What kind of features can you extract from a single time series to use it with neural networks?

I'm new to ML, and I'm trying to get my head around feature extraction... So if I have a time series of >1000 numbers, what kind of features I can extract from it to help improve a neural network prediction?