论文部分内容阅读
作物最优施肥量与土壤养分含量、产量之间存在复杂的非线性关系。为更加准确地模拟这种关系,提出一种改进的的BP神经网络集成方法。该方法采用K-均值聚类优选神经网络个体,采用拉格朗日乘子方法计算待集成的神经网络个体的权值。然后,基于农田肥料效应试验数据,以土壤养分含量和施肥量作为神经网络的输入,以产量作为神经网络的输出,建立了作物精准施肥模型。该模型通过求解一个非线性规划问题,能同时获得最大产量和最优施肥量。试验结果表明,在施肥模型的拟合精度方面,改进的神经网络集成方法(其均方根误差为64.54)明显优于单个神经网络方法(其均方根误差为169.74)。而且,作为一种定量模型,基于改进的神经网络集成的施肥模型优于传统施肥模型,能有效地指导精准施肥。
There is a complicated nonlinear relationship between the optimal fertilizer amount and soil nutrient content and yield. In order to simulate this relationship more accurately, an improved BP neural network integration method is proposed. The method uses K-means clustering to optimize neural network individuals and uses Lagrange multiplier method to calculate the weights of individual neural networks to be integrated. Then, based on the experiment data of fertilizer effect of farmland, the soil nutrient content and fertilizer amount were taken as the input of neural network, and the yield was used as the output of neural network to establish the precise crop fertilization model. By solving a nonlinear programming problem, the model can simultaneously obtain the maximum yield and the optimal fertilization rate. The experimental results show that the improved neural network ensemble method (the root mean square error is 64.54) is obviously superior to the single neural network method (the root mean square error is 169.74) in fitting accuracy of fertilization model. Moreover, as a quantitative model, the integrated fertilization model based on improved neural network is better than the traditional fertilization model, which can effectively guide the precise fertilization.