基于森林空间收获问题的模拟退火算法邻域搜索技术比较

邻域搜索是当前提高启发式算法求解效率的核心技术之一，然而近期关于该搜索策略的性能却产生了较大争议。模拟退火算法作为一种典型的启发式算法，已广泛应用于一系列的林业规划问题。为此，本研究以模拟退火算法为例，系统评估2种不同邻域搜索技术在森林空间收获安排问题中的应用效果。规划模型以50年规划周期(10个规划分期)内的最大化木材收获为目标函数，以蓄积均衡收获、蓄积期末存量、单位限制模型和绿量限制等为主要约束条件。测试方法以模拟退火算法为原型，以每次优化过程中随机选择的小班数量为标准，共包括1-邻域和2-邻域2种不同的搜索技术。模拟规划数据由3个假设的栅格数据集组成，其共产生了3293个(林分Ⅰ)、29536个(林分Ⅱ)和81625个(林分Ⅲ)0-1型决策变量。研究结果表明:模拟退火算法2-邻域搜索技术能够提高各规划问题的最大目标函数值；但当规划问题的决策变量数(或小班数量)较大时(即林分数量≥3600)，单纯增加邻域范围并不能提高规划问题的平均目标函数值。因此，鉴于模拟退火算法的优化结果对规划问题具有较高的敏感性，因此森林经营决策人员应慎重选择模拟退火算法邻域搜索作为相关规划问题的优化求解技术。 

Evaluating neighborhood search techniques of simulated annealing based on forest spatial harvest scheduling problems

Neighborhood search techniques have become one of the most important strategies to improve the resolution efficiency of heuristics in forestry, however a drastically debate on the resolution efficiency of this search strategy has been put forward recently. Simulated annealing algorithm, as an example of heuristics, has been employed in a wide set of forestry planning problems. Therefore, the overall goals of this research were to evaluate the performances of different neighborhood search techniques of simulated annealing in forest spatial planning problems. The objective function was to maximize the harvest volume over ten 5-year planning periods, which mainly included timber volume flow constraints, ending inventory constraints, unit restriction model and green-up constraints. The tested neighborhood search techniques were 1-opt moves, and 2-opt moves of simulated annealing which have been widely used in forestry planning, in which the candidate solutions of 1-opt moves were generated by randomly changing the treatment of just one unit, however the candidate solutions of 1-opt moves were generated by randomly changing the treatments of two units simultaneously. The planning problems were applied to three hypothetical datasets, which encompassed 3293 ( forestⅠ) , 29536 ( forestⅡ) and 81625 ( forest Ⅲ) binary decision variables. The results showed that the 2-opt technique of simulated annealing can locate the maximum solutions for all the three planning problems, however increasing the number of units for changing the treatment schedule simultaneously in more than one unit did not improve the performance of simulated annealing if the combinatorial problems were very large ( i. e. , the number of management units within a forest was larger than 3600 ) . Since the planning results highly depend on the sizes of planning problems, thus forest managers and planners should pick up the optimization techniques carefully when they plan to make forest plans in practices.