Atmospheric Environment Data Generation Method Based on Stacked LSTM-GRU

At present, most of the meteorological prediction research is mainly aimed at one weather parameter, and in this paper, a LSTM-GRU-multi network model is proposed to generate multiple weather parameters simultaneously. Training with the LSTM model alone will take a large amount of time due to the atmospheric environment data samples that have many features and huge data volume. Compared with LSTM, GRU has a simpler structure, reduced number of network parameters, and faster convergence speed. In this paper, LSTM and GRU are combined to accelerate the training speed of the network under the premise that LSTM effectively processes a large amount of atmospheric environment data. At the same time, the single-layer network structure is too simple to meet the requirements of processing multiple features of atmospheric environment data, thus a multi-layer network structure is used in this paper. In the experiment, it also includes the processing of outliers and anomalies in the meteorological data sets, the data standardization, etc. The experimental results show that the MSE of the LSTM-GRU-multi model has decreased by at least 11.65% as compared with the traditional multi-layer LSTM model on the test set, meanwhile, the training time of the proposed model is reduced by 2.3%. Thus, the LSTM-GRU-multi model can generate multi-parameter atmospheric environment data accurately and fast.