Research On Precipitation Prediction Method Of LSTM Based On Evolutionary Algorithm

Precipitation prediction is an important part of the normal operation of human society,and accurate prediction results are important to transportation,agriculture and public safety.Long Short-Term Memory(LSTM),a data-driven time series prediction model,has good performance in precipitation prediction.However,it is necessary to configure the hyperparameter of the model carefully and continuously adjust and optimize them by experts with rich field experience,which not only increases the difficulty of using the model,but also brings high labor cost.For this problem,the LSTM precipitation prediction method system based on evolutionary algorithm is proposed.This method system defines the hyperparameter search space of LSTM precipitation prediction model from the perspective of optimization theory and takes precipitation prediction accuracy as the optimization target,intelligently selecting the best hyperparameter combination in hyperparameter space to determine the LSTM precipitation prediction model with the best prediction effect and optimal generalization ability by respectively utilizing Genetic Algorithm and a hybrid of Cross Entropy with Genetic Algorithm,then the precipitation prediction model named GA-LSTM based on Genetic Algorithm and the precipitation prediction model named CEGA-LSTM based on a hybrid of Cross Entropy with Genetic Algorithm are formed respectively.The results of numerical experiments show that GA-LSTM model and CEGA-LSTM model have high accuracy in precipitation prediction,and have the advantages of adaption,simplicity and popularization,The precipitation prediction method based on the two models not only improves the prediction accuracy of traditional methods,but also realizes the intelligent optimization of hyperparameter combination,which are of great theoretical and practical significance.