基于深度学习的沙尘天气智能预测系统设计研究

Design of Dust Weather Intelligent Prediction System Based on Deep Learning

针对使用模拟沙尘输送的数值模式预测方法、参数化方法受到因变量影响而导致预测结果精准度不高问题,提出了基于深度学习的沙尘天气智能预测系统设计。根据系统硬件结构,读入所需因变量初值采用Noah模式模拟沙尘暴陆面模式,利用GIS系统对数据分类,结合输送模式完成物理量描述。构建基于深度学习预测模型,通过训练多层神经网络沙尘天气数值,去除噪声数据。计算沙通量和垂直尘通量,预测沙尘源地。结合国家卫星气象中心实际观测数据,进行实验验证分析,具有精准预测结果。

In order to solve the problem of inaccuracy of prediction results caused by the influence of dependent variables in numerical model prediction method and parameterization method of simulating sand and dust transport,this study designed the intelligent prediction system of sand and dust weather based on deep learning.According to the hardware structure of the system,the Noah model is used to simulate the land surface model of sandstorm by reading the initial values of dependent variables.The GIS system is used to classify the data,combined with the transport model to complete the physical quantity description.The prediction model based on deep learning is constructed,and the noise data is removed by training the sand and dust weather data of multi-layer neural network.Sand flux and vertical dust flux are calculated to predict the source of sand and dust.Combined with the actual observation data of the National Satellite Meteorological Center,the experimental verification analysis is carried out to obtain more accurate prediction results.