Time series prediction method supporting data using convex hull knowledge distillation technology.

The data includes KDConv model training scripts, experimental configuration files, prediction result files, and raw evaluation data to ensure that all experiments are reproducible and the results are verifiable The dataset used in this paper's experiment is: ① ETT (Electric Transformer Temperature): divided into hourly level dataset (ETTh) and 15 minute level dataset (ETTm), which records the load characteristics of seven types of oil immersed and power transformers from July 2016 to July 2018. ② Exchange_rate: This dataset collects daily exchange rate information for 8 countries from 1990 to 2016. ③ Weather: Contains 21 weather indicators such as air temperature and humidity, collected in 2020 and recorded every 10 minutes. ④ Electricity: 321 customers' hourly electricity consumption was collected from 2012 to 2014. ⑤ Traffic: describes the road occupancy rate. It contains hourly data recorded by sensors on San Francisco highways from 2015 to 2016. It should be noted that the training/testing set partitioning of all datasets in this experiment adopts a time continuous segmentation strategy, without artificially introducing distribution mutations. Therefore, the main purpose is to verify the effectiveness of the method under stable distributions