基于多值神经元复数神经网络的土壤墒情预测

发布时间：2018-02-27 20:46

本文关键词： 土壤水分模型神经网络多值神经元春玉米　出处：《农业工程学报》2017年S1期 　论文类型：期刊论文

【摘要】：为指导节水灌溉策略的制定,利用基于多值神经元的复数神经网络(multilayer neural network with multi-valued neurons,MLMVN)方法,建立了土壤墒情多步预测模型。首先,利用均值法替换样本中的异常值并对缺失值进行补充,并由数据分析知土壤墒情数据为非平稳的非线性时间序列。然后,根据土壤墒情与环境因素(降雨量、气温和风速)的相关性分析结果选择降雨量为关键环境因素。最后将土壤墒情、降雨量及目标土壤墒情复数化,作为网络输入和期望输出建立MLMVN预测模型。结果表明,网络结构为240-15-1200-1时单步预测精度为0.883,采用循环预测法进行步长为72的多步预测,平均预测精度为0.853,比实数域误差反向传播神经网络BP提高了9.1%。研究表明,MLMVN模型多步预测误差累计小,预测结果可作为该地区节水灌溉策略制定的理论依据。
[Abstract]:In order to guide the formulation of water-saving irrigation strategy, the multilayer neural network with multi-valued neuronsn (MLMVN) method based on multi-valued neuron was used to establish a multi-step prediction model of soil moisture. The method of mean value is used to replace the abnormal value in the sample and to supplement the missing value, and the data analysis shows that the soil moisture data is a nonstationary nonlinear time series. Then, according to the soil moisture and environmental factors (rainfall), The correlation analysis results of air temperature and wind speed selected rainfall as the key environmental factor. Finally, the soil moisture, rainfall and target soil moisture were plural as network input and expected output to establish MLMVN prediction model. When the network structure is 240-15-1200-1, the precision of single-step prediction is 0.883.The cycle prediction method is used to predict multi-step with step size of 72. The average prediction accuracy is 0.853, which is 9.1higher than the BP neural network in real field error backpropagation. The research shows that the multi-step prediction error of MLMVN model is small, and the prediction result can be used as the theoretical basis for the formulation of water-saving irrigation strategy in this area.
【作者单位】：中国农业大学信息与电气工程学院;
【基金】：国家自然科学基金资助项目(31471409)
【分类号】：TP183;S152.7

【相似文献】