Accurate prediction of production decline in carbonate gas reservoirs is crucial for effective management and sustainable development of natural resources. Traditional prediction methods often perform poorly when handling complex production data, failing to capture the intricate trends and variations in reservoir production decline. To address this issue, this paper proposes a carbonate gas reservoir production decline evaluation model based on the Transformer algorithm. As a powerful deep learning model, the Transformer algorithm is renowned for its excellent sequence modeling capabilities and efficient capture of long-term dependencies, enabling more accurate prediction of the production decline trends in carbonate gas reservoirs. Experimental results show that the Transformer-based model not only excels in prediction accuracy but also surpasses traditional methods in stability and long-term prediction capabilities. This finding highlights the potential of deep learning in analyzing complex time series data, especially in the application prospects of resource development management. The proposed model is expected to provide more scientific and reliable technical support for the sustainable development and management of carbonate gas reservoirs.