Earticle

This study proposes a novel transmit antenna selection (TAS) method for improving communications between the unmanned aerial vehicle (UAV) and ground station (GS). By selecting an appropriate UAV antenna, the signal-to-noise ratio (SNR) at the GS can be significantly enhanced. However, obtaining the necessary channel state information between the UAV and GS is challenging due to the UAV’s movement and resulting channel variations. To overcome this challenge, we propose an innovative approach that leverages a convolutional neural network to predict SNRs and a missing SNR completion method. The numerical evaluation demonstrates that the proposed method can effectively enhance TAS accuracy.

A simple relay selection scheme is proposed for analog network coding using differential modulation (ANC-DM). We propose a relay selection scheme based solely on the received signal power in which their own information is canceled out at each source node. The proposed power-based relay selection scheme is further applied to analog network coding using double-differential modulation (ANC-DDM), which is suitable for channels with carrier offset. Simulation results show that our proposed simple power-based criterion for relay selection in ANC-DM system has little loss compared to the optimal relay selection scheme using ideal BER.

Device-to-device (D2D) or Sidelink aided communication is regarded as one of the most promising technologies to improve the spectral efficiency of the 5G and beyond communication system. However, two main challenges exist: 1) the selection of the optimal number of devices for improving the spectral efficiency, and 2) improving the physical layer security of such a communication system. The optimal device improves the secrecy capacity, and the selection of optimal devices enhances the physical layer security. Therefore, in this paper, we propose a double threshold-based optimal device selection (ODS) scheme for a cooperative wireless network with amplify and forward (AF) protocol in the presence and absence of an eavesdropper to enhance the physical layer security for the D2D network. The proposed scheme is analyzed with different distance cases, device scenarios, and modulation schemes. The bit error rate (BER) performance analysis concludes that a performance gain of more than 4 dB is achieved at the BER of 0.004 for the proposed double threshold-based ODS scheme compared to the existing ODS scheme at a high signal to noise ratio (SNR). Furthermore, the proposed scheme enhances the physical layer security.

The performance of traditional routing protocols for mesh networks can decrease significantly in disaster recovery situations due to dynamic changes to the network environment. A number of multi-path routing protocols have been proposed to address such issues. MMQR is a multi-path routing protocol for OFDM?TDMA mesh networks which use multiple paths that can be replaced immediately by one another in case of link failures. In this paper we extend the MMQR such that the alternative paths are reordered adaptively as the communication environment changes based on the heuristic model of attractor selection mechanism by which biological entities are known to adapt to dynamically changing environment. Simulation results show that the multi-path routing protocol with attractor selection mechanism outperforms existing multi-path routing protocols in terms of packet delivery ratio, throughput and QoS.

We develop an intelligent information system in a multilayer electronic supply chain network. Using the internet for supply chain management (SCM) is a key interest for contemporary managers and researchers. It has been realized that the internet can facilitate SCM by making real time information available and enabling collaboration between trading partners. Here, we propose a multi-agent system to analyze the performance of the elements of a supply network based on the attributes of the information flow. Each layer consists of elements which are differentiated by their performance throughout the supply network. The proposed agents measure and record the performance flow of elements considering their web interactions for a dynamic route selection. A dynamic programming approach is applied to determine the optimal route for a customer in the end-user layer.

Enhanced Oil Recovery (EOR) has gained more attention as the results of increasing oil prices and higher oil demands. A large number of EOR methods have been developed, but the selection of EOR method is still a challenging task to the petroleum engineer. One of the main tools used to predict the EOR efficiency is a reservoir simulation which requires detailed reservoir description that may not be available or unreliable at the initial evaluation stage, and also needs extensive time. The other method is to use an expert opinion, but it tends to be biased by the expert's operational experience. This paper proposes an Artificial Neural Network (ANN) approach to enable the petroleum engineer to select the appropriate EOR method based on the reservoir characteristics. The structure of the ANN is optimized to be consisting of four layers by the repeated trial and error during the training. After being trained successfully with the successful EOR field data, the ANN is tested against the new data which are not used for the training, and contain the certain level of error, respectively. The results showed that the ANN model developed in this study can be used to select the most appropriate EOR process in a time and cost effective way.

이 연구에서는 거리기반 후보군 선정법을 인공신경망과 결합하여 생산 히스토리매칭을 수행하고, 생산추이 를 신뢰도 높게 예측하는 기법을 개발하였다. 생산추이를 입력층으로 한 역산모델 인공신경망을 구성하고, 거리기반 군집화를 통해 실제 생산자료와 유사한 추이를 보이는 후보모델을 인공신경망의 학습집단으로 활용하여 지도학습효 과를 개선하였다. 4개의 수압파쇄대를 갖는 셰일가스전에서 2년간 생산이력을 히스토리매칭하여 향후 3년간의 생산 이력을 예측하였다. 인공신경망 학습을 통해 매칭한 학습집단의 일가스생산량 오차는 초기 29.2%에서 마지막 세대 에는 4.3%로 개선되었다. 최적해는 참값과 비교하여 일가스생산량은 1.2%, 총생산량은 0.7%의 오차를 보였다. 최적 해를 이용한 민감도 분석결과, 모암의 공극률과 유체투과도가 생산추이예측에 민감하였다. 이 연구의 유사생산추이 후보군을 이용한 인공신경망 학습능력 개선기술은 효율적인 저류층 특성화와 신뢰도 높은 생산량 예측기법으로 활 용할 수 있다.

This paper presents a reliable history matching method that improves supervised-learning performances of an artificial neural network using the candidate models selected by distance-based clustering. The artificial neural network is used as an inverse method instead of a proxy model. The reservoir models within the cluster containing the true dataset update the training module. The developed method matches 2-year production histories and then estimates 3-year performances followed at shale gas reservoirs. The training performance improves the mean absolute error of daily gas rates from 29.2% to 4.3% while the optimum solutions, obtained by k-medoids clustering, show about 1.2% error of daily gas rates and 0.7% error of total production. The sensitivity analysis made by the optimal solutions reveals that the matrix porosity and its permeability influence the shale-gas production. The developed approach is able to estimate unknown production rates with reliable predictability.

Forest in the our country is in the age that needs positive operation in order to foster economical forest. Multiple operations for making valuable forest should be conducted steadily and timely from afforestation to harvesting. In order to execute these kinds of forest operations, the construction of skid trail network that can be effectively used as a pathway for forestry machine and working space is necessary. To investigate facility effect of skid trail network, we executed the location of skid trail network through centroid method by GIS for 50ha of harvesting workplace in mechanized model forest located in Hongcheon, Gangwon Province. As a result of this research, skid trail density in this area changed from 79m/ha with current method to 42m/ha with improved method. It appeared that skid trail density with improved method is nearly half of current method even though the cutting area is the same as the current cutting area. Also, skidding distance changed from 117m with current method to 57m with improved method. It appears that skidding distance with improved method is nearly half of current method even though cutting area was enlarged in adjacent tending cutting area.

최근 들어 인공 지능을 이용한 영상 객체 인식에 대한 연구 및 개발이 활발하게 진행되고 있다. 그 연장선상에서 도로 유지 및 관리 분야에도 관련 연구의 활용도가 크게 향상될 것으로 기대된다. 그 중에서도 특히 도로 노면 파손 객체 인식 (Object Detection) 을 위한 인공 지능 모델이 지속적으로 개발되고 있다. 이러한 객체 인식 알고리즘을 개발하려면 우선적으로 특징 지도를 생성하는 Backbone Network가 반드시 필요한데, 본 논문에서는 이를 선정하는 방법을 제안하고자 한다. 이를 위해 6,000여 장의 도로 노면 파손 영상 데이터를 확보하고, 근래에 많 이 사용되는 4종류의 심층 신경망을 활용하여 성능을 비교한다. 3가지의 성능 평가 방법을 적 용하여 심층 신경망의 특징을 분석하고 최적의 심층 신경망을 결정한다. 또한 하이퍼 파라미 터의 최적 조율을 통해 성능을 향상시키고, 최종적으로 도로 노면 파손 영상 분류를 위하여 85.9%의 정확도로 수행이 가능한 경량화된 Backbone Network용 심층 신경망을 제안한다.

In recent years, research and development on image object recognition using artificial intelligence have been actively carried out, and it is expected to be used for road maintenance. Among them, artificial intelligence models for object detection of road surface are continuously introduced. In order to develop such object recognition algorithms, a backbone network that extracts feature maps is essential. In this paper, we will discuss how to select the appropriate neural network. To accomplish it, we compared with 4 different deep neural networks using 6,000 road surface damage images. Based on three evaluation methods for analyzing characteristics of neural networks, we propose a method to determine optimal neural networks. In addition, we improved the performance through optimal tuning of hyper-parameters, and finally developed a light backbone network that can achieve 85.9% accuracy of road surface damage classification.

인공지능 기술의 발달로 인해 사물인터넷 기기와 스마트 기기들의 사용이 급속도로 증가하였다. 기존의 인공지능 모델 학습 방법은 보안 문제 등 많은 단점이 존재한다. 따라서 기존의 단점을 보완하면서 원 데이터 송 신이 필요 없는 연합학습(FL: Federated Learning)의 연구가 활발히 이루어지고 있다. 연합학습은 많은 장점들을 가지고 있지만 참여 기기 간 통신 및 네트워크의 상황에 차이가 존재하기 때문에 일부 기기로 인해 전체적인 학 습이 지연될 수 있다는 한계를 가진다. 따라서 본 논문에서는 통신과 네트워크 상황에 따라 적절하게 학습 참여 기기를 선택하는 기법을 제안한다. 제안한 기법을 활용할 경우 측정된 네트워크와 통신 성능 지표를 바탕으로 각 기기들의 통신 및 네트워크 상황을 예측하고 이를 통해 기기들을 선택함으로써 보다 효율적인 연합학습을 수행한 다. 이는 기존의 학습기법과 비교하였을 때 전체적인 학습시간을 감소시키는 결과를 보였으며 더욱 짧은 시간에 더 높은 성능을 가지는 모델을 구축하는 결과를 보였다.

Supported by the development of artificial intelligence (AI) technology, the use of Internet of Things (IoT) and smart devices has increased rapidly. Existing methods of training AI models have many disadvantages, such as security problems. Therefore, many researches on federated learning (FL) that does not require transmission of raw data have been actively conducted for overcoming the existing shortcomings. FL has many advantages, but there is a limit that some devices may delay the FL system’s overall training due to the differences in communication and network situations between participating devices. To overcome such limitation, in this paper, we propose the participant selection scheme based on network situation prediction of server-to-device communication paths. Using the proposed scheme, more efficient FL can be performed by selecting participant devices intelligently based on the predicted communication and network information of each device.

딥러닝(Deep learning) 기법은 패턴분석, 이미지분류 등 다양한 분야에서 높은 성과를 나타내고 있다. 특히, 주식시장 분석문제는 머신러닝 연구분야에서도 어려운 분야이므로 딥러닝이 많이 활용되는 영역이다. 본 연구에서는 패턴분석과 분류능력이 높은 딥러닝의 일종인 합성곱신경망(Convolutional Neural Network) 모델을 활용하여 주가방향 예측방법을 제안한다. 추가적으로 합성곱신경망 모델을 효율적으로 학습시키기 위한 속성선택(Feature Selection, FS)방법이 적용된다. 합성곱신경망 모델의 성과는 머신러닝 단일 분류기와 앙상블 분류기를 벤치마킹하여 객관적으로 검증된다. 본 연구에서 벤치마킹한 분류기는 로지스틱 회귀분석(Logistic Regression), 의사결정나무(Decision Tree), 인공신경망 (Neural Network), 서포트 벡터머신(Support Vector Machine), 아다부스트(Adaboost), 배깅(Bagging), 랜덤포레스트(Random Forest)이다. 실증분석 결과, 속성선택을 적용한 합성곱신경망이 다른 벤치마킹 분류기보다 분류 성능이 상대적으로 높게 나타났다. 이러한 결과는 합성곱신경망 모델과 속성선택방법을 적용한 예측방법이 기업의 재무자료에 내포된 가치를 보다 정교하게 분석할 수 있는 가능성이 있음을 실증적으로 확인할 수 있었다.

Recently, deep learning has shown high performance in various applications such as pattern analysis and image classification. Especially known as a difficult task in the field of machine learning research, stock market forecasting is an area where the effectiveness of deep learning techniques is being verified by many researchers. This study proposed a deep learning Convolutional Neural Network (CNN) model to predict the direction of stock prices. We then used the feature selection method to improve the performance of the model. We compared the performance of machine learning classifiers against CNN. The classifiers used in this study are as follows: Logistic Regression, Decision Tree, Neural Network, Support Vector Machine, Adaboost, Bagging, and Random Forest. The results of this study confirmed that the CNN showed higher performancecompared with other classifiers in the case of feature selection. The results show that the CNN model effectively predicted the stock price direction by analyzing the embedded values of the financial data

네트워크 효과가 큰 산업에서 호환성의 선택은 기업의 중요한 전략이다. Katz and Shapiro (1985) 이후 호환성에 대해 다양한 연구가 진행되었으나, 수평적으로 차별화된 제품 간 호환성에 대해서는 연구에 따라 엇갈린 결론이 제시되고 있다. 본 연구는 게임이론 모형을 통해 수평적으로 차별화된 제품을 생산하는 기업들이 호환성을 선택할 유인이 있는가, 기업의 호환성 선택은 사회적으로 바람직한가 등의 문제를 고찰한다. 네트워크 효과에 따라 시장 규모가 확대될 수 있도록 일반적인 역수요함수를 사용했다는 점, 호환성과 제품 차별화 간의 상충관계를 명시적으로 고려했다는 점 등이 기존 연구와 차별화되는 부분이다. 분석 결과를 보면, 호환성이 제품 차별화를 크게 감소시킬 것으로 예상되면 기업은 호환성을 선택하지 않는다는 것이고, 이와 같은 기업의 선택은 호환성이 사회적으로 바람직한 상황에서도 나타날 수 있다는 것이다. 비디오 게임기 시장, 국내 모바일 인터넷 시장 등의 실제 사례에서도 이와 같은 사실을 확인할 수 있다.

Brands of products are said to be ‘compatible’ if they can work interchangeably, or the output of one brand can be operated or used by the other brands. Though often considered as a technical issue, product compatibility turns out to be one of key strategic decisions in industries exhibiting network effect. Katz and Shapiro (1985) pioneered to examine compatibility decisions between homogeneous products in the presence of network effect. Many authors attempted to extend it into either vertically or horizontally differentiated products. As for vertically differentiated products, it is generally known that a firm producing a low-quality product prefers compatibility, while its rival producing a high-quality product does not. When it comes to horizontal differentiation, however, there is little consensus whether firms have sufficient incentive to pursue compatibility, and their decisions are socially desirable. Using a simple game-theoretic model, I analyze equilibrium production quantities, firms’ incentives toward compatibility, and their welfare consequences. By employing Cournot type competition, I explicitly consider network effect in the analysis. Further, I explicitly concern myself with the trade-off between compatibility and horizontal differentiation. The main conclusions are as follows; first, there exist equilibria where firms select compatibility; second, firms tend to achieve socially insufficient compatibility. Several real world examples including the video game industry and the mobile Internet service industry in Korea support my argument. Previous literature mainly attributed the failure of achieving compatibility to asymmetry between firms (product quality, installed base of customers, etc). However, even symmetric firms may not provide compatibility when compatibility decreases product differentiation. Finally, firms should pay more attention to the characteristics of their products such as the trade-off between compatibility and differentiation when they formulate compatibility strategy.

본 논문에서는 이동하는 통신 환경에서 합성곱 신경망에 기반하여 최적의 송신 안테나를 선택하는 방법 을 제안한다. 고려하는 통신시스템은 다중 안테나를 가지고 있으며 양방향 통신은 시분할 이중화로 수행한다. 수신 시에는 모든 안테나를 이용하지만 송신할 때는 최적의 안테나를 선택하여 전송한다. 합성곱 신경망의 입력은 과거 수신 시 측정한 신호대잡음비이다. 기존의 방법으로는 과거 신호대잡음비들의 평균값에 기반한 방법과 가장 최근에 수신할 때의 신호대잡음비에 기반하는 두 가지 방법을 고려한다. 제안하는 방법과 기존의 두 가지 방법은 컴퓨터 모의실험을 통해 비교한다. 이동속도와 수신할 확률을 바꿔가며 모의실험 한 결과 제안하는 방식이 광대역 신호에서 가장 높은 안테나 선택 정확도를 보이고 협대역 신호에서는 가장 최근 신호대잡음비를 사용하는 기존의 방법이 미세하지만 가장 우수하다.

This paper proposes a new technique for selecting optimal transmit antennas using convolutional neural network(CNN) in mobile communication environments. The communication system considered in this paper has multiple antennas and two-way communication is performed in a time-division duplexing(TDD) manner. The system uses all of the antennas in receiver mode but only one in transmitter mode. The input of the CNN is the signal to noise ratios(SNRs) for the past received signals. The conventional method selects the optimal antenna based on the average of the past received SNRs or the most recently received SNR. We compare the proposed method with two conventional methods through computer simulation. According to the results, by changing the mobile speed and the probability(or frequency) of receiving, the proposed CNN method has the highest accuracy in wideband signals while the conventional method using the recently received SNR has the highest accuracy in narrowband signals.