The invention relates to a voice recognition method for power grid dispatching, which comprises the following steps that: a voice input interface is configured in a log system, input voice is acquired, the voice is recognized and converted into characters, the characters are recognized, an electric power semantic library of dispatching operation logs is established, and semantic understanding is automatically performed on the recorded logs by utilizing machine learning; after semantic understanding is carried out on the log, different voice commands are generated according to different scenes, a running event short message is automatically sent according to the voice sending commands, and an addressee number is pre-established in a database; post-processing is performed on the operation event by utilizing machine learning in combination with a knowledge graph in the electric power semantic library, and a scheduling command is generated and issued to the outside; and through incoming call voice collection, character conversion and semantic understanding, machine learning is combined with an electric power semantic library to automatically understand an incoming call intention, so that a dispatching command short message is automatically generated and sent by combining different templates according to different scenes. According to the invention, automatic scheduling can be realized, and the problem of insufficient scheduling personnel is solved.