基于神经网络算法的果树需水预测研究

【目的】准确预测果树需水量。【方法】对采集地果园环境数据进行主成分分析,筛选出影响果树蒸腾量的关键因子。建立以长短时记忆(LSTM)神经网络为基础的预测模型来预测果树蒸腾量。为提高预测的精度,在LSTM神经网络的基础上加入了注意力(Attention)机制,形成Attention-LSTM预测模型。【结果】将改进的模型与其他模型的预测精度进行对比,仿真试验表明,该模型的预测精度最高,RMSE和MSE分别为0.487和0.062。【结论】该预测模型可以准确预测果树蒸腾量,从而实现果园精准灌溉并提高水果产量,具有一定的实际意义。

Calculating Demands of Fruit Trees for Water Using Neural Network Algorithm

【Objective】Knowing the demand of crops for water is a prerequisite in designing saving-water irrigation and improving water management. The aim of this paper is to study the accuracy and reliability of the neural network algorithm for estimating water demands of fruit trees.【Method】Principal component analysis method was used first to analyze the environmental and meteorological data to find key factors that affect the evapotranspiration of the fruit trees in orchards most. They were then used to derive a model(LSTM) based on the long-term and short-term memory neural network to estimate the water demand of the fruit trees. For improving estimation accuracy, we added an attention algorithm to the LSTM. The superiority of the model was tested against those used in the literatures and practices.【Result】 Comparing with existing models for estimating demands of the fruit trees for water, the proposed model improved estimation accuracy, with its MAE, MAPE, RMSE being 0.387, 0.148, 0.487 and 0.062 respectively.【Conclusion】 We proposed a neutral neural network method to estimate water demand of fruit trees. Adding an attention algorithm to the model improved its accuracy considerably, compared with the existing models used in the literature. It has practical implications for estimating evapotranspiration not only for orchards but also for other natural and managed ecosystems.