Strategic capacity decision‐making in a stochastic manufacturing environment using real‐time approximate dynamic programming

In this study, we illustrate a real‐time approximate dynamic programming (RTADP) method for solving multistage capacity decision problems in a stochastic manufacturing environment, by using an exemplary three‐stage manufacturing system with recycle. The system is a moderate size queuing network, which experiences stochastic variations in demand and product yield. The dynamic capacity decision problem is formulated as a Markov decision process (MDP). The proposed RTADP method starts with a set of heuristics and learns a superior quality solution by interacting with the stochastic system via simulation. The curse‐of‐dimensionality associated with DP methods is alleviated by the adoption of several notions including “evolving set of relevant states,” for which the value function table is built and updated, “adaptive action set” for keeping track of attractive action candidates, and “nonparametric k nearest neighbor averager” for value function approximation. The performance of the learned solution is evaluated against (1) an “ideal” solution derived using a mixed integer programming (MIP) formulation, which assumes full knowledge of future realized values of the stochastic variables (2) a myopic heuristic solution, and (3) a sample path based rolling horizon MIP solution. The policy learned through the RTADP method turned out to be superior to polices of 2 and 3. © 2010 Wiley Periodicals, Inc. Naval Research Logistics 2010     

Optimal policy for a dynamic multi-echelon inventory model

A general multiperiod multi-echelon supply system consisting of n facilities each stocking a single product is studied. At the beginning of a period each facility may order stock from an exogenous source with no delivery lag and proportional ordering costs. During the period the (random) demands at the facilities are satisfied according to a given supply policy that determines to what extent stock may be redistributed from facilities with excess stock to those experiencing shortages. There are storage, shortage, and transportation costs. An ordering policy that minimizes expected costs is sought. If the initial stock is sufficiently small and certain other conditions are fulfilled, it is optimal to order up to a certain base stock level at each facility. The special supply policy in which each facility except facility 1 passes its shortages on to a given lower numbered facility called its direct supplier is examined in some detail. Bounds on the base stock levels are obtained. It is also shown that if the demand distribution at facility j is stochastically smaller (“spread” less) than that at another facility k having the same direct supplier and if certain other conditions are fulfilled, then the optimal base stock level (“virtual” stock out probability) at j is less than (greater than) or equal to that at facility k.     

Lagrangian solution of maximum dispersion problems

We address the so‐called maximum dispersion problems where the objective is to maximize the sum or the minimum of interelement distances amongst a subset chosen from a given set. The problems arise in a variety of contexts including the location of obnoxious facilities, the selection of diverse groups, and the identification of dense subgraphs. They are known to be computationally difficult. In this paper, we propose a Lagrangian approach toward their solution and report the results of an extensive computational experimentation. Our results show that our Lagrangian approach is reasonably fast, that it yields heuristic solutions which provide good lower bounds on the optimum solution values for both the sum and the minimum problems, and further that it produces decent upper bounds in the case of the sum problem. For the sum problem, the results also show that the Lagrangian heuristic compares favorably against several existing heuristics. © 2000 John Wiley & Sons, Inc. Naval Research Logistics 47: 97–114, 2000     

Rolling‐horizon replenishment: Policies and performance analysis

We consider a rolling‐horizon (RH) replenishment modeling framework under which a buyer can update demand information and inventory status, modify order quantities committed previously, place an advanced order for a new period at the end of the RH, and move along in time seamlessly. We show that the optimal order policy for the two‐period RH problem is a dual‐threshold type for updating period(s) plus a base‐stock type for the advanced order. We provide analytical formulas and algorithms to compute the optimal thresholds and the optimal base‐stock level exactly. With our analytical results and numerical procedures, we demonstrate the significant value of RH replenishment in matching supplies to demands more closely. We also show that with RH updating (flexibility), the value of additional demand information beyond the RH diminishes quickly. © 2010 Wiley Periodicals, Inc. Naval Research Logistics, 2010     

A forward network simplex algorithm for solving multiperiod network flow problems

An optimization model which is frequently used to assist decision makers in the areas of resource scheduling, planning, and distribution is the minimum cost multiperiod network flow problem. This model describes network structure decision-making problems over time. Such problems arise in the areas of production/distribution systems, economic planning, communication systems, material handling systems, traffic systems, railway systems, building evacuation systems, energy systems, as well as in many others. Although existing network solution techniques are efficient, there are still limitations to the size of problems that can be solved. To date, only a few researchers have taken the multiperiod structure into consideration in devising efficient solution methods. Standard network codes are usually used because of their availability and perceived efficiency. In this paper we discuss the development, implementation, and computational testing of a new technique, the forward network simplex method, for solving linear, minimum cost, multiperiod network flow problems. The forward network simplex method is a forward algorithm which exploits the natural decomposition of multiperiod network problems by limiting its pivoting activity. A forward algorithm is an approach to solving dynamic problems by solving successively longer finite subproblems, terminating when a stopping rule can be invoked or a decision horizon found. Such procedures are available for a large number of special structure models. Here we describe the specialization of the forward simplex method of Aronson, Morton, and Thompson to solving multiperiod network network flow problems. Computational results indicate that both the solution time and pivot count are linear in the number of periods. For standard network optimization codes, which do not exploit the multiperiod structure, the pivot count is linear in the number of periods; however, the solution time is quadratic.     

An application of servomechanisms to inventory

In recent years, some attention has been devoted to the application of techniques of control theory to inventory management. In particular, H. Vassian (1955) developed a model for a periodic review inventory system utilizing techniques of discrete variable servomechanisms to analyze the system in a cost-free structure. The resulting model is inherently deterministic, however, and emphasizes the control of inventory fluctuation about a safety level by selecting an appropriate order policy. Such an order policy is defined only up to an arbitrary method of forecasting customer demands. The present paper is a continuation of the model developed by Vassian in which exponential smoothing is used as a specific forecasting technique. Full recognition of the probabilistic nature of demand is taken into account and the requirement of minimizing expected inventory level is imposed. In addition, explicit formulas for the variance in inventory are derived as functions of the smoothing constant and the tradeoff between small variance and rapid system response is noted. Finally, in an attempt to remove the bias inherent in exponential smoothing, a modification of that technique is defined and discussed as an alternate forecasting method.     

Inspection games with local and global allocation bounds

This article discusses a two‐player noncooperative nonzero‐sum inspection game. There are multiple sites that are subject to potential inspection by the first player (an inspector). The second player (potentially a violator) has to choose a vector of violation probabilities over the sites, so that the sum of these probabilities do not exceed one. An efficient method is introduced to compute all Nash equilibria parametrically in the amount of resource that is available to the inspector. Sensitivity analysis reveals nonmonotonicity of the equilibrium utility of the inspector, considered as a function of the amount of resource that is available to it; a phenomenon which is a variant of the well‐known Braess paradox. © 2013 Wiley Periodicals, Inc. Naval Research Logistics, 2013     

Robust tower location for code division multiple access networks

Designing Code Division Multiple Access networks includes determining optimal locations of radio towers and assigning customer markets to the towers. In this paper, we describe a deterministic model for tower location and a stochastic model to optimize revenue given a set of constructed towers. We integrate these models in a stochastic integer programming problem with simple recourse that optimizes the location of towers under demand uncertainty. We develop algorithms using Benders' reformulation, and we provide computational results. © 2006 Wiley Periodicals, Inc. Naval Research Logistics, 2007