Electricity Consumption Behavior and Load Forecasting Analysis Coupling Meteorological Factors Using BK-Means and NRBO-XGBoost Algorithms

To address the challenges of extracting user electricity consumption behavior features and insufficient load prediction accuracy in multi-energy coupling scenarios, this study proposes an electricity behavior analysis and forecasting methodology integrating data cleansing with meteorological correlations. Firstly, the Akima interpolation method is employed to rectify abnormal load data points, combined with a highly robust Z-M-ESD algorithm (Z-score Median-based Extreme Studentized Deviate) incorporating median identification and seasonal adjustment for iterative data cleansing, achieving an average 64.765% reduction in outlier correction errors. Secondly, BIRCH pre-clustering is utilized to adaptively determine optimal cluster numbers and initial centroids, thereby improving the traditional K-Means algorithm for joint meteorological clustering analysis of wind-photovoltaic power outputs and load coupling, as well as user clustering analysis. This enhancement elevates the user classification silhouette coefficient to 0.4679, representing a 24.04% improvement over conventional methods. Finally, an NRBO-XGBoost based load forecasting model incorporating meteorological parameters such as temperature and humidity is developed. Experimental results demonstrate that the proposed approach reduces the root mean square error to 49.2 kW and mean absolute error to 38.6 kW when considering meteorological factors. This methodology provides a theoretical foundation for demand-side management in power systems.