Abstract: In view of the phenomenon of seed blocking and uneven supply during the operation of quantitative seed feeding device, Xiuyou 5 super rice was taken as the research object to study the seed supply theory and seed yield prediction. Using Python as the algorithm framework, BP neural network, decision tree and XGboost algorithm model were used to predict the performance of vibrating rice sowing device. In order to verify the validity of the model, combined with the data of the last 14 times of the test set, the determination coefficient R2 and relative error were used as evaluation indexes to check the prediction accuracy of each model, and the optimal seed supply prediction model was obtained by comparative analysis. The results showed that the R2 of BP neural network model was 0.87 and the relative error was 18%, the R2 of decision tree model was 0.91 and the relative error was 11%, and the R2 of XGboost model was 0.95 and the relative error was 5%. Compared with the other two models, XGboost model predicted the seed supply with higher degree of fit and more significant prediction effect. It can provide the basis for determining the working parameters of the quantitative seed feeder and facilitate the producers to make scientific decisions.

Key words: super rice, quantitative seed supply, BP neural network, decision tree, XGboost

摘要： 针对定量供种装置工作时易发生堵种、供种不均匀等现象，以秀优5号超级稻为研究对象，进行供种理论及供种量预测研究。以Python为算法构架，应用BP神经网络、决策树以及XGboost算法模型对振动式水稻播种装置进行性能预测。为验证模型的有效性，结合测试集后14次数据，以决定系数R2和相对误差为评价指标，检验各模型的预测精度，对比分析得出最优供种量预测模型。结果表明：BP神经网络模型的R2为0.87，相对误差为18%，决策树模型的R2为0.91，相对误差为11%，XGboost模型的R2为0.95，相对误差为5%，较其他两种模型相比，XGboost模型预测供种量，拟合程度更高，预测效果更显著，可为定量供种器确定工作参数提供依据，以利于生产者科学决策。

关键词: 超级稻, 定量供种, BP神经网络, 决策树, XGboost