Human Flesh Search (HFS), originated in China, has become an explosive Web phenomenon. HFS episodes typically start with news events. Participants pay close attention to the events, get together online, investigate, try to solve the real world problems and find the truth. In HFS episodes, participants form online communities, share information and collaborate with each other. Such online communities are unique subjects of social network study. This dissertation presents the first systematic empirical study and data-driven modeling of HFS. We give the formal definition of HFS, summarize the typical HFS process and classify the episodes based on their topics. We study network measurements of the social networks corresponding to the communities of individual HFS episodes. The communities are strongly centralized, and have small world property. Information diffusion within the communities restricts by the central hubs of the networks. To understand the overall properties of HFS communities, HFS core network is built to study the connections of participants in all episodes. The result shows that HFS core network is a small world and scale-free network. Since the HFS communities do not follow any existing network model, a modified network model is purposed to explain the characteristics of HFS episode networks.

Grafted vegetable seedlings have been proven to possess higher seed/non-seed diseases resistance and yields compared with non-grafted ones. Owing to the seasonality of vegetable planting and labor intensiveness of grafted seedling production (e.g., grafting operation), U.S. vegetable seedling supply chains suffer from high grafted seedling cost. To make grafted seedlings affordable for vegetable growers, low-cost production systems and cost-efficient grafting capacity must be achieved via optimal design of a grafting operation system and supply chain collaboration, respectively. Toward this end, a two-level simulation-based framework is proposed in this work for improving the overall performance of the grafted seedling supply chain by supporting both the grafted seedling production system design and supply chain collaboration decisions. The considered supply chain consists of a single grafted seedling producer that produces grafted seedlings and multiple vegetable growers that seasonally purchase grafted seedlings and produce vegetables to meet price-sensitive demand from the downstream market. More specifically, the low level of the proposed framework focuses on the grafted seedling production system design by integrating discrete event simulation (DES) together with a fuzzy analytic hierarchy process (AHP) for multiple criteria (i.e. production cost, capital investment, production throughput time, resource utilization, and product quality). A Unified Modeling Language (UML)-based simulation modeling and generation approach is developed to automatically generate simulation models of various production system design alternatives. UML information models are developed to provide the system structural information for simulation model generation, production information for simulation execution, and output requirement information for defining simulation outputs. The performance of the production system design alternatives for the aforementioned criteria is evaluated via the generated simulation models, and the corresponding simulation results together with decision makers' judgments on the criteria are used to select the best system design via AHP. A best alternative search (BAS) procedure is proposed for the adopted AHP approach to search for the best system design against ranking impreciseness caused by simulation randomness. At the high level, the proposed framework focuses on the optimal supply chain decisions for early order commitment (EOC) to reduce the amortized production capacity cost. EOC is a supply chain collaboration mechanism, where the grafted seedling producer encourages the vegetable growers to commit their orders earlier than their regular ordering times by providing certain benefits (e.g., price discount). Based on the optimal design of a grafted seedling production system and the corresponding production cost obtained at the low level, we first derive analytical solutions for the grafted seedling producer's optimal capacity, vegetable grower's optimal order quantity, and ordering time under a basic supply chain structure (i.e., single-seedling producer and single-vegetable grower). We then introduce capacity competition by extending the basic structure to a multi-vegetable grower structure. The existence of the N-person game equilibrium and the corresponding relationships between the grafted seedling producer's profit and the vegetable growers' early order decisions are provided. In addition, a capacity reservation mechanism is proposed for the seedling producer to motivate the vegetable growers to release order information in advance. To identify the convergence of the vegetable growers' ordering times, a Cellular Automata simulation model is developed, where each vegetable grower is modeled as a Pavlovian or greedy agent making an ordering time decision so as to receive the higher profit over iterations. The proposed framework is demonstrated for grafted seedling supply chains in North America. The experiment results reveal the benefits of the proposed framework in reducing the grafted seedling cost, as well as in increasing the entire supply chain's profit.

Reliability assurance of hierarchical systems is crucial for their health management in many mission-critical industries. Due to the limited/absent reliability data and engineering knowledge available at the system level and the complex system structure, system-level reliability assurance is challenging. To meet with these challenges, the dissertation proposes a generic, flexible and recursive multi-level information aggregation framework by systematically utilizing multi-level reliability information throughout a system structure to improve the performance of a variety of system reliability assurance tasks. Specifically, the aggregation approach is first present to aggregate complex reliability data structure (e.g., failure time data with covariates and different censoring) with less distribution assumptions to improve accuracy of system-level reliability modeling. The system structure is mainly restricted to the hierarchical series-and-parallel system with independent intra-level components and/or sub-systems. Then, the aggregation approach is extended to accommodate multi-state hierarchical system by considering both probabilistic inter-level failure relationships and cascading intra-level failure dependency. Last, the aggregation approach is incorporated into the design of system-level reliability demonstration testing to achieve the potential sample size reduction. Different demonstration testing strategies with and without information aggregation are comprehensively compared with closed-form conditions obtained. A series of case studies have also been conducted to demonstrate that the proposed aggregation methodology can successfully improve the system reliability modeling accuracy and precision, and improve the cost-effectiveness of the system reliability demonstration tests.

Planning and control of unmanned vehicles play a major role in multi-vehicle systems since accomplishing challenging missions requires not only an extensive decision-making process but it also demands execution of those decisions based on the received information from multiple sensors. In this dissertation, a simulation-based planning and control system is designed, developed and demonstrated for effective and efficient crowd surveillance via collaborative operation of unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). The dissertation research works involve three phases. At phase one, a dynamic data driven adaptive multi-scale simulation (DDDAMS)-based planning and control framework is designed and developed, where the major components include 1) integrated controller, 2) integrated planner, 3) decision module for DDDAMS, and 4) real system. Moreover, crowd detection, tracking, and motion planning modules are implemented in this framework to perform the crowd surveillance mission. This framework adopts dynamic data driven application system (DDDAS) paradigm, where the integrated planner is invoked on a temporal or event basis to incorporate dynamic data from onboard sensors of unmanned vehicles into the simulation and select the best control strategy. At phase two, a testbed is designed and constructed using agent-based hardware-in-the-loop simulation, which involves various hardware components (i.e. real UAVs and UGVs containing onboard sensors and computers) and software components (agent-based simulation and hardware interface). The agent-based simulation, a major component of this testbed, is developed by modeling the behavior of the unmanned vehicles while utilizing the terrain elevation data obtained from GIS. Moreover, a social force model is used to mimic the crowd dynamics in the simulated environment. The constructed testbed is used to evaluate the effectiveness and computational efficiency of the proposed planning and control framework. At phase three, a team formation approach is proposed for allocating unmanned vehicles to different crowd clusters using their geometry and available number of resources based on two different criteria (i.e. overall coverage of all clusters and uniform assignment of resources among clusters). This approach is used in crowd splitting scenarios when the crowd starts to divide into clusters, and the existing team of unmanned vehicles is not able to continue following all the clusters. Moreover, control strategies for team formation, information aggregation, and motion planning of unmanned vehicles are introduced, and a method for determining the value of the control strategy parameter for the information aggregation of UAVs and UGVs is proposed. In conclusion, we believe this work has a profound impact on both the research community and practitioners using unmanned vehicles. Also, the developed hardware-in-the-loop DDDAMS system has the potential to be deployed in real-world applications such as border patrol.

This work focuses on the analysis and application of various nonlinear, autonomous guidance algorithms that utilize sliding mode control to guarantee system stability and robustness. While the basis for the algorithms has previously been proposed, past efforts barely scratched the surface of the theoretical details and implications of these algorithms. Of the three algorithms that are the subject of this research, two are directly derived from optimal control theory and augmented using sliding mode control. Analysis of the derivation of these algorithms has shown that they are two different representations of the same result, one of which uses a simple error state model (Δr/Δv) and the other uses definitions of the zero-effort miss and zero-effort velocity (ZEM/ZEV) values. By investigating the dynamics of the defined sliding surfaces and their impact on the overall system, many implications have been deduced regarding the behavior of these systems which are noted to feature time-varying sliding modes. A formal finite time stability analysis has also been performed to theoretically demonstrate that the algorithms globally stabilize the system in finite time in the presence of perturbations and unmodeled dynamics. The third algorithm that has been subject to analysis is derived from a direct application of higher-order sliding mode control and Lyapunov stability analysis without consideration of optimal control theory and has been named the Multiple Sliding Surface Guidance (MSSG). Via use of reinforcement learning methods an optimal set of gains has been found that make the guidance perform similarly to an open-loop optimal solution. Careful side-by-side inspection of the MSSG and Optimal Sliding Guidance (OSG) algorithms has shown some striking similarities. A detailed comparison of the algorithms has demonstrated that though they are nearly indistinguishable at first glance, there are some key differences between the two algorithms and they are indeed not identical. Finally, this work has a large focus on the application of these various algorithms to a large number of space based applications. These include applications to powered-terminal descent for landing on planetary bodies such as the moon and Mars and to proximity operations (landing, hovering, or maneuvering) about small bodies such as an asteroid or a comet. Further extensions of these algorithms have allowed for adaptation of a hybrid control strategy for planetary landing, and the combined modeling and simultaneous control of both the vehicle's position and orientation implemented within a full six degree-of-freedom spacecraft simulation.

When customers buy a new product, the reliability related issues such as the warranty/post-warranty service and the performance of the product are becoming important factors in the customers' decision-making process. In this dissertation, some important aspects in estimating both warranty and post-warranty repair demands have been studied. To ensure the new product will have a good performance, the availability of a repairable k-out-of-n:G system considering spare parts logistics is investigated. The installed base of the product varies with time due to both new sales and units being taken out of service is considered. We explicitly address the fact that customers may not always request repairs for failed units and formulate the corresponding warranty and post-warranty repair demands for a general failure process. For the case where the product failure time is exponential, we derive the closed-form expressions for the two types of repair demands for both an individual unit and the installed base. A step further, we develop an integrated model to estimate the gross profit for a new durable product to be sold in a fixed sales period at a fixed price. The sales over time is characterized by a stochastic Bass model and the production system is a make-to-order type of system. An approximate, yet accurate approach is developed to quantify the expected total cost of production involving a learning effect. The analysis of key parameters that affect the optimal gross profit is carried out in numerical examples. Finally, the availability management of a new product (or system) is studied. We introduce a collection of operational availability maximization problems, in which the component redundancy levels and the spares stock quantities are to be determined simultaneously under economic and physical constraints.

Water is an essential natural resource for life and economic activities. Water resources management is facing major challenges due to increasing demands caused by population growth, increased industrial and agricultural use, and depletion of fresh water sources around the world. In addition to putting stress on our civilization, factors such as water supply availability, spatial population changes, industrial growth, etc. are all sources of major uncertainty in water resources management. There are also uncertainties regarding climate variability and how it affects both water demands and supplies. Stochastic programming is a mathematical tool to help make decisions under uncertainty that models the uncertain parameters using probability distributions and incorporates probabilistic statements in mathematical optimization. This dissertation applies stochastic programming to water resources management. In particular, we focus on reclaimed water distribution network design to effectively reuse water in a municipal system and a water allocation problem in an integrated water system under uncertainty. We first present a two-stage stochastic integer program with recourse for cost- effective reclaimed water network design. Unlike other formulations, uncertain demands, temporal, and spatial population changes are explicitly considered in our model. Selection of pipe and pump sizes are modeled using binary variables in order to linearize the nonlinear hydraulic equations and objective function terms. We then develop preprocessing methods to significantly reduce the problem dimension by exploiting the problem characteristics and network structure. We analyze the sensitivity of the network design under varying model parameters, present computational results, and discuss when the stochastic solution is most valuable. Next, we investigate the use of risk-averse approach in water resources management using the so-called conditional value-at-risk as a risk measure. We develop a multistage risk-averse stochastic program with recourse for long-term water allocation under uncertain demands and water supply variability. We propose a specialized decomposition-based algorithm to solve multistage risk-averse stochastic programs, and present both the single-cut and the multicut version of the algorithm. We then compare the solution methodologies with different ways of decomposing the resulting problem. We solve the multistage risk-averse water allocation problem with different risk aversion levels and model assumptions, present computational results to demonstrate the potential benefits of risk-averse approach, and provide a guideline for risk aversion level selection.

A two-level agent-based modeling framework is proposed for the electric power system to solve the problems of renewable energy utilization and demand-side management. While in the detailed level of the framework the customers and utility companies are modeled as agents to represent electricity demand and supply performances, respectively, the high level reflects the aggregated performance of the considered electricity market via state space models. To connect the two levels, a social network is introduced as a dynamic medium for the interactions among customer agents. While the customers' consumption behaviors are modeled at lower level and affected by each other, their individual performances contribute to the system performance in the high level. This dissertation concerns three problems. First, the problem of renewable energy adoption concerns penetration process of distributed solar systems with various incentive policies (i.e., Income Tax Credits and Feed-in Tariff) for renewable energy. The proposed hybrid model incorporates agent-based modeling and system dynamics to simulate the solar system diffusion process among the residential customers. Second, the demand-side management problem focuses on scheduling the Plug-in Hybrid Electric Vehicles (PHEV) charging under different scenarios of demand response programs (i.e., Time-of Use and Real-time Pricing). For the Time-of Use (TOU) program, the decision-support analysis results from simulation-based optimization for both customers and the utility company. For the Real-time Pricing (RTP) program, the discussion is to find proper pricing functions according to different customers. Third, the problem concerns the agent interaction based on different architectures of social network (i.e., small-world and scale-free) and the network evolution based on triadic closure. Such interaction is applied to the first two problems with the effect of changing the customers' social connections, preferences in consumption behaviors and acceptable grid prices. Furthermore, to extend the demand-side management problem, this research also discusses the energy management at individual households integrating PV generation system, battery storage and electric vehicle under demand response programs. The conceptual model is based on the threshold method to suggest residential customers when to use the electricity from which sources (PV generation, storage, or local grid).

We consider multistage, stochastic production systems using pull control for production authorization in discrete parts manufacturing. These systems have been widely implemented in recent years and constitute a significant aspect of lean manufacturing. Extensive research has appeared on the optimal sizing of buffer inventory levels in such systems. However the issue of control points, i.e. where in the multistage sequence to locate the output buffers, has not been addressed for pull systems. Allowable container/batch sizes, optimal inventory levels, and ability of systems to automatically adjust to stochastic demand depend on the location of these control points.We begin by examining a serial production system producing a single part type. Two models are examined in this regard. In the first, container size is independent of the control section, while in the second, container sizes are section dependent. Additionally, a nesting policy is introduced which introduces the additional constraint that the container size in a section is related to the container size in any other section by a power of two.Necessary and sufficient conditions are derived for ensuring that a single, end-of-line accumulation point is optimal. When this is not the case, an algorithm is provided to determine the optimal control points. Effects of factors such as value added structure, fixed location cost, setup and material handling cost, kanban collection time, and material transportation time on the control structure are investigated. Results are extended to determine the optimal container size when lead time at a stage is a concave function of container size.The study is then extended to a multi-product case. Queuing aspects are introduced to account for the interaction between the different part types. The queuing model used is a modification of the Decomposition/Recomposition model described in Shantikumar and Buzacott (1981). The models in the chapter do not assume a serial structure any longer. Additionally, general interarrival and service time distributions are considered. The effect of number of products, demand arrival distribution, value added structure, and number of stages on the control structure and system cost is investigated.Finally, a simulation model is developed in Chapter 5 to verify and validate the mathematical models described in Chapters 3 and 4.

A vehicle-based service system might be susceptible to unexpected costs and delays due to unforeseen events, such as a vehicle breakdown, a traffic accident, a medical emergency, etc. In such situations, a priori algorithmic solution may be deteriorated and fleet plans may need to be adjusted in real-time as a function of the dynamic system state. I consider real-time logistics management problems where a vehicle breaks down in the midst of operations. First, a backup vehicle needs to be determined to pick up the passengers/cargo from the breakdown vehicle, and from the breakdown point completing the remaining portion of the planned trip. This backup vehicle can be dispatched from the depot or from the vehicles currently in service. In the former case, it may impose a significant delay if the depot is far away from the breakdown location. In the latter case, the vehicle used as backup may have to change its own schedule. Trips uncompleted by this backup vehicle may have to be further covered by other vehicles. Thus, a good solution should be acquired in conjunction with the status of all other vehicles in the entire network.Yet, the new schedule may be considerably different from the original one after rescheduling is done. These changes may make the crew-rescheduling problem challenging, since it is essential to ensure that all crews know the itinerary of their new trips. Furthermore, the vehicle breakdown may not only delay the current trip that is directly affected by the disruption but also other trips that the breakdown vehicle has to cover in the network. Some of the delayed trips may have to be cancelled. A good approach should consider operating cost, fixed vehicle cost, delay cost, schedule disruption cost as well as trip cancellation cost simultaneously. This real-time logistics management problem has not been properly addressed in the literature.The major contributions of this study are the modeling and formulation of this vehicle rescheduling problem, and the development of some fast algorithms to solve it quickly. The exact algorithms or heuristics are proposed based on the different requirements and assumptions of the problem.