Cloud-Base Intellingent Inversion Algorithm Based On Ensemble Learning

Cloud is one of the important factors affecting weather and climate.At present,more than two-thirds of the world's regions are covered by clouds.The existence of clouds can directly affect the transmission process of atmospheric radiation,thus changing the temporal and spatial distribution of atmospheric radiation.Moreover,the cloud layer can directly participate in the water cycle process,so the precipitation distribution around the world is also related to the cloud layer distribution.On the inversion calculation problem of the cloud base,the inversion calculation based on the cloud height observer is more affected by subjective factors,and there is a certain deviation between the calculation result and the actual cloud height.In order to solve the problem of cloud base height inversion calculation with better construction performance and higher intelligence,this paper combines classical machine learning algorithm and convolutional neural network through integrated learning strategy to design one.A cloud-intelligent inversion calculation model MCHIIC with multiple algorithms.The main innovations of this paper include:1.In the cloud high-level element data acquisition module of the MCHIIC model in this paper,in order to be able to screen the relevant data given by the Hunan Provincial Meteorological Observatory in real time to extract the relevant live element data suitable for cloud height inversion calculation,this paper designs An adaptive screening algorithm CLHF is used to extract the cloud high-level feature data in dynamic data.When the related data changes,CLHF can automatically adapt to the data changes,and can be time-limited and geographically unlimited.The limitation of the number of terrains and observation sites accurately selects the cloud height data required by the model,which greatly reduces the error and workload caused by the human screening work.2.In the convolution processing module of the MCHIIC model in this paper,the fully connected layer is generally placed in the last layers of the network,and its purpose is to weight the features of the convolutional layer and the pooled layer,since this paper studies Multi-classification problem,and inversion calculation of cloud height in a certain height range,so this paper improves the output matrix ? and offset b on Softmax loss by improving the classic loss function Softmax loss.The loss function CLH-Softmax loss for cloud high inversion is designed,so that the inversion result can be maintained in a more reasonable interval,thereby further improving the accuracy of cloud high classification output.3.In the cloud high classification output module of MCHIIC model,in order to further improve the classification accuracy and generalization ability of the model on the cloud high-level feature dataset,this paper proposes a weighted fusion algorithm(WF)and a weight-based correction unit.Weighted-Regulate-UnitStackingBagging(WRUSB)algorithm.In the WF algorithm,the results of the output of the CLH-CNN classifier and the Adaboost classifier are weighted and input to the Softmax layer for probability analysis.In the WRUSB algorithm,the output weight of the base classifier is adjusted before the voting of the Bagging algorithm.Each output is given different correction coefficients according to the output weight,and the cloud bottom height inversion is obtained by voting method.As a result,the classification accuracy of the output is improved while improving the generalization ability of the model.