Abstract: The energy structure in rural areas is becoming increasingly diversified, resulting in load demands that are highly seasonal and unstable. To better adapt to such load variations and ensure the stable operation of the power system, this paper proposes a load pattern switching detection and model calibration approach based on the CEBoosting algorithm. By introducing a causation entropy enhancement strategy, the method can effectively identify transition points in user load patterns and dynamically calibrate and update the online prediction model based on the latest consumption states. The algorithm computes a robust cumulative causation entropy using multiple short-time-batch data streams, thereby improving the accuracy of load pattern switching detection. Once a state change is identified, the model parameters are further adjusted using prediction residuals to enhance both the adaptability and accuracy of the model. Comparative experiments conducted on a rural electricity load dataset demonstrate that the CEBoosting algorithm can effectively detect load pattern switching and significantly reduce prediction errors (MAPE and RMSE), outperforming conventional time series forecasting models in both adaptability and stability. The proposed method offers a feasible solution for addressing complex load variations in rural power.