基于粗糙集与最小二乘支持向量机的密闭鼓风炉故障诊断方法分析-控制科学与工程专业论文.docxVIP

摘要密闭鼓风炉是铅锌熔炼过程中的主要设备之一，其炉况稳定对整 摘要 密闭鼓风炉是铅锌熔炼过程中的主要设备之一，其炉况稳定对整 个熔炼过程的安全运行及提高铅锌产量起重要的作用。因此保持密闭 鼓风炉炉况稳定，及时诊断并排除故障对于确保铅锌熔炼过程的稳定 运行及提高企业的经济效益都具有重大意义。 论文在分析铅锌熔炼密闭鼓风炉的工艺过程及相关的故障诊断 理论的基础上，为了从铅锌熔炼过程数据中获得故障特征信息，针对 铅锌熔炼过程中数据信息量大，故障诊断存在大量冗余信息的特点， 将粗糙集理论应用于密闭鼓风炉故障诊断。首先采用基于属性依赖度 的遗传属性约简算法及启发式属性值约简算法求得最简约简，去除冗 余知识，然后运用混合策略方法提取诊断规则，最后将得到的诊断规 则应用于密闭鼓风炉故障诊断。实例说明该方法具有一定的诊断效 果。 针对采用单一粗糙集理论进行故障诊断时容错性较差的不足，论 文研究基于粗糙集理论与最小二乘支持向量机相结合的密闭鼓风炉 故障诊断方法。首先用粗糙集对故障信息进行预处理，然后用最d-- 乘支持向量机对预处理后的故障信息进行学习曾ll练及分类诊断。该方 法不仅充分发挥粗糙集对知识的约简能力和支持向量机优良的分类 能力优点，同时克服最小二乘支持向量机对冗余知识和有用知识识别 的局限性，有效地降低支持向量机的输入信息空间维数，弥补粗糙集 方法对输入信息中的噪声较敏感，抗干扰能力差的不足，进而达到提 高诊断准确性的目的。最后通过实例验证，结果表明了该诊断方法的 有效性。 关键词：故障诊断，密闭鼓风炉，粗糙集，最小二乘支持向量机 ABSTRACTImperial ABSTRACT Imperial Smelting Furnace(ISF)is one of the major equipments in Pb—Zn smelting process，whose stability plays all important role in the smelting process．Therefore keeping the ISF situation stability, diagnosing and eliminating the fault in time is significant to guarantee the smelting process stable and enhance economic emciency． Based on analyzing the technical process and faults mechanism of the ISP(imperial smelting process)，and obtaining fault information from process data，the Rough Set(RS)is applied in data reduced and rules generated for the data information content and redundant information is large in IS P．Firstly,The attribute reduction method based on genetic algorithm with dependability and heuristic attribute value reduction use to remove the redundant information．and then mixed strategy method iS used to extract diagnosis rules．finally the fault of ISF iS diagnosed．The example proves that this method has a certain diagnosis． Against the fault tolerance of RS in fault diagnosis．the fault diagnosis method based on RS and LS．——SVM(Least Squares Support Vector Machine)iS studied．Firstly,the fault information is reduced and the fault rule iS generated on RS．Then the information iS diagnosed by LS SVM．This method．not only fully develops the reduction ability of RS and the h