In recent years, with the rapid growth of electric vehicle (EV) ownership, accurate forecasting EV charging loads has become a crucial research topic for power grid planning and charging infrastructure optimization. To address the challenges of traditional forecasting methods in handling the complex nonlinear characteristics and dynamically coupled influencing factors in load data, this paper proposes a charging load forecasting model that integrates data preprocessing, variational mode decomposition-sample entropy (VMD-SE) data reconstruction, and a convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) deep learning framework. First, the Gaussian mixture model-K-nearest neighbor (GMM-KNN) method is utilized to detect and fill abnormal and missing values in the data, improving data quality. Next, VMD is employed to decompose the load data, and SE is applied to select and reconstruct important modal signals to extract multi-scale features. Finally, a hybrid deep learning framework combining CNN and BiLSTM models is constructed to capture local features and temporal dependencies for accurate forecasting. Experimental results demonstrate that the proposed method exhibits high accuracy and robustness in multi-season load forecasting, significantly outperforming traditional methods. This provides an effective solution for EV charging load forecasting.