Method for Data Retrieval Intent Recognition Targeting Complex Grid Control Operations

As the power system develops and the power grid scales up, power grid control becomes more complex. This paper proposes a data retrieval intent recognition method for the complex power grid control business. Through TF-IDF and SimHash algorithms, a high-quality word family is constructed, considering keyword frequency and importance in the corpus, while retaining similarity information between keywords. By building an associative database and performing TF-IDF+Word2Vec text feature extraction, the system establishes an associative graph between historical data and the word family, forming superior text feature vectors for robust intent recognition support. Finally, emoBERT and TextCNN algorithms are used for intent recognition and classification, swiftly and accurately identifying input text intent, providing crucial information for power dispatch decision-making. Experimental results demonstrate an accuracy of over 94.7% on the dataset, showing feasibility and effectiveness.