Armagan Tarim

Cost-based filtering for stochastic inventory control

An interesting class of production/inventory control problems considers a single product and a single stocking location, given a stochastic demand with a known non-stationary probability distribution. Under a widely-used control policy for this type of inventory system, the objective is to find the optimal number of replenishments, their timings and their respective order-up-to-levels that meet customer demands to a required service level. We extend a known CP approach for this problem using a cost-based filtering method. Our algorithm can solve to optimality instances of realistic size much more efficiently than previous approaches, often with no search effort at all.

A Global Chance-Constraint for Stochastic Inventory Systems Under Service Level Constraints

We consider a class of production/inventory control problems that has a single product and a single stocking location, for which a stochastic demand with a known non-stationary probability distribution is given. Under the widely-known replenishment cycle policy the problem of computing policy parameters under service level constraints has been modeled using various techniques. Tarim and Kingsman introduced a modeling strategy that constitutes the state-of-the-art approach for solving this problem. In this paper we identify two sources of approximation in Tarim and Kingsman’s model and we propose an exact stochastic constraint programming approach. We build our approach on a novel concept, global chance-constraints, which we introduce in this paper. Solutions provided by our exact approach are employed to analyze the accuracy of the model developed by Tarim and Kingsman.

A Steady-State Genetic Algorithm with Resampling for Noisy Inventory Control

Noisy fitness functions occur in many practical applications of evolutionary computation. A standard technique for solving these problems is fitness resampling but this may be inefficient or need a large population, and combined with elitism it may overvalue chromosomes or reduce genetic diversity. We describe a simple new resampling technique called Greedy Average Sampling for steady-state genetic algorithms such as GENITOR. It requires an extra runtime parameter to be tuned, but does not need a large population or assumptions on noise distributions. In experiments on a well-known Inventory Control problem it performed a large number of samples on the best chromosomes yet only a small number on average, and was more effective than four other tested techniques.

Cost-Based Domain Filtering for Stochastic Constraint Programming

Cost-based filtering is a novel approach that combines techniques from Operations Research and Constraint Programming to filter from decision variable domains values that do not lead to better solutions [7]. Stochastic Constraint Programming is a framework for modeling combinatorial optimization problems that involve uncertainty [9]. In this work, we show how to perform cost-based filtering for certain classes of stochastic constraint programs. Our approach is based on a set of known inequalities borrowed from Stochastic Programming -- a branch of OR concerned with modeling and solving problems involving uncertainty. We discuss bound generation and cost-based domain filtering procedures for a well-known problem in the Stochastic Programming literature, the static stochastic knapsack problem. We also apply our technique to a stochastic sequencing problem. Our results clearly show the value of the proposed approach over a pure scenario-based Stochastic Constraint Programming formulation both in terms of explored nodes and run times.

Event-Driven probabilistic constraint programming

Real-life management decisions are usually made in uncertain environments, and decision support systems that ignore this uncertainty are unlikely to provide realistic guidance. We show that previous approaches fail to provide appropriate support for reasoning about reliability under uncertainty. We propose a new framework that addresses this issue by allowing logical dependencies between constraints. Reliability is then defined in terms of key constraints called “events”, which are related to other constraints via these dependencies. We illustrate our approach on two problems, contrast it with existing frameworks, and discuss future developments.