The invention relates to a pavement service performance self-optimization prediction method, which comprises the following steps: acquiring pavement vibration data, extracting vibration features, constructing a
time sequence feature
tensor as model input, and constructing a feature decoupling multi-head deep
network model by taking a vibration
time domain root mean square as model output; dividing the current pavement vibration data into a teacher stage and at least one increment stage, calculating a comprehensive drift
score of a single vibration feature, and calculating a drift
score of a
single sample based on the comprehensive drift
score; a teacher stage data training model is used as a basic teacher model,
incremental learning is carried out by using data in an incremental stage to obtain a student model, and a
loss function of
incremental learning is combined with a drift score of a
single sample for construction; outputting a prediction result by using the student model; and when new data is accumulated, taking the student model as a new basic teacher model, and carrying out
incremental learning by utilizing the accumulated new data so as to realize continuous evolution of the model. Compared with the prior art, the method has the advantages of long-term and accurate prediction and the like.