A cementing formula intelligent algorithm based on a large model and RAG technology

Abstract

The application discloses a cementing formula intelligent algorithm based on a large model and RAG technology and belongs to the field of recommendation methods. The application comprises the following steps: acquiring cementing slurry formula test data and constructing a cementing slurry formula database; extracting key features of each cementing slurry formula in the cementing slurry formula database; determining well conditions based on the key features of each cementing slurry formula; pre-processing the extracted key features and converting feature values thereof into numerical types; texturing the key features in numerical types, converting the same into mark sequences through a word segmenter, generating hidden states, and obtaining formula embedding vectors after pooling processing; based on the formula embedding vectors, constructing an index based on L2 distance by using a FAISS library; searching for top-k formulas with the highest similarity after inputting well conditions; and generating a new cementing slurry formula based on the searched similar formulas and the determined well conditions.

Images