Modeling And Application Of Bird Migration Time Series Data Based On Deep Network

Migratory bird migration is one of the most striking biological phenomena in nature.The migration process of migratory birds is accompanied by many problems,for example:the migration of migratory birds carrying infectious viruses may bring health and safety problems to humans; human infrastructure may cause casualties to migrating migratory birds and other issues.Therefore,the study of migratory bird migration time series data is of great significance for the prevention of bio-health security problems and the protection of biodiversity.The work in this thesis is all research on the time series data of migratory bird migration.The specific content and contributions are as follows:(1)Prediction of migratory activity status of migratory birds.Accurately predicting the activity status of migratory birds is crucial for the protection of migratory birds.This paper compares the effects of various traditional machine learning models and deep learning models on the prediction of migratory bird migration activities,and proposes a TCN-based algorithm model CT-TCN,which introduces the SE attention mechanism,and makes separate calculations for the time dimension and the channel dimension.Fusion,through the ablation experiment,the proposed model has the best effect,and has also achieved good results in the prediction of migratory bird migration activity status.(2)Prediction of migratory stop points of migratory birds.Migratory birds stop and rest where there are food sources during migration,and these sites provide opportunities for these migratory birds to interact with ecosystems to transmit pathogens and viruses.Therefore,the accurate prediction of the migratory stop points of migratory birds will provide assistance for the prevention and control policies of relevant health and safety departments.In this thesis,the prediction problem of migratory bird migration stop points is defined,and a model integrating TCN and MLP-Mixer is proposed,which fuses the features extracted by the two network structures.Through comparative experiments,it is proved that the fusion model is better than other models.Then the prediction problem was improved,and the direct prediction of latitude and longitude was changed to the offset of longitude and latitude,and the prediction effect was improved.Finally,time difference was added as the multi-task stop point prediction for auxiliary training.Task learning outperforms single-task.(3)Developed a visualization software that can assist migratory bird data analysis and predict the migratory stop point of migratory birds.The software can extract the longitude and latitude from the migratory bird data to calculate the speed and flight direction of individual migratory birds,and can obtain the location of the stop point based on the algorithm rules,and can also visualize the migration data and the stop point data.These functions can assist researchers in the study of data.The visualization tool can also visualize the prediction results of the stop points of migratory birds,so that the predicted positions can be displayed on the map,which can provide assistance for the related work of the health security department and the biodiversity protection department.