Short-term load forecasting of industrial customers based on SVMD and XGBoost

The electricity consumption by industrial customers in the society accounts for a significant proportion of the total electrical energy. Thus, it is of great significance for demand-side electrical energy management to develop an accurate method for short-term load forecasting for industrial customers. Unlike traditional load forecasting on system-level, the load forecasting of individual industrial customer is more challenging due to its significant volatility and uncertainty. We propose an adaptive decomposition method based on VMD and SampEn (SVMD) to decompose the raw load data into a trend series and a set of fluctuation sub-series, and then establish the corresponding prediction model (line regression model for the trend series and XGBoost regression model for each fluctuation sub-series). The hyper-parameters of XGBoost are optimized by bayesian optimization algorithm (BOA). Furthermore, relevant factors that affect the electricity consumption behavior of industrial customers are considered in order to further improve the accuracy of the hybrid method. The proposed method is tested in multiple scenarios with different industrial customers of China and Irish. The results show that the proposed model has significantly improved performance over the contrast models in state-of-the-art load forecasting.

Automated summary

An adaptive decomposition method based on VMD and SampEn (SVMD) is proposed for short-term load forecasting for industrial customers, combining XGBoost and linear regression models to establish prediction models. Relevant factors influencing industrial customers' electricity consumption behavior are considered to enhance accuracy, and test results show that the method significantly outperforms existing load forecasting models.