基于深度学习的气象要素预测方法

METEOROLOGICAL ELEMENTS FORECASTING METHOD BASED ON DEEP LEARNING

关键词:  气象预测  深度学习  神经网络  长短时记忆

Key words:  meteorological forecast  deep learning  neural network  long short term memory

摘要:

针对气象预测内容繁多且影响因素多样的问题，提出了一种基于长短时记忆（LSTM）的气象预测方法。方法能够对繁杂的气象数据进行自动预处理，提取相应的特征信息。通过神经网络的前向训练、长短时记忆反馈学习，经过多隐藏层地自主训练，对能见度、温度、露点、风速、风向以及压力气象信息实现准确预测。通过实验以及与经典机器学习预测方法的比较，验证了本文方法在气象预测中的有效性，进一步提升了气象预测的准确性，各项预测值的均方检验误差平均值为0.35。

Abstract:

Aiming at the problems of various weather forecasting content and various influencing factors, a weather forecasting method based on long short-term memory (LSTM) is proposed.The method can automatically preprocess the multifari-ous meteorological data and extract the corresponding characteristic information. Through the neural network forward training, long and short time memory feedback learning, through the multi-layer neural network autonomous training, to achieve the visibility, temperature, dew point, wind speed, wind direction and pressure meteorological information accu-rate prediction. Through simulation experiments and comparison with classical machine learning prediction methods, we verify the effectiveness of the method proposed in this paper in weather prediction, and further improve the accuracy of weather prediction. The mean square test error of the predicted values is 0.35.