Abstract: The speech recognition of power grid dispatching is an important part of intelligent virtual power dispatching, and the accuracy of the language model is directly related to the effect of speech recognition. A language model for speech recognition of power grid dispatching based on BERT is proposed. First, the principle of BERT model and the method of constructing a language model for speech recognition of power grid dispatching based on BERT are introduced. Then, according to the characteristics of power grid dispatching language, a method of extracting the semantic features, the keyword features and the named entity features from dispatching sentence input to the language model for speech recognition of power grid dispatching is proposed, which can improve the adaptability of the language model to the power grid dispatching language. Finally, the language model is tested by case studies and compared with other common language models. The results of the case studies show that the language model for speech recognition of power grid dispatching, which considers the characteristics of the power grid dispatching language, has obvious advantages in the performance of language model and the speech recognition accuracy of power grid dispatching.