Precipitation Prediction Based On Machine Learning Algorithm

Precipitation prediction has great social value.However,due to the absence and distortion of meteorological data in the process of collection and transmission,some features of the original data are difficult to be captured by the prediction model.At the same time,the complex meteorological causes,the dynamic changes of each element over time and space,and the difficult formulation description of the relationship among the elements,all of which restrict the establishment of accurate precipitation model.Therefore,we focus on two aspects: improving the quality of data samples and exploring machine learning algorithms suitable for establishing precipitation models.In the aspect of data set processing,based on the comparison of applicable conditions of various popular data cleaning algorithms,we will carry out a data cleaning algorithm combining neighboring point mean interpolation and CART according to the characteristics of the sample data,in order to improve the quality of the data set;In the precipitation prediction modeling research,according to the continuous changes of meteorological elements,and in order to distinguish the influence of different feature vectors on precipitation prediction,the attention mechanism is introduced into the Bi-LSTM training network,we will propose an improved Bi-LSTM-Attention learning model,which enables the network to learn the weight allocation strategy of each feature vector independently,and improves the prediction accuracy of the model.In the simulation comparison,the improved data interpolation algorithm and two commonly used data preprocessing methods are used to process the original data set respectively,and three data sets are formed as comparison samples;Based on each sample,the SVR algorithm,RFR algorithm,GRU algorithm and the proposed Bi-LSTM-Attention algorithm are used to establish the precipitation prediction models.The results show that the improved data interpolation algorithm effectively improves the quality of the data set,and the prediction accuracy of the Bi-LSTM-Attention model is higher.