基于多尺度时间窗口的核电运行数据关联性分析方法研究

Research on Correlation Analysis Method for Nuclear Power Operation Data Based on Multi-Scale Time Window

[目的]核电运行数据具有维度高、体量大等特点,而核电厂内部系统的复杂性导致难以构建相应的机理模型,因此依靠人工从核电数据中筛选出具有关联性的参数非常困难,而非关联性参数的混入将极大地影响模型精度,通过提高模型精度以达到精准建模的目的。[方法]文章提出了一种基于多尺度时间窗口的关联分析方法,该方法对目标参数进行状态切换点提取,依据不同传感器所记录数据的特点对各个传感器进行分类,再针对不同种类的传感器设计符合其特点的检测窗口,利用从目标参数所提取到的状态切换时间点,对各个传感器的相应时间邻域进行状态切换检测,计算各个传感器与目标传感器的关联匹配率来判断其关联性大小。[结果]利用真实的核电厂历史运行数据展开实验,通过建立的关联匹配率规则,成功地筛选出了与目标传感器具有关联性的传感器参数。[结论]实验结果表明,文章所提出的方法可以更为准确地筛选出关联性参数,与常用的皮尔逊相关系数相比,文章所提出的方法准确性更高。

[Introduction] Nuclear power operation data is characterized by high dimension and large volume, and the complexity of the internal system of nuclear power plant makes it difficult to build a corresponding mechanism model. Therefore, it is very difficult to manually screen out relevant parameters from nuclear power data, and the introduction of non-relevant parameters will greatly affect the accuracy of the model. By means of improving the model accuracy, the purpose of accurate modeling can be reached. [Method] This paper proposed a correlation analysis method based on multi-scale time window. This method extracted state switch points for target parameters, classifies each sensor according to the characteristics of the data recorded by different sensors, and then designs detection windows for different kinds of sensors that meet their characteristics. The state switch detection was carried out in the corresponding time neighborhood of each sensor, and the correlation matching rate between each sensor and the target sensor was calculated to judge the correlation. [Result] Based on the actual historical operation data of nuclear power plant, the sensor parameters associated with the target sensor are selected successfully by the established correlation matching rate rule. [Conclusion] The experimental results show that the proposed method can screen out the correlation parameters more accurately. Compared with the commonly used Pearson correlation coefficient, the proposed method is more accurate.