| Aiming at the urgent requirements of accurate analysis and early warning of tuberculosis transmission and PM2.5 pollution,the paper plans to carry out research on tuberculosis incidence trend analysis and prediction and PM2.5 prediction by using spatiotemporal big data analysis technology,geographic information system and deep learning artificial intelligence technology.The main results are as follows:(1)Analysis of spatiotemporal distribution of pulmonary tuberculosis based on GIS technologyIn the national region,based on the collected data of disease,medical treatment and environment,and based on the geographic information system(GIS)method,tuberculosis was studied,and its spatial distribution,spatial-temporal evolution,epidemic situation monitoring and spatial-temporal aggregation characteristics were analyzed.The spatial autocorrelation analysis method is used to study and analyze the spatial characteristics of pulmonary tuberculosis.The main statistical indicators are Moran’s I,Geary’s C and Getis-Ord G.(2)Analyze and excavate the key influencing factors and their contribution to the spread of tuberculosisIn the whole country,based on the collected disease,medical,meteorological,environmental,economic and other data,the spatio-temporal analysis model such as multiple geographical weighted regression MGWR is used to deeply analyze and mine the impact characteristics and degree of each influencing factor on the incidence rate of tuberculosis.At the same time,the spatial heterogeneity of the impact of air quality,social life and medical level factors on the incidence rate of pulmonary tuberculosis was studied,and the least squares(OLS)model and geographical weighted regression(GWR)model were used for comparative analysis.(3)Tuberculosis prediction method based on deep learningBeijing Tianjin Hebei region was selected as the study area,and CNN(convolutional neural network model)and LSTM(long-term and short-term memory model)were used to predict the incidence rate of pulmonary tuberculosis,and a CNN BiLSTM Attention hybrid model combining CNN model,BiLSTM model and attention mechanism was designed.Through the Bayesian optimization algorithm,the optimal fitting parameters of the LSTM model,LSTM-CNN model,BiLSTM-CNN model,and NN-LSTM-CNN-Attention hybrid model were determined.Based on this,comparative experiments were conducted on the predictions of the four models.The results showed that the combined model outperformed other comparative models in the evaluation indicators of mean absolute error(MAE),root mean square error(RMSE),and mean absolute percentage error(MAPE),It can be used to predict the incidence rate of pulmonary tuberculosis.In addition,ablation experiments were conducted on each combination model to demonstrate the rationality and effectiveness of the NN-CNN-LSTM-Attention Attention hybrid model proposed in this paper.(4)WLSTME PM2.5 concentration prediction method based on dynamic wind fieldBeijing-Tianjin-Hebei region was selected as the study area,and a WLSTME model was developed to predict the daily average PM2.5 concentration at a specific site,taking into account the uneven distribution of monitoring points.First,use MLP to combine the historical wind speed and direction with the PM2.5 data of the corresponding day of adjacent stations to generate a weighted PM2.5.Secondly,the weighted PM2.5 data of the neighboring stations for nearly 10 days and the historical PM2.5 data of the central station are combined and input into the LSTM layer,while solving the temporal and spatial dependence and extracting the temporal and spatial characteristics.Finally,another MLP is used to adjust the deviation by combining the spatiotemporal characteristics with the meteorological data and timestamp data of the central station.The prediction results show that the prediction accuracy and reliability of WLSTME are higher than those of STS VR,LSTME and GWR in each season and region,especially in spring and autumn,which can be used to predict PM2.5 concentration. |
PDF Full Text Request