Chung Cheng University Institutional Repository
Not a member yet
889 research outputs found
Sort by
探討勞工運用健康照護資訊平台進行過勞預防之使用意圖;A Study on Labors' Intention to use in using Healthcare Information Platform for the Burnout Prevention
No full text[[abstract]]隨著媒體的報導，職場過勞問題逐漸受到社會大眾的重視。為了加強勞工對自己身體健康之重視，我國政府設計出一套過勞風險評估及個人健康管理系統，幫助勞工掌握自己的身體健康狀況。因此，本研究期望透過保護動機理論，結合整合性科技接受模式理論為基礎，並加入信任及隱私顧慮，探討哪些因素才是影響勞工使用健康照護資訊平台的意圖。本研究以網路問卷發放，以曾有使用健康照護資訊平台的勞工為調查對象，總共蒐集210份有效問卷，並以統計軟體 SPSS 20.0 和 Smart PLS 3.2 進行資料分析，整體研究架構的解釋力為 66.5%。研究結果顯示，勞工對於健康行為、科技使用、知覺易感性、績效期望、努力期望、社會影響、信任皆會正向影響使用者使用意圖。研究發現，社會影響透過勞工健康照護資訊平台影響使用意圖最高。
Through the media reports, burnout problem is valued gradually by more people. In order to strengthen labor's attention to their own health, Taiwan’s government has designed risk assessment for burnout and personal health management system, helping labors control their health.Therefore, this study expects the Protection motivation theory (PMT) combine with the Unified Theory of Acceptance and Use of Technology (UTAUT), and join the trust and privacy concern, to investigate which factors will impact willingness to use the healthcare information platform.This study used the Internat questionnaire, choseing the labor for the sample who used the healthcare information platform. Used SPSS 20.0 and Smart PLS 3.2 statistical software to data analysis. The study result is that Healthy Behavior (HB), Technology Use (TU), Perceived Vulnerability (PV), Performance Expectancy (PE), Effort Expectancy (EE), Social Influence (SI), Trust (T) positive impact Behavior Intention (BI). This study found that the main factor impact on the labor use in healthcare information platform is Social Influence
使用深度Q網路來學習玩超級瑪利歐兄弟;Learning to play Super Mario Bros using Deep Q-Networks
No full text[[abstract]]超級瑪利歐兄弟是任天堂於1985年出品的著名橫向卷軸動作闖關遊戲，而The Mario AI Benchmark提供了我們一個模擬超級瑪利歐兄弟遊戲的平台，在這平台上可以透過產生隨機的關卡來提供人們評估Agent的優劣。由於目前的遊戲控制Agent幾乎是透過遊戲引擎提供的狀態作為輸入，而不是透過遊戲畫面中的原始影像作為決策的判斷因素，因此我們使用Deep Q-Networks作為Q-learning中計算Q值的值函數近似。我們可以透過遊戲影像中的高維狀態輸入經過Deep Q-Networks得到低維的動作輸出，根據環境所給予的獎勵來更新網路。在我們的實驗結果中說明了有無Experience Replay的差異，以及隨機關卡的訓練資料有助於提升agent對關卡的適應程度。最後我們使用了Double Deep Q-Networks來改良agent，最後也在限定的條件下得到不錯的表現。
Super Mario Bros is the famous side-scrolling video game produced by Nintendo in 1985, and the Mario AI Benchmark provides us with a benchmark software of Super Mario Bros game. The benchmark can generate random levels to let people estimating their agents’ performance.As the current game control agent is almost through the state provided by the game engine as input rather than through the original image of screenshots as a decision-making factor, we use Deep Q-Networks as Q-learning value function approximation. We can use the high-dimensional states as inputs in the game to get low-dimensional outputs for possible actions through Deep Q-Networks, and update the network according to the reward given by the environment. In our experimental results, we showed the difference of whether the agent used experience replay, and that the training data of the random levels help to improve the adaptability of the agent. Finally, we used Double Deep Q-Networks to improve the agent, and finally got good performance under the limited condition
藉由擴增休眠檔縮短開機後反應時間的快速開機技術;Fast Booting Technique for Reducing After Booted Response Time by Extended Hibernation File
No full text[[abstract]]現今大多數的智慧裝置搭載的功能愈來愈強大，設計愈來愈複雜，開機所需預載的服務也進而更多，導致整個開機時間變長，而使用者想使用產品，則是希望能馬上使用，所以有了快速開機的需求，利用休眠式快速開機(hibernate)，可以優化系統開機時間，讓系統開機時間減少85%。休眠式快速開機在進入休眠之前，會把應用程式記憶體大量的swap out到swap space或是secondary storage，使得製作出來的休眠檔的大小非常小，讓系統能在很短的時間內讀取完休眠檔，讓系統回復到休眠前的狀態，優化了系統的開機速度，但是因為休眠前大量的swap out應用程式的資料，導致系統回復以後，應用程式需要大量從swap space中換進所需的page而導致應用程式反應的時間變長，進而影響到使用者體驗。本論文基於羅習五等人的SBH快速開機技術提出一個方法，來優化應用程式在快速開機後的反應時間，系統休眠回來後，標記因為應用程式從swap space中swap in進來所需要的page，在休眠的時候保留那些被標記的page來擴增休眠檔大小，讓系統在回復以後減少從swap space中換進來頁面的數量，進而優化應用程式反應的時
支援ARM資料處理指令之SIMD技術為基礎的軟體實現錯誤容忍之研究;SIMD-Based Software Fault Tolerance for ARM Data-Processing Instructions
No full text[[abstract]]指令層級的軟體容錯技術是一種比使用硬體實作更具有低花費且高彈性的技術，此技術是當瞬時錯誤發生時，透過重複執行指令來複製至少三份的副本以進行錯誤的偵測與修復，然而重複執行相同指令卻也是造成執行效能下降的主要原因之一。現今大部分的嵌入式系統和手持移動裝置都運行在ARM處理器架構上，然而目前很少研究是針對ARM處理器的容錯技術。在這篇論文中，我們針對ARM處理器的資料處理指令，開發以SIMD為基礎的軟體容錯技術，其核心技術在於透過SIMD指令優化重複執行指令的效能，以提升程式的可靠度。最後，我們透過LLVM編譯器來實作，並能夠自動地產生具有容錯能力的程式。實驗結果顯示，我們所提出的方法可以正確的產生容錯程式，而執行效能比原始版本下降4%-70%，程式碼大小比原始版本大上1.1-2.5倍。
Software-based instruction-level redundant execution is a cost-efficient and flexible technique to tolerate transient faults compared to use hardware implementation. It duplicates at least three copies of data to recover errors when system failure. However, redundancy execution is the main reason for performance degradation. The ARM architecture is widely used in nowadays embedded systems and mobile devices. However, few studies focus on instruction-level redundant execution of ARM processor. In this thesis, we develop a SIMD-based software fault tolerance technique for ARM data-processing instructions. The core technique employs instruction-level redundant execution with SIMD optimization to tolerate single bit-flipping errors. The proposed approach is implemented in the LLVM compiler to generate protected code automatically. For the tested benchmark programs, preliminary experiments show that the proposed approach can correctly generate fault-tolerant codes at costs of 4%–70% performance degradation on average compared with the original versions. The fault-tolerant versions of code are 1.1–2.5 times larger than the original versions
叢集生成樹之演算法問題;On the clustered spanning tree problem
No full text[[abstract]]在很多網路拓樸的應用中，終端機常會因為安全或效率的關係而分成很多群，因此在同一群的終端機可以在群內選擇路徑。在這個情況下，我們希望找到一顆生成樹，使得群內的每一個點都叢集在一起。在這篇論文中，我們研究兩個叢集生成樹問題，它們的點都分成數群，第一個是最小路由叢集生成樹問題，另一個是最小史坦納叢集生成樹問題。在本論文的第一個部分，我們研究最小路由叢集生成樹問題。給定一個點分成k 群的輸入圖，我們定義一個叢集生成樹的子樹彼此都不相交。我們在最小路由叢集生成樹這個問題證明了不可近似的結果。在輸入圖為metric graph時，我們提出了一個兩倍的近似演算法。我們也研究它的延伸問題，當目標為找不同群間的路徑時，我們證明了在輸入圖分成兩群時可以多項式函數時間求得解，當分成三群時則是NP-hard。當分成三群時我們提出了一個兩倍的近似演算法。在本論文的第二個部分，我們研究當輸入圖為metric graph 的最小史坦納叢集生成樹問題，它是最小史坦納樹的延伸問題。我們證明了在叢集生成樹的條件下，給定了群與群之間的拓樸和群內的拓樸後，這個問題仍然是NP-hard。我們證明了叢集生成樹的史塔納樹ratio 是介於三到四之間。我們也提出了一個(ρ+2)倍的近似演算法，ρ是當前最好的史塔納樹近似演算法。當給定了群內的拓樸後，可以找到(ρ+1)倍的近似演算法。我們也研究了其他的延伸問題，一個是只找群與群之間的史坦納叢集生成樹，另一個是在已知群與群之間的最小史塔納叢集生成樹，找群內的最小史塔納叢集生成樹。
In many network applications, terminals may be grouped into clusterssuch that the communications between terminals of the same cluster shouldbe routed \locally" for the sake of efficiency and safety. In such case, welook for a spanning tree in which vertices of each cluster should be clusteredtogether. In this thesis, we study two different kinds of clustered spanningtree problem where the set of vertices is partitioned into clusters: one is the minimum routing cost clustered tree problem, the other one is the clustered Steiner tree problem.In the rst part of this thesis, we investigate the minimum routing costclustered tree problem. For an edge-weighted graph G = (V,E,w), in whichthe vertices are partitioned into k clusters, a spanningtree is a clustered spanning tree if the subtrees spanning the clusters are mutually disjoint. We show the inapproximability of finding a clustered spanning tree with minimum routing cost, where the routing cost is the total distance summed over all pairs of vertices. We present a 2-approximation for the case that the input is a metric graph. We also study a variant in whichthe objective function is the total distance summed over all pairs of verticesof different clusters. We show that the problem is polynomial-time solvablewhen the number of clusters k is 2 and NP-hard for k = 3. We propose apolynomial-time 2-approximation algorithm for the case of 3 clusters.In the second part of this thesis, we study the Clustered Steiner tree prob-lem on metric graphs, which is a variant of Steiner minimum tree problem. Inthis problem, the required vertices are partitioned into clusters, and the sub-trees spanning different clusters must be disjoint in a feasible clustered Steiner tree. We show that the problem is NP-hard even if the local-topologies and the inter-cluster tree are given. We show that the Steiner ratio of this problem is lower and upper bounded by three and four, respectively. We also propose a (ρ+2)-approximation algorithm, where is the approximation ratio for the Steiner minimum tree problem, and the approximation ratio can be improvedto ρ+ 1 if the local topologies are given. Two variants of this problem arealso studied. When the goal is to minimize the inter-cluster cost and ignorethe cost of local trees, the problem can be solved in polynomial time. But itis NP-hard if we ask for the minimum cost of local trees among all solutionswith minimum inter-cluster cost
以 FM-index 為基礎之第三代定序自我型錯誤修正法;A self-error correction algorithm for third-generation sequencing using FM-index
No full text[[abstract]]因為第三代定序技術所產生出的序列為較長的序列，定序的偏差也較低還有定序分布平均等特質，使得第三代定序技術成為現有基因組裝(de novo assembly)的受歡迎選項。 但是由於它所產出的序列錯誤率較高，所以在進行基因組裝前都先必須進行序列的錯誤修正。目前錯誤修正的方法可以分為比對序列分析法和非比對序列分析法。比對序列分析法比較費時但可以在高相似度和低覆蓋率的區域修正。另一方面，分比對錯誤修正法比較快速但敏感度較低。在這篇論文裡，我們研發出一個新的非比對錯誤修正法，藉由FM-index試著把錯誤修正問題轉化成路徑搜尋問題。為了能夠在高相似度和低覆蓋率的區域進行錯誤修正，研發出了使用多種長度子字串的可適性種子搜尋演算法。最後實驗結果指出我們的方法比現有的比對序列分析法和非比對序列分析法還要快在大腸桿菌跟酵母菌之下。在大物種線蟲我們的方法比現有的比對序列分析法還要慢但還是比現有的非比對序列分析法還要快速。
The 3rd-generation sequencing technologies are becoming the popular choice in de novo assembly projects, because of long reads, less sequencing bias, and more uniform coverage. But it comes at the cost of much higher error rates and thus error correction is often performed prior to assembly. Currently, error correction methods can be divided into alignment-based and alignment-free approaches. Alignment-based methods are more time-consuming but able to correct reads in repetitive and low-coverage regions. On the other hand, alignment-free methods are much faster but have less sensitivity. In this thesis, we develop a novel alignment-free algorithm which reduces the correction problem to a path-searching problem via FM-index extension. In order to correct reads in low-coverage and repetitive regions, an adaptive seeding algorithm using multiple sizes of k-mers is developed. The experimental results indicated that our method is faster than existing alignment-based and alignment-free methods in E. coli and S. cerevisiae datasets. For large genome datasets, our method is slower than alignment-based methods but still faster than existing alignment-free method
基於廣泛即時數位交通資訊地圖之低訊息量車輛路徑規劃方法;Message-Efficient Route Planning Based on Comprehensive Real-time Traffic Map for Infrastructure-less VANETs
No full text[[abstract]]交通壅塞是許多地區常見的問題，特別是在市區的尖峰時段中，可能會大量增加駕駛在通勤上所花費的時間，甚至增加油耗量以及廢氣的排放量。因此，如何有效解決道路壅塞是目前很重要的議題。 而近年來有許多使用車載隨意網路(Vehicular Ad Hoc Netwoks, VANETs)技術來解決道路壅塞的研究被提出。在這些研究中，主要是利用車輛間能夠互相通訊的能力，藉由車輛來收集各路段交通資訊，並與其他車輛進行資訊交換。而為了解決道路壅塞，車輛可藉由收集到的即時交通資訊來選擇能夠避開塞車區域且較快到目的地的路徑。而在這些方法中，為了加快取得路段資訊，車輛會透過主動發送訊息的方式向其他車輛要求資訊，雖然能加快取得交通資訊，但會相對增加大量封包交換成本(packet overhead)。 因此在本論文中，我們提出使用被動式結合主動式的資訊收集方式，使車輛能夠以較少封包成本來取得廣泛且即時的交通資訊，並且將車輛收集的即時交通資訊應用在我們提出的路徑規劃方法上。車輛會將收集到的交通資訊紀錄於每台車內的數位地圖中，並於路徑選擇時根據數位地圖所記錄的各路段數據，考量各路段預計行駛車輛數、目前所需旅行時間以及交通資訊可靠度等因素，為車輛建議可靠且能快速到達目的路段的路徑。 在模擬實驗中，與先前的研究相比，結果顯示我們所提出的方法能夠以更少的封包成本來收集交通資訊並能夠為車輛找出更快到達目的地的行車路徑。
Traffic congestion is a critical problem in many countries, especially in the urban area, which may greatly increase the commuting time and even increase fuel consumption. Therefore, how to alleviate the road congestion problem effectively is an important issue. In recent years, many studies using vehicular ad hoc networks to solve the traffic congestion problem have been proposed. Vehicle utilizes the communication ability to collect traffic information from other vehicles. To alleviate the traffic jam, vehicle can use the collected real-time traffic information to select a driving path, which can avoid the congestion area. In these studies, to speed up the traffic information collecting time, vehicles will proactively send the request message to acquire traffic information from other vehicles. However, the proactive request scheme will increase the extra packet overhead. Therefore, we propose a real-time traffic collection method, which combines with proactive and passive scheme. Vehicles passively collect traffic data and proactively broadcast the traffic data in three special situation to speed up spreading traffic information. Thus, vehicles can obtain a real-time traffic information with lower packet overhead and record the information into the digital map in each vehicle. Then, we consider the travel time, reliability, expected traffic flow and other factors of each road based on the digital map to select a reliable and fast route. The simulation results show that, compared with previous study, our method can collect traffic information with lower packet overhead and find a faster route to the destination
智慧交通系統中用以舒緩交通壅塞的交通燈號連鎖排程策略;Traffic Signal Chain Scheduling to Relieve Traffic Jam in Intelligent Transportation System
No full text[[abstract]]隨著科技日新月異，大量生產製造技術越來越成熟，許多已開發國家開始大力發展汽車工業，而私人汽車的數量也越來越多。汽車對於兩地的往返雖然方便，但過多的車輛會導致城市中道路漸漸無法負擔這些車輛，進而發生交通堵塞現象，每逢尖峰時刻，車流將從四面八方大量湧入城市中心，進而使交通堵塞現象愈趨嚴重。而過去其中一種應對交通壅塞方式大多是在道路交叉處增設交通號誌，但目前交通號誌採用的是固定週期方式，每個紅綠燈會被分配到固定的秒數，此種方式雖然可以緩解塞車情況，但是無法根據當前交通情況進行應對。所以當某些路段有過多車輛，或是某些路段車輛數較小，固定週期的號誌仍會照著既有週期去開放綠燈，容易造成車輛多的路段無法排除大量車輛，而車輛數小的路段會有浪費綠燈秒數的現象發生。在本論文中，我們將提出解決交通壅塞的機制來有效解決交通壅塞問題。首先，我們會將路段進行區分，分成壅塞路段和非壅塞路段。在壅塞路段中，我們會將壅塞路段進行分組，讓鄰近的壅塞路段形成一個壅塞區域，對於這些壅塞區域，我們將內部的交通號誌透過互相搭配，讓促使交通號誌形成連續綠燈，將當下發生的壅塞狀況的路段內的車輛能夠有效排除壅塞區域。而那些非壅塞路段，我們則使用一般的動態交通號誌，在考量因素中，加入當下車道密度、鄰近車道密度、車輛等待時間來動態調整信號燈，避免讓高密度路段的車潮先行進入高密度路段。在我們的實驗模擬中，與ITLC[22]和固定式週期紅綠燈進行車輛總旅程時間、車輛等待時間、車輛平均速度和道路車輛密度變異數比較，結果顯示我們所提出的方法雖然會讓整體的車輛等待時間上升，但是我們的方法的車輛總旅程時間卻可以比較低，而從車輛平均速度指標可以看出，我們的方法可以讓車輛維持較高速行進，避免在壅塞道路有走走停停現象，而道路密度變異數則可以顯示出在我們的方法中，車輛都是較平均分散在道路上，較不會造成有些車道有過多車輛，而有些車道車輛過少。關鍵字—車聯網; 車輛對車輛通訊; 車輛對基礎設施通訊; 基礎設施對基礎設施通訊; 智慧交通系統; 智慧燈號連鎖政策;
For the rapid number increasing of the vehicles, traffic congestion has become a serious problem to people’s life. Traffic congestion may leads some problems e.g., travel delay, fuel consumption, air pollution. Traditional traffic signal based on fixed cycles, which has fixed period schedule, are used to smooth the traffic problem.However, traditional traffic signal can not efficiently deal with the congestion problem. In this paper, we propose method to smooth the traffic congestion problem. In our method, we distinguish the road between congestion or non-congestion. In congestion road, we combine the road with their neighbor road to form a congestion area .First, we configure those traffic signal in the congestion area to become traffic signal chain. The vehicles in congestion area can drive in the traffic signal chain to leave the congestion area. In the road which is distinguished into non- congestion, we consider the road density of current road and neighbor road to schedule to traffic signal to smooth the vehicle in the non-congestion area.The simulation results show that our method can reduce the travel time of vehicles as they travel in the congestion area, as compared with the fixed signal and ITLC[22]. Our method also perform well in the metric of road density variance, road density variance show that our method can spread the vehicle evenly than other methods. Keywords—VANET; Inter-Vehicle Communication; Vehicle-to-Roadside-Unit Communication; Inter-Roadside-Unit Communication; Intelligent Transportation System; Traffic Signal Chain
以空間、時間、深度、轉換、及時空特徵作無參考雙眼立體視訊品質評估;No-reference Stereoscopic Video Quality Assessment Using Spatial, Temporal, Depth, Transform, and Spatiotemporal Features
No full text[[abstract]]近年來，3D應用技術越來越普遍，因此人類對於雙眼立體視訊的品質越來越重視，換句話說雙眼立體視訊的品質評估方法將會被廣泛的使用，其中無參考的雙眼立體視訊的評估技術為最實用且有效的方式，因此本研究著重於無參考的技術，首先在雙眼立體視訊的五個不同領域中擷取能反映雙眼立體視訊品質的特徵，其中在空間域上擷取模糊程度、區塊程度、局部二值模式及邊緣的資訊，在時間域上擷取亮度、DC值及視差變化的資訊，在轉換域上擷取離散餘弦轉換及小波轉換中的特徵，在時空域上擷取方向梯度直方圖及三維離散餘弦的特徵，接著以直方圖量化的方式來統計每一種特徵，並對左右視角的特徵取平均，接著將所有特徵正規化到相同的分布，再利用特徵選取來刪除不必要的特徵以提升效能，最後透過支持向量回歸的技術來預估最後的雙眼立體視訊品質分數，由實驗結果可以表示，本研究所提出的方法在NAMA3DS1_COSPAD1資料庫中較優於另一個無參考的方法，並且可媲美其它三種完全參考及部分參考的方法。
Recently, 3D technology application is more and more widespread. Thus, humans pay more attention on stereoscopic video quality. In other words, the stereoscopic video quality assessment approaches will be widely used. No-reference stereoscopic video quality assessment technology is the most useful and effective way. Hence, no-reference stereoscopic video quality assessment technology is mainly focused in this study. First, five domain features including spatial, temporal, depth, transform, and spatiotemporal features are extracted. On the spatial domain, blurriness, blockiness, local binary pattern (LBP), and edge information are extracted. On the temporal domain, variation information of luminance, DC values, and disparity are extracted. On the depth domain, disparity and depth motion are extracted. On the transform domain, discrete wavelet transform (DWT) and discrete cosine transform (DCT) information are extracted. On the spatiotemporal domain, histogram of gradient (HOG) and 3D-DCT information are extracted. Each feature vector is obtained by using histogram statistics and normalization to the same distribution. Then, the feature vectors from the left-view and right-view videos are averaged. Here, feature selection is applied to strike out the unnecessary features to improve the performance. Here, support vector regression (SVR) is applied to estimate the stereoscopic video quality score. Finally, experimental results show that the proposed approach is better than the other NR approach on NAMA3DS1_COSPAD1 database and compares favorably with others FR and RR approaches
探討G20國家的油價變動率對股價報酬率的影響;The relationships between oil price change and stock return : evidence from G20 countries
No full text[[abstract]]本文探討WTI油價變動率對G20國家的股價報酬的影響，採用共整合檢定、向量自我迴歸模型（VAR）、Granger因果關係檢定以及門檻自我相關結構模型（TAR）等實證方法探討其關聯性，並且進行線性實證與非線性實證的對比。WTI油價與G20國家（除歐盟外）的股價指數資料選取從1990年1月1日至2016年12月31日的日資料，資料期間會因為數據庫的原因和個別國家的特殊情況而略有不同。實證結果指出：（1）在VAR模型檢定下，日本、澳大利亞、南非、俄羅斯、印度尼西亞和沙烏地阿拉伯的當期的股價變動率會受到前一期的WTI油價變動率的影響，並且均為正相關；當期的WTI油價油價變動率會受到加拿大前一期的股價變動率的影響，並且呈現負相關的關係；而土耳其的股價變動率會受到前一期和前二期的WTI油價變動率的影響，分別呈現正相關和負相關關係的WTI油價變動率的影響，而WTI油價變動率也會也會受到土耳其前一期及前兩期的股價變動率的影響，並且分別呈現正相關與負相關的關係。（2）在Granger因果檢定下，結果皆與VAR模型得到的結果基本吻合，即日本、澳大利亞、南非、俄羅斯、印度尼西亞和沙地阿拉伯的油價變動率對股價報酬率有單向的因果關係；美國和加拿大的股價報酬率分別對WTI油價變動率有單向的因果關係；而土耳其的股價報酬率對油價變動率有雙向的因果關係。（3）中國、印度、法國、德國、加拿大和墨西哥義大利的股價報酬率與油價變動率之間非線性的關係存在；而美國、韓國、土耳其和巴西的股價報酬率有非線性關係的存在，並且關係為負相關；日本、澳大利亞、南非、英國、印度尼西亞和沙烏地阿拉伯的股價報酬率與油價變動率存在正相關的非線性關係；阿根廷的股價報酬率與油價變動率不僅存在正相關的非線性關係，同時還存在負相關的非線性關係。（4）在進行線性模型和非線性模型的實證對比時，我們發現韓國、阿根廷、英國與巴西在VAR模型中與Granger因果檢定結果中，油價變動率與股價報酬率無相關關係或影響，但是在TAR模型中，WTI油價變動率對這些國家的股價報酬卻是有影響的。此外，在VAR模型中WTI油價變動率對俄羅斯股價報酬率有正相關關係，而TAR模型中俄羅斯股價報酬與油價變動率在三個體制下會有不同的正相關關係。而VAR模型中土耳其的股價變動率與WTI油價變動率的影響，分別呈現負相關的關係和正相關關係，但是在TAR模型中，卻只有WTI油價變動率對土耳其股價報酬率有負向的影響的結果。在全球油價與各國股市市場不斷波動的情況下，投資人應該持續關注其走勢，才能對市場作出更靈敏的決斷與預測。關鍵字：G20國家、門檻自我相關結構、向量自我迴歸、因果關係檢定、線性與非線