基于RBF神经网络的大豆种植密度和施肥量优化

Optimization of Soybean Planting Density and Fertilizer Application Rate Based on RBF Neural Network

为解决使用传统回归模型对大豆种植密度及施肥量进行优化时存在的拟合精度低、优化结果不准确等问题,提出一种基于RBF神经网络的优化方法。将大豆种植密度、N、P2O5、K2O施用量作为试验因素,产量作为影响指标,选取黑河43作为试验材料,进行四因素五水平的正交旋转试验,获得各处理下大豆产量数据。对种植密度、施肥量与产量关系构建RBF神经网络拟合模型,对模型进行优化,得到最优种植密度42.65×104株·hm^-2、施N量61.82 kg·hm^-2、施P2O5量106.05 kg·hm^-2、施K2O量19.81 kg·hm^-2,该配比下大豆产量为3 821.48 kg·hm^-2。对优化结果进行试验验证,最优配比下大豆实际产量为3 742.29 kg·hm^-2,与优化结果相对误差为-2.17%,表明该方法有效,且优化结果准确。

In order to solve the problems of low fitting accuracy and inaccurate optimization results when soybean planting density and fertilizer application rate was optimized with the traditional regression model, this study proposed an optimization method based on RBF neural network. Soybean planting density, fertilizer application rate of N, P2O5, K2O were taken as experimental factors, and soybean yield was taken as impact indicator. An experiment of 4 factors and 5 levels was designed by the orthogonal rotation method on the seed of Heihe 43. The data of soybean yield under each treatment was obtained. The RBF neural network fitting model was constructed for the relationship between planting density, fertilizer application rate and yield, and the optimization method proposed in this paper was used to optimize this model. The optimization result was planting density 42.65×104 plants·ha^-1, N fertilizer application rate 61.82 kg·ha^-1, P2O5 fertilizer application rate 106.05 kg·ha^-1, K2O fertilizer application rate 19.81 kg·ha^-1, the yield of soybean under this combination was 3 821.48 kg·ha^-1. Another experiment was carried out to verify the optimization result. The actual soybean yield at the optimal ratio was 3 742.29 kg·ha^-1. The relative error between actual yield and optimum yield was-2.17%. It showed that the optimization method was effective and the optimization result was accurate.