Data Processing Strategies in Short Term Electric Load Forecasting

At present, the support vector machine (SVM) has been successfully applied in the field of electric load forecasting, but most of the load forecasting models are based on the day characteristics of meteorological factors, without considering a real-time weather factors which are valuable real-time information, while the prediction accuracy and generalization ability are influenced by the sample set of input variables. This paper aims to propose a selection strategy based on a sample of real-time weather factors. Firstly, data processing methods deal with abnormal points, secondly, then we use the day meteorological feature vectors to reduce the sample set, and use FP-Growth algorithm to select the training sample similar to prediction day based on real-time weather factors, finally prediction model based on SVM is established. The practical application shows that the text of the prediction models and processing strategies can be more accurate predictions.