基于自适应增强的BP模型的浙江省茶叶产量预测

本文采用1999—2018年浙江省59个县市的茶叶产量数据和地面气象要素驱动数据,提出了基于产量等级因子的自适应增强的反向传播（BP）神经网络模型的茶叶产量预测机制。首先分析提取了种植面积、年平均气温、3—7月的平均相对湿度、年平均相对湿度等11个影响因子,然后构建浙江省茶叶产量预测模型。试验结果表明,基于产量等级因子的自适应增强的BP模型算法相关系数达到0.893,相对误差的平均值和方差分别为0.187和0.136。在试验数据选取方面,相较于距离预测年份较远的数据,采用临近预测年份的数据,预测精度较高。根据本研究的茶叶产量预测机制,建立了浙江省茶叶产量预测误差空间分布图,其中1级优势区的平均误差为18.32%,2级次优势区为16.73%,3级一般产区为22.69%。预测模型能够实现浙江省各县市的茶叶产量预测,对茶叶生产的宏观管理具有一定指导意义。

Tea Yield Prediction in Zhejiang Province Based on Adaboost BP Model

The study proposed the tea yield prediction mechanism using the adaboost BP network model with the tea yield level factor and China meteorological forcing dataset in 59 counties of Zhejiang in 1999-2018. We extracted 11 factors including the planting area, the yearly average temperature, the average relative humidity from March to July in the sensitivity analysis. The tea yield prediction model was established then. The result shows that the adaboost BP method with the yield level factor could reach the correlation coefficient as 0.893 and the average of the relative error as 0.187 and the variance of the relative error as 0.316. When selecting history data, the prediction error was lower when the data was closer to the prediction years. Based on the proposed method, the distribution of the prediction error was made. The average relative errors were 18.32%, 16.73% and 22.69% in level 1 high production area, level 2 medium area and level 3 general production area, respectively. The proposed model could realize the tea yield prediction in the counties of Zhejiang Province and could be used in the management of tea production process.