A statistical learning framework for the intelligent imputation of offshore wind farm missing SCADA data

The Supervisory Control and Data Acquisition (SCADA) system collects operational data of wind turbines in a wind farm and monitors operational conditions of them. Yet, the SCADA system is often affected by various factors, such as the sensor failure and network congestion, resulting in the data loss. In this paper, a statistical learning framework is developed to intelligently impute the SCADA missing data of an offshore wind farm, which is of a high dimension and a long time span. The framework mainly consists of three parts: (1) the data pre-processing based on correlation analysis; (2) the model training and verification based on a novel integration of the feature selection, candidate models, and virtual sensor principles; as well as (3) the missing data estimation and data post-processing. The experimental results demonstrate the effectiveness and generality as well as the industrial application value of the developed statistical learning framework.