34254 research outputs found
Sort by
Impact Evaluation of Various Factors on Terrain-Aided Localization for Underwater Vehicles
No full textRecent advancements in key technologies have significantly improved the operational capabilities and application domains of autonomous underwater vehicles (AUVs). The operational environments for AUV missions are becoming increasingly complex, dynamic, and hazardous, trending towards long-duration and large-scale missions. To address these developments, AUVs must evolve to be more intelligent and autonomous, allowing them to adapt operational modes to varying environmental conditions. For example, in the context of long-duration and large-scale missions, AUVs need to acquire reliable underwater positioning information with minimum cost. Terrain-Aided Localization (TAL) proves to be a valuable approach in mitigating the substantial costs associated with deploying surface vessels or calibrating underwater baseline transponders. Additionally, TAL enhances the AUV's ability to navigate autonomously and maintain stealth during exploration and exploitation activities. Particle Filter (PF) is commonly employed in TAL for AUV position estimation, yet the performance of TAL is subject to various influencing factors. Current research on TAL performance primarily focuses on isolated exploration of one or two factors, lacking a comprehensive investigation into the broader range of factors impacting TAL performance. This study addresses this gap by examining the influence of seven factors on TAL position estimation performance, considering the employed PF algorithm, digital terrain model (DTM), and measurement instruments. These factors encompass particle elevation interpolation, DTM resolution, initial AUV position error, initial particle distribution range, particle filter process noise, measurement noise, and terrain elevation variation along the route. Tests are conducted using both real and synthetic DTM to discern differences in results between the two. Given the computational challenges of a full factorial experiment, this study employs the Taguchi method to estimate the impact of the seven factors on TAL position estimation performance through a partial factorial experiment. Simultaneously, the study deduces the optimal combination of factor levels for TAL. The research findings underscore that, whether utilizing real or synthetic DTM, crucial factors influencing AUV TAL estimation performance include particle elevation interpolation, initial AUV position error, and particle filter process noise
Identification of unreported molecules for aggregation-induced emission of glutathione-capped gold nanoclusters: polyethyleneimine and surfen
No full textGold nanoclusters are a novel type of fluorescent material. By using stabilizing agents such as thiol molecules, polymers, peptides, and proteins, it's possible to synthesize gold nanoclusters that exhibit fluorescence. In this study, thiolated gold nanoclusters were synthesized. They emit light at specific positions: 610 nm in visible light and 810 nm in near-infrared light. These nanoclusters possess the property of aggregation-induced emission (AIE) and are stabilized using glutathione, a molecule known as a tripeptide that's found in living organisms. These stabilized nanoclusters are referred to as glutathione-gold nanoclusters (GSH-AuNCs).
We utilized polymers, proteins, amino acids, metal ions, and molecules containing amino groups to promote the enhancement of aggregation-induced emission (AIEE) in GSH-AuNCs. From experimental results, it was observed that molecules with freely movable amino groups at their ends, such as polyethylenimine (PEI) and surfen, had a stronger effect on promoting AIEE when interacting with GSH-AuNCs. By using surfen and glutathione to generate electrostatic attraction, surfen-stabilized GSH-AuNCs (surfen@GSH-AuNCs) were formed.
Additionally, by taking advantage of the unique binding affinity between surfen and heparin, a polysaccharide found in the body, the prepared surfen@GSH-AuNCs were used in creating a fluorescent sensor for heparin detection. This sensor relied on the change in fluorescence intensity proportional to the variation in emitted light wavelengths between surfen (490 nm) and GSH-AuNCs (610 nm). The linear detection range spanned from 0.5 to 20 \uce\ubcM, with an R2=0.97. The calculated detection limit (S/N=3) was 0.51 \uce\ubcM
Machine Learning-Driven Strategies for Optimizing Next-Gen Vehicular Network Resources
No full textThis dissertation represents an integrated progression of research that systematically enhances the Internet of Vehicles (IoV) using advanced machine learning strategies to address commu- nication and computational challenges. The foundational work begins with the adaptation of deep reinforcement learning (DRL) for managing IoV resources and facilitating effective vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication. It introduces a novel priority-sensitive task offloading scheme using DRL algorithms to optimize resource allocation in a vehicular fog computing setup. Progressing this, multi-access edge computing (MAEC) and reconfigurable intelligent surfaces (RIS) are integrated into the IoV framework to enhance latency-sensitive applications like autonomous driving, employing a multi-agent DRL for improved resource allocation and reduced delays.
Advancing into the application of digital twins (DT) and unmanned aerial vehicles (UAVs), the dissertation explores the adoption of these technologies to enhance IoV network adaptability. A DT-informed IoV framework is outlined, leveraging a framework with two joint parallel processing DRL-based algorithms, termed RADiT, for resource allocation that dynamically responds to the quality of service (QoS) demands. This framework marks an evolution in vehicular networks, enabling smart decision-making under the pressures of high mobility and fluctuating connectivity.
Next, this framework is extended to incorporate decentralized learning by underscoring a dual-strategy approach combining a novel hybrid asynchronous federated learning (HAFL) method and a multi-agent DRL algorithm, MARS. This innovative confluence enhances the IoV's efficiency, equipping it to handle the unpred- ictable nature of vehicular networks while maintaining QoS across various conditions.
Expanding the scope, this research progresses beyond task offloading to a holistic optimization of network resources. It utilizes asynchronous federated learning (AFL) and a multi-agent DRL approach termed DMAAC algorithm, to improve network functionality, particularly focusing on the stringent latency and reliability standards required by the rigorous demands of ultra-reliable low-latency communications (URLLC) in vehicular networks. Simultaneously, a pioneering cooperative caching scheme, Co-Ca, is introduced to manage the caching resources efficiently, playing a crucial role in URLLC support. These strategies, when applied collectively, augment the robustness and operational efficacy of IoV networks, preparing the infrastructure to meet future smart transportation needs.
The final segment utilizes quantum computing within the vehicular metaverse for future-proofing the IoV network. This study introduces the quantum-based learning framework tailored for the vehicular metaverse termed QV-MetaFL framework with two sub-algorithms, the quantum sequential-training-program (Q-STP) and the quantum vehicle-context-grouping (Q-VCG), for cost optimization and data heterogeneity management, marking a transition to quantum federated learning (QFL). Extensive simulations for each of the frameworks demonstrate the superiority of all the proposed frameworks, underscoring the potential for the evolution of smart transportation systems
Application of Machine Learning in Real Estate Appraisal Models: A Case Study of Tainan City
No full textThe Financial Supervisory Commission (FSC) regulates the calculation methods for banks' own capital and risk-weighted assets. In cases of residential and commercial real estate exposure, the adoption of the "Loan-to-Value (LTV) ratio" is emphasized, highlighting the crucial importance of accurate real estate appraisal for credit risk management in banks. Through the application of machine learning, this study provides banks with real-time and effective property appraisals, enhancing risk management and enabling a more precise assessment of credit risk associated with real estate collateral.
This study utilizes real estate transaction data from Tainan, Taiwan, categorizing the data into four building types: residential buildings, apartment complexes, condominiums, and houses. Four different machine learning methods, including linear regression, LASSO, random forest, and support vector machine regression, are employed for predicting the price per square meter of each building type. The performance of these models is evaluated based on RMSE, MAE, and MAPE metrics. The results show that the random forest model performs exceptionally well across all building types, exhibiting the lowest RMSE and MAE, and relatively lower MAPE values. The performance of the Linear Regression and LASSO models is comparable, while the Support Vector Machine Regression model tends to perform less favorably in most building types.
The applicability of the models varies across different building types, with the random forest model standing out due to its non-linear modeling capabilities and feature selection advantages. It effectively handles the complex characteristics of diverse building types and provides detailed feature importance analysis. The recommendation is to prioritize the use of the random forest model in practical applications, especially for real estate price predictions involving different building types and multivariate characteristics, offering a more accurate and robust choice
How Equalization of Land Rights Act Amendment Affects the Abnormal Return of Taiwanese Banks: Event Study Analysis
No full textThis research uses the event study analysis to explore how the announcements on Equalization of Land Rights Act Amendment affect the abnormal return of Taiwanese banks. The announcements include the draft approval by the Ministry of the Interior on December 9, 2021, the approval by the Executive Yuan on April 7, 2022, and the approval by the Legislative Yuan on January 10, 2023. The empirical results are as follows.
These three announcements do have overall negative impacts on Taiwanese banking industry. There is a greater negative impact on the stocks of private banks than government-owned banks on the announcement of the draft approval by the Ministry of the Interior. However, there is a greater negative impact on the stocks of government-owned banks than private banks on the announcement of the approval by the Executive Yuan and the Legislative Yuan. Additionally, concerning the negative impact on the housing loans of Taiwanese banks on the three announcements, whether considering all banks or dividing them into government-owned and private banks, the negative abnormal returns were highest for the announcement of the approval by Executive Yuan, indicating the most significant negative impact. On the other hand, the announcement of approval by the Legislative Yuan resulted in the smallest negative returns, indicating the most insignificant impact
Design of Parameter Smoothing for Equivalent Circuit Model of Lithium-ion Battery
No full textLithium-ion batteries, known for their high energy density and lightweight properties, are widely used in smartphones, laptops, electric vehicles, and renewable energy storage systems. Accurately estimating their battery voltage is crucial for their efficient management, safety, and battery life.
Traditional methods for estimating battery capacity have limitations in accuracy and stability. Researchers have developed equivalent circuit models to simulate how lithium-ion batteries charge and discharge. These models use components like resistors and capacitors to describe battery behavior.
Parameter estimation methods often face overfitting issues. This thesis improves accuracy by optimizing parameters in the circuit model through solving optimization problems. Additionally, employing a smoothing approach for optimized parameters reduces errors in voltage estimation.
Research on battery equivalent circuit models is vital for better battery management, increased energy efficiency, and longer battery life. This thesis aims to enhance the reliability, safety, and efficiency of lithium-ion battery applications, promoting advancements in renewable energy and electric transportation fields
Single-Phase Identical Bipolar-Type Buck-Boost AC-AC\uc2\ua0Converter
No full textThis thesis introduces a single-phase identical bipolar AC-AC buck-boost converter. The proposed converter provides identical non-inverting and inverting buck-boost operations, resulting in a voltage gain of \uc2\ub1D\ue2((1-D)). It incorporates dual-bridge inverter modules, input and output LC filters (C_in L_in and C_o L_o), and a capacitor (C). Similar to single-phase matrix converters, it allows for discrete adjustments in the output voltage frequency. Additionally, it delivers continuous input and output currents and performs effectively with loads having unity and non-unity power factors. The proposed converter is particularly well-suited for dynamic voltage restorer (DVR) applications. It adeptly corrects a wide range of grid voltage fluctuations, eliminating the need for a low-frequency voltage injection transformer. Furthermore, its ability to adjust frequency in distinct steps makes it applicable in various scenarios, including high-gain AC-DC converters and traction systems. This thesis places its primary focus on understanding the operational principles of the proposed converter. It then delves into discussions regarding the design of circuit parameters and substantiates its theoretical framework through simulations. In the experimental stage, a 300 W-rated circuit demonstrates its capability to provide the required load voltage of 75 Vrms using appropriate duty cycles. The theoretical and experimental validations ensure the feasibility and effectiveness of the proposed converter
Analysis of the remuneration response of Taiwan energy industry stocks to the amendment of the Green Energy Ordinance
No full textCurrently, the issues of greenhouse effect and energy shortage are affecting the world. Governments around the world have begun to investigate the impact of petrochemical energy on the environment in the past and have amended laws to develop new energy sources and promote carbon neutrality. Taiwan is also keeping up with the trend. In 2016, it launched the non-nuclear homeland policy and set a goal for 20% of Taiwan's total power supply to come from renewable energy by 2025.
It has implemented new energy plans and continuously revised renewable energy regulations and green energy subsidies to ensure sustainable use of resources and create a circular economy.
This study primarily explores the government's modification of renewable energy development regulations. The three stages of events, including the government's announcement on June 21, 111, the Executive Yuan's approval on December 8, 111, and the Legislative Yuan's third reading approval on May 29, 112, have declared the impact on the content of the regulations on wind power, solar energy, geothermal energy, small hydropower, and biomass energy-related companies' stock prices. The study uses event study methodology to analyze the impacts, and the empirical results show that the energy-related industries mostly have significant and positive abnormal returns on the announcement day and the Executive Yuan's approval day. However, in the empirical study of the Legislative Yuan's third reading approval declaration, there are significant but mostly negative average abnormal returns
The Impact of pandemic Insurance on Taiwan Property and Casualty Insurance Company.
No full textIn 2020, due to the COVID-19, life and economy have been severely impacted around the world, and people's lives and health have been severely threatened.At this time, the insurance industry, which has the power to stabilize society, has responded to the current situation and designed a new product "pandemic insurance" that meets the current needs of the people. This allows policyholders to obtain compensation for their living expenses after being diagnosed with COVID-19 or being quarantined, resulting in financial losses.
However, because it does not have sufficient past experience during this epidemic to calculate appropriate premiums and claim amounts, the property and casualty insurance industry has been severely impacted by pandemic insurance. Faced with the rapid changes in epidemic prevention policies, under the changes, the number of confirmed cases has risen rapidly, and claims applications have come like a tsunami. In the end, they are faced with huge claims amounts, causing property and casualty insurance companies to suffer serious financial losses in 2022.
This study will look at the mutual impact between a series of government policies since the outbreak of the COVID-19 in 2020 and the epidemic prevention policies of the property and casualty insurance industry, as well as some financial performance consequences for the property and casualty insurance industry after the outbreak of the epidemic prevention insurance crisis in 2022. Analyzing the impact, we found that the reasons for the chaos in epidemic prevention insurance are not only related to government policies, but also that epidemic prevention insurance policies have indeed had a considerable impact on the property and casualty insurance industry
A sorrow shared is a sorrow halved? Effects of daily customer cyber incivility on employees' work and non-work outcomes
No full textAlthough previous research has confirmed the adverse effects of customer incivility occurring in face-to-face communication, little is known about customer incivility that manifested itself in online interactions (i.e., customer cyber incivility). Drawing on attributional ambiguity theory and conservation of resource (COR) theory, the present study theorized and examined how and when customer cyber incivility affects employees\ue2 work and non-work outcomes. Data collected from 102 full-time employees across 10 consecutive workdays were analyzed using a reliability-corrected single-indicator LMS (RCSLMS). The results showed that employees exposed to daily customer cyber incivility at work reported an increased level of ego depletion at night which, in turn, resulted in a higher level of work-family conflict. The study also confirmed that two types of social support exert stress-alleviating effects on the negative outcomes of customer cyber incivility. Specifically, both intrapersonal cognitive social sharing and interpersonal job social support can diminish the time-lag effect of customer cyber incivility on ego depletion, and further weaken its positive impact on work-family conflict via ego depletion. However, the conditional indirect impact of customer cyber incivility on the next day\ue2s extra-role service behavior through ego depletion was not supported. This study extends prior research on workplace incivility by providing empirical evidence for the conceptualization of customer cyber incivility and the buffering role of both social supports at different levels. The theoretical and practical implications were discussed