Convolutional network insensitive to the input data variations

Question

I have a 1 D convolutional network that is designed to predict time series data. It has 11 layers, as it is basically designed like an AutoEncoder, with a bottleneck in between.