The topic of this course practice is multivariate time series forecasting based on sequence models. We use three methods to accomplish this task: LSTM, Transformer and Enhanced RNN. The relevant models and codes are in the corresponding folders.
The dataset used in this project is from [code]. Among them, we choose ETTh1.csv as our dataset. This dataset contains 14,400 entries in total, each of which contains 8-dimensional features, including date, the predicted value "Oil Temperature (OT)", and 6 different types of external load values. Train/Val/Test is splited into 6:2:2, which means that the training set has 8,640 entries, and the validation and test sets have 2,976 entries each.
We would like to thank Hujiang Huang and Zhiwei Zhu for their support and advices in our course project.
If you have any questions, please feel free to contact [email protected]. Have fun!