SlienceG Team Solution for KDDCUP2022

for each turbine, we have 3 models:
1 for lightgbm(namely lgn)
1 for first 144 timestamps prediction(namely GRU-FH)
1 for 288 timestamps prediction(namely GRU-ALL).
more information about KDDCUP2022

model training:

before training, you need to upload the dataset into folder LGB_train and GRU_train.

for LGB_train:

1. modify the file prepare.py to your path.
   
2. data proprecess

python datapreprocess.py #Bash

3. LGB Train

python train_split_smooth.py #Bash

for GRU_train:

1. modify the file prepare.py to your path.
   
2. if you want to train GRU-ALL, set the output_len 288,
   elif you want to train GRU-FH, set the output_len 144.
   
3. model training

python train.py #Bash

model testing:

before testing, put the pretrained model into folder checkpoints of prediction

1. modify the file prepare.py and predict.py to your own model path.
   
2. submit the prediction folder for online test.
   

contact:

Any problems please contact me at jackie64321@gmail.com