Based on a variety of heterogeneous data of energy enterprises, this paper provides an intelligent search method based on the Bert preprocessing model. This paper provides a method of intention recognition based on the combination of template matching and text classification, which involves machine learning, deep learning and other fields of artificial intelligence. First, the preprocessing model Bert is used to rewrite the natural language into a vector based on syntax. Then, the information extraction technology is used to extract the structured and machine understandable information from the problem to provide parameters for intention processing. The intention recognition technology used in this paper first uses template matching method, and then uses text classification method. Finally, the effects of various methods are compared through experimental examples.