基于协同训练半监督学习的干旱灾害天气预测
Drought Disaster Weather Forecast Based on Co-Training Semi-Supervised Learning

近年来，气象研究领域中存在大量高价值信息的无标签数据，然而对这些无标签数据进行高置信度标记是非常困难的，同时这些数据对于建立准确的气象预测模型又是十分重要的。基于此，本文研究了一种基于协同训练的半监督学习方法并用于干旱灾害天气分析预测，该方法使用重复标记策略，根据训练过程中的数据变化动态的计算置信度阈值作为约束条件，并提出了一种增强协同训练方法来评估气象领域中的无标签样本数据的置信度。为了评估所提出方法的性能进行了实验分析，结果表明，该方法的性能优于原始协同训练方法，有效地提高了分类器的分类精度，并验证了该方法用于干旱灾害天气预测的有效性和显著性。
In recent years, a large amount of unlabeled data with high-value information exists in the field of meteorological research. However, it is very difficult to mark these unlabeled data with high confidence, and these data are very important for establishing an accurate meteorological prediction model. Based on the situation, this paper studies a semi-supervised learning method based on co-training and used for drought disaster weather analysis and prediction. This method uses a repeated labeling strategy to calculate the confidence threshold dynamically according to the data changes during the training process as a constraint condition, and proposes an enhanced co-training method to evaluate the confidence of unlabeled sample data in the meteorological field. In order to evaluate the performance of the proposed method, experimental analysis is carried out. The results show that the performance of this method is better than that of the original co-training method, which effectively improves the classification accuracy of the classifier, and verifies the effectiveness and significance of this method for drought disaster weather prediction.