相对湿度环境是农业生产监测与预测的重要内容，关系到植物的生长状况、多种病害的生态防治和灌溉措施的调节。针对相对湿度变化规律相当复杂，影响因素间非线性程度相当高，为了提高相对湿度预测精度，提出一种基于BP人工神经网络的相对湿度预测模型。该模型采用气象要素（日照时数、降水量、最小温度、平均温度和最大温度）实测数据作为神经网络的输入样本，并根据试验观测资料对模型进行了检验。结果表明：利用此模型分别对1987~1998年和1999~2000年陵水地区月平均相对湿度进行模拟和预测，相对湿度拟合值与实测值的相对误差为0.21%，相对湿度预测值与实测值的相对误差为0.28%。改进的BP人工神经网络能准确地捕捉相对湿度的变化趋势。运用BP人工神经网络方法进行相对湿度的研究，方法简洁，结果直观易懂，同时也为其他区域相对湿度研究提供借鉴。

Relative humidity environment is an important item of agricultural production monitoring and forecasting. Relative humidity concerns to plant growth condition, ecological prevention and control of several plant diseases, irrigation regulation. Because relative humidity has many affecting factors, there are nonlinear relations between relative humidity and its influence factors. An improved BP artificial neural network was put forward as the relative humidity prediction model, in order to improve the relative humidity prediction accuracy. The model can predict the air relative humidity with such inputs as the weather data (hours of sunlight, precipitation, the lowest temperature, the average temperature, the highest temperature). Experiments were carried out in Lingshui area to collect data to validate the model. The simulation results indicate that the monthly average relative humidity for the period of 1987~1998 was simulated and predicted for the period of 1999~2000. Relative error between the simulation and measured air relative humidity was 0.21%, and relative error between the prediction and measured air relative humidity was 0.28%. The improved BP artificial neural network has higher prediction precision. It was simple that using the BP model method to study relative humidity. And also it can provide reference for the other area’ research about relative humidity.

国家重点基础研究发展规划项目(2010CB833406)；国家自然科学基金项目(40825008,40975020,41075067)；中国科学院重要方向项目(KZCX2-EW-114)