• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.89-100
• /
• 2023

Recently, on-device artificial intelligence (AI) solutions using mobile devices and embedded edge devices have emerged in various fields, such as computer vision, to address network traffic burdens, low-energy operations, and security problems. Although vision transformer deep learning models have outperformed conventional convolutional neural network (CNN) models in computer vision, they require more computations and parameters than CNN models. Thus, they are not directly applicable to embedded edge devices with limited hardware resources. Many researchers have proposed various model compression methods or lightweight architectures for vision transformers; however, there are only a few studies evaluating the effects of model compression techniques of vision transformers on performance. Regarding this problem, this paper presents a performance evaluation of vision transformers on embedded platforms. We investigated the behaviors of three vision transformers: DeiT, LeViT, and MobileViT. Each model performance was evaluated by accuracy and inference time on edge devices using the ImageNet dataset. We assessed the effects of the quantization method applied to the models on latency enhancement and accuracy degradation by profiling the proportion of response time occupied by major operations. In addition, we evaluated the performance of each model on GPU and EdgeTPU-based edge devices. In our experimental results, LeViT showed the best performance in CPU-based edge devices, and DeiT-small showed the highest performance improvement in GPU-based edge devices. In addition, only MobileViT models showed performance improvement on EdgeTPU. Summarizing the analysis results through profiling, the degree of performance improvement of each vision transformer model was highly dependent on the proportion of parts that could be optimized in the target edge device. In summary, to apply vision transformers to on-device AI solutions, either proper operation composition and optimizations specific to target edge devices must be considered.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.7
• /
• pp.11-18
• /
• 2023

In this paper, we evaluate deep learning time series forecasting models. Recent studies show that those models perform better than the traditional prediction model such as ARIMA. Among them, recurrent neural networks to store previous information in the hidden layer are one of the prediction models. In order to solve the gradient vanishing problem in the network, LSTM is used with small memory inside the recurrent neural network along with BI-LSTM in which the hidden layer is added in the reverse direction of the data flow. In this paper, we compared the performance of Informer by comparing with other models (LSTM, BI-LSTM, and Transformer) for real Nitrogen dioxide (NO2) data. In order to evaluate the accuracy of each method, mean square root error and mean absolute error between the real value and the predicted value were obtained. Consequently, Informer has improved prediction accuracy compared with other methods.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.14 no.1
• /
• pp.84-92
• /
• 2001

In this papers, we proposed the new method in order to diagnosis aging state of transformers. For wavelet transform, Daubechies filter is used, we can obtain wavelet coefficients which is used to extract feature of statistical parameters (maximum value, average value, dispersion skewness, kurtosis) about each acoustic emission signal. Also, these coefficients are used to identify normal and fault signal of internal partial discharge in transformer. As improved method for classification use neural network. Extracted statistical parameters are input into an back-propagation neural network. The number of neurons of hidden layer are obtained through Result of Cross-Validation. The network, after training, can decide whether the test signal is early aging state, alst aging state or normal state. In quantity analysis, capability of proposed method is superior to compared that of classical method.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.64 no.1
• /
• pp.23-28
• /
• 2015

Two old synchronous condensers in Jeju are being replaced by new machines to operate Jeju AC network with Haenam-Jeju HVDC system stably. Before new synchronous condensers operate on site, voltage stability analysis is conducted to verify stable operation of jeju AC network. Through impedance analysis of the synchronous machine, transformer and ac network, the equivalent circuit is constructed and the voltage drop during start-up is calculated. Then, PSS/E fault analysis is performed to acquire short-circuit capacity according to the generator operation scenarios. Voltage variation when starting synchronous condenser is simulated in PSCAD/EMTDC and satisfies the operating condition of jeju AC network and HVDC #1 system.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.67 no.9
• /
• pp.1146-1151
• /
• 2018

Recently, researches on the intelligence of electric power facilities have been trying to apply artificial intelligence techniques as computer platforms have improved. In particular, faults occurring in substation should be able to quickly identify possible faults and minimize power fault recovery time. This paper presents fault location technique for 154kV substation using neural network. We constructed a training matrix based on the operating conditions of the circuit breaker and IED to identify the fault location of each component of the target 154kV substation, such as line, bus, and transformer. After performing the training to identify the fault location by the neural network using Weka software, the performance of fault location discrimination of the designed neural network was confirmed.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.88.4-88
• /
• 2001

In this paper, we propose a solver for differential equations, using a multi-layer neural network. The multi-layer neural network is a transformer function originally where the function is differential and the explicit representation has been developed. The learning determines the response of neural networks; however, the response is not equal to the output values. The differential relations are also the response. The differential conditions can be also set as teaching data; therefore, there is a possibility to reach a new solver for the differential equations. Since it is unknown how to define the input data for the neural network solver during long terms, we could not derive the expressions. Recently, the analogue type neural network is known and it transforms any vector to another The "any" must be...

• Journal of Power Electronics
• /
• v.17 no.1
• /
• pp.20-30
• /
• 2017

A novel zero-voltage-switching (ZVS) push-pull forward converter with a parallel resonant network is presented in this paper. The novel topology can provide a releasing loop for the energy storage in a leakage inductor for the duration of the power switching by the resonant capacitors paralleled with the primary windings of the transformer. Then the transformer leakage inductor is utilized to be resonant with the parallel capacitor, and the ZVS operation is achieved. This converter exhibits many advantages such as lower duty-cycle losses, limited peak voltage across the rectifier diodes and a higher efficiency. Furthermore, the operating principles and key problems of the converter design are analyzed in detail, and the ZVS conditions are derived. A 500W experimental converter prototype has been built to verify the effectiveness of the proposed converter, and its maximum efficiency reaches 94.8%.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.59 no.10
• /
• pp.1741-1747
• /
• 2010

The development of overload management systems for distribution transformers offers new opportunities for improving the reliability of distribution systems. It allows network planners to optimize the system resource utilization and investment cost. Such an improvement in the flexibility of the distribution network is only possible if the operator has more accurate knowledge of the realtime conditions of distribution transformers. In this paper, we present an improved overload decision system for distribution transformers using realtime monitoring data. Our study can be categorized into two parts: (a) improvement in the criteria for judging the overload conditions of distribution transformers and (b) development of an overload evaluation system using realtime monitoring data. In order to determine the overload criteria, overload experiments are performed on sample transformers; the results of these experiments are used to define the relationship between the transformer overload and the increase in the top-oil temperature. To verify the accuracy of the experimental results, field tests are performed using specially manufactured transformers, the loads and top-oil temperatures of which can be measured. For arriving at online overload decisions, we propose methods whereby the measured load curve can be converted into an overload characteristic curve and the overload time can be calculated for any load condition. The developed system is able to evaluate the overload for individual distribution transformers and calculate the losses using realtime monitoring data.

• The Transactions of the Korean Institute of Electrical Engineers C
• /
• v.54 no.12
• /
• pp.578-584
• /
• 2005

This paper describes development of management software for transformers based on artificial intelligent analysis technology of dissolved gases in oil. Fault interpretation using the artificial intelligent analysis is performed by the artificial neural network and a rule based on the analysis of dissolved gases. The used gases are acetylene($C_{2}H_{2}$), hydrogen($H_2$), ethylene($C_{2}H_{4}$), methane($CH_4$), ethane($C_{2}H_{6}$), carbon monoxide(CO) and carbon dioxide($CO_2$). This software is mainly composed of gases input, fault's causes, expected fault's phenomena in detail, the decision on maintenance as well as report and gas trend windows. It is indicated that this is very powerful software for the efficient management of oil-immersed transformers using data analysis of gas components.

• Proceedings of the KIEE Conference
• /
• 1998.07f
• /
• pp.2067-2069
• /
• 1998

A prototype semiconductor switch for the command resonant charging system has been developed for a line type modulator, which charges parallel pulse forming network(PFN) up to voltage of 5 kV at repetition rates of 60 Hz. A phase controlled power supply provides charging of the 4.7 ${\mu}s$ filter capacitor bank to voltage up to 5 kV. A solid state module of series stack array of sixe matched SCRs(1.6 kV, 50 A) is used as a command charging switch to initiate the resonant charging cycle. Both resistive and RC snubber network are used across each stage of the switch assembly in order to ensure proper voltage division during both steady state and transient condition. A master trigger signal is generated to trigger circuits which are transmitted through pulse transformer to each of the 6 series switch stages. A pulse transformer is required for high voltage trigger or power isolation. This paper will discuss trigger method, protection scheme, circuit simulation, and test result.