[ICASSP 2026] An Adaptive Sampling Method Based on Reinforcement Learning for Wind Power Forecasting under Extreme Weather

Authors:

Ruibo Guo, Mengjun Xu, Lei Liu1,⋆ Rui Yang, Hongwei Zhao, Tengyuan Liu, Bin Li1

1. Abstract

Accurate wind power forecasting (WPF) is essential for reliable energy management, especially during extreme weather events. However, existing WPF methods often compromise overall model performance when adapting to extreme events, as they tend to overfit to rare, extreme samples. To address this challenge, we propose an adaptive sampling method (ARS-WPF) based on reinforcement learning (RL), which enhances forecasting for extreme events without sacrificing accuracy under normal weather. Specifically, ARS-WPF formulates the selection of training samples as an RL problem, optimized using a novel fine-grained reward function. This reward encourages the selection of informative samples to enhance forecasting, penalizes overfitting performance and supports precise timing of crucial turbine shutdowns. Experimental results show that ARS- WPF achieves error reductions of 4.3% and 11.7% in MAE under cold wave and typhoon events, respectively.

2. Requirements

The version of python is 3.10. The version of torch is 1.13.1 .

torch==1.13.1
pytorch-forecasting==0.10.3
pytorch-lightning==1.9.0
scipy==1.13.0
numpy==1.24.2
pandas==1.5.3
sympy==1.12.1
einops==0.8.0
tqdm==4.67.1

3. Datasets

The dataset from Goldwind Cup Challenge is already placed in the datasets folder.

The data preprocessing code is provided in utils.py.

4. Usage

• an example for train and evaluate a new model：

bash run.sh

• You can get the following output:

[Epoch 14, Dataloader num 3630] Rewards = -0.2046, Val loss = 0.0807, Event Val Loss = 0.106390, 
Epoch 14: Train Loss = 0.031636, Val Loss = 0.080682, Event Val Loss = 0.106390, 
采样记录已成功保存至 ./typhoon_OurNetModel_1700_seed52.csv
test on normal dataset
acc:85.40%, MAE:11.91%, RMSE:14.60%, SMAPE:38.04%, MAPE1:92.79%, MAPE2:35.60%, r2:0.8073, pearson:0.9373, quality_rate75:0.9140, quality_rate80:0.8275, quality_rate85:0.6835  

test on event extreme dataset
acc:78.61%, MAE:17.68%, RMSE:21.39%, SMAPE:73.26%, MAPE1:293.48%, MAPE2:68.60%, r2:0.2837, pearson:0.8299, quality_rate75:0.7526, quality_rate80:0.6351, quality_rate85:0.4722  

模型已保存到: /data/guorb/pre/data_base/results/OurNetModel_1700_1_in_48_out_48_typhoon.pth

5. Citation

If you have any problems, contact me via [email protected].