摘要
本研究借助MapReduce强大的并行运算能力和良好的扩展性,尝试解决传统BP神经网络在处理大数据训练集时的瓶颈问题。通过农田肥料效应实验数据,以施肥量作为神经网络输入,以最终产量作为输出,建立精准施肥模型。该模型通过求解非线性规划问题,能同时获得最大产量和最优施肥量,解决需要预估产量的问题。在预测精准性方面,通过大数据集学习得到的施肥模型效果明显优于小数据集的学习结果,能够有效地指导精准施肥。
With the help of powerful parallel computing ability and good expansibility called MapReduce, this research tries to solve the bottleneck problem of BP neural network when BP neural network processes big data training sets. On the basis of experimental data of farmland fertilization effect, taking fertilizing amount as neural network inputs and the final yield as outputs, then to model the precision fertilization. By solving the problem of nonlinear programming, this model can gain the maximum yield and optimum fertilizing amount so that it can solve the needs of estimated yield. In prediction accuracy, results of big data training set of fertilization model is much better than the results of small data sets, and the model can effectively guide precision fertilization.
出处
《中国农机化学报》
2016年第2期191-195,共5页
Journal of Chinese Agricultural Mechanization
基金
吉林省教育厅科学技术研究项目(201363)
吉林省教育厅科学技术研究项目(201248)
