Compare several methods including LSTM, GRU, TCN
For more information about TCN, check https://github.com/philipperemy/keras-tcn
https://github.com/locuslab/TCN
HQQ.csv is the data measured each 30 min automatically in one location by online water quality monitor system
Water Quality Record.csv is the data measured daily in one location by lab
What we find:
1. When dataset is big enough, for the single variable time series prediction, TCN is obviously not as good as LSTM & GRU
2. LSTM/GRU get the best result when we set the batch size as 1, but when handling large dataset, that would need a long time to train.
In the field of deep learning, time series prediction has strong connection withe the problem of NLP. I generated some notes about current popular models like transformers, BERT, GPT, and more in Zhihu.com. Here are my link:https://www.zhihu.com/people/bei-luo-shi-men-84-2/posts