(AIMS 2023) The Load Forecasting of Special Transformer Users Based on Unsupervised Fusion Model

Abstract. Recently, load forecasting based on machine learning and artificial intelligence has shown great promise. As an individual with high power load, the accurate load forecasting of the special transformer users can guide the planning of future power generation and transmission, which is the key to ensuring the safe, stable and efficient running of the power system. However, most forecasting models do not consider the actual power consumption data of users and the power load characteristics of different types of users, and their application practicability and stability are insufficient. In this paper, we present a load forecasting syncretic model based on user unsupervised classification. Firstly, in the unsupervised classification of users, multiple features and similarities are introduced at the same time to realize the one-time classification of special users. Secondly, with the time series clustering method as the core, the secondary classification of special users is realized. Finally, different time series forecasting models are used for the power load curve characteristics of different user categories. Our model has achieved the best results in multiple metrics (RMSE: 0.359, MSE: 0129, MAE: 0.079, CR: 0.881), which can effectively improve the stability and accuracy of the load forecasting of the special transformer users.