Abstract: With the gradual opening of the day-ahead markets under the background of double carbons, a lot of new energy generation grid connection increases the risks of bidding decision-making of Generation Companies (GENCOs) obviously. This paper regards the new energy producer as one of the price makers and considers that the uncertainty of new energy output brings about extra costs to the producers, thus a bidding game model for the producers is established and an intelligent data-driven decision-making method based on deep learning for the GENCOs is proposed. By studying the historical market transaction results, this method is able to solve the optimal bidding without referring to the GENCOs' private information such as their costs. Firstly, the optimal bid labels in the historical data are selected by using the Nash equilibrium conditions. These labels are treated as the training data for the data-driven model with load power, weather and other factors. Then, a deep learning model which combines the Auto Encoder (AE) with the Long/Short-Term Memory (LSTM) network is proposed. Since there are few bidding samples in the new energy market, which is unable to meet the training requirements of deep learning, the LSTM network is allowed to pre-learn with a large amount of unlabeled data and is fine-tuned with some labeled samples. After the training, the LSTM model makes the decision online and submits the optimal bidding of GENCOs to the independent system operators for clearing. Finally, the simulation results show the effectiveness of the proposed data-driven model in different scenarios.