For the purpose of accurately and efficiently extracting grid related value information from mixed big data, an automatic power grid data summarization algorithm based on long and short term memory network and artificial colony optimization algorithm is studied. Design the contextual information of the bidirectional LSTM learning target words, increase the attention mechanism, and extract the electric power category words and terms. The conditional random field model performs training tasks on embedded sequences to predict whether sentences can be classified into the category of electricity. With the support of the improved artificial clustering optimization algorithm, the problem of extracting power abstracted from big data is optimized, and the most valuable power related data is determined from the mixed big data. The proposed algorithm is validated based on actual grid data, and the results show that the proposed algorithm achieves good results.