

Power cables in the urban power supply system are laid in cable trenches or directly buried underground, and the insulation is susceptible to corrosion and penetration, causing local insulation aging or abrasion. Early detection of soft faults helps prevent power system faults from occurring. This paper proposes a method that can detect cable multi-point soft faults and distinguish fault types based on Frequency Domain Reflectometry (FDR). Firstly, we establish a numerical calculation model of distributed parameters to analyze the impedance spectroscopy characteristics of different types of defects and the positioning methods and errors. Next, a simulation study was carried out for different characteristics of aging and damage, which means different types of soft faults were successfully distinguished. Then, based on the results of the simulation, experiments are carried out to verify the effectiveness of the positioning algorithm and the feasibility of distinguishing defect types. Simulation and experiment show that the positioning error is within 0.5% and 1% simultaneously. Compared with traditional location methods, the defect location method proposed in this paper can determine the type and severity of the defect while locating multiple defects. The experimental results show that the defect identification method proposed in this paper has great potential for field application.

Infrared small target detection under complex background is of great significance in the field of remote sensing, such as optical remote sensing, infrared precise guidance, infrared surveillance, and night navigation. Because of low-contrast and complex-background clutters of infrared images, infrared small target detection under complex background is difficult and has a serious false alarm. In this paper, a difference-based local contrast method is proposed to improve the detection performance. First, a median filtering process is used to reduce pixel-sized noises, and infrared image after filtering is divided into a series of sub-images. Then, a difference-based local contrast measure based on contrast mechanism is calculated for generating the saliency map, which can exceedingly improve the detection rate and reduce the false alarm rate. Eventually, an adaptive threshold is used to extract the target sub-image. Experiments on five real infrared small target image datasets show that the proposed method is robust and effective with great respect to detection accuracy. Our method can achieve a higher detection rate and lower false alarm rate under complex background with better performance in target enhancement and background suppression compared with the traditional algorithms.

Traditional methods cannot detect hot spot temperature, mechanical vibration and partial discharge signal and other state parameters of the transformer in real time. Optical fiber sensing technology has a wide range of application prospects due to its high accuracy, strong anti-interference ability and distributed measurement. However, the premise for its application is that the embedded optical fiber does not affect the safe operation of the transformer. This paper explores the interact between embedded fibers and transformers. In this paper, four different types of optical fibers were initially selected to prepare fiber-oil composite insulation. Samples under different aging time were obtained through accelerated thermal aging experiments. The mechanical strength and optical loss characteristics of fibers with different aging degrees were tested separately. At the same time, moisture content, acid value, oil chromatogram, dielectric loss factor and volume resistivity of insulating oil were tested. The results show that the mechanical strength of the PVC tight fiber decreases significantly as the aging time increases. PVC tight fiber significantly improves the dielectric loss factor, CO 2 /CO and total hydrocarbon content in insulating oil. The mechanical properties of the other three optical fibers remain basically unchanged and have little effect on the insulating oil. Thermal stress does not affect the optical transmission characteristics of all optical fibers. The PVC tight fiber is not suitable as the embedded optical fiber in transformers, while the other three optical fibers meet the basic requirements of the embedded optical fiber. The experimental results provide solutions for the safety risk assessment of the embedded optical fiber in transformers.

Compared with high-voltage transmission cables, 10kV distributed cables are currently receiving less attention. Contrary to the continuously improving national overall power supply reliability, the failure rate of 10kV distribution cables is not optimistic. Thermal aging is an important cause of defects in distributed cables, which further develop into failures. This paper conducts research on defects of different severity, uses FDR to determine the severity of defects, and analyzes the impact of defect development on positioning accuracy. Firstly, we establish a simulation model of the aging cable, and use the windowed Fourier transform to locate the multi-point aging. Then, we carry out relevant experiments based on the simulation results to further verify the accuracy of the algorithm. The simulation and test results show that for cables of the same length, the aging time has no effect on the defect location accuracy. The positioning accuracy of simulation and test algorithms are within 0.46%.

Stereolithography (SL) based 3D printing is promising to produce of next-generation dielectric materials because of its high geometric accuracy, less internal defect and high flexibility. Polymer matrix composites (PMCs) are widely used in the 3D printing of dielectric materials. However, the loading of high-refractive-index ceramic filler in PMCs significantly reduces their process compatibility, making it difficult to successfully fabricate the 3D printing materials. In this study, silane-coupling-agent (SCA) treated barium titanate microspheres (BT-S) were prepared using radio-frequency thermal plasma spheroidization (RF-TPS) and KH570 silane treatment. Experimental results showed that the prepared BT-S particles have spherical shape and impact structure. This enhances the 3D printability of UV curable PMCs by the improvement of UV curing performance and fluidity. Finally, 3D printing of UV/BT-S composites were successfully printed using stereolithography (SLA) 3D printing technique.

In this paper, a HLS-based convolutional neural network (CNN) accelerator is designed for FPGA and channel-wise low-bit quantization is applied to YOLOv3- Tiny, whose weights are quantized to 2-bit while activations are quantized to 8-bit. The quantization range is learnable in training to prevent severe accuracy loss. The accelerator uses sliding window technique to improve data reusability and efficient process element (PE) is designed to utilize low-bit calculation. This design makes full use of DSP and LUT resources and exploits optimal parallelism on embedded FPGA. The performance of our design can reach 90.6 GOP/s on PYNQ-Z2 at 150 MHz, which outperforms other accelerators implemented on the same platform in terms of peak performance and power efficiency.

Video denoising is of great significance in video processing when shooting conditions are complex such as dynamic scenes and low light. Although existing algorithms have already achieved remarkable denoising performance, the inference time of them is usually impractical for real-time applications. In this paper, we propose a convolutional neural network architecture for video denoising. In contrast to other existing CNN-based methods, our approach utilizes different proportion convolutional kernel numbers in a block for extracting enriched features. Channel attention mechanism is integrated in the network to enhance the denoising performance. The network only needs three contiguous frames and noise map as inputs, which leads to a similar excellent running time to the state-of-the-art. We compare our method with different conventional algorithms VBM4D, VNLB and the state-of-the-art CNN-based method FastDVDnet. Experiment results indicate that our method outputs more convincing results in visual and more robustness than others in both peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) indexes.

Electromigration-induced reliability (EM) is the major concern for design of modern power grids, characterized by high current densities and long metal lines. Due to their high inaccuracy of handling structures with large scale and complexity, traditional empirical EM prediction methods are not applicable for modern power grids. In this paper, we propose a novel analytical model of hydrostatic stress evolution during the void growth phase for general multi-segment wires, common interconnect structures in power grids, based on the adoption of Laplace transformation on Korhonen’s equation with coupled boundary conditions (BCs). The analytical solution is expressed with a set of auxiliary basis functions using the complementary error function. With analysis of the analytical expression form, the compact model is presented for practical EM analysis. Compared with the finite element analysis (FEA) results, our compact model can lead to less than 0.50% error on average for multi-segment wires extracted from IBM power grid benchmarks.

For distributed Edge Computing, we propose to leverage the deterministic data delivery in optical Determinist Dynamic Networks to jointly optimize compute and network resource allocation. Numerical results show that our DDN-based approach can accelerate the application total execution time by 50–70% from baseline approaches.

Object detection and tracking are critical computer vision tasks because of the broad needs in society; however, deep neural network-based methods cost many computational resources that hinder them from real scene applications. Quantization is a widely adopted technique to reduce the storage space and memory footprint which makes deep learning models more energy-efficient and resource-friendly. Traditional network quantization methods directly quantize neural networks layer-wise, which means the parameters in different channels take the same quantization range. In this paper, we propose a low-bit learning method for convolutional neural network object detector quantization. Different from previous methods, we quantize the detector channel-wisely to avoid accuracy loss in the low-bit framework. We use progressive quantization, progressive batch normalization fusion, and cut the unnecessary long-tail weights and activations to reduce quantization loss. Moreover, based on the object detector and long short-term memory network (LSTM), we develop a high-performance tracking system. We leverage the tensor decomposition to compress weights in LSTM for getting a higher compression ratio. Experiments are conducted on public datasets and our infrared aerial dataset for object detection and tracking. The experimental results show that our approach obtains better performance compared with the state-of-the-art methods in terms of accuracy and compression ratio.

Cables play an extremely important role in the transmission of electrical energy and communication signals in modern large-scale electrical systems. Early cable insulation defects often cause cable insulation breakdown, leading to permanent insulation damage. However, it is often very difficult to accurately detect these potential local defects. This paper proposes a novel non-destructive detection method for cable defect based on broadband impedance spectroscopy. Firstly, a simulation model of cable impedance spectrum is established. The correlation between impedance parameters and cable defects is well analyzed, and the influence of different defect locations on the cable impedance spectrum is studied. The effect of frequency spectrum on the accuracy of fault location is also analyzed. Finally, several cables of different lengths are simulated and the results show that this method has great spatial resolution within 0.3% for cables from several meters to hundreds of meters. It indicates that this non-destructive default location method has great potential in field application of cable diagnosis.

Oil-paper insulation high-voltage bushings are one of the most important power equipment in substations, and their insulation status is critical to the safety and the stabilization of the power system operation. During the production and operation, the bushing may be dampened because of the insufficient drying and impregnation or the intrusion of external moisture, which leads to the deterioration of oil paper insulation and threatens the reliability of bushing. Due to the special insulation structure of the bushing capacitor core, the distribution of moisture inside the bushing is usually non-uniform, especially when the moisture is intruded from the external environment. In this paper, a numerical simulation model of dual-unit laminated oil-paper and 110kV capacitor bushing is established. Then the frequency domain dielectric spectrum (FDS) characteristics of oil-paper insulation with different non-uniform moisture distribution are obtained by simulation. Different types of non-uniform moisture distribution include: non-uniform moisture distribution in different directions of the bushing, and different degrees of non-uniform moisture distribution. When the non-uniform moisture distribution of bushing is along the axial direction, different degrees of non-uniformity of moisture distribution have little effect on the FDS results; when the non-uniform moisture distribution of bushing is along the radial direction, the $\text{tan}\delta-f$ curve shows a clear dielectric loss peak, the characteristics of which are closely related to the degree of non-uniformity of the moisture distribution. Partial damping at the outermost side of the bushing or the end screen are more common in practice, and the FDS results of these two different types of damped bushing show different characteristic. This paper carried out a comprehensive FDS simulation study of different types of non-uniform dampened oil-paper insulation, which is of great significance for the evaluation of the insulation status of the oil-paper bushing.

At present, protection devices such as low-voltage circuit breakers and fuses are commonly used in low-voltage distribution networks, which can effectively prevent short circuits, overloads, and ground leakage. However, this method is out of work in detecting series arc faults caused by poor contact, insulation failure, etc. Therefore, how to achieve accurate detection of series arc faults has become a hot issue in current research. Wavelet transform is usually used for series arc fault detection. But it exists the problem of spectral aliasing, the false detection rate is still high. This paper uses detection method based on the current waveform to carry out research. By building an arc fault platform to simulate series arc faults, normal and arc fault data under different loads have been obtained. The structure of deep learning algorithm can be established through these experimental data. The accuracy of the algorithm is improved by using mini-batch gradient descent, exponential decay learning rate and Adam's optimization algorithm. By establishing test data for diagnostic verification, it was found that the algorithm has an excellent recognition rate.

Bushing is the most important part of transformer to connect external component. Bushing fault is one of the prime reasons causing transformer faults. Moisture is an important factor leading to aging of bushing insulation, so it is necessary to evaluate the moisture content of bushing. In the current research, it can confirm whether the oil-paper insulation bushing is uniform damp. But there is no research related to the moisture location of oil-paper insulation bushing. We use Comsol software to carry out a simulation study of the finite element analysis method for nonuniform damp bushing. And the distribution characteristics of hydronium ions and hydroxide ions under electrode polarization can be obtained from the simulation analysis. In this paper, the polarization and depolarization current (PDC) experimental platform was set up and models of nonuniform damp bushings were made. The position information of the high-moisture capacitive screen can be obtained through two different wiring methods. The validity of the method can be verified by the distribution law of hydrolysis ions in this simulation.

Network virtualization enables Internet service providers to flexibly provision heterogeneous virtual networks (VNs) over shared substrate networks (SNs). As more and more applications are carried by VNs, the survivability of VNs becomes a critical challenge under large-scale disaster risks. In this paper, we propose a synchronous evacuation strategy for VNs with dual virtual machines (VMs) inside a disaster risk zone (DRZ), which suffer higher risks than the VNs with single VM inside the DRZ. The evacuation strategy exploits post-copy technique to sustain the online service alive and enhances synchronous VM migrations to shorten the dual-VM evacuation time. Numerical results show that the proposed strategy can achieve satisfactory performances in terms of average and total evacuation times of dual-VMs.

Optical fibers allow more than high bandwidth for common services such as video and data transmission. The scientific community investigates optical fibers due to their delay value for photonic services, which require minimum conversions from the optical to the electrical domain. Optical fibers are used widely around the world, and with the decreasing price of optical components, it is possible to use optical fibers for infrastructure protection without any interruption of service. This article presents the measurement results for simultaneous transmission of two services with a sensor system according to the 100-GHz grid defined by the International Telecommunication Union. The results confirm that the 100-GHz grid is not sufficient.

Anomaly comprehension in surveillance videos is more challenging than detection. This work introduces the design of a lightweight and fast anomaly comprehension neural network. For comprehension, a spatio-temporal LSTM model is developed based on the structured, tensorized time-series features extracted from surveillance videos. Deep compression of network size is achieved by tensorization and quantization for the implementation on terminal devices. Experiments on large-scale video anomaly dataset UCF-Crime demonstrate that the proposed network can achieve an impressive inference speed of 266 FPS on a GTX-1080Ti GPU, which is 4.29 faster than ConvLSTM-based method; a 3.34% AUC improvement with 5.55% accuracy niche versus the 3D-CNN based approach; and at least 15k× parameter reduction and 228× storage compression over the RNN-based approaches. Moreover, the proposed framework has been realized on an ARM-core based IOT board with only 2.4W power consumption.

Video detection and classification constantly involve high dimensional data that requires a deep neural network (DNN) with huge number of parameters. It is thereby quite challenging to develop a DNN video comprehension at terminal devices. In this paper, we introduce a deeply tensor compressed video comprehension neural network called DEEPEYE for inference at terminal devices. Instead of building a Long Short-Term Memory (LSTM) network directly from raw video data, we build a LSTM-based spatio-temporal model from tensorized time-series features for object detection and action recognition. Moreover, a deep compression is achieved by tensor decomposition and trained quantization of the time-series feature-based spatio-temporal model. We have implemented DEEPEYE on an ARM-core based IOT board with only 2.4W power consumption. Using the video datasets MOMENTS and UCF11 as benchmarks, DEEPEYE achieves a 228.1× model compression with only 0.47% mAP deduction; as well as 15k× parameter reduction yet 16.27% accuracy improvement.