土壤水分预测神经网络模型和时间序列模型比较研究

土壤水分运动是一个复杂的时间序列系统，其变化与区域气候条件和生态环境密切相关，具有明显的随机性波动。建立土壤水分动态变化模型可以使田间土壤水分的适时适量调节方便可行，有利于农田水利工程的规划和管理。该文利用人工神经网络方法和时间序列自回归(AR)模型进行了土壤水分预测建模研究，试验结果表明：在数据量较少的情况下，AR模型具有较好的预测效果；在数据量较多的情况下，神经网络模型能够获得较好的预测效果。

Comparison of autoregression and neural network models for soil water content forecasting

Soil water dynamics is a complex time series system with obviously random fluctuation, closely related to regional climate and ecological environment. Establishing the model of soil water dynamics can not only modulate real time farm soil water, but also is available to farm irrigation works. In this paper, the autoregression and neural network were applied to establish the model of purple soil water forecast in hilly region. The result showed that: in the case of less data, the autoregression model can preferably fit the soil water time series and its forecasting was available. In the case of enough data, the neural network model could do it better.