Due to domain mismatch in far-field environments, far-field automatic speaker verification (ASV) shows inferior performance. To address this issue, we propose a far-field ASV method based on adaptive feature alignment (AFA-FSV). Initially, this method employs RepVGG model to extract general speaker features separately from both the source and target domains. To accurately capture the overall distributions of the source and target domains while further considering the paired distances of the same speaker features between the two different domains, a speaker embedding domain distance metric method is designed. It adaptively allocates different weights to generic speaker features based on the magnitude of distribution differences between the source and target domains, thereby aligning the deep features of the same speaker across domains to achieve domain-invariant feature extraction. Finally, precise speaker classification is performed based on the extracted domain-invariant features. Experimental results on the HI-MIA dataset demonstrate that AFA-FSV outperforms other competitive domain-adaptive methods.