This study presents and evaluates one-day-ahead electricity price forecasting (EPF), using several hybrid deep learning (DL) models in the EPEX German electricity market. Several architectures were compared, such as Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), CNN and GRU (CNN-GRU) and CNN-GRU combined with Variational Autoencoder (CNN-GRU-VAE), with the intention to assess overall predictive performance and delineate the best model amongst the described short-term models applied to EPF. Each of the above hybrid models were evaluated using typical performance measures of Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). From all assessed models the proposed hybrid deep learning (CNN-GRU-VAE) model achieved the greatest predictive utility with the smallest MAE and RMSE values of 0.615 and 0.851 respectively. As an architecture, CNN-GRU-VAE model successfully integrated the main ideas from CNN spatial feature extraction/learning, GRU temporal learning and (VAE) probabilistic capabilities for integrative learning in EPF. Of note, both GRU and CNN-GRU models (without variational elements), produced independent strong predictive performance. In contrast, the ANN and CNN models, produced the largest error numbers, and characteristics of both models indicated limited ability to capture both non-linear and temporal complexities of predicting electricity prices. Overall the findings suggest advanced hybrid deep learning models may have a role in improving accuracy in EPF, and provide knowledge and new opportunity for research in deep learning-based energy forecasting.