The digital transformation of the power industry generates vast technical documents requiring efficient text summarization. Existing methods face challenges in computational cost and hyperparameter optimization, as large language models demand substantial resources while relying on manual tuning. This paper proposes a parameter-efficient approach combining LoRA with particle swarm optimization for Chinese power-domain text summarization. Our method applies LoRA to BART’s key layers to reduce trainable parameters, while introducing PSO and QPSO algorithms for automated hyperparameter optimization. Experiments on the text summarization dataset show our approach outperforming full fine-tuning and manual tuning baselines. Furthermore, power-domain case studies validate the method’s practical applicability and domain adaptation capabilities.