Research on Short-Term Load Forecasting Using K-means Clustering and CatBoost Integrating Time Series Features

Better understanding of the actual power consumption pattern of industrial customers is critical for improving load forecasting to enhance the operation and energy management of electric power systems. This paper introduced a hybrid model to forecast the short-term load function, which consists of K-means clustering and CatBoost model integrating time series features. K-means clustering is executed for partitioning industrial customers with similar load function into the same cluster. The improved CatBoost can be applied to forecast the load demand using small training dataset and overcome the local offset problem. The model is evaluated on the load data collected from Yangzhong High-tech Zone in China, and is compared with models based on Autoregressive Moving Average model, Long Short-Term Memory, Gradient Boosting Decision Tree and CatBoost, respectively. The results demonstrate that the proposed model outperforms all the other mentioned models.