Chung Cheng University Institutional Repository
Not a member yet
889 research outputs found
Sort by
以同步擠壓轉換為基礎之混合方法進行電壓閃爍評估;A Synchrosqueezing Transform-based Hybrid Approach for Voltage Flicker Assessment
No full text[[abstract]]隨著電力系統中非線性負載使用量逐漸增加，使得各項電力品質問題日趨嚴重，電壓閃爍是電力品質的重要指標之一。由於電力系統中具有快速變動的負載造成了電壓變動，而電壓變動不但會造成電子儀器設備功能失常，更會影響照明燈具輸出的穩定度，燈光閃爍會造成視覺干擾、視覺疲?甚至傷害視?。準確的量測為改善電力品質問題的首要工作，因此該如何建立有效且正確的數據分析是相當重要的。IEC 61000-4-15電壓閃爍量測方法係以模擬視覺對光線變化之反應，評估閃爍的嚴重程度，其設計涵蓋了多項功能方塊以模擬人眼對燈光閃爍以及產生之腦部響應，並且以統計方式計算電壓閃爍之短期與長期閃爍指標。目前為最多國家所採用的標準。本論文提出一套以同步擠壓轉換為基礎之分析方法來計算出閃爍汙染問題的嚴重度，利用IEC 61000-4-15規範中所制定的閃爍分析儀量測功能方塊圖做為基本架構，並針對其中解調出電壓包絡線的方塊加以改良。本論文提出之方法以希爾伯特轉換擷取電壓變動之包絡線，再以同步擠壓轉換與均值移動聚類法分析電壓變動源之閃爍頻率成份，透過帶通濾波器組求得精確之閃爍頻率之振幅，模擬與量測之結果皆顯示提出之方法在電壓閃爍的量測相當正確且有效。
With the widespread existence of nonlinear loads in the power system, the power quality disturbances are increasingly present. Flicker is one of the major power quality event. The impacts of flicker disturbance have been commonly seen in the power distribution networks nowadays. Voltage fluctuations caused by rapid reactive power consumption of loads and photovoltaic or wind generator output variations may produce flickers and cause undesired effects on electric power components and human eyes. IEC 61000-4-15 is the most popular flickermeter standard providing basic information for the design based on human eye-brain perception intended to indicate the correct flicker perception level for all practical voltage fluctuation waveforms, furthermore, performing an on-line statistical analysis of the flicker level.This dissertation presents a hybrid approach for voltage flicker assessment by using synchrosqueezing transform-based algorithm. The approach is an improved demodulation method to extract the voltage envelope which can replace the squaring demodulation suggested by IEC 61000-4-15. The proposed method firstly gives characterization of voltage fluctuations through accurate extraction of the measured voltage envelope by Hilbert transform. The synchrosqueezing transform and an unsupervised clustering method called mean shift, are then applied to determine the number of frequency components and corresponding frequencies. It is followed by the implementation of the bandpass filters to accurately detect the magnitude of each flicker component. The proposed hybrid method is tested by both simulations and field measurements validation. Results compared with other commonly seen methods show that the proposed method provides a more accurate flicker assessment
手勢追蹤及其應用;Hand Gesture Tracking and its Application
No full text[[abstract]]由於人機交互系統的發展，用於一般相機鏡頭的手部追蹤技術受到很多的關注。對於這項工作來說，我們改良了近期有名的物件追蹤演算法，稱Tracking-Learning-Detection (TLD)，以利於進行高效率的手部追蹤應用。具體來說，我們結合了反向投影方法來適應其中的追蹤及偵測機制以達成手部追蹤的各種情況。同時，我們也提出利用手部追蹤軌跡來提高TLD分類的準確度。在實驗結果表明了此改良的TLD成功地實現人機交互應用的目標。
Hand tracking on general camera has received a lot of attention due to numerous applications of human computer interaction. In this work, we improve a recently famous object tracking algorithm, named Tracking-Learning-Detection (TLD), in order to do efficient and effective hand tracking. Specifically, we incorporate the back projection method to adapt the tracking and detection mechanism for hand tracking. We also propose to use hand trajectories to advance the accuracy of the TLD classifier. Experimental results show that the revised TLD successfully accomplishes the target human computer interaction
配電系統之分散式源位置與容量最佳化;Optimal Placement and Sizing of DG Units in a Distribution System
No full text[[abstract]]Nowadays, the integration of distributed generation (DG) units is common in the distribution system with many potential benefits. However, the benefits depend on the placement and sizing of DG units which are selected for connecting the network. This thesis presents a biogeography-based optimization (BBO) method to optimally place and size DG units considering multiple objective functions in the unbalanced distribution system. The objective functions are total power loss, voltage unbalanced index reduction, and voltage profile improvement while total harmonic distortion (THD) and individual harmonic distortion (IHD) are kept within the limits set by the IEEE 519 harmonic standard. The load flow and harmonic power flow are performed by using co-simulation between engines OpenDSS and MATLAB through component object model server dynamic-link library. Two test cases are under study: the first one is the unbalanced distribution system of IEEE 123-bus test feeder and the second one is a distribution system of Taipower. Finally, the results obtained by the proposed method are shown and compared with other methods.
Nowadays, the integration of distributed generation (DG) units is common in the distribution system with many potential benefits. However, the benefits depend on the placement and sizing of DG units which are selected for connecting the network. This thesis presents a biogeography-based optimization (BBO) method to optimally place and size DG units considering multiple objective functions in the unbalanced distribution system. The objective functions are total power loss, voltage unbalanced index reduction, and voltage profile improvement while total harmonic distortion (THD) and individual harmonic distortion (IHD) are kept within the limits set by the IEEE 519 harmonic standard. The load flow and harmonic power flow are performed by using co-simulation between engines OpenDSS and MATLAB through component object model server dynamic-link library. Two test cases are under study: the first one is the unbalanced distribution system of IEEE 123-bus test feeder and the second one is a distribution system of Taipower. Finally, the results obtained by the proposed method are shown and compared with other methods
一個基於堆疊式降噪自動編碼器的光體積變化描記圖專注辨識系統;A Photoplethysmography Attention Recognition System Based on Stacked Denoising Autoencoders
No full text[[abstract]]本研究提出一個利用光體積變化描記圖(Photoplethysmography, PPG)作為數據並使用堆疊式降噪自動編碼器(Stacked Denoising Autoencoders, SDAE)分類的專注辨識系統。本研究中我們自行設計專注誘發實驗並同時量測數據，在專注辨識系統的研究內容方面主要有兩大方向，分別為傳統機器學習(Traditional Machine Learning, TML)及深度學習(Deep Learning, DL)。在傳統機器學習實驗架構過程可分為訊號前處理、特徵擷取、特徵選取和分類，訊號前處理為濾除雜訊和偵測重要點；特徵擷取的特徵主要有四大類：心率變異(Heart Rate Variability, HRV)、時域、波形和個體差異，總共224個；特徵選取的方式有三種：P值法(P-value)、基因演算法(Genetic Algorithm, GA)和改良式基因演算法(Modified Genetic Algorithm, MGA)，用以降低架構複雜度及提升辨識正確率；分類使用倒傳遞類神經網路(Back Propagation Neural Network, BPNN)。在深度學習實驗架構過程可分為訊號前處理和分類，訊號前處理為濾除雜訊、波形擷取、頻譜處理和個體差異處理，總共12種波形；分類使用堆疊式降噪編碼器，堆疊方式分為類別堆疊(Stacked Categories)和編碼堆疊(Stacked Encoding)兩層次。結果顯示，在傳統機器學習方面，以倒傳遞類神經網路搭配基因演算法於留一人測試交叉驗證分類辨識正確率為92.23 %，特徵數為96；搭配改良式基因演算法於留一人測試交叉驗證分類辨識正確率為86.25 %，特徵數為20。在深度學習方面，以調整過的全階層堆疊式降噪自動編碼器於留一人測試交叉驗證辨識正確率為84.25%。
This study proposes an attention recognition system using photoplethysmography(PPG) as data and using stacked denoising autoencoders(SDAE) for classification.In the study, we design experiment, to induce attention and measure data. The main development of attention recognition system has two major directions, namely, traditional machine learning(TML) and deep learning(DL). The TML experiment architecture processes can be divided into signal pre-processing, feature extraction, feature selection, and classification. Signal pre-processing is to filter out noise and detect important points. There are four main types of features to be extracted, including heart rate variability(HRV), time domain, waveform, and individual differences, resulting in a total of 224 features. There are three methods of feature selection, namely, P-value method, genetic algorithm(GA), and modified genetic algorithm(MGA) to reduce the complexity of the architecture and improve the accuracy of recognition. The classification uses back propagation neural network(BPNN) as classifier. The DL experimental architecture processes can be divided into signal pre-processing and classification. Signal pre-processing is to filter noise, extract waveform, have spectrum processing and have individual differences processing. A total of 12 kinds of waveforms can be extracted. The SDAE is used as classifier. The stacked method can be further divided into stacked categories and stacked encoding.The results show that with TML, the recognition accuracy using BPNN and GA is 92.23% and the number of features is 96 by using leave one person out cross-validation. The recognition accuracy using BPNN and MGA is 86.25% and the number of features is 20 using in leave one person out cross-validation. With DL, the recognition accuracy is 84.25% using adjusted all hierarchy SDAE with leave one person out cross-validation
靜態雜訊容忍度基準之臨界電壓靜態隨機存取記憶體細胞元與陣列設計;SNM-Aware Subthreshold SRAM Cell and Array Designs
No full text[[abstract]]近年來隨著可攜式電子產品的發展，各式降低功率消耗的硬體設計不斷推陳出新，而常被用於當嵌入式記憶體的靜態隨機存取記憶體(Static Random Access Memory, SRAM)，其設計也就極為重要。本論文將從細胞元架構的角度，了解靜態隨機存取記憶體在次臨界電壓操作的重要性，接著舉出現今次臨界電壓下的設計議題並一一探討，目的為設計出低功耗且面積小的四十奈米製程下的次臨界電壓靜態隨機存取記憶。首先將會敘述在四十奈米製程下傳統6T靜態隨機存取記憶的次臨界電壓設計考驗，接著選出兩篇作品符合次臨界電壓下的所有設計議題，藉由重新改良後的一套記憶體細胞元設計方法，來決定最佳的細胞元尺寸，最後將以較具優勢的一方，在四十奈米下進行完整的記憶體電路設計並於最後提出改善，而未被實做的另一方作品，則將引用文獻之數據，來比較各個次臨界電壓靜態隨機存取記憶體的優劣，在針對其優劣做出評估與分析。
In recent years, with the development of portable electronic products, all kinds of hardware are designed for low-power consumption. SRAM, it’s very important part in SOC.This paper will introduce from the point of view of cell structure to realize the importance of SRAM in sub-threshold voltage region. And then, for low-power consumption and small-area designs in 40nm process, we will discuss all the design issues about sub-threshold voltage design.First, we will list the problem of conventional 6T SRAM about sub-threshold voltage design in 40nm process, Then select the two works to meet all the sub-threshold voltage design issues. To determine the best cell size by improvement of cell design flow and the more advantageous one will be implement the complete memory circuit and improve in 40nm process. To compare advantages of each work , the another work will reference the data of the paper. Finally, we will analyze the data fully and make a conclusion
類神經網路式案例適應方法之研究;A Study of ANN-Based Case Adaptation in a CBR System
No full text[[abstract]]案例式推理 (Cased-Based Reasoning, CBR) 源自於各個領域的專家，在處理相似性高的日常工作內容時，習慣仰賴既有的經驗來解決問題。其中案例適應功能的實現需要考慮到多個面向，包含系統本身對於案例背後的實質意義之理解。目前已經有多個致力於實現案例式推理的軟體，但案例適應之效果普遍不佳。如此一來，更加突顯了案例適應的重要性與難度。而本研究方法旨在遵循CBR 框架的同時，加入案例差異訣竅的概念並透過類神經網路之優勢來完成案例適應功能。系統會依據目標問題之特徵進行案例搜索，尋找最相似案例以及其解答。此一搜索到的解答可被視?新解答之基礎，並且隨著案例適應階段的預測結果稍作修改。?了證實本研究之方法是否有效，在實驗階段所採用 的資料庫來源?加州大學爾灣分校(University of California, Irvine)之開放資料庫，內容?大量風洞實驗所收集到的NACA 0012翼型之噪音資料。最後，本論文對於實驗結果所採用的檢驗標準?平均絕對誤差以及學生t檢定(Student’s t-Test)，用於對比本論文實驗之預測方法與真實解答之相似程度。以統計結果而言，對於執行了案例適應後解答沒有改善的假設是不成立的。故本研究所提出之案例適應方法，其效果能以此?根據。
Case-based reasoning method, CBR in short, is an inspiration from experts of various domains, who prefer to rely on their experience when dealing with solving similar problems. However, experience may not be exactly the same as the target problem that they are facing. To make use of experience, case adaptation is necessary. There are several issues that have to be considered when implementing case adaptation in a CBR system, including the comprehension to each case. There also exist various research working on CBR, but applying limit cases in a case base to the target problem is still a difficult task. Thus, the importance and difficulty of case adaptation are obvious. The goal of this study is to design and implement case adaptation, with advantages of artificial neural networks and the concept of case difference heuristic. The system retrieves the most similar case depending on attributes of the target case. The solution part of the retrieved case will then be refined after case adaptation.The dataset used in our experiment is a noise spectrum database collected from a family of NACA 0012 airfoils. It was obtained from a series of aerodynamic and acoustic tests of airfoil blade sections conducted in a wind tunnel.To show the effectiveness of our method, Student’s t-Test are used to determine that the CBR system with case adaptation is significantly better than the CBR system without case adaptation. As the result, the hypothesis of non-improvement after case adaptation is rejected. This implies that case adaptation designed in this study is effective
電壓驟降指標視覺化實現;Implementation of Voltage Sag Indexes Visualization
No full text[[abstract]]隨著科技的發展，高科技產業倚重精密設備，電壓驟降的發生會使得精密設備故障，造成龐大的經濟損失，有必要了解電網中各地的電壓驟降情形，除了計算電壓驟降指標，電壓驟降指標視覺化的實現，能改善大量指標數據不易於使人理解實際電網中發生情形的缺陷。本論文使用IEEE Std 1564-2014 所建議的電壓驟降計算方法，對台灣電網中量測到的所有異常電壓事件進行分析，篩選符合驟降定義的異常電壓事件，計算每一筆單一驟降事件的特徵化，並以一年為單位，計算各個測站的電壓驟降位置指標，再進行視覺化的呈現。本研究使用D3.js為電壓驟降指標視覺化的工具，分別嵌入微軟所開發的線上地圖Bing Map以及自行繪製的台灣電力系統高壓線路地圖，利用測站的地理空間資訊，將電壓驟降指標視覺化呈現於真實地圖上，實現台灣電壓驟降指標的視覺化。
n recent years, within the different possible disturbances, the voltage sags have the highest frequency of occurrence. Additionally, due to a high sensitivity of electronic equipment to this disturbance, high economic losses have been generated. For this reason, voltage sags are considered as having greater impact nowadays. It is necessary to understand the real situation of voltage sag event in power system. Except the voltage sag calculation, the Implementation of voltage sag visualization is needful. It improves the problem that voltage sag isn’t easy to understand. In thesis, using the method provided by IEEE Std 1564 calculate the voltage sag index. And visualized the voltage sag index data. We used D3.js as a tool of visualization. Combine the real map and voltage sag geography information and visualize the voltage sag in Taiwan
基於光流及深度學習的超車偵測技術;Overtaking Vehicle Detection Techniques Based on Optical Flow and Deep Learning
No full text[[abstract]]近年來，智慧型車輛的興起，讓車用影像成為熱門的研究項目，本研究提出一個利用安裝於車輛正後方的行車紀錄器偵測超車並協助駕駛變換車道的即時系統。在這項工作中，我們提出基於運動線索定位出可能的目標物。接著，用卷積類神經網路(Convolutional Neural Network)辨識車輛，並用一段時間的追蹤判斷車輛的行為，確認為超車之後才對駕駛發出警告。此外，我們針對運動線索和卷積類神經網路同時在重複物件問題上發生的錯誤提出一個可靠的算法，我們實驗於白天的市區、高速公路以及夜晚的場景中，總計共有180,000個幀，每一張幀所需時間約為10ms，這證實此系統能良好的使用於真實的道路上。
In recent years, the rise of intelligent vehicles makes the car images become popular research issues. This thesis proposes a real-time system that uses a monocular camera mounted on the rear of a vehicle to detect overtaking. In this work, we propose to locate vehicles based on motion cues, then identify the vehicle with the Convolutional Neural Network(CNN) and track the behavior of the vehicle for a period of time. In addition, we propose a reliable algorithm for the issue of repetitive patterns. Experimental results are presented with real scene video sequences. The performance evaluation has demonstrated the effectiveness of the proposed techniques
微振動體振動率與位移量之檢測雷達設計;Design of a Micro-Vibration Rate And Displacement Detection Radar
No full text[[abstract]]本論文運用駐波波包解調雷達技術，設計一高靈敏度非接觸式微波微振動檢測器。本檢測器整合駐波波包解調雷達、微控制器、無線藍芽傳輸模組、與手機App軟體，配備有LCD模組、Wi-Fi模組、GPS模組等擴充孔，成為一完整之感測與無線回報功能之感測器。其中核心技術是微振動訊號處理演算法:(1)電路校正演算法，用於消除雷達電路中耦合器以及相移器所造成之溢漏信號。(2)自動相位調整演算法，透過調控相移器量測空間中之駐波波型，並改善量測靈敏度和正確度。(3)相對位移估算演算法，透過訊號強度真值表之建立用來估算待測物相對位移。此系統尺寸為53×41 mm2，約2/3張健保卡大小，在基本配備使用時，功率消耗最大值為450 mW。在12×8 cm2金屬平板的10 m、1.50 Hz振動實驗中，偵測誤差值為0.01 Hz (0.6%)。另外，針對穿戴式手腕脈搏量測的應用，此系統的量測結果與血壓計和手按計數進行比對，方均根誤差為0.8-1.3 BPM。
In this thesis, a high-sensitive non-contact microwave micro-vibration detector was designed, based on standing-wave envelope demodulation technique. This highly-integrated detector includes a standing-wave envelope demodulation radar, microprocessor, bluetooth, and mobile-phone App software. This detector is also incorporated with LCD display, Wi-Fi and GPS modules for further application extension. The core signal processing in this detector are: first, system calibration algorithm for elimination of hardware RF impairments; Second, automatic phase shifting algorithm for maximal detection accuracy and sensitivity; and Third, the relative displacement computation for vibration displacement estimation. The size of demonstrated detector is 5.3×4.1 cm2 and the dc power consumption is 450 mW. The experimental results show that for a 1.50 Hz, 10 m vibration metal plate of 12×8 cm2, the detection error rate is less than 0.01 Hz (0.6%). Moreover, for human wrist pulse non-contact wearable measurement, the root-mean-squared error is 0.8-1.3 BPM, compared with hand count and with commercial blood-pressure and pulse meter
應用T-S模糊控制之永磁同步發電機驅動器研製;Design and Implementation of Permanent-Magnet Synchronous Generator Drive Using T-S Fuzzy Control
No full text[[abstract]]本論文旨在應用T-S模糊控制法於永磁同步發電機之驅動器研製。T-S模糊控制法常應用於非線性系統之控制，其中使用解模糊化過程、平行分佈補償(PDC)設計、Lyapunov穩定度分析，並採用線性矩陣不等式(LMI)得到控制增益值。永磁同步發電機驅動系統等效電路藉由兩相調變(TPM)推導以減少切換損失。本論文首先敘述系統之架構以及發電機各區間之狀態方程式推導，接著推導T-S模糊控制法及LMI型式。本系統將永磁同步發電機之三相交流電轉換為直流電供應至負載端，經由類比/數位轉換回授直流鏈端電壓以及三相電流至微控制器，代入控制器以求得開關之責任週期比。最後以實驗結果驗證控制器性能。
The main purpose of this thesis is to design and implement the permanent-magnet synchronous generator (PMSG) drive using T-S fuzzy control. The T-S fuzzy control is generally applied to the control of nonlinear systems. The defuzzification, parallel distributed compensation (PDC) and Lyapunov stability analysis are implemented in the design of T-S fuzzy control. The Linear Matrix Inequality (LMI) is adopted to obtain the controller gais. In order to reduce switching losses, the equivalent circuit of the PMSG drive system is derived from two phase modulation (TPM). First, the system configuration and state equation of each region of the PMSG are introduced in this thesis. Next, the T-S fuzzy control and the LMI are derived. The three-phase AC power of PMSG is converted to DC power for loads. The DC link voltage and three-phase current are sensed to the microcontroller through analog-to-digital convertion (ADC). The duty ratios of the power switches are calculated via the controller. Finally, the controller performance is verified by experimental results