摘要
为解决在智能化生产预警方法应用的过程中原始数据维度高、数据结构复杂、数据量大的问题,提出了基于关联度分析的生产异常模式挖掘方法。该方法建立了预警目标与影响特征之间的关联关系,通过计算关联度筛选出重要特征。在均值化方法处理数据的过程中,通过引入时间序列、选取时间粒度来截取距离数据,通过计算关联度、摒弃无效影响特征和降低数据维度来完成数据的准备过程。结合损耗性异常的业务数据特点,采用了基于时间序列的G-R分段拟合方法拟合数据,并利用均方根误差方法校验模型的准确性。实验验证选取了三次采油生产的异常情况为实例,采用G-R模型对特征集的元素进行分段拟合以求解相关参数。实例验证结果表明,该方法的预测数据与原始观测数据的吻合度高,且预测准确度较高。
In order to solve the problem of high original data dimension,complex data structure and large data volume in the process of application of the intelligent production alarming method, a mining method of abnormal production pattern based on relevancy analysis is proposed. It establishes the incidence relation between early warning target and influential characteristics and screens out important features through relevancy calculations. In the process of data processing by equalization method the distance data is extracted by introduction of time series and selection of time granularity and preparation process of data is completed by calculation of relevancy, elimination of inva- lid influential features and reduction of data dimension. Combined with the data characteristic of abnormal loss, the G-R segmentation fit-ting method based on time series to fit the data and root mean square error method to verify the accuracy of the model. In the process of experimental verification, the abnormal situation of tertiary recovery production is taken as an example and the G-R model is adopted to carry on segmentation fitting towards the elements of the feature setting for solution of relevant parameters. The experimental results show that the proposed method agrees well with the original observation data, and its prediction accuracy is high.
出处
《计算机技术与发展》
2017年第9期124-128,132,共6页
Computer Technology and Development
基金
黑龙江省自然科学基金面上项目(F2015020)
黑龙江省教育科研规划重点课题(GJB1215013)
黑龙江省2016年教育科研课题(16Q117)
关键词
特征筛选
时间序列
函数拟合
关联分析
feature selection
time sequence
function fitting
relevancy analysis
