Intelligent Question-Answering System and Inference Mechanism for Grid Dispatching with Energy Storage Access Scenarios Based on Semantic Models

INTRODUCTION: In the power grid energy storage dispatch system, after the large-scale integration of new energy storage systems, the demand for inquiries regarding the operation and maintenance of storage equipment as well as the coordinated power regulation has significantly increased. The intelligent question-answering function is of vital importance in enhancing the efficiency of dispatchers. OBJECTIVES:The research focuses on the field of scheduling, integrating expertise in power dispatching and energy storage grid connection operation with natural language processing technology, and has constructed an efficient question-answering generation framework. A semantic model optimized by dynamic masking and large-batch training is combined with a dispatching knowledge graph to enhance semantic understanding. METHODS: The model achieves an intent classification accuracy of 0.97 and an F1 score of 98.1%. The accuracy of terminology disambiguation and complex sentence analysis reaches 92.5% and 84.8%, respectively. Based on this framework and a reasoning mechanism, an intelligent question answering system is developed using multi-algorithm collaborative reasoning to support answer generation and optimization. RESULTS: Experimental results show that the system maintains an accuracy of 91.3% even at difficulty level 5. The violation rate of dispatching procedures is as low as 22.5%, with a maximum memory usage of 805 MB and a response time of 63.2 seconds. CONCLUSION: This system can automatically generate effective responses, optimize the accuracy and rationality of question-and-answer interactions, and precisely output compliant control plans for abnormal handling issues related to energy storage grid connection. It demonstrates significant advantages in enhancing response efficiency and accuracy.