Accurate power load forecasting is crucial to grid security and economic dispatch. However, the forecasting process is influenced by multiple factors including temporal, meteorological, and economic variables. This leads to high-dimensional data with significant redundancy, which increases complexity and reduces forecasting precision. To address this issue, a hybrid forecasting method that integrates optimal feature selection with a combined model of Attention-Bi-LSTM and XGBoost was proposed, referred to as the "ABL+XGB" model. This approach considered 27 influencing factors, employed 6 feature selection methods to identify the optimal feature subset, and utilized the proposed "ABL-XGB" model for prediction. Experimental results based on datasets from the Singapore and Norway electricity markets demonstrate that the proposed hybrid forecasting model has a high accuracy, and the prediction results of Lasso regression method combined with the prediction model are closer to the real data and the error is smaller than other feature selection methods, which indicates that a good feature selection method is essential for achieving accurate power load prediction.