欢迎访问《农学学报》，
农学学报

农学学报 ›› 2026, Vol. 16 ›› Issue (5): 85-91.doi: 10.11923/j.issn.2095-4050.cjas2025-0036

• 农业信息 农业气象 • 上一篇    下一篇

基于BP神经网络的玫瑰盛花始期气象预报

张瑶1,2(), 王夏添2, 梁海瀚3, 张佳桐2, 王琪珍1()   

  1. 1 济南市气象局， 济南 250102
    2 平阴县气象局， 山东平阴 250400
    3 商河县气象局， 山东商河 251600
  • 收稿日期:2025-02-26 修回日期:2025-10-08 出版日期:2026-05-20 发布日期:2026-05-15
  • 通讯作者:
    王琪珍，女，1967年出生，山东莱芜人，正研级高级工程师，本科，主要从事生态与农业气象研究工作。E-mail：。
  • 作者简介:

    张瑶，女，1997年出生，山东临沂人，助理工程师，硕士，研究方向：生态与农业气象。通信地址：250102 山东省济南市历城区舜华南路1806号 济南市气象局，Tel：0531-67719205，E-mail：。

  • 基金资助:
    山东省气象局特色农业气象专项平阴玫瑰气象服务关键技术研究(2023SDBD10)

Weather Forecast in Initial Stage of Full Flowering of Rose Based on BP Neural Network

ZHANG Yao1,2(), WANG Xiatian2, LIANG Haihan3, ZHANG Jiatong2, WANG Qizhen1()   

  1. 1 Jinan Meteorological Bureau, Jinan 250102
    2 Pingyin Meteorological Bureau, Pingyin, Shandong 250400
    3 Shanghe Meteorological Bureau, Shanghe, Shandong 251600
  • Received:2025-02-26 Revised:2025-10-08 Online:2026-05-20 Published:2026-05-15

RichHTML

1

PDF (PC)

2

摘要：

研究旨在探寻更精确有效的玫瑰盛花始期预报方法。基于1994—2024年山东省平阴县平阴玫瑰物候资料和气象资料，分析盛花始期的时间变化特征，并通过皮尔逊相关系数筛选出影响花期早晚的关键气象因子，用于BP神经网络法建立花期预报模型，并与逐步多元线性回归法作比较，利用均方根误差(RMSE)、相对误差(RE)和决定系数(R2)对模型预报精度进行评价。结果表明，1994—2024年平阴玫瑰盛花始期呈波动提前趋势，平均每10 a提前0.4 d。1994—2020年有16个气象因子与盛花始期日序数呈极显著相关(P<0.01)，其中4月中旬的温度因子是影响盛花始期的主要气象因子。BP神经网络模型训练集的RMSE为0.75 d，RE为0.62%，R2为0.92，平均绝对误差为0.44 d；逐步多元线性回归模型的RMSE为1.31 d，RE为1.08%，R2为0.77，平均绝对误差为1.04 d。两种模型均能在4月下旬对玫瑰盛花始期进行预报。通过2021—2024年的资料对模型预报效果进行检验，其中，BP神经网络模型预报值与实际值相一致的年份占75.0%；逐步多元线性回归模型预报值与实际值一致的年份占25.0%。综上所述，BP神经网络模型较逐步多元线性回归模型的预报精度更高，在平阴玫瑰盛花始期预报中具有更高准确性和实际应用价值。

关键词: 玫瑰, 盛花始期, 气象预报, BP神经网络, 逐步多元线性回归

Abstract:

In order to explore a more accurate and effective method for predicting the initial stage of full flowering of rose, based on the observation data and meteorological data in Pingyin County, Shandong Province from 1994 to 2024, the time changing trend of the initial stage of full flowering of rose was analyzed. The key meteorological factors were selected through correlation analysis, which was used to establish the prediction model by BP neural network, and was compared with stepwise multiple linear regression method. Root mean square error (RMSE), relative error (RE) and coefficient of determination (R2) were used to evaluate the prediction accuracy of the model. The results showed that the initial stage of full flowering of rose was advanced in 1994-2024, with an average advance of 0.4 days per 10 years. From 1994 to 2020, 16 meteorological factors were significantly correlated with the ordinal number of the initial stage of full flowering (P<0.01), among which the heat condition in mid-early April was the main meteorological factor affecting the flowering period. The RMSE, RE and R2 of the training set of BP neural network model were 0.75 d, 0.62% and 0.92, and the mean absolute error was 0.44 d. The RMSE, RE and R2 of the stepwise multiple linear regression model were 1.31 d, 1.08%, 0.77, and the mean absolute error was 1.04 d. Both models can forecast the initial stage of full flowering of roses in late April. Data from 2021-2024 were used to verify the prediction effect of the model. The years in which the forecast values of the BP neural network model were consistent with the actual values accounted for 75.0%; the years in which the forecast values of the stepwise multivariate linear regression model were consistent with the actual values accounted for 25.0%. In summary, BP neural network model has better prediction effect than stepwise multiple linear regression model, and has higher reliability and application potential in the initial stage of full flowering of rose forecast.

Key words: rose, the initial stage of full flowering, meteorological prediction, BP neural network, stepwise multiple linear regression

中图分类号: