To improve the accuracy of extracting complex and multi-level causal relationships in aquaculture disease prevention and control, this study introduces a Bidirectional Focus Enhancement Mechanism and proposes the RoBFM model. The model integrates the pretrained language model RoBERTa-wwm-ext-large to generate high-quality word embeddings and utilizes a bi-directional long short-term memory network to enhance the modeling of long-range dependencies. The core innovation of RoBFM lies in the BFM, which explicitly focuses on and enhances the features of "cause" and "effect" entities in causal relationships through a unique design of forward and backward weight allocation. This effectively addresses the shortcomings of traditional methods in handling complex causal relationships. Experimental results show that the model outperforms existing models on metrics such as the F1 score on both the public dataset DuIE and the self-built domain-specific dataset DLOU-CRE. Particularly on the DLOU-CRE dataset, the F1 score reached 68.23%, fully validating the effectiveness of this method for the task of causal relationship extraction in aquaculture disease prevention and control, as well as its superiority on domain-specific data.