The invention discloses a
data stream processing method and
system based on deep
reinforcement learning, and relates to the technical field of
industrial Internet of Things. Comprising the following steps: S1, constructing a
topological graph neural network, determining a feature similarity weight of equipment topological features through a multi-head graph attention mechanism, and dynamically aligning the equipment topological features of a source domain and a target domain; s2, setting a
main channel through a deep
reinforcement learning model, setting an auxiliary channel through a
time sequence contrast element learning model, and obtaining health degree evaluation and fault mode characteristics of the equipment; and S3, monitoring a data flow according to the health degree evaluation and the fault mode characteristics, and recombining a
network structure through a neural architecture search model. According to the method, the transfer
learning problem caused by equipment isomerism in the
industrial Internet of Things is solved, the physical distance, the process
coupling and the feature similarity can be dynamically balanced, the initial accuracy of the source
domain model in the target domain is improved, and the
annotation data volume required by the target domain is reduced.