A Combined Quartile-Regression-Interpolation Data Preprocessing Strategy for Short-Term Wind Power Prediction

To address the characteristics that wind power is volatile and difficult to predict, firstly, the abnormal data in the original wind power data set are identified and removed by the quadrature-regression method; then the K-Nearest Neighbor method is used to interpolate the data set with the abnormal values removed to restore the temporal order of the data set; finally, a Long Short-Term Memory neural network is built and the actual wind farm data is used as an example for comparison and analysis. The results show that the described method can better remove the anomalous data from the original data set, and interpolate the data set with the outliers removed, and the prediction accuracy is effectively improved by using the data set processed with the outliers to predict wind power compared with the original data set.