The invention discloses a new method for performing fault prediction on a complex industrial process. The method comprises two stages of fault state
feature extraction and fault prediction. The fault state
feature extraction comprises the following steps: firstly, screening out features related to faults from data of a complex industrial process by using a
random forest algorithm; secondly, introducing a stack
noise reduction self-encoding network for feature reconstruction, then constructing a square prediction error (SPE) statistic to serve as a fault state feature, and by utilizing a
kernel density estimation method, determining a control limit; and finally, substituting new data into the model, calculating the statistical magnitude and judging whether the statistical magnitude is normal or not. The fault prediction comprises the following steps: forming a
time sequence by using the SPEs, and then realizing
trend prediction of the SPEs by using a prediction model of the SFTCN. According to the invention, the
random forest algorithm is adopted to reduce the training cost of the stack
noise reduction self-encoding network, the improved time
convolution network is adopted to effectively extract the
time sequence characteristics of the fault state, and the fault prediction precision is high.