The vigorous construction of new power systems entails higher requirements for the efficient and accurate identification technology for power grid monitoring signals. This paper first analyzes the basic principles of the Soft-Masked BERT language model, and establishes a signal text error correction model based on Soft-Masked BERT. According to the typical information table of the State Grid, the rule dictionary of "signal semantics-grid events" in normal and fault conditions is analysed. Based on the above models, a power grid situation awareness model based on RNN is established, and a semantic analysis of power grid monitoring signals and a situation awareness solution process based on deep learning are proposed. Finally, taking the actual monitoring signal of a 110 kV substation as the test data, the proposed RNN model is used to analyze the semantic analysis and situation awareness simulation analysis of the monitoring signal of the power grid in this area by combining the Pycorector toolkit and the Pytorch software. The validity and correctness of the model are verified. This work is supported by the National Natural Science Foundation of China (No. 52077017).