启发式 FF-NN 模型在随机动态批量问题中的应用

摘要

For the mathematical intractability caused by the complex structure of multi-period single-item lot sizing problem under stochastic environment,we propose the heuristic-based feed forward neural networks (FF-NN)model.By studying an optimal lot sizing strategy which is based on the minimum total relevant cost price and uncertainty demand,we construct four FF-NN models,they are based on Taguchi method,back propagation (BP),genetic algorithm (GA)and bee colony algorithm (BA)respectively.We also compare all combinations of various methods and models by using heuristic cost calculation methods in three specific domains,including the revised silver and meal (RSM),revised least unit cost (RLUC)and cost benefit (CB).Experimental results show that the combination of BA-based FF-NN model and RLUC method is the heuristic combination to help the decision makers the best,which well settles the mathematical intractability of stochastic dynamic lot-sizing problem.%针对随机环境下多期单项批量问题的复杂结构导致的数学难解性，提出基于启发式的前馈神经网络 FF-NN（Feed For-ward-Neural Network）模型。通过研究一种基于最小总相关成本价格和不确定性需求的最优批量策略，构建基于 Taguchi 方法、反向传播（BP）、遗传算法（GA）、蜂群算法（BA）的四种前馈神经网络模型，使用三种特定领域的启发式成本计算方法，包括修正银餐（RSM）、修正最小单位成本（RLUC）、成本效益（CB），比较各种方法及模型的组合。实验结果表明，基于 BA 算法的 FF-NN 模型与RLUC 方法的组合是对决策者最有帮助的启发式组合，很好地解决了随机动态批量问题中的数学难解性。