Search reversion within s-metaheuristics: impacts illustrated with a forest planning problemBettinger P., Demirci M., Boston K. (2015). Search reversion within s-metaheuristics: impacts illustrated with a forest planning problem. Silva Fennica vol. 49 no. 2 article id 1232. 20 p. Highlights• The interruption of the sequence of events used to explore a solution space and develop a forest plan, and the re-initiation of the search process from a high-quality, known starting point (reversion) seems necessary for some s-metaheuristics.• When using a s-metaheuristic, higher quality forest plans may be developed when the reversion interval is around six iterations of the model. AbstractThe use of a reversion technique during the search process of s-metaheuristics has received little attention with respect to forest management and planning problems. Reversion involves the interruption of the sequence of events that are used to explore the solution space and the re-initiation of the search process from a high-quality, known starting point. We explored four reversion rates when applied to three different types of s-metaheuristics that have previously shown promise for the forest planning problem explored, threshold accepting, tabu search, and the raindrop method. For two of the s-metaheuristics, we also explored three types of decision choices, a change to the harvest timing of a single management unit (1-opt move), the swapping of two management unit's harvest timing (2-opt moves), and the swapping of three management unit's harvest timing (3-opt moves). One hundred independent forest plans were developed for each of the metaheuristic / reversion rate combinations, all beginning with randomly-generated feasible starting solutions. We found that (a) reversion does improve the quality of the solutions generated, and (b) the rate of reversion is an important factor that can affect solution quality.

Based on our work with ConAgra Foods (http://www.conagrafoods.com), a leading U.S. food manufacturer, we study a large-scale production-planning problem. The problem incorporates several distinguishing characteristics of production in the processed-food industry, including (i) production patterns that define specific combinations of weeks in which products can be produced, (ii) food groups that classify products based on the allergens they contain, (iii) sequence-dependent setup times, and (iv) manufacture of a large number of products (typically, around 200–250) on multiple production lines (typically, around 15–20) in the presence of significant inventory holding costs and production setup costs. The objective is to obtain a minimum-cost four-week cyclic schedule to resolve three basic decisions: (a) the assignment of products to each line, (b) the partitioning of the demand of each product over the lines to which it is assigned, and (c) the sequence of production on each line. We show that the general problem is strongly NP-hard. To develop intuition via theoretical analysis, we first obtain a polynomially solvable special case by sacrificing as little of its structure as possible and then analyzing the impact of imposing production patterns. A mixed-integer programming model of the general problem allows us to assess the average impact of production patterns and production capacities on the cost of an optimal schedule. Next, to solve practical instances of the problem, we develop an easy-to-implement heuristic. We first demonstrate the effectiveness of the heuristic on a comprehensive test bed of instances; the average percentage gap of the heuristic solution from the optimum is about 3%. Then, we show savings of about 28% on a real-world instance (283 products, 17 production lines) by comparing the schedule obtained from the heuristic to one that was in use (at ConAgra) based on an earlier consultant's work. Finally, we discuss the IT infrastructure implemented to enable the incorporation of optimized (or near-optimized) solutions for ongoing use.

We study the problem of (re)designing the regional network by which cadaveric livers are allocated. Whereas prior research focused mainly on maximizing a measure of efficiency of the network that was based on aggregate patient survival, we explicitly account for the trade-off between efficiency and a measure of geographical equity in the allocation process. To this end, we extend earlier optimization models to incorporate both objectives and develop an exact branch-and-price approach to solve this problem, generalizing a solution approach studied for the case where only efficiency is taken into account. In addition, we propose an effective solution algorithm that approximates the (generally nonconcave) frontier of Pareto-efficient solutions with respect to the two objectives by simultaneously generating and successively improving upper and lower bounds on this frontier. We implement and test our approach on observed data and show that solutions significantly dominating the current configuration in both efficiency and equity can be found. Of course, other subjective criteria are needed to choose among the different Pareto-efficient candidate solutions.

No abstract

Long-term management plans have been developed for nearly all of the forests in Turkey. These plans are applied at a sub-district management unit level and may contain guidance for both intermediate yield and final yield harvests. To implement an intermediate yield plan, which involves the scheduling of forest thinnings (stand tending), consideration in Turkey is given to the advantages of working in the same terrain and the same general area each year. Therefore, compartments are often clumped together to create thinning blocks, taking into consideration the thinning priority of the stands, road conditions, site index, age, and proximity of the compartments. Further, when preparing annual budgets and planning to meet the market’s needs, forest enterprises require an even flow of intermediate wood volume each year. In this paper, we introduce a new approach in stand tending planning designed to schedule an equal amount of intermediate wood volume each year and to create thinning blocks by minimizing the distance to pre-defined ramps (landings). We developed both linear and nonlinear goal programming models to minimize both the deviations from a harvest volume (annual intermediate yield allowable cut) target and the deviations from a target value determined for the distances (total and average) of the centroid of each compartment to the hypothetical forest ramps. By using the extended version of Lingo 16, we solved the problem with different weights for the deviations in volume and distance that ranged from 0.0 to 1.0, in 10% intervals, which created 11 scenarios. We carefully analyzed the results of each scenario by taking into consideration the wood volume and distance of compartments to the ramps. The best scenario using the linear model produced a deviation in volume scheduled for the entire decade of 6 m3, while the deviation in total distance between harvest areas and ramps was 59.7 km. Scenario 5, with weights of 0.6 for volume and 0.4 for distance, produced these results, where compartments were closest to one another. The best scenario using the nonlinear model also produced a deviation in volume of 0 m3 and the total average deviation in distance between harvest areas and ramps was 8.7 km. Scenario 3, with weights of 0.8 for volume and 0.2 for distance, produced these results. The approach and models described through this study may be appropriate for further integration into forest management planning processes developed for the planning of Mediterranean forests.