Introduction: The rapid proliferation of power internet of things (PIoT) devices has led to increasing protocol diversity and the emergence of unknown protocols, complicating the identification and management of large numbers of legacy terminals and challenging efficient PIoT operation. This study addresses these challenges by introducing an advanced classification method to improve the efficiency and accuracy of protocol identification. Methods: The dataset is first preprocessed, and 61 key features are selected using the ReliefF algorithm to improve the model performance. Thereafter, a stacking ensemble model is established by integrating three base learners. Lastly, the improved elite ant colony optimization (IEACO) algorithm is employed to optimize the parameters of the constructed stacking ensemble model. With this framework, the IEACO algorithm demonstrates its pivotal role by implementing a mechanism for dynamically adjusting the number of ants and integrating both elite strategy and adaptive weighting in the pheromone updates. Results: Experimental findings demonstrate that the proposed algorithm can reduce total computation time by 20% compared to the conventional elite ant colony optimization (EACO) approach. With the proposed IEACO optimization, the stacking model achieved an average accuracy of 96.06%, representing a 1.6% improvement over manual tuning. The F1-score is 2.54% higher than that of random forest, the leading base learner in the stack. Deployment of the method on an intelligent gateway device further demonstrates the model’s effectiveness, particularly in fine-- grained category-level evaluations. Discussion: The proposed method improves accuracy and efficiency in protocol identification for complex PIoT environments. However, further validation on diverse protocols and adaptation to limited labeled data remain important future directions. Conclusion: The proposed method enhances protocol identification for legacy PIoT terminals, thereby laying the foundation for subsequent intelligent management.
Wang et al. (Fri,) studied this question.