With the high access proportion of renewable energy and power electronic equipment, the control decision optimization and dispatching methods of microgrids are faced with great problems and challenges. China Southern Power Grid is transforming into an "energy value chain integrator". It is possible for the company to operate and maintain tens of thousands of microgrids, and the traditional model-driven, plan-based control, and manual dispatch mode will be difficult to meet the demand of optimal scheduling for microgrids. Faced with the demand of artificial intelligence in the field of microgrid automatic operation, an auxiliary decision-making method of optimal dispatching for miucrogrids based on deep learning is proposed in this paper. First, the typical mathematical programming model of dayahead optimization scheduling for microgrids is introduced and the difficulties and limitations of the model-driven modeling and solution methods are analyzed in this paper. Then, a deep learning model of dayahead optimization scheduling for microgrid based on deep bidirectional long-short memory neural network is established and the principle of revision and processing of the output of the model is given. Finally, the effectiveness of the model and algorithm in this paper is verified by an example analysis.