The invention relates to a multi-time scale 
convolutional neural network soft measurement method based on an attention mechanism, and belongs to the technical field of soft measurement. The method comprises the following steps: 1, determining an auxiliary variable, carrying out 
data processing, selecting an easily-measured variable related to a difficult-to-measure parameter as the auxiliary variable of a soft measurement model, and collecting a 
time sequence of the auxiliary variable and the difficult-to-measure parameter; abnormal value 
elimination is carried out on the collected 
time sequence; 2, selecting an attention mechanism and an attention area, and dividing the attention area according to the time 
delay and the 
effective time scale of each auxiliary variable relative to the difficult-to-measure parameters; 3, constructing the input of a soft measurement model, forming a matrix by the 
time sequence of each auxiliary variable, and determining the input of the soft measurement model by combining the attention area of the attention mechanism; 4, establishing a time sequence 
convolutional neural network soft measurement model; 5, training a time sequence 
convolutional neural network soft measurement model; and 6, carrying out real-
time estimation on unmeasured parameters by utilizing the time sequence convolutional neural 
network model trained in the step 5.