1,720,994 research outputs found
Combining simulation experiments and analytical models with area-based accuracy for performance evaluation of manufacturing systems
Get PDFSimulation is considered as one of the most practical tools to estimate manufacturing system performance, but it is slow in its execution. Analytical models are generally available to provide fast, but biased, estimates of the system performance. These two approaches are commonly used distinctly in a sequential approach, or one as alternative to the other, for assessing manufacturing system performance. This article proposes a method to combine simulation experiments with analytical results in a single performance evaluation model. The method is based on kernel regression and allows considering more than one analytical methods. A high-fidelity model is combined with low-fidelity models for manufacturing system performance evaluation. Multiple area-based low-fidelity models can be considered for the prediction. The numerical results show that the proposed method is able to identify the reliability of low-fidelity models in different areas and provide estimates with higher accuracy. Comparison with alternative approaches shows that the method is more accurate in a studied manufacturing application
Dynamic programming for energy control of machine tools in manufacturing
No full textControlling the machine state towards the optimum trade-off between cycle time and power demand improves machine sustainability in manufacturing. This paper presents a dynamic programming approach for optimally controlling a single server manufacturing system with finite capacity and Poisson arrivals (M/M/1/K). The admission of parts into the system is controlled together with the service rate of the machine tool. Further, the machine can be triggered in a low power consumption state when the service is interrupted. The structural properties of the optimal control are analyzed when the transition off/on is instantaneous and when a transitory - i.e., warm-up - is necessary to resume the service. The transitory length is considered exponentially distributed. Numerical results are based on value iteration approximation
Extended kernel regression: A multi-resolution method to combine simulation experiments with analytical methods
No full textSimulation is widely used to predict the performance of complex systems. The main drawback of simulation is that it is slow in execution and the related compute experiments can be very expensive. On the other hand, analytical methods are used to rapidly provide performance estimates, but they are often approximate because of their restrictive assumptions. Recently, Extended Kernel Regression (EKR) has been proposed to combine simulation with analytical methods for reducing the computational effort. This paper has different purposes. Firstly, EKR is tested on different cases and compared with other techniques. Secondly, two different methods for calculation of confidence band are proposed. Numerical results show that the EKR method provides accurate predictions, particularly when the computational effort is low. Results also show that the performance of the two confidence band methods depends on the case analyzed. Thus, further studies are necessary to develop a robust method for confidence band calculation
Essays on nudging customers\u27 behaviors: Evidence from online grocery shopping and crowdfunding
No full textThe dissertation consists of three essays that employ predictive analytics, structural modeling techniques and field experiments to understand and nudge customers’ behaviors in two types of online engagement platforms. The first one is customers’ purchase behaviors in an online grocery store and the other is customer’ contribution behaviors in a reward-based crowdfunding platform. In both contexts, we study how to actively nudge their behaviors. In Chapter 2, we investigates how, when dealing with products that are available in limited quantities, customers may be nudged to purchase them. Specifically, our main problem is to identify targeted customers to receive the limited number of coupons. We develop a Support Vector Machines (SVM) based approach to rank order customers. We conduct a field experiment in an online grocery store to evaluate how well the identified customers are nudged through information and/or couponing. We find that, in terms of the successful nudges, our SVM-based approach performed better than other approaches. We are not just focusing on nudging customers to purchase but also on nudging them to contribute. In Chapter 3, we examine how to leverage the project reward structure (PRS) to nudge backers to contribute on reward-based crowdfunding platforms. We develop a structural model of the backer’s dynamic pledging and learning behaviors. We use it to test a variety of behavioral theories of how PRS and intertemporal changes in the PRS influence backers’ pledging decisions over the course of a project’s funding period. We also use the model to run market simulations and shed lights on how to offer the PRS changes and what is the optimal timing to make such changes. Coupons often act as price discrimination tools to nudge low willingness-to-pay customers to purchase. However, in our context where there is limited product availability, strategies other than just sending coupons may be desirable. For some customers, it is sufficient that we provide information alone but no coupons. Also coupons of different discount depths might play a different role as customers might update their expectations. Particularly, in Chapter 4, we investigate the impact of different nudging strategies on customers’ purchase behaviors. We evaluate the effectiveness of those different nudging strategies via a randomized field experiment. Consistent with the prior literature, we found coupons could serve as a form of “advertisement”. Furthermore, our findings show that coupons with a low discount rate could have a longer information carryover effect than those with a higher discount one. The experiment also generated insights about when couponing as opposed to information is more effective when nudging
Learning and Decision Making Under Uncertainty
No full textIn practice, we often make decisions under uncertainties with known distributions or even without knowing distributions. This study explores decision learning and decision making in not-for-profit operations and supply chain management. We first study dynamic staffing under volunteer supply uncertainty, then explore how to dynamically balance uncertain supply with uncertain demand under lost sales, and finally study decision learning with limited data when distributions are unknown. We provide a brief description of the results obtained from the specific problems considered in this study.The dynamic staffing problem under volunteer supply uncertainty is explored in Chapter 2. We model a finite-horizon staffing problem in nonprofit organizations making hiring and assignment decisions for paid workers and volunteer, given a budget constrain, a capacity constraint, and uncertainties of volunteer supply and part-time worker turnover. Although the optimal staffing policy is computationally challenging to identify in general, we show that an intuitive prioritization assignment policy for all staff and a simple hire-up-to policy for part-time workers can be conveniently applied and close to optimal. Based on the theoretical properties of the optimal policy, we further suggest two easy-to-implement heuristics, both of which have low relative optimality gaps. We also provide performance lower bounds of both heuristics.The dynamic problem of balancing uncertain supply with uncertain demand under lost sales is explored in Chapter 3. We study the dynamic inventory replenishment and product pricing policy aiming to mitigate both supply uncertainty and demand loss. Since the dynamic planning problem is highly non-concave and thus intractable, we propose an approach that focuses on a class of intuitively appealing and practically plausible policies that require the amount of stock allocated for meeting the demand to be increasing and the product price to be decreasing in the available inventory level. We show that, under general conditions for the stochastic supply and demand functions, over a restricted monotone policy class, the dynamic problem become a concave optimization problem . We further reduce the restricted class to a refined class which can be easily computed, and appropriately selected refined policies produce optimal or close-to-optimal profits.The decision learning with limited data problem is explored in Chapter 4. We study how to utilize the data from related systems for decision making with limited data, which underscores the role of domain knowledge, the statistical similarity among the related systems and the structural relationships between inputs and outputs. When a related system has ample data, we demonstrate, through the application of newsvendor systems, that transfer learning can improve decision performance in the focal system, and cross-learning solutions can significantly improves the performance of the focal system over the transfer-learned solution and are asymptotically optimal. When there are multiple related systems with limited data, we transform the data from different systems to create a generic stochastic environment for the decision making problem, and show that the derived co-learning solution is asymptotically optimal for each involved system, as well as the aggregate system
The patient assignment problem in home health care: using a data-driven method to estimate the travel times of care givers
Get PDFHome health care is one of the recent service systems where human resource planning has a great importance. The assignment of patients to care givers is a relevant issue that the home health care service provider must address before generating the daily routes. The assignment decision is typically made without knowing the visiting sequence, which creates some uncertainties and disparities regarding the effective workload of care givers. However, taking into account travel times in the care giver workload while solving the assignment problem is not straightforward, because travel times can also be affected by clinical conditions of patients and their homes. Providing good travel time estimates that would be used in the assignment decision is the specific topic this paper focuses on. In particular, we propose a data-driven method to estimate the travel times of care givers in the assignment problem when their routes are not available yet. The method, based on the Kernel regression technique, uses the travel times observed from previous periods to estimate the time necessary for visiting a set of patients located in specific geographical locations. The main advantage offered by this technique is the empirical modelling of the travel routes generated by care givers. Numerical results based on realistic problem instances indicate that the proposed estimation method performs better than the average value and k-nearest neighbor search methods and can be successfully used in a two-stage approach that first assigns patients to care givers and then defines their routes
Recommended from our members
Topics in Modeling Uncertainty with Learning
Get PDFIt is fair to say that in many real world decision problems the underlying models cannot be accurately represented. Uncertainty in a model may arise due to lack of sufficient data to calibrate the model, non-stationarity, or due to wrong subjective assumptions. Hence optimization in presence of model uncertainty is a very important issue. In the last few decades, there has been a lot of work on finding robust solutions to model uncertainty in operations research. With advances in the field of convex optimization, many robust optimization problems are efficiently solvable. Still there are many challenges and open questions related to model uncertainty, specially when learning is also involved. In this thesis, we study the following challenges and problems related to model uncertainty with learning:First, defining and computing a robust solution to problems with model uncertainty is a challenging task, specially in dynamic optimization problems. Many dynamic problems are tractable only because the structure of the solutions can be analyzed and therefore the dimension of a solution space can be reduced. It is therefore important to design and study robust equivalents of dynamic optimization problems and analyze the structure of robust solutions.Second, in many situations the need for robust solution arises because of lack of sufficient data to calibrate model parameters. It is important to study the properties of traditional robust solutions. Are these robust solutions good for decision making in long run as compared to non-robust solutions? If not, is it possible to design alternative approaches?Third, it is not possible have robustness to model uncertainty if the attention is restricted to a wrong model class. One may use a relatively model free learning and decision making methodology such as reinforcement learning, which can achieve an optimal solution asymptotically. However, these approaches may take a long time to learn and therefore it is important to look at non-parametric methodologies which can achieve good small sample performance.The main contribution of this thesis can be summarized as follows:We study an infinite-time queuing control system, where both arrival and departure rates can be controlled. We consider the case when arrival and departure processescan not be accurately modeled. We prove that a threshold policy is optimal under a max-min robust optimization objective when the uncertainty in the processes is characterizedby a novel notion of relative entropy. The new notion of relative entropy accounts for different levels of modeling errors in arrival and departure processes.We perform numerical tests to study the performance of traditional robust optimization solutions with learning using past few data points. It is shown that the performance of a robust solution may even be worse than a classical point estimate based non-robust solution. We introduce the notion of generalized operational statistics that guarantees a better solution than a classical solution over a set of uncertain parameters, while incorporating subjective prior information. We apply operational statistics approach to mean-variance portfolio optimization problem with uncertain mean returns. We show that the operational statistics portfolio problem can be efficiently solvable by reformulating it as a semi-definite program. Various extensions are discussed and numerical experiments are done to show the efficacy of the solution.We introduce objective operational learning, a new non-parametric approach that incorporates structural information to improve small sample performance. We show how structural and objective information can be incorporated in the objective operational learning algorithm. We apply the algorithm to an inventory control problem with demand dependent on inventory level and prove convergence
- …
