Classification prediction | MATLAB realizes time series classification prediction of TCN time convolution neural network

Classified forecast | MATLAB Realization TCN Time series classification prediction of time convolution neural network

Basic introduction

The potential benefit of using convolutional networks is better parallelism 、 Better control of receptive field size 、 Better control the memory occupation and more stable gradient of the network during training . Like a circular network , Convolution networks can operate on variable length input sequences , It can be used to model sequence to sequence or sequence to task .

Model describes

TCN The main building block of is the causal convolution of expansion , It runs on the time step of each sequence . under these circumstances ,“ Causal relationship ” It means that the activation calculated for a particular time step cannot depend on the activation from the future time step .
In order to build the context from the previous time steps , Multiple convolution layers are usually stacked on top of each other . In order to achieve a large receptive field size , The expansion factor of subsequent convolution layer increases exponentially , As shown in the figure below .

• （1）TCN They may not have such strong adaptability in transfer learning . This is because in different fields , Historical information required for model prediction