45 research outputs found
Recommended from our members
Multi-objective Optimization for Equitable Post-Disaster Relief Supply Distribution
No full textHumanitarian logistics in the post-disaster phase of an earthquake requires detailed planning about the relief distribution network including assigning available distribution centers (DCs) to the affected areas, distribution of the relief commodities demanded by the affected population, and efficient allocation of the available vehicle fleet for the distribution in a short span of time. As the demand for relief commodities changes dynamically, the allocation of relief commodities requires a multi-period emergency plan to fully utilize the emergency resources efficiently. Furthermore, as a disaster occurs suddenly without any warning, relief supplies are insufficient in the initial phase of the disaster. At such times, the decision makers face difficulties distributing the available supplies equitably across all the affected areas without putting any particular community at risk.
This study focuses on two different dimensions: efficiency and equity by minimizing the total unmet demand as well as minimizing the total travel time to satisfy demand at different nodes across different time periods while requiring that the percentage of satisfied demand at each node is within a specified deviation range from the average demand satisfaction rate for all nodes.
To address this problem, a deterministic multi-objective mathematical programming formulation is developed to model the design of a disaster relief distribution network with the primary objective of minimizing the total unmet demand across all demand nodes and the secondary objective of minimizing the total transportation time. The model is solved using the lexicographic method for a problem instance of a Cascadia Subduction Zone (CSZ) earthquake in the state of Oregon. Four scenarios are evaluated for two different earthquake magnitudes with different levels of damage to candidate DCs. Pareto optimal frontiers are obtained to determine the trade-off between the unmet demand and the total travel time for these scenarios. The model results show that an equitable distribution of relief commodities is possible at a relatively high demand satisfaction rate when supplies are still limited but the number of vehicles for two different modes of transportation is large. Moreover, shortages in vehicles significantly increase the unmet demand across different demand nodes. Overall, this research provides useful insights about the characteristics of the relief distribution network and provides a method for trade-off analysis that decision-makers can use to improve the efficiency of humanitarian logistics in a post-disaster setting
Recommended from our members
Applications of Bayesian Optimization and Classification Surrogate Modeling to User-Guided Design
No full textIn engineering problems not all constraints can be explicitly written as mathematical functions. There needs to be a way integrate designer knowledge into the design process and preferably use that to guide an optimization problem. In this thesis, these constraints are modelled using classification surrogate models and integrated with Bayesian optimization. This method is applied to situations where a user, or a design team, can work to construct a feasible design space throughout an optimization process. This proposed method is first proven with test optimization problems to show viability, both using the surrogate model constraint and a mix of known constraints. Next, the method is extended to include user-feedback with a program that allows users to define regions of feasible and infeasible design space, creating a surrogate model that guides the optimization process. The proposed method shows promise in situations where design constraints are difficult to quantify mathematically but are known to the design team, such as biological concerns, ergonomic issues, or complex geometric considerations
Recommended from our members
Selecting Probability Distributions for Use in Complex Discrete Event Simulation Models
No full textIn discrete event simulation, probability distributions are used as models of various unpredictable system components. The times between equipment failures and job processing times are examples of such random system components. The commonly recommended and taught practice for selecting a probability distribution to represent a random component is a multistep process that concludes with a recommended “best fit” distribution. In this research, a more straightforward approach for selecting a probability distribution to represent a random component is examined. In this approach a standard or “default” distribution is used, with parameters set so that the distribution moments match the first two sample moments of collected data. Two default distributions were examined, which were the two parameter Lognormal, and Mixed Empirical-Exponential distribution. Comparisons of the use of default distributions was compared to the use of best fit distributions in three models of increasing complexity. Measurements of complexity were developed that show a clear difference in the complexity of the models developed. Statistically significant differences in estimated performance measures when using best, or default distributions are less frequent as the systems become more complex. The results also indicate that using two-parameter lognormal default distributions results in fewer differences when compared to results using best fit distributions than the Mixed Empirical-Exponential default distribution
Recommended from our members
When Normal Distributions Fail: An Application of Feature Engineering to YouTube Videos
No full textYouTube has turned creating, watching, and interacting with videos into a mainstream activity and has substantial impact on modern society as the most popular video sharing site in the world. However, there is a lack of published research using statistically designed experiments which can aid content creators in making and optimizing thumbnails and meta-level features such as the title, description, and length of videos. This research explores the effect specific features of YouTube videos have on video views. Understanding which features have a significant impact will allow creators to fine-tune and select a subset of features which maximize views. In addition, a conceptual model can be used to further understand human-computer interaction, decision making, and variability of a process. Results from this research can be used to identify which videos features should be included to maximize views and identify non-value-added features or steps in the production process.Keywords: Industrial engineering, YouTube, video sharing, ANOVA, feature engineering, experimental design, signal detection theor
Recommended from our members
Mining Temporal Patterns for Prediction: A Mixed-Integer Programming Approach
No full textIn the past two decades, the advancement in data collection and storage have led to the accumulation of complex datasets. Consequently, various industries have sought data-driven solutions to predict and detect anomalies. Temporal patterns have emerged as potential features in prediction models that could improve the performance of the identification of anomalies.
Existing state-of-the-art approaches in pattern mining for prediction follow a top-down approach, where pattern-defining parameters (e.g. support thresholds and gap parameters) are first chosen arbitrarily or through expert opinion. Subsequently, patterns are identified based on these predefined parameters, and then a prediction model is constructed using these identified patterns as features.
In contrast, our research focuses on a novel and wholistic approach to the discovery of patterns. We propose a methodology that enables the simultaneous and optimal discovery of both pattern-defining parameters and prediction model coefficients.
This approach allows these parameters to be determined based on the characteristics of the data and the outcome of interest. To achieve this, we develop a mixed-integer programming (MIP) framework, which optimizes the pattern discovery process effectively called RTP-MIP. We compare our model's results with the existing recent temporal pattern mining (RTP) algorithm using random (RTP-Random) and grid search (RTP-Grid) techniques to select parameters value. Experimental results show that MIP can accurately predict the outcome and optimal values of parameters
Models for predicting the evolution of influenza to inform vaccine strain selection
No full textInfluenza vaccine composition is reviewed before every flu season because influenza viruses constantly evolve through antigenic changes. To inform vaccine updates, laboratories that contribute to the World Health Organization Global Influenza Surveillance and Response System monitor the antigenic phenotypes of circulating viruses all year round. Vaccine strains are selected in anticipation of the upcoming influenza season to allow adequate time for production. A mismatch between vaccine strains and predominant strains in the flu season can significantly reduce vaccine effectiveness. Models for predicting the evolution of influenza based on the relationship of genetic mutations and antigenic characteristics of circulating viruses may inform vaccine strain selection decisions. We review the literature on state-of-the-art tools and prediction methodologies utilized in modeling the evolution of influenza to inform vaccine strain selection. We then discuss areas that are open for improvement and need further research
Recommended from our members
Destructive measuring instrument precision estimation
No full textMeasuring instrument precision/repeatability is one variance component in a measurement system. Assessing measuring instrument precision is necessary to evaluate its capability for a particular application. Well-known approaches for assessing precision/repeatability rely on repeated measurements of items. However, a destructive measurement destroys measured items, so that repeated measurements are not possible. This research focuses on estimating measuring instrument precision for a destructive measurement instrument. The methodology is based on an assumed polynomial relationship between the squared means and the variance of the measured item/part dimension. A specific dimension of different part types (with different means) are independently measured by a single operator and measuring instrument, and the measurements are assumed to be normally distributed. Additionally, the repeatability is assumed constant. Formulas for repeatability estimation, the variance of the estimator, and confidence intervals are developed for a quadratic polynomial relationship. These formulas are derived under two scenarios. In the first scenario, part type means are assumed known and exact formulas are obtains. In the second scenario, part type means are estimated from measurements and approximate formulas are obtained. Confidence interval coverage evaluation is performed with Monte Carlo simulation to verify the known mean scenario, and to evaluate the accuracy of the estimated mean scenario. A general formula for higher order polynomial relationships is developed and evaluated for cubic and quartic relationships between true means squared and variance. The results indicate that the derived formulas give confidence interval coverage close to 95%
Recommended from our members
Measuring Instrument Precision Estimation in Destructive Testing
No full textThe variability introduced into measurements by a measuring instrument is referred to as measurement instrument precision. Experimental procedures and analysis methods exist when measurements are repeatable and can be repeated on the same item. However, when the measurements are destructive and repeated measurements are not possible, estimating measuring instrument precision is difficult since measuring instrument precision is confounded with part variance. In this research, formulas are developed for estimating measuring instrument precision and the measuring instrument precision estimate variance, from which confidence intervals can be obtained. The results are obtained by measuring two different part types, assuming the part measurement coefficient of variation is constant, the measurement instrument precision is constant, and that part measurements are normally distributed and independent. Equations are derived to estimate measuring instrument precision and its standard error when part type means are assumed known, and also when part type means are estimated from the measurement data. The results are validated using Monte Carlo simulation
