基于遗传优化BP神经网络的水稻气象产量预报模型

利用1951—2010年江苏省水稻产量及同期14个气象站点的逐日平均气温、降水资料,采用因子膨化及相关分析,研究了水稻气象产量的影响因子及影响时段。在此基础上建立了逐步回归、PCA-BP神经网络以及PCA-GA-BP神经网络3种产量预报模型。结果表明:(1)7—9月份是水稻产量形成的关键时期,对气温、降水的变化最为敏感,气温对气象产量的影响大于降水;(2)两种神经网络模型预报效果好于回归模型;(3)遗传优化的神经网络模型比未优化模型的训练速度提高了70%左右,预报精度也提高了4.3%。

A prediction model of rice meteorological yield based on neural networks optimized by genetic algorithm

With rice yield of Jiangsu Province during 1951 to 2010 and daily average temperature and precipitation data from 14 meteorological stations during the same period, this paper investigated the impact of hydrothermal resources on rice meteorological yield and the influence periods by using factors expanded method and related analysis, then established three kinds of yield forecast models on the basis of the meteorological yield: stepwise regression, PCA-BP neural networks and PCA-GA-BP neural network. The results showed that: (1)The critical period of yield formation of rice was July, August and September, rice yield was most sensitive to the changes of hydrothermal resources during this time, and the impact of temperature on meteorological yield was greater than precipitation; (2)two neural network models perform better than the regression model; (3)The training speed of network model which has been optimizated by genetic algorithm was about 70% faster than the model which has not been optimizated, while the forecast accuracy was also improved by 4.3%.