Accurate short-term load forecasting is crucial for efficient power grid operation, especially in dynamic environments with evolving energy sources and consumption patterns. This challenge is often exacerbated by concept drift, where the underlying data distribution changes over time. This paper addresses these issues by proposing a comprehensive methodology for enhanced short-term load forecasting considering concept drift. The proposed approach leverages an ensemble learning strategy incorporating Long Short-Term Memory (LSTM) models. LSTMs are well-suited to capture temporal dependencies in data and adapt to concept drift. Additionally, we incorporate a feature selection mechanism to identify the most informative features for fore casting. This combined approach optimizes forecasting performance in dynamic grids. The proposed methodology is evaluated on diverse energy datasets. The results demonstrate its effectiveness in improving forecasting accuracy, highlighting the importance of concept drift adaptation, feature selection, and ensemble learning for achieving robust and reliable energy consumption forecasts.