Objective Aiming at the increasingly complex coupling between urban peak power load and meteorological conditions, this paper explores the internal mechanism of peak load responding to real-time and cumulative meteorological factors.

Method Based on hourly load data and meteorological observations from workdays in Guangzhou from 2020 to 2025, an hourly-scale accumulated temperature effect was introduced. Correlation analysis and the extreme gradient boosting (XGBoost) modeling method were employed to quantify the complex synergistic effects and contribution differences of real-time and cumulative meteorological factors on peak loads.

Result The results indicate that the accumulated temperature effect, temperature, humidity, air pressure, and meridional wind speed are the critical meteorological drivers for peak loads in Guangzhou. The meteorological combination of high temperature, high humidity, pronounced heat accumulation effect, low air pressure, and weak wind speed is highly likely to trigger peak power loads. The effects of meteorological factors exhibit a significant temperature-dependent pattern. As temperature rises, the correlation between real-time meteorological factors and peak-hour power load follows a "rising then falling" trend, whereas the correlation between the accumulated temperature effect and peak-hour power load strengthens continuously. Seasonal XGBoost load forecasting models, constructed using hourly-scale features, achieved a mean absolute percentage error (MAPE) of less than 2.9% and an explained variance ratio (R²) exceeding 0.92 for all seasons. SHAP-based feature importance analysis for peak-period load forecasting indicates that the accumulated temperature effect and historical load are core factors across spring, summer and autumn. Real-time humidity plays a prominent role during the transitional seasons of spring and autumn, while winter load is primarily regulated by historical load inertia.

Conclusion This research supports optimized factor selection and dispatch strategies for climatically similar regions, while offering insights into load prediction methodologies that incorporate both the immediate and cumulative influences of weather factors.