• 한국전자통신학회논문지
• /
• 제15권3호
• /
• pp.465-472
• /
• 2020

최근 들어, 오디오 이벤트 검출을 위하여 다양한 딥뉴럴네트워크 기반의 방법들이 제안되어 왔다. 본 연구에서는 베이스라인 CRNN(Convolutional Recurrent Neural Network) 구조에 attention 방식을 도입함으로서 오디오 이벤트 검출의 성능을 향상시키고자 하였다. 베이스라인 CRNN의 입력단에 context gating을 적용하고 출력단에 attention layer을 추가하였다. 또한, 프레임(frame) 단위의 강전사 레이블(strong label)정보 뿐만 아니라 클립(clip) 단위의 약전사 레이블(weakly label) 오디오 데이터를 이용한 학습을 통하여 보다 나은 성능을 이루고자 하였다. DCASE 2018/2019 Challenge Task 4 데이터를 이용한 오디오 이벤트 검출 실험에서 제안된 attention 기반의 CRNN을 통하여 기존의 CRNN 방식에 비해서 최대 66%의 상대적 F-score 향상을 얻을 수 있었다.

• 전자공학회논문지CI
• /
• 제38권4호
• /
• pp.10-21
• /
• 2001

본 논문은 컴퓨터를 통해 가상세계를 구축하기 위한 논리적 시뮬레이션 방법을 개발한다. 기존의 일정한 패턴으로 전개되는 게임, 유아용 교육 시스템은 다수의 사용자를 가상의 공간으로 유도하여 상황을 전개하나 환경자체의 변화를 유발시키는 것에는 미흡하다. 따라서 본 논문에서는 개념적이고 논리적인 상황 변화를 가능하게 하는 가상 환경 시스템을 개발한다. 환경의 논리적인 구현은 시공간적인 상황속에서 시뮬레이션함으로써 이루어진다. 본 논문에서는 상황계층구조를 통해 시뮬레이션 가능한 상황을 정의하고 단일 사건을 정의한다. 그리고 인과관계를 세분화하여 사건 발생을 전개해 나간다. 자연 현상에서 사건은 물리법칙에 기반하여 발생하며, 현상은 사건들간의 연관성을 이용하여 표현한다. 이러한 방법은 초기 사건의 세분화 작업이 어려우나 사건의 재사용의 장점을 가져온다. 그리고 동일 패턴의 사건에서는 새로운 조건을 이용함으로써 보다 현실적이고 논리적인 성황의 구현을 가능하게 한다. 특히 자연현상에서 사건의 원천을 정의하고 객체의 생명 시간에 의한 객체의 존재유무가 사건의 주된 요인으로 취급된다. 제안하는 방법은 자연현상 중 계절의 변화에 적용하여 표현 가능함을 보인다.

• 한국경영과학회지
• /
• 제20권2호
• /
• pp.159-178
• /
• 1995

Supervisory control is an approach based on formal language. it is used to model and control discrete event systems in which each discrete event process is represented as an automation. A supervisor is a generator that switches control patterns in such a way that a given discrete evenet process behaves in obedience to various constraints. A flexible manufacturing cell (FMC) is one of discrete evenet systems. Functions necessary for the operation of an FMC are characterized by operational components and informational compoments. The operational components can be modeled using the finite state machines and the informational components can be modeled using the abstract formalism which describes supporting operations of the cell controller. In this paper, we addressed function required for FMC control specification, software engineering aspects on FMC control based on supervisory control, a concept of event queue for resolving synchronization problem, and complexity reduction. Based on the mathematical model of an FMC. we synthesized the controller by integrating a supervisor for FMC with control specification that specifies event-driven operation of the cell controller. The proposed control scheme is stable mathematically so that the system always behaves on a controlled way even under the existence of uncontrollable events. Furthermore, using an event queue concept, we can solve a synchronization problem caused by the violation of instantaneity assumption of supervisory control theory in real life situation. And also, we can propotype a control software rapidly due to the modularity of the proposed control scheme.

• 천문학회보
• /
• 제38권2호
• /
• pp.33-33
• /
• 2013

We are developing empirical space weather (solar flare, solar proton event, and geomagnetic storm) forecast models based on solar data. In this talk we will review our main results and recent progress. First, we have examined solar flare (R) occurrence probability depending on sunspot McIntosh classification, its area, and its area change. We find that sunspot area and its increase (a proxy of flux emergence) greatly enhance solar flare occurrence rates for several sunspot classes. Second, a solar proton event (S) forecast model depending on flare parameters (flare strength, duration, and longitude) as well as CME parameters (speed and angular width) has been developed. We find that solar proton event probability strongly depends on these parameters and CME speed is well correlated with solar proton flux for disk events. Third, we have developed an empirical storm (G) forecast model to predict probability and strength of a storm using halo CME - Dst storm data. For this we use storm probability maps depending on CME parameters such as speed, location, and earthward direction. We are also looking for geoeffective CME parameters such as cone model parameters and magnetic field orientation. We find that all superstorms (less than -200 nT) occurred in the western hemisphere with southward field orientations. We have a plan to set up a storm forecast method with a three-stage approach, which will make a prediction within four hours after the solar coronagraph data become available. We expect that this study will enable us to forecast the onset and strength of a geomagnetic storm a few days in advance using only CME parameters and the WSA-ENLIL model. Finally, we discuss several ongoing works for space weather applications.

• 한국군사과학기술학회지
• /
• 제14권5호
• /
• pp.913-919
• /
• 2011

In this paper, we present a new efficient event detection algorithm for sensor networks with faults. We focus on multi-attributed events, which are sets of data points that correspond to interesting or unusual patterns in the underlying phenomenon that the network monitors. Conventional algorithms cannot detect some events because they treat only their own sensor readings which can be affected easily by environmental or physical problem. Our approach exploits spatio-temporal correlation of sensor readings. Sensor nodes exchange a fault-tolerant code encoded their own readings with neighbors, organize virtual sensor readings which have spatio-temporal correlation, and determine a result for multi-attributed events from them. In the result, our proposed algorithm provides improvement of detecting multi-attributed events and reduces the number of false-negatives due to negative environmental effects.

• Wind and Structures
• /
• 제3권3호
• /
• pp.177-191
• /
• 2000

This paper presents an approach for evaluating directionality effects for both wind speeds and wind loads in hurricane-prone regions. The focus of this study is on directional wind loads on low-rise structures. Using event-based simulation, hurricane directionality effects are determined for an open-terrain condition at various locations in the southeastern United States. The wind speed (or wind load) directionality factor, defined as the ratio of the N-year mean recurrence interval (MRI) wind speed (or wind load) in each direction to the non-directional N-year MRI wind speed (or wind load), is less than one but increases toward unity with increasing MRI. Thus, the degree of conservatism that results from neglecting directionality effects decreases with increasing MRI. It may be desirable to account for local exposure effects (siting effects such as shielding, orientation, etc.) in design. To account for these effects in a directionality adjustment, the factor described above for open terrain would need to be transformed to other terrains/exposures. A "local" directionality factor, therefore, must effectively combine these two adjustments (event directionality and siting or local exposure directionality). By also considering the direction-specific aerodynamic coefficient, a direction-dependent wind load can be evaluated. While the data necessary to make predictions of directional wind loads may not routinely be available in the case of low-rise structures, the concept is discussed and illustrated in this paper.

• Advances in Computational Design
• /
• 제3권4호
• /
• pp.405-418
• /
• 2018

Building Information Modelling (BIM) and Discrete Event Simulation (DES) are tools widely used in the context of the construction industry. While BIM is used to represent the physical and functional characteristics of a facility, DES models are used to represent its construction process. Integrating both is beneficial to those interested in the field of construction management since it has many potential applications. Game engines provide a human navigable 3D virtual environment in which the integrated BIM and DES models can be visualised and interacted with. This paper reports the experience obtained while developing a simulator prototype which integrates a BIM and a DES model of a single construction activity within a commercial game engine. The simulator prototype allows the user to visualise how the duration of the construction activity is affected by different input parameters interactively. It provides an environment to conduct DES studies using the user's own BIM models. This approach could increase the use of DES technologies in the context of construction management and engineering outside the research community. The presented work is the first step towards the development of a serious game for construction management education and was carried out to determine the suitable IT tools for its development.

• 산업경영시스템학회지
• /
• 제42권4호
• /
• pp.126-134
• /
• 2019

Nowadays, since there are so many big data available everywhere, those big data can be used to find useful information to improve design and operation by using various analysis methods such as data mining. Especially if we have event log data that has execution history data of an organization such as case_id, event_time, event (activity), performer, etc., then we can apply process mining to discover the main process model in the organization. Once we can find the main process from process mining, we can utilize it to improve current working environment. In this paper we developed a new method to find a final diagnosis of a patient, who needs several procedures (medical test and examination) to diagnose disease of the patient by using process mining approach. Some patients can be diagnosed by only one procedure, but there are certainly some patients who are very difficult to diagnose and need to take several procedures to find exact disease name. We used 2 million procedure log data and there are 397 thousands patients who took 2 and more procedures to find a final disease. These multi-procedure patients are not frequent case, but it is very critical to prevent wrong diagnosis. From those multi-procedure taken patients, 4 procedures were discovered to be a main process model in the hospital. Using this main process model, we can understand the sequence of procedures in the hospital and furthermore the relationship between diagnosis and corresponding procedures.

• 한국안전학회지
• /
• 제15권2호
• /
• pp.126-135
• /
• 2000

Probabilistic Safety Assessment (PSA) is an engineering analysis method to identify possible contributors to the risk from a nuclear power plant and now it has become a standard tool in safety evaluation of nuclear power plants. PSA consists of three phases named as Level 1, 2 and 3. Level 2 PSA, mainly focused in this paper, uses a step-wise approach. At first, plant damage states (PDSs) are defined from the Level 1 PSA results and they are quantified. Containment event tree (CET) is then constructed considering the physico-chemical phenomena in the containment. The quantification of CET can be assisted by a decomposition event tree (DET). Finally, source terms are quantitatively characterized by the containment failure mode. As the main benefit of PSA is to provide insights into plant design, performance and environmental impacts, including the identification of the dominant risk contributors and the comparison of options for reducing risk, this technique is expected to be applied to the industrial safety area.

• 한국정보시스템학회지:정보시스템연구
• /
• 제27권4호
• /
• pp.115-128
• /
• 2018

Purpose The purpose of this study is to predict future behaviors of business process. Specifically, this study tried to predict the last activities of process instances. It contributes to overcoming the limitations of existing approaches that they do not accurately reflect the actual behavior of business process and it requires a lot of effort and time every time they are applied to specific processes. Design/methodology/approach This study proposed a novel approach based using deep learning in the form of dynamic recurrent neural networks. To improve the accuracy of our prediction model based on the approach, we tried to adopt the latest techniques including new initialization functions(Xavier and He initializations). The proposed approach has been verified using real-life data of a domestic small and medium-sized business. Findings According to the experiment result, our approach achieves better prediction accuracy than the latest approach based on the static recurrent neural networks. It is also proved that much less effort and time are required to predict the behavior of business processes.