Associated Prediction Of Temperature And Precipitation By Combining Attention And LSTM Modle

Weather change has a great impact on industrial and agricultural production and people's daily life.Accurate prediction of future weather conditions is of great significance for people to make use of advantages and avoid disadvantages,arrange industrial and agricultural production,and prevent and reduce disasters.Short term weather prediction methods can be divided into radar detection,satellite observation and neural network methods,etc.However,the traditional neural network methods for weather numerical prediction generally use the shallow network,and only forecast one weather value.The accuracy of these methods also needs to be improved.In order to achieve multi-valued weather prediction and to improve the prediction accuracy,it is necessary to propose a neural network model which can extract the depth features of data and with high prediction accuracy.Based on the above considerations,it has been done the following research work:1.A weather multi-value prediction algorithm based on Attention and LSTM combined association model(ALSTM)is proposed to predict future temperature and precipitation.The ALSTM model integrates LSTM,Attention,Dropout and other mechanisms,and realizes the prediction of multiple weather elements.In the algorithm,the data is normalized first,then the processed data is used to train the ALSTM model,and finally the trained model is obtained.Based on the hourly weather data of Beijing from 2010 to 2015,ALSTM,traditional LSTM algorithm,BP neural network algorithm and deep recurrent neural network based on LSTM(DRNN)algorithm are used to predict the temperature and rainfall of the next hour.The experimental results show that the ALSTM algorithm can not only predict temperature and rainfall simultaneously,but also has better prediction accuracy than the other three models,and the prediction accuracy of each value is more than 97%.In addition,the performance of ALSTM is compared with that of the association prediction algorithm without attention mechanism(NALSTM).The experimental results show that the prediction accuracy of ALSTM is significantly better than that of NALSTM,which indicates that the addition of attention mechanism can improve the prediction effect.2.The combination association model(ALSTM)based on Attention and LSTM is optimized,and a feature segmentation and combination optimized ALSTM model(SALSTM)is proposed.It is applied to process the daily weather dataset between 2014 and 2019 of Chenzhou city in Hunan province,and to predict the temperature and rainfall next day.SALSTM is compared with the deep recurrent neural network model based on LSTM(DRNN)and ALSTM,and the experimental results show that the prediction effect of SALSTM is better than that of DRNN and ALSTM.This shows the feasibility and good performance of Attention and LSTM combined association model based on feature segmentation and combination optimized.