基于灰关联与批量回归的水轮发电机温度数据重构改进OA

Aiming at the engineering needs of abnormal detection of hydro-generator temperature data,this study proposes a BPOD-IGRA-MLR-BIR cleaning framework based on multi-modal data fusion.The traditional threshold and statistics methods suffer from insufficient real-time performance and weak capability in multi-source data collaboration.Innovations of this framework include:(1)constructing a box plot outlier detection(BPOD)mechanism to rapidly identify obvious outliers through dynamic threshold calculation;(2)improved grey relational interpolation algorithm(IGRA)by introducing a time-weighted factor to optimize the calculation of relevance and enhance reconstruction precision of time series data;(3)developing a multivariate linear regression interpolation model(MLR-BIR)to establish a multi-point temperature nonlinear correlation model for collaborative interpolation.Verification using measured data from a hydropower plant showed the framework increased abnormal coverage to 98.7%,with a reconstruction error of±1.2℃,outperforming single methods in both sensitivity and reconstruction precision.Engineering practice shows this method effectively solves issues such as missing temperature data and abnormal interference in unit operation monitoring,providing highly reliable data support for health assessment and fault early warning.