The electricity consumption of large enterprises, apart from their inherent characteristics, is also influenced by various factors such as climate, economy, and international situations. By leveraging big data from more diverse and abundant sources (including historical electricity consumption data, financial data, weather data, holiday data, etc.), it is possible to identify relevant factors and establish higher-dimensional prediction models, thereby improving the accuracy of electricity consumption forecasts. This paper uses the electricity consumption prediction of an industrial chain in a chemical park as a case study. By employing a sliding window method based on big data random matrix theory, and identifying highly correlated influencing factors, we use XGBoost to build a high-dimensional electricity prediction model, effectively enhancing the prediction accuracy. This prediction method has significant application value in practical engineering.