Research on Burrs Processing Method in Load Data and Electric Load Forecasting

Electric load forecasting is a very important task, but there are often many abnormal data in the load data (Burrs). This paper proposes a load forecasting method in view of the large number of burrs existing in load forecasting. We first used the preprocessed load data to cluster the courts and got the 7050 and the 3033 these two categories (7050 and 3033 are the numbers of the two categories respectively, here we use the numbers as their indexes). Next, we use two methods the sliding box filter method and the comparison method to remove burrs. After extracting the features, we use XGBoost and LightGBM for load prediction. Finally, we analyzed the courts with large prediction errors.