Abstract: Accurate prediction of agricultural output value and agricultural product prices plays a driving role in promoting efficient utilization and development of agricultural resources, adjusting agricultural structure, and strengthening agricultural informatization construction. Based on the  data of agricultural output value and key influencing factors in major tea producing provinces, as well as tea commodity trading data from three kinds of e‑commerce platforms. The classic stepwise regression methods were used to determine the key influencing factors and weights of agricultural output value and tea commodity prices, and a simulation model for agricultural output value and tea commodity prices based on Elman neural network was constructed. The results showed that the key influencing factors of agricultural output value of major tea producing provinces included accumulated temperature, precipitation, grain crop sown area, cash crop sown area, proportion of cash crop output, total power of agricultural machinery, machine cultivated area, machine sown area, machine harvested area, rural electricity consumption, fertilizer application amount (net amount), rural population and rural employees. The key influencing factors of tea commodity prices in major tea producing provinces included platform, province, tea category, picking season, product level, and value‑added services. The simulated and measured root mean square error (RMSE) of the agricultural output value model of the main tea producing provinces based on Elman neural network algorithm were in the range of 6.21-27.51 million yuan, and the normalized root mean square error (NRMSE) was in the range of 3.10%-12.23%. The simulated and measured RMSE values of three kinds of e‑commerce platform tea product price models based on Elman neural network algorithm were in the range of 81.94-98.26 yuan/kg, and the NRMSE was in the range of 8.42%-35.66%.

Key words: tea, Elman neural network, stepwise regression, agricultural output value, tea commodity price, simulation model

摘要： 精准预测农业产值和农产品价格对高效利用发展农业资源、调整农业结构和加强农业信息化建设等起推动作用。基于茶叶主产省农业产值及关键影响因素数据和3种电商平台的茶商品交易数据，利用经典的逐步回归方法确定农业产值和茶商品价格的关键影响因素及权重，构建基于Elman神经网络算法的农业产值和茶商品价格模拟模型。结果表明，茶叶主产省农业产值的关键影响因素包括活动积温、降水量、粮食作物播种面积、经济作物播种面积、经济作物产量占比、农业机械总动力、机耕面积、机播面积、机收面积、农村用电量、化肥施用量(折纯量)、乡村人口数和乡村从业人员数；茶叶主产省茶商品价格的关键影响因素包括平台、省份、茶类、采摘季节、商品级别和增值服务。基于Elman神经网络算法的茶叶主产省农业产值模型模拟值与实测值的均方根误差为6.21~27.51亿元，归一化均方根误差为3.10%~12.23%；基于Elman神经网络算法的3种电商平台茶商品价格模型模拟值与实测值的均方根误差为81.94~98.26元/kg，归一化均方根误差为8.42%~35.66%。

关键词: 茶叶, Elman神经网络, 逐步回归, 农业产值, 茶商品价格, 模拟模型