Nowadays, directional communication represents a high potential solution for Vehicle-to-Everything (V2X) communication in Beyond-Fifth-Generation (B5G) vehicular networks, using the millimeter wave (mmWave) technique. Due to the high mobility in the vehicular environment and to ensure a higher data rate to establish a reliable V2X communication link, an efficient beam alignment is needed to point the optimal direction between the moving vehicle and the mmWave base station. To deal with this issue, we propose in this paper a new beam alignment method to select the optimal beam angle of Departure (AoD) from the mmWave base station, used further to transmit data toward the moving vehicle directly. The novelty of our proposal concerns the suggestion of an original hybrid beam alignment approach, combining a 128-filters-based Convolutional Neural Network (CNN) and 4-layers based Bidirectional Long Short-Term Memory (BiLSTM). The selected angle is performed automatically for each vehicle position with the lowest error probability and the highest received signal power. The performance of our proposed CNN-BiLSTM model was evaluated using popular regression metrics such as mean squared error (MSE), mean absolute error (MAE), and root-mean-square error (RMSE). The results show that our model achieved values of at least 0.0107, 0.0765, and 0.103 for MSE, MAE, and RMSE, respectively, outperforming other machine learning algorithms such as KNN regressor, SVR, BiLSTM, and CNN-LSTM.