基于最大熵神经网络算法的柔性制造系统调度策略研究  

Research on scheduling strategy of flexible manufacturing system based on maximum entropy neural network algorithm

作　　者：解明利[1] 胡占齐[1] 马宁 Xie Mingli;Hu Zhanqi;Ma Ning(College of Mechanical Engineering,Yanshan University,Qinhuangdao Hebei 066004,China;Dept.of Mechanical ＆ Aerospace Engine-ering,Politecnico di Torino,Torino 10129,Italy)

机构地区：[1]燕山大学机械工程学院,河北秦皇岛066004 [2]都灵理工大学机械与航天工程系,意大利都灵10129

出　　处：《计算机应用研究》2018年第12期3697-3700,3705,共5页Application Research of Computers

基　　金：河北省科技计划资助项目(15211823);国家自然科学基金资助项目(51775480)

摘　　要：为了解决传统神经网络BP梯度下降算法在解决柔性制造系统调度策略时易陷入局部最优的问题,在规则化神经网络结构的基础上,提出了一种基于最大熵的神经网络权值优化算法。利用神经网络隐层节点变量的条件概率,在计算寻优过程中,通过改变收敛算子求解熵函数的期望,进而迭代求解网络的最优权重向量。对比实验表明,相较BP梯度下降算法,采用最大熵权值调整算法,数据搜索空间范围大,能保证系统准确收敛到全局最优解,算法鲁棒性好。在实际的调度策略应用中,该算法能明显缩短整体生产任务的加工周期,达到提高企业生产效率的目的。To solve the problem that the traditional BP gradient descent algorithm is easy to fall into the local optimum on the scheduling strategy of flexible manufacturing system,this paper proposed a neural network weight optimization algorithm based on maximum entropy. In the process of optimization,the calculation for the expectation of the entropy function could get the optimal weight vector of the network by changing the convergence operator. The comparison experiment shows that,compared with the traditional BP gradient descent algorithm,the maximum entropy weight adjustment algorithm has a large range of data search space,which can guarantee the system to converge to the global optimal solution accurately and the algorithm has good robustness,in the actual application of scheduling strategy,the algorithm can shorten the processing cycle of the overall production task and achieve the purpose of improving the production efficiency.

关 键 词：最大熵 神经网络 柔性制造系统 调度策略 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]