With the rapid development of the power Internet of Things (IoT) system, it has also brought about a more complex protocol adaptation process, posing more challenges to the power system. In this context, feature matching technology under the background of machine learning has increasingly become an important component in the field of protocol adaptation. The IoT protocol adaptation technology based on feature matching can automatically extract features from different network protocols and achieve optimal matching of features, thereby improving the accuracy of protocol adaptation in the entire IoT system. This paper proposes a feature-matching-based power IoT protocol adaptation framework. The framework processes and extracts the original features of IoT protocols through a fine-tuned Transformer model, and uses the Support Vector Machine (SVM) algorithm to match the core features with the protocols, enabling the final adaptation of the protocols with a predefined protocol feature library, greatly enhancing the accuracy and flexibility of power IoT protocol adaptation. The experimental results demonstrate that the accuracy has been achieved up to 89.9% on the PAWS-X dataset and 96.9% on the IoT-23 dataset, indicating a significant improvement in both the accuracy and efficiency of the power IoT protocol adaptation under this framework.