Fixed-quantity dynamic lot sizing using simulated annealing

In this paper, simulated annealing (SA) is applied to the deterministic dynamic lot-sizing problem with batch ordering and backorders. Batch ordering requires orders that are integer multiples of a fixed quantity that is larger than 1. The performance of the developed SA heuristic is compared to that of a genetic algorithm (GA) and a modified silver-meal (MSM) heuristic developed in the literature, based on the frequency of obtaining the optimum solution and the percentage average deviation from the optimum solution. In addition, the effects of three factors on the performance of the SA, GA, and the MSM are investigated in a 2³ factorial experiment. The investigated factors are the demand pattern, the batch size, and the length of the planning horizon. Results indicate that the SA heuristic has the best performance, followed by GA, in terms of the frequency of obtaining the optimum solution and the average deviation from the optimum solution. SA is also the most robust of the investigated heuristics as its performance is only affected by the length of the planning horizon.     

基于遗传神经网络的人脸识别分类器设计

神经网络由于具有抗干扰性和容错性,常被用在人脸识别的分类器中.但是,神经网络的训练过程花费的时间很长,训练的次数多,而且不易收敛.这里提出一种合作式的遗传神经网络作为新的分类器.使分类器的训练次数明显的减少,收敛性得到了提高.实验仿真得到了很好的效果.     

A new algorithm for determining production schedules when solving the multi-product economic lot sizing problem in flow shops

In this paper, we study the generation of production schedules in the multi-product economic lot sizing problem in flow shops (i.e., the FS-ELSP) under the power-of-two policy. To investigate this problem, we first review the mathematical model for the FS-ELSP and Ouenniche and Boctor’s (O&B’s) heuristic, and we comment that there exist several problems in O&B’s heuristic if we try to use it to generate feasible production schedules for the FS-ELSP. Therefore, we propose two new heuristics, namely, the modified O&B’s heuristic and Huang and Yao’s (H&Y’s) heuristic to improve O&B’s heuristic. To compare the performance of these three heuristics, we randomly generate 500 instances for each of the seven levels of utilization rate from 0.4 to 0.75. Based on our numerical experiments, we conclude that H&Y’s heuristic significantly outperforms O&B’s heuristic and the modified O&B’s heuristic.     

A genetic algorithm for the simultaneous lot sizing and scheduling problem in capacitated flow shop with complex setups and backlogging

In this paper the capacitated lot sizing and scheduling problem with sequence-dependent setups, setup carryover, and backlogging has been studied. The problem can be formulated as a mixed-integer program. Most lot sizing problems are hard to solve, especially in medium and large scale. In recent years, to deal with the complexity and find optimal or near-optimal results in reasonable computational time, a growing number of researchers have employed metaheuristic approaches to lot sizing problems. One of the most popular metaheuristics is genetic algorithm which has been applied to different optimization problems successfully. Therefore, we have developed a genetic algorithm to solve this model. To test the accuracy of the genetic algorithm, a lower bound is developed and compared against the genetic algorithm. In computational experiments, proposed genetic algorithm performed extremely well. It is concluded that the genetic algorithm is efficient and effective for this problem.     

Genetic algorithms for the multiple-machine economic lot scheduling problem

This paper focuses on an extension of the Economic Lot Scheduling Problem, which schedules productions of products on multiple identical machines. The objective is to minimize the total average production and inventory costs per unit time for all products. We develop a genetic algorithm under the Common Cycle policy and compare it with an existing heuristic under the same policy. Computational results show that our genetic algorithm outperforms the existing heuristic and its running time does not increase much even for high utilization problems, while the latter requires substantial time to solve most of the high utilization problems. In addition, a genetic algorithm under the Extended Basic Period and Power-of-Two policy is proposed. This new heuristic performs much better, especially when the number of machines is small and the machine utilization is not very high.     

求解生产批量计划问题的改进量子进化算法

针对遗传算法和粒子群算法在求解生产批量计划问题中易陷入局部最优解的问题,提出了改进的量子进化算法.对各周期项目计划产量的决策变量进行基于概率幅的量子比特个体编码,在迭代求解的过程中通过约束违反度比较个体的支配关系,有效指导种群向合理解进化,并根据当前迭代次数动态调整旋转角机制控制基因位的坍塌速度,在进化后期尽量保留最优个体的基因信息以提高算法的收敛速度和求解精度.实验结果表明了该算法的有效性.     

用Hopfield神经网络实现触觉图像恢复

通过对安装于机器人手爪上的阵列触觉传感器所采集到的触觉图像的分析,提出了用Hopfield神经网络实现触觉图像恢复的技术.Hopfield神经网络把触觉图像的每一个像素作为一个处理单元(神经元),像素之间的关系即神经元之间的权值作为储存单元.首先利用Hebb规则计算权值矩阵,用以存储所有样本的标准模式,然后利用网络的联想记忆能力恢复被抓物体的触觉图像.实验结果表明,该方法能达到很好的触觉图像恢复和识别效果.     

Discrete lot sizing and scheduling problem under batch processing constraints in the semiconductor manufacturing

Highly advanced technology products are manufactured on the series of equipment having various characteristics and requirements. The deposition process of organic light emitting diode, for example, is performed in chambers for a long manufacturing time, while various product devices belonging to the same family can be processed in sequence before the preventive maintenance schedule or the refill of chemical gas required. The deposition process plays as a bottleneck, and its productivity of the schedule is critical to the upstream low-temperature poly silicon process and the downstream encapsulation and module processes as well. The batch family scheduling problem is formulated using the mixed-integer programming (MIP) in consideration of the family setups, lot sizing, supplementary mask tools requirements, and material exhaustion requirements. It is shown that the planning and scheduling decisions can be made simultaneously in an integrated model, and that it can be implemented in the actual manufacturing line. Through the optimized model analysis, the capacity can be enhanced by 20～30 % without losing the throughput and demand satisfaction as well.     

基于BP网络的电厂发电机组动态控制算法

本文将BP神经网络模型应用到电厂发电机组动态控制中,并对发电机组进行数学建模,给出了发电机组在70%和100%不同负荷下的动态特性数学模型,提出了发电机组动态控制系统,并为减少系统的稳态误差,采用了FUZZY-PID复合控制结构,仿真结果表明,本文提出的动态控制算法比传统的PID控制算法在准确性和稳定性都更优.     

基于神经网络的车辆主减速器混合遗传算法优化设计

车辆驱动桥的组成包括主减速器、差速器和半轴,其中汽车主减速器的结构和尺寸极大的影响着汽车的动力学性能和经济性,因此采用优化设计方法来设计汽车主减速器是非常重要的。在满足主减速器接触强度、弯曲强度和边界约束的条件下,建立了优化设计数学模型。由于传统的优化方法存在着求解过程复杂和寻优过程容易陷入局部最优解的问题,故通过神经网络方法拟合待求系数,应用遗传算法工具箱调用混合遗传算法寻求最优解,使求解过程得到简化,确保可靠地获得全局最优解。     

Mathematical model and parallel genetic algorithm for hybrid flexible flowshop lot streaming problem

Lot streaming is the technique of splitting a given job into sublots to allow the overlapping of successive operations in multi-stage manufacturing systems thereby reducing production makespan. Several research articles appeared in literature to solve this problem and most of these studies are limited to pure flowshop environments where there is only a single machine in each stage. On the other hand, because of the applicability of hybrid flowshops in different manufacturing settings, the scheduling of these types of shops is also extensively studied by several authors. However, the issue of lot streaming in hybrid flowshop environment is not well studied. In this paper, we aim to contribute in bridging the gap between the research efforts in flowshop lot streaming and hybrid flowshop scheduling. We propose a mathematical model and a genetic algorithm for the lot streaming problem of several jobs in multi-stage flowshops where at each stage there are unrelated parallel machines. The jobs may skip some of the stages, and therefore, the considered system is a complex generalized flowshop. The proposed genetic algorithm is executed on both sequential and parallel computing platforms. Numerical examples showed that the parallel implementation greatly improved the computational performance of the developed heuristic.     

应用Hopfield神经网络和小波域隐Markov树模型的图像复原

为了解决传统的Hopfield神经网络图像复原算法对噪声抑制和图像细节保护不能很好兼顾的问题,提出了一种基于改进的连续Hopfield神经网络和小波域隐Markov树(HMT)模型的复原算法。将小波域HMT模型作为图像小波系数统计关系的先验知识,并以正则化项的形式引入到神经网络模型中,最终利用Hopfield神经网络的能量收敛特性完成图像复原。同时,提出了一种高度并行的网络权值矩阵计算方法,通过对模板图像进行算子操作,分批求取网络权值,避免了大型矩阵的乘法运算。实验结果表明,无论是对真实图像还是人工生成图像,算法复原结果的视觉效果均有明显改善,提高信噪比(ISNR)较传统同类算法增加0.3dB以上,达到了同时抑制噪声和保护图像细节的目的。     

基于Agent的分布式协商策略在供应链动态计划中的应用

在分布式协作环境中，多个商业实体或执行单元，为达成某一共同决策而进行有效地协商，提出一种基于Agent的分布式协商机制和算法。采用智能Agent技术，使各Agent代表各自商业实体或执行单元的利益与其他商业实体Agent为达成某一共同决策进行协商。协商机制中使用了偏好模型的概念，并将Agent的偏好及约束看作私有信息。算法中采用了建议的全局偏好评估和冲突消解。通过实例对提出的分布式协商算法进行了仿真。仿真结果说明，使用全局偏好评估可使协商结果从全局来看更为理想；使用冲突消解明显提高了协商的成功率。最后，分析并扩展了该分布式协商算法的应用范围。     

Disassembly leveling and lot sizing for multiple product types: a basic model and its extension

We consider two interrelated problems that occurred in disassembly systems: disassembly leveling and lot sizing. Disassembly leveling, one of disassembly process planning decisions, is to determine disassembly structures that specify parts and/or subassemblies to be obtained from disassembling used/end-of-life products, and disassembly lot sizing is the problem of determining the amounts of disassembly operations required to satisfy the demands of their parts and/or subassemblies. Unlike the existing studies, this study considers the two problems at the same time for the objective of minimizing the sum of disassembly setup and operation costs. In particular, we consider a generalized version in which disassembly levels may be different even for products of the same type. Two types of the problem are considered in this study. The first one is the basic problem without parts commonality, i.e., products do not share their parts or subassemblies, for which a polynomial time optimal algorithm is suggested after developing a mathematical programming model. The second one is an extended problem with parts commonality. After developing another mathematical programming model for the extension, we prove that it is NP hard. Then, a heuristic algorithm is suggested together with its computational results.     

一种基于改进遗传算法的神经网络优化算法研究

遗传算法是目前优化搜索算法中应用比较广泛的一种,但基本遗传算法存在收敛速度慢、易于陷入局部最优等缺点。针对上述问题对遗传算法(GA)的选择算子进行改进,在最优保存策略的基础上将每代种群按照适应度由小到大排序,平均分成前中后3段,按照0.6、0.8、1的比例进行选择;从尾段中随机抽取个体来补足种群由于选择操作而损失的个体;既利用了最优保存策略的全局收敛特性同时也保持了种群的多样性;用改进的遗传算法调整神经网络的权值形成了新的改进遗传算法优化BP神经网络(IGA-BP);通过与选择算子为适应度比例选择算子的GA-BP网络进行比较,结果表明算法改进后缩短了收敛时间同时减少了运行误差;最后将该改进算法应用于水泥回转窑的故障诊断中,验证了算法的可行性。     

两级分销网络选址——配送问题的模型及算法

为合理设计企业分销网络,提出两级分销网络选址一配送问题,研究在由工厂、配送中心和客户组成的且只销售一种产品的两级分销网络中,从若干候选点里确定已知个数的配送中心的位置以及工厂到配送中心和配送中心到客户的配送方案,使得在满足工厂和配送中心能力限制、客户产品需求、配送中心最小转运量和最小配送量的前提下,包括配送中心选址固定费用、从工厂到配送中心和配送中心到客户的配送费用在内的总费用最小,构建该问题的混合整数线性规划模型,并设计0-1和优先权混合编码的遗传算法。通过对1个演示算例和1组包括2～10个工厂、5-100个配送中心候选点、20-400个客户、6～20个需建设配送中心数的随机生成算例的计算分析,表明了所提方法的可行性和有效性。     

基于改进粒子群算法的生产批量计划问题研究

为求解基于成组单元有能力约束的生产批量计划问题,提出了一种基于二进制粒子群算法和免疫记忆机制相结合的方法,并阐明了该方法的具体实现过程。在该方法中,采用罚函数法处理约束条件,每个粒子都代表一组可用于描述具体批量计划方案的规则组合。通过对其他文献中一个仿真实例的计算和结果比较,表明该算法在寻优能力、求解速度和稳定性等方面都明显优于文献中的遗传算法。     

Road profile estimation using neural network algorithm

This paper more specifically focuses on the estimation of a road profile (i.e., along the “wheel track”). Road profile measurements have been performed to evaluate the ride quality of a newly constructed pavement, to monitor the condition of road networks in road management systems, as an input to vehicle dynamic systems, etc. The measurement may be conducted by a slow-moving apparatus directly measuring the elevation of the road or using a means that measures surface roughness at highway speeds by means of accelerometers coupled with high speed distance sensors, such as laser sensors or using a vehicle equipped with a response-type road roughness measuring system that indirectly indicate the user’s feelings of the ride quality. This paper proposes a solution to the road profile estimation using an artificial neural network (ANN) approach. The method incorporates an ANN which is trained using the data obtained from a validated vehicle model in the ADAMS software to approximate road profiles via the accelerations picked up from the vehicle. The study investigates the estimation capability of neural networks through comparison between some estimated and real road profiles in the form of actual road roughness and power spectral density.     

基于RBF神经网络的底盘测功机控制算法

介绍了底盘测功机的类型并对电涡流测功机工作原理进行了说明,介绍了RBF(Radial Basis Function),神经网络的算法并分析了其控制系统模型,利用其在线整定PID(Proportion Integration Differentiation)控制技术对系统控制器进行设计,并通过MATLAB仿真给出了系统控制模式的控制效果.经试验验证,控制算法控制效果良好,完全满足底盘测功机对控制品质的要求.