Residual life prediction method based on improved pulse separable convolution enhanced Transform encoder

The invention discloses a residual life prediction method based on improved pulse separable convolution enhanced Transform encoder, which is characterized in that an improved pulse separable convolution module converts a time sequence signal into a pulse signal to capture the mutual relationship among different sensor data, reduces the information loss in the encoding process, improves the network by using a substitution gradient method, and improves the prediction accuracy of the residual life of the sensor. And an ECA attention mechanism is added to re-calibrate the pulse feature map, and useful pulse features are highlighted. And then, comprehensively controlling the dependency relationship of the fault features in each time step on the long time span of the whole life cycle by adopting a multi-head self-attention mechanism mode on the basis of the effective features. And finally, a final RUL result is obtained through an improved Transform encoder algorithm and an MLP (Markup Language Protocol) linear lay