| As one of the most important basic data of water resources,precipitation observation is a dynamic input project to study the hydrological cycle system in a basin or region.It plays a great role in industrial and agricultural production,water conservancy development,river flood control and project management.The observation and recording of the amount of liquid(rain)water falling in the sky.All forms of precipitation are represented by the depth of the water layer accumulated on the horizontal surface of the place where it is borne.Its unit of measurement is mm,usually measured to 0.1mm(0.1mm and above as rainfall,and vice versa).Previous studies on rainfall were mostly based on conventional factors affecting rainfall as explanatory variables.In this paper,the meteorological data of cloud cover is selected,which has better characteristics and higher explanatory effect on rainfall.Based on the hourly rainfall data of Fujian Province from June 2021 to May 2022,Fujian Province is divided into five regions according to latitude and longitude.Firstly,the random forest classification algorithm was used to train the classification model of rainfall in the five regions successively,and the results showed that the prediction accuracy of each region reached more than 96%.Besides,according to the output of the top 30 features affecting rainfall,the ranking of regional features was compared,and the importance of the features affecting rainfall in different regions was obtained.Secondly,the rainfall data(rainfall of 0.1mm or above)of the 5 regions were screened out.The LASSO regression was used to reduce the dimensionality of the data.The lamba value with the minimum mean square error was selected by cross-validation to obtain the rainfall data after dimensionality reduction.Then,the random forest regression algorithm was used to model the data,and the results showed that the mean absolute error(MAE)on the test set of rainfall in each region was below 0.01,but the prediction error of large rainfall was larger.Next,according to the classification standard of rainfall grade,the rainfall data were divided into light rain and moderate rain,and the random forest algorithm was used to establish the classification model.The model results show that the accuracy of the rainfall classification test set in each region reaches more than 94%.In addition,random forest regression models were established for the rainfall data of each regional level respectively.First,feature selection was carried out for the regional level data by Lasso regression,and the regression model was established by using the random forest regression algorithm.The results show that the mean square error and mean absolute error are reduced and the prediction ability of the model is improved.However,the mean square error and the mean absolute error are both large when establishing the regression model for the moderate rainstorm,so the prediction ability is greatly improved after removing the rainstorm data and establishing a new model. |
