基于SVM的复杂制造系统动态调度方法研究与应用(终稿).doc

硕士学位论文 基于SVM的复杂制造系统动态调度方法研究与应用 摘要 在制造业及自动化领域，复杂生产过程调度问题一直是研究的热点。复杂制造系统具有大规模、多目标、工艺复杂、不确定性等特点，合理选择调度策略对整个生产调度过程产生重要影响。 在给定的生产状态下，为了能快速较优地选择调度策略(即动态调度)，可以充分利用与调度相关的历史生产数据来挖掘出相应的调度策略。同时，为了提高调度的效率，应当对生产数据进行去冗余处理，即对生产属性进行特征选择。 本文研究了复杂制造系统动态调度策略选择方法，提出了一种基于SVM)作为数据挖掘工具，采用二进制粒子群优化算法(BPSO)对生产属性(特征)子集进行寻优，获得基于SVM的动态调度策略分类模型。在此基础上，对于任意给定的生产状态，通过该模型，能实时地获取当前生产状态下近似最优的调度策略。 另外，本文还对基于多目标的调度策略综合评价方法进行了研究，并分别应用功效函数法和信息熵法实现多目标决策评价和指标权重系数选择，并应用于调度策略选择方法的研究中。 最后，本文针对某实际复杂制造系统，对所提出的动态调度方法进行实验验证。研究结果表明，该方法很好地提高了调度效率，保证了调度的实时性和准确性。 关键字：动态调度，特征选择， SVM，参数优化，多目标决策评价 Abstract The scheduling of complex production process is popular in the field of manufacturing and automation. Characterized with large-scale, multi-objective and uncertain environments, the complex manufacturing systems should make reasonable choice of scheduling strategies that has great effect on their operational performance. In order to choose a better scheduling strategy quickly under a given condition of production (dynamic scheduling), it is a good way to make full use of on-line and off-line production data related with scheduling. Meanwhile, in order to improve the scheduling efficiency, some redundant data should be removed, namely a process of feature selection of production attributes. In this thesis, a dynamic scheduling strategy selection method for complex manufacturing systems is studied and a scheduling strategy selection approach based on production attributes is proposed. This approach based on historical data uses support vector machine (SVM) as a data mining tool and binary particle swarm optimization algorithm (BPSO) to optimize production attributes (feature subsets). Thus, the best scheduling strategy can be achieved under any given production status in real-time through a SVM classifier after optimization. In addition, the multi-objective comprehensive evaluation methods for scheduling strategies are studied in this thesis. In the study of the dispatching strategy selection approach, an efficacy funct