基于PCA和多元回归算法构建大棚黄瓜霜霉病预警系统

Construction of Early-warning System of Greenhouse Cucumber Against Downy Mildew with PCA and Multiple Regression Algorithm

采用大棚气象条件检测系统记录了大棚黄瓜在不同种植季的环境条件变化,将其与不同季节的黄瓜霜霉病发病率的数据进行了相关研究。利用机器学习领域的主成分分析(PCA),使用50%的数据作为训练集,找出了2种权重较高的特征值,并针对其与霜霉病发病率进行了多元回归分析,计算出最佳的回归模型。利用该模型,对剩余50%的数据进行预测。结果表明,该降维分析和多项式回归得到的预警系统可根据环境检测数据,有效预测霜霉病的发病率,准确度达到85%,为黄瓜霜霉病的预防提供了重要的预警信息。

By using the environmental condition record system, data of greenhouse cucumber planting were obtained. Its relationship with the plant disease rate of downy mildew was analyzed. PCA dimension reduction algorithm in machine learning was used to treat 50% of the total data as training data to construct an early-warning system. As a result, 2-dimension matrix accompanied with a regression equation was confirmed effective to predict the disease rate at similar conditions. The predicting of the remaining infection rate indicated this method had accuracy over 85% compared to the recorded data. In general, this research provided a novel method on predicting the infection rate of downy mildew of greenhouse cucumber.