The power dispatching event knowledge graph can provide comprehensive historical dispatching knowledge to assist dispatchers in decision-making. However

the complex logical structure of power dispatching texts

combined with data sparsity and few-shot scenarios

makes event argument extraction challenging

which severely hampers the construction of knowledge graphs. To address these issues

this paper proposes a novel multiple structural schema instructor- universal information extraction (MSSI-UIE) framework for power dispatching event extraction. First

the UIE's original generative architecture is improved with the bidirectional encoder representations from transformers (BERT) model

integrating power domain-specific term masking pre-training and dual-pointer network to achieve precise event argument localization. Then

a DeepSeek-based text data augmentation approach is developed to effectively expand the training corpus. Finally

three types of structural schema instructor (SSI) are designed to collaboratively extract event arguments from multiple semantic dimensions. Experimental results demonstrate that this method exhibits strong performance in data-sparse and few-shot scenarios

effectively enhancing the efficiency and accuracy of power dispatching event extraction.