基于神经网络的短期土壤墒情预测模型研究

Research on Short-term Soil Moisture Prediction Model Based on Neural Networks

为实现短期的土壤墒情预测,根据天津市蓟州区、静海区、宁河区、滨海新区的10个气象墒情自动监测站3年的数据,对短期土壤墒情预测模型进行研究。选取站点编号、空气温度、空气湿度、风速、风向等28项影响因子,用包含天气预报和不含天气预报的2组数据分别训练BP神经网络和Elman神经网络,并对4组预测模型结果进行对比分析。结果表明:不含天气预报的BP神经网络模型和包含天气预报的BP神经网络模型精度分别为94.79%、95.54%,不含天气预报的Elman神经网络模型和包含天气预报的Elman神经网络模型精度分别为96.85%、96.64%。研究认为,Elman神经网络具有稳定性好、精度高的特点;理论认为,含天气预报的模型精度比不含天气预报的模型精度高,BP神经网络表现出这一相关性,而Elman神经网络并没有表现出这一相关性。

To achieve short-term soil moisture predication,a study was conducted on the short-term soil moisture forecasting model based on three years of data from 10 automatic meteorological soil moisture monitoring stations in Jizhou District,Jinghai District,Ninghe District,and Binhai New Area of Tianjin City.A total of 28 influencing factors,including station number,air temperature,air humidity,wind speed,and wind direction,were selected.Two sets of data,one including weather forecasts and one without weather forecasts,were used to train BP neural networks and Elman neural networks respectively.The results of the four prediction models were compared and analyzed.The results showed that the accuracy of the BP neural network model without weather forecasts and the BP neural network model with weather forecasts were 94.79%and 95.54%respectively,while the accuracy of the Elman neural network model without weather forecasts and the Elman neural network model with weather forecasts were 96.85%and 96.64%respectively.The study concluded that the Elman neural network has the characteristics of good stability and high accuracy.The theory suggests that the model accuracy with weather forecasts should be higher than that without weather forecasts,and the BP neural network shows this correlation,while the Elman neural network does not show this correlation.