Research On Long Sequence Forecasting Method Of Nearshore Aerosol Extinction Coefficient

The aerosol extinction coefficient is an index that measures the intensity of light absorption and scattering by aerosol particles at a specific wavelength.It represents the degree to which aerosols attenuate radiation and is an important parameter that affects the Earth-atmosphere transmission system and laser propagation.Additionally,the aerosol extinction coefficient can provide important basis and data support for air pollution detection and climate change research.Therefore,predicting the aerosol extinction coefficient is of great significance for applications such as optical remote sensing,optoelectronic engineering,and environmental monitoring.Due to limitations such as detection technology and measurement environment,long-term continuous acquisition of the aerosol extinction coefficient has low efficiency and uncontrollable data quality.To address this,it is necessary to establish a long sequence prediction method for the aerosol extinction coefficient.Based on the analysis and research of meteorological data,visibility,and extinction coefficient measurements of Maoming area in 2020,this paper establishes single-step and long sequence forecasting models for aerosol extinction coefficient.The main research contents are as follows:(1)Using temperature,humidity,pressure,visibility and other parameters,and combining attention mechanism with Bidirectional Long Short-Term Memory network,an Attention-BiLSTM model is established to predict the extinction coefficient at a single-step.Through the analysis and comparison of the prediction accuracy of MLP,RNN,LSTM,BiLSTM,BiLSTM-Attention,and Attention-BiLSTM models,the single-step prediction model for aerosol extinction coefficient proposed in this paper has improved the accuracy by 23.7%compared to the classical RNN.Compared with other different LSTM variants,the accuracy has also been improved,and the trend of data changes can be accurately captured.(2)In order to enhance the reliability of long sequence time series forecasting,this paper proposes a model for predicting aerosol extinction coefficient,by adding global timestamps to the input data in the input representation stage,using convolutional aggregation to calculate query and key vectors,optimizing the self-attention mechanism,and using generative decoding to output the predicted results at once.Experimental results show that compared to other Transformer models,the proposed model has significantly improved accuracy,memory usage,and speed.In the prediction accuracy experiment,the MAE of the proposed model decreased from 0.237 to 0.103,and the MSE decreased from 0.345 to 0.241.In the memory usage experiment,the model can effectively alleviate memory overflow problems when the input length is greater than 720.In the speed experiment,when the input length is 672,the training time per round decreased from 15.32 seconds to 12.39 seconds.The experimental results demonstrate the effectiveness and reliability of the proposed model,providing a new approach and method for predicting long sequence time series of aerosol extinction coefficients.(3)Design and implement an atmospheric database software.Firstly,the SQL Server database system is used to create the database according to the design ideas and table structure.ERA5 data in the South China Sea region and observational data from multiple locations are collected to implement basic operations such as data input,query,modification,and deletion.Secondly,Qt Creator is used to develop the software to store,manage,and analyze atmospheric data,providing data support for future research.Finally,the long sequence prediction module of aerosol extinction coefficient is integrated with the atmospheric database software to realize the engineering application of long sequence forecasting.