基于改进LSTMs模型的区域中长期气温预测方法研究

Regional Medium and Long-term Temperature Prediction Method Based on Improved LSTMs Model

结合残差网络阶跃连接的优点,基于长短期记忆网络模型(LSTM)和双向长短期记忆网络模型(BiLSTM),提出了对区域中长期气温预测准确率较高的DeepLSTMs网络模型.利用主成分分析对哈尔滨2007-2018年逐时气象资料进行降维,得到温度预测的主要影响因素,对气象要素进行预处理和重构,并结合DeepLSTMs网络模型对哈尔滨市中长期气温进行了大量的预测实验.结果表明,利用DeepLSTMs网络模型对该地区中长期气温的预测精度高于比较所用方法.

Combined with the advantages of step connection of residual network,and based on long-term and short-term memory network model and two-way long-term and short-term memory network model,the DeepLSTMs model with high accuracy for medium and long-term temperature prediction is proposed.The principal component analysis is used to reduce the dimension of hourly meteorological data in Harbin from 2007 to 2018,and the main influencing factors of temperature prediction are obtained,The meteorological elements are pre-processed and reconstructed,and a large number of prediction experiments are carried out on the medium and long-term temperature in Harbin combined with the DeepLSTMs network model.The results show that the prediction accuracy of the medium and long-term temperature in this area using the DeepLSTMs network model is higher than that of the method used.