This invention relates to the field of IoT device identification and
machine learning technology, and proposes an IoT device identification method and
system based on incremental
federated learning. The method includes: Step S1, each gateway node preprocesses and extracts features from network traffic to generate local feature vectors; Step S2, a single-layer self-organizing
incremental learning neural network is used to dynamically cluster the feature vectors, and prototype data is selected to construct a local prototype set; Step S3, each node uploads the prototype data to a central
server, which aggregates and distributes the global prototype set; Step S4, each node combines local data and global prototype data to perform local incremental training on the
global model and uploads the
model parameters; Step S5, the
server uses a federated averaging
algorithm to aggregate the
model parameters, generate a new
global model, and distributes it, repeating S4 to S5 until convergence; Step S6, if identification fails when a
new device connects, the local clustering model is updated and a new round of training is triggered, enabling the
global model to acquire the ability to identify new devices. The
system includes: multiple gateway nodes (including data preprocessing,
feature extraction, incremental clustering, and local training modules) and a central
server (including data aggregation,
model aggregation, and model distribution modules). Through the above scheme, the present invention achieves collaborative device identification that dynamically adapts to
new device access, avoids catastrophic forgetting, and protects data privacy.