基于自动机器学习的辽宁地区雷雨大风天气预测
Thunderstorm Gale Weather Prediction in the Liaoning Area Based on Automatic Machine Learning

针对辽宁地区雷雨大风天气的不确定性和时空差异性的特点，本文提出了一种基于自动机器学习的雷雨大风天气预测方法。首先由历史再分析数据集和地面实况数据集构建了需要的雷雨大风数据集；其次对经过预处理后的数据进行特征工程；然后使用基于多层堆栈集成、重复k-折交叉装袋策略的AutoGluon自动机器学习方法建立雷雨大风预测模型。最后，通过实验结果表明，使用AutoGluon方法构建的最佳模型在多项评估指标中，命中率为96.72%，漏报率为0.46%，误报率为1.62%。
A method for predicting thunderstorm and gale weather in Liaoning area using automatic machine learning to address uncertainties and spatial-temporal differences is proposed in this paper. To begin with, we construct the necessary dataset for thunderstorms and gales using both the historical reanalysis data set and the ground live data set. Then we perform feature engineering on the preprocessed data. After that, we establish the thunderstorm gale prediction model using the AutoGluon automatic machine learning method, which is based on multi-layer stack integration and a repeated k-fold cross-bagging strategy. Finally, the experimental results show that the best model constructed by the AutoGluon method has a hit rate of 96.72%, a false negative rate of 0.46%, and a false positive rate of 1.62% in several evaluation indicators.