Application Of EMD-LSTM Model Integrating Attention Mechanism In Weather Prediction

Daily weather condition is closely related to every field of production and life.Weather condition forecast plays an important role in social development.At present,weather forecast software has become an important carrier to check the information of short-term weather conditions in the future.In this paper,relevant research is made to improve the accuracy of weather prediction.Based on the analysis of the main research methods and research status of weather conditions,an EMD-LSTM model integrating attention mechanism is proposed.The application process of Empirical Mode Decomposition and Long Sort Term Memory neural network in weather conditions is analyzed.Through the framework algorithm structure of encoder and decoder,the attention mechanism and the algorithm process of five common attention mechanisms are introduced.According to the data characteristics of urban weather conditions,a long and short time memory network model is designed to predict weather conditions,and the algorithm structure and implementation process are integrated with attention mechanism.On this basis,the feasibility of the proposed model is verified by experiments.The global attention mechanism commonly used in LSTM,that is,all input hidden states are scanned in each computation,which will adversely affect the operational efficiency and accuracy.In order to improve this problem,the improved attention mechanism algorithm is used to combine Scaled Dot-Product Attention,Dual-Stage Attention,and Muti-Head Attention with LSTM neural network.The method is applied to the daily weather data of Shenzhen in the past five years,and the short-term weather is forecasted with multivariate.By comparing with the traditional deep learning model and regression model,the feasibility and superiority of EMD-ALSTM algorithm are verified.The experimental results show that the proposed EMD-LSTM model with Dual-Stage Attention mechanism has higher prediction accuracy and efficiency than the traditional model,which provides a new idea for weather prediction.