Child respiratory disease early warning method and device based on atmospheric environment monitoring and deep learning model

Abstract

The invention discloses a children respiratory disease early warning method based on atmospheric environment monitoring and a deep learning model. The children respiratory disease early warning method comprises the following steps: step S1, acquiring past atmospheric environment monitoring and children respiratory infection admission data of a certain region; s2, an LSTM model and a DLNM model are constructed and trained; s3, predicting atmospheric environment data of a next time sequence based on the LSTM model; s4, inputting the predicted atmospheric environment data of the next time sequence into the DLNM model; and S5, outputting a prediction result of the children respiratory tract infection epidemic condition of the next time sequence. According to the invention, through combination of the long short-term memory network (LSTM) and the distributed hysteresis nonlinear model (DLNM), accurate prediction of the number of children respiratory disease patients is realized.

Description

Technical Field

[0001] The present invention relates to a warning method for children's respiratory diseases based on atmospheric environment monitoring and deep learning models, and also relates to a corresponding warning device for children's respiratory diseases based on atmospheric environment monitoring and deep learning models, belonging to the field of public health and being a multidisciplinary cross-research of environmental science, epidemiology and statistics, data science and artificial intelligence. Background Art

[0002] Research shows that there is a close association between air pollutants such as PM2.5, PM10, SO2, NO2 and O3 and the incidence of children's respiratory diseases. Therefore, accurately predicting the incidence trend of children's respiratory diseases is of great significance for public health management and disease prevention.

[0003] At present, existing research has used time series analysis methods to predict the incidence trend of children's respiratory diseases. For example, through the time series analysis of pediatric emergency asthma disease visit data, it is found that the incidence of such diseases has an obvious seasonal trend. However, traditional statistical methods have limitations in dealing with complex non-linear relationships and time lag effects, and it is difficult to fully explore the complex associations between atmospheric environmental factors and children's respiratory diseases.

[0004] In recent years, deep learning technology has made remarkable progress in the field of time series prediction. As a special recurrent neural network, the long short-term memory network (LSTM) can effectively handle the long-term dependence relationships in time series data. In addition, the distributed lag non-linear model (DLNM) can flexibly characterize the lag effects of environmental factors on health effects, providing a new tool for analyzing the relationship between air pollution and children's respiratory diseases. However, there is no systematic solution in the prior art that combines LSTM and DLNM for predicting the number of children suffering from respiratory diseases based on atmospheric environment monitoring data.

[0005] To sum up, the prior art has the following deficiencies in using atmospheric environment monitoring data to predict children's respiratory diseases: First, it fails to fully consider the lag effects of atmospheric pollutants on children's respiratory diseases; second, it lacks the effective modeling ability for complex non-linear relationships; third, there is no systematic warning method and device. Therefore, developing a time series analysis method based on atmospheric environment monitoring data, combining LSTM and DLNM models, for accurately predicting the number of children suffering from respiratory diseases, has important practical significance and innovation value. Summary of the Invention

[0006] In view of the problems in the background art, the present invention proposes a warning method for children's respiratory diseases based on atmospheric environment monitoring and deep learning models, specifically including the following steps: The present invention first discloses a method for warning children's respiratory diseases based on atmospheric environment monitoring and deep learning models, which includes the following steps: Step S1, obtain the historical atmospheric environment monitoring and children's respiratory infection hospitalization data in a certain area; Step S2, construct and train the LSTM and DLNM models; Step S3, predict the atmospheric environment data of the next time series based on the LSTM model; Step S4, input the predicted atmospheric environment data of the next time series into the DLNM model; Step S5, output the prediction result of the epidemic situation of children's respiratory infections in the next time series.

[0007] Preferably, obtaining the historical atmospheric environment monitoring and children's respiratory infection hospitalization data in a certain area includes the following steps: Step S11, obtain the historical atmospheric environment monitoring data in a certain area, with days as the unit, including 5 environmental variables: HCHO, humidity, PM2.5, ozone, and temperature; this part of the data requires the support of authoritative databases, such as: (1) official platforms: China National Environmental Monitoring Centre, Data Centre of the Ministry of Ecology and Environment of China, websites of local environmental protection departments; (2) professional data centres: National Earth System Science Data Centre, National Meteorological Science Data Centre; (3) satellite remote sensing platforms: combining satellite remote sensing, atmospheric models and large data of pollution sources and other technologies, can monitor the main atmospheric pollutants nationwide, with high spatial resolution and fast update capabilities.

[0008] Step S12, obtain the children's respiratory infection hospitalization data, with days as the unit; this part of the data requires the support of authoritative databases, such as: (1) official public health institutions: Chinese Center for Disease Control and Prevention; (2) medical institution and hospital databases: Hospital Information System (HIS), multi-centre research registration system.

[0009] Preferably, constructing and training the LSTM and DLNM models includes the following steps: Step S21, construct and train the LSTM for each of the 5 environmental variables: HCHO, humidity, PM2.5, ozone, and temperature; Step S22, construct the DLNM model using the 5 environmental variables: HCHO, humidity, PM2.5, ozone, and temperature combined with the outcome variable of the number of children's respiratory infection hospitalizations.

[0010] Preferably, the construction steps of the LSTM model specifically include: Step S211, standardize / normalize: scale the data to a specific range (such as 0 to 1 or mean of 0 and standard deviation of 1), which helps to accelerate the convergence of the model; Step S212, Data Segmentation: Divide the data into a training set and a validation set in a ratio of 8:2; Step S213, Serialization: Convert the data into a format suitable for LSTM input; for time series data, the data needs to be divided into input sequences (X) and target values; Step S214, Model Training: Design an LSTM model using the deep learning framework PyTorch and define the model structure; Step S215, Save the model.

[0011] Preferably, after hyperparameter tuning, the model structure is: number of input layer nodes = 1; number of hidden layer nodes = 64; number of hidden layers = 3; number of output layer nodes = 1; learning rate = 0.001.

[0012] Preferably, the construction steps of the DLNM model specifically include: Step S221, Convert the data into a format suitable for DLNM analysis, such as arranging it in a time series, and calculate lag variables; the model emphasizes the impact of HCHO lag exposure, with a lag of 7 days, and uses natural cubic spline cross-basis functions to generate cross-bases; Step S222, Fit the model using the Generalized Additive Model (GAM), considering multiple environmental factors and their non-linear effects.

[0013] Preferably, when predicting the atmospheric environment data of the next time series based on the LSTM model, since the 5 environmental variables have their own time variation laws, the 5 environmental variables are predicted separately for a specified number of days X, where X ≤ 14.

[0014] Preferably, X = 4.

[0015] The present invention also discloses a warning device for children's respiratory diseases based on atmospheric environment monitoring and a deep learning model, including a processor, a memory, and a display. The processor reads the computer program in the memory to enable the warning device to execute the warning method, and the results are displayed on the display.

[0016] Advantages of the present invention:

[0017] The present invention provides a warning system and method for children's respiratory diseases based on atmospheric environment monitoring data. By combining the Long Short-Term Memory Network (LSTM) and the Distributed Lag Nonlinear Model (DLNM), accurate prediction of the number of children suffering from respiratory diseases is achieved.

[0018] Compared with the prior art, the present invention has the following remarkable beneficial results: (1)Accurately predicting the disease incidence trend: The present invention utilizes the powerful time series modeling ability of the LSTM model to effectively capture the complex dynamic relationship between atmospheric environmental factors and the incidence of childhood respiratory diseases. At the same time, the DLNM model is used to model the lag effect of air pollutants, further improving the prediction accuracy. This combined method can more accurately predict the incidence trend of childhood respiratory diseases, providing a scientific basis for public health decision-making.

[0019] (2)Early warning and preventive measures: Through the warning system of the present invention, the high-incidence periods and regions of childhood respiratory diseases can be predicted in advance, providing sufficient time for medical institutions and public health departments to allocate resources and deploy preventive measures. For example, issuing air quality health warnings in advance to remind parents to reduce children's outdoor exposure time, or guiding medical institutions to make preparations in advance, thereby effectively reducing the incidence and severity of childhood respiratory diseases.

[0020] (3)Improving the utilization efficiency of public health resources: The present invention can provide data support for the rational allocation of public health resources. By accurately predicting the disease incidence trend, medical institutions can arrange medical staff, equipment, and drugs in advance, avoiding waste and shortage of resources. At the same time, government departments can formulate more targeted environmental protection policies and public health intervention measures based on the prediction results, improving resource utilization efficiency and reducing social medical costs.

[0021] (4)Enhancing the ability of environmental health risk assessment: The present invention combines atmospheric environmental monitoring data and deep learning technology to more comprehensively evaluate the potential risks of air pollutants to children's health. By analyzing the short-term and long-term effects of different pollutants and their interactions, the present invention provides new tools and methods for environmental health risk assessment, contributing to promoting scientific research and policy-making in the field of environment and health.

[0022] (5)Strong adaptability and scalability: The warning system of the present invention has strong adaptability and scalability. It can be applied to atmospheric environmental monitoring data in different regions and seasons, and can also be extended to other types of environmental health research as needed. For example, by adjusting and optimizing the model parameters, it can be applied to the prediction of other environment-related diseases such as adult respiratory diseases and cardiovascular diseases, with broad application prospects.

[0023] (6)Promoting multidisciplinary integration and technological innovation: The present invention integrates technologies from multiple disciplinary fields such as environmental science, epidemiology and statistics, data science, and artificial intelligence, demonstrating the great potential of interdisciplinary research. Brief Description of the Drawings

[0024] Figure 1Schematic flow chart of the method for warning children's respiratory diseases based on atmospheric environment monitoring and deep learning model provided by the embodiments of the present invention.

[0025] Figure 2 Model diagram provided by the embodiments of the present invention.

[0026] Figure 3 Example diagram of the prediction result of HCHO based on the LSTM model provided by the embodiments of the present invention.

[0027] Figure 4 Example diagram of the prediction result of humidity based on the LSTM model provided by the embodiments of the present invention.

[0028] Figure 5 Example diagram of the prediction result of O3 based on the LSTM model provided by the embodiments of the present invention.

[0029] Figure 6 Example diagram of the prediction result of PM2.5 based on the LSTM model provided by the embodiments of the present invention.

[0030] Figure 7 Example diagram of the prediction result of temperature based on the LSTM model provided by the embodiments of the present invention.

[0031] Figure 8 Example diagram of the prediction result of the number of children admitted to hospital with lower respiratory tract infections in each city provided by the embodiments of the present invention.

[0032] Figure 9 Example diagram of the prediction result of the total number of children admitted to hospital with lower respiratory tract infections in 10 cities provided by the embodiments of the present invention.

[0033] Figure 10 Device diagram of the method for warning children's respiratory diseases based on atmospheric environment monitoring and deep learning model provided by the embodiments of the present invention. Detailed implementation manners

[0034] The technical content of the present invention will be described in detail below with reference to the accompanying drawings and specific examples. Note: This example is implemented based on the environmental monitoring and data of children's lower respiratory tract infections in 10 cities in Jiangsu Province.

[0035] A method for warning children's respiratory diseases and allocating public health resources, such as Figure 1 As shown, the method for warning children's respiratory diseases based on atmospheric environment monitoring and deep learning model provided by the present invention includes the following steps: Step S1, obtaining the past atmospheric environment monitoring and children's respiratory infection admission data in a certain area; Step S2: Construct and train an LSTM (Long Short-Term Memory) and a DLNM (Distributed Lag Non-linear Model); Step S3: Predict the atmospheric environment data for the next time series based on the LSTM model; Step S4: Input the predicted atmospheric environment data for the next time series into the DLNM model. For the specific data flow and operation logic, see Figure 2 ; Step S5: Output the prediction results of the prevalence of childhood respiratory infections for the next time series. The predicted numerical results and daily stability are shown in Table 1: Table 1: Evaluation Table of the Prediction Effect of the LSTM-DLNM Model

[0036] Among them, LRI represents the number of children admitted to the hospital due to lower respiratory tract infections, and MAPE represents the number of children admitted to the hospital due to lower respiratory tract infections.

[0037] When obtaining the historical atmospheric environment monitoring and childhood respiratory infection hospitalization data in a certain area, the following steps are included: Step S11: Obtain the historical atmospheric environment monitoring data in a certain area, with days as the unit. This model mainly includes 5 environmental variables: HCHO, humidity, PM2.5, ozone, and temperature. This part of the data requires the support of authoritative databases, such as: (1) Official platforms: China National Environmental Monitoring Centre, Data Centre of the Ministry of Ecology and Environment of China, websites of local environmental protection departments; (2) Professional data centres: National Earth System Science Data Centre, National Meteorological Science Data Centre; (3) Satellite remote sensing platforms: Combining satellite remote sensing, atmospheric models, and big data of pollution sources and other technologies, the main atmospheric pollutants across the country can be monitored, with high spatial resolution and fast update capabilities.

[0038] Step S12: Obtain the childhood respiratory infection hospitalization data, with days as the unit. This part of the data requires the support of authoritative databases, such as: (1) Official public health institutions: Chinese Center for Disease Control and Prevention; (2) Medical institution and hospital databases: Hospital Information System (HIS), multi-centre research registration systems.

[0039] When constructing and training the model used in the present invention, the following steps are included: Step S21: Construct and train an LSTM for each of the 5 environmental variables of HCHO, humidity, PM2.5, ozone, and temperature respectively; Step S22: Use the 5 environmental variables of HCHO, humidity, PM2.5, ozone, and temperature combined with the outcome variable of the number of children admitted to the hospital due to respiratory infections to construct a DLNM model.

[0040] Build an LSTM model. The structure of the LSTM is shown in Figure 2 , including the following steps: Step S211, Standardization / Normalization: Scale the data to a specific range (such as 0 to 1 or with a mean of 0 and a standard deviation of 1), which helps to accelerate model convergence; Step S212, Data Splitting: Split the data into a training set and a validation set in a ratio of 8:2; Step S213, Serialization: Convert the data into a format suitable for LSTM input. For time series data, the data needs to be divided into input sequences (X) and target values (y); Step S214, Model Training: Design an LSTM model using the deep learning framework PyTorch, and define the model structure (after hyperparameter tuning: the number of input layer nodes = 1; the number of hidden layer nodes = 64; the number of hidden layers = 3; the number of output layer nodes = 1; learning rate = 0.001); Step S215, Save the model.

[0041] When building a DLNM model, it includes the following steps: Step S221, Convert the data into a format suitable for DLNM analysis, such as arranging it in a time series, and calculate lag variables; the model emphasizes the impact of HCHO lag exposure. The number of lag days is 7, and natural cubic spline cross-basis functions are used to generate cross-bases; Step S222, Fit the model using a Generalized Additive Model (GAM), considering multiple environmental factors and their non-linear effects; When predicting the atmospheric environment data of the next time series based on the LSTM model, since the 5 environmental variables have their own time variation laws, the 5 environmental variables are predicted separately for a specified number of days, which is generally 14 days (X ≤ 14) is appropriate. An increase in the prediction time limit will cause prediction instability. The separate prediction results of the 5 environmental variables are as Figures 3 - 7 shown, where the city codes are: CZ = Changzhou; HA = Huai'an; NJ = Nanjing; SQ = Suqian; SZ = Suzhou; WX = Wuxi; XZ = Xuzhou; YC = Yancheng; YZ = Yangzhou; ZJ = Zhenjiang. The specific verification results are as follows: Figure 3 Among them, the verification errors of each city (in the above order) are: 0.0221, 0.0105, 0.0205, 0.0175, 0.0074, 0.0172, 0.0217, 0.0211, 0.0256, 0.0227, indicating that the model is relatively accurate in predicting HCHO; Figure 4Among them, the verification errors of each city (in the above order) are: 0.0263, 0.0251, 0.0056, 0.0234, 0.0302, 0.0054, 0.0055, 0.0051, 0.0052, 0.0227, indicating that the model has relatively accurate predictions for humidity; Figure 5 Among them, the verification errors of each city (in the above order) are: 0.0138, 0.0097, 0.0091, 0.0048, 0.0113, 0.0095, 0.0064, 0.0041, 0.0038, 0.0064, indicating that the model has relatively accurate predictions for O3; Figure 6 Among them, the verification errors of each city (in the above order) are: 0.0081, 0.0061, 0.0061, 0.0180, 0.0062, 0.0162, 0.0064, 0.0041, 0.0038, 0.0064, indicating that the model has relatively accurate predictions for PM2.5; Figure 7 Among them, the verification errors of each city (in the above order) are: 0.0055, 0.0055, 0.0115, 0.0097, 0.0118, 0.0134, 0.0106, 0.0099, 0.0128, 0.0120, indicating that the model has relatively accurate predictions for temperature.

[0042] The sequences of the 5 environmental variables for 14 days predicted by the above LSTM are input into the DLNM model.

[0043] The model automatically outputs the prediction results of the prevalence of childhood respiratory infections at the next time series, Figure 8 showing the specific prediction results for each city. The trend indicates that: this model has good prediction effects and has a certain warning role. Figure 9 Showing the total prediction results of 10 cities, it is experimentally proved that the prediction effects within 4 days are better (where the error (MAPE) on the first day is 0.0974, the error on the second day is 0.0915, the error on the third day is 0.1083, the error on the fourth day is 0.1030, and the errors in the first 4 days are all within 20%, indicating that the model has good prediction effects in the short term), and it is experimentally proved that the prediction effects within 4 days are better, and its prediction error MAPE (Mean Absolute Percentage Error) is about 10%.

[0044] To implement the method for warning childhood respiratory diseases based on atmospheric environment monitoring and deep learning model provided by the present invention, the present invention also provides a device for warning childhood respiratory diseases based on atmospheric environment monitoring and deep learning model. As Figure 10As shown, the warning device includes a memory 31, a processor 32, and a display 33. The processor 32 reads a computer program in the memory 31 to enable the warning device to execute a warning method, and the result is displayed on the display 33.

[0045] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Those skilled in the art to which the present invention pertains can make various modifications or supplements to the described specific embodiments or use similar means for substitution, but will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A method for warning of children's respiratory diseases based on atmospheric environment monitoring and deep learning models, characterized in that It includes the following steps: Step S1: Obtain the historical atmospheric environment monitoring and children's respiratory infection hospitalization data in a certain area; Step S2: Construct and train the LSTM and DLNM models; Step S3: Predict the atmospheric environment data of the next time series based on the LSTM model; Step S4: Input the predicted atmospheric environment data of the next time series into the DLNM model; Step S5: Output the prediction result of the children's respiratory infection epidemic situation of the next time series.

2. The method according to claim 1, wherein Obtaining the historical atmospheric environment monitoring and children's respiratory infection hospitalization data in a certain area includes the following steps: Step S11: Obtain the historical atmospheric environment monitoring data in a certain area, with days as the unit, including 5 environmental variables: HCHO, humidity, PM2.5, ozone, and temperature; Step S12: Obtain the children's respiratory infection hospitalization data, with days as the unit.

3. The method according to claim 2, wherein Constructing and training the LSTM and DLNM models includes the following steps: Step S21: Respectively construct and train the LSTM for 5 environmental variables: HCHO, humidity, PM2.5, ozone, and temperature; Step S22: Use the 5 environmental variables: HCHO, humidity, PM2.5, ozone, and temperature, combined with the outcome variable of the number of children's respiratory infection hospitalizations, to construct the DLNM model.

4. The method according to claim 3, characterized in that The construction steps of the LSTM model specifically include: Step S211: Standardization / Normalization: Scale the data to a specific range; Step S212: Data splitting: Divide the data into a training set and a validation set in a ratio of 8:2; Step S213: Serialization: Convert the data into a format suitable for LSTM input; for time series data, the data needs to be divided into input sequences and target values; Step S214: Model training: Design the LSTM model using the deep learning framework PyTorch and define the model structure; Step S215: Save the model.

5. The method according to claim 4, characterized in that After the model structure is adjusted by hyperparameters, it is: the number of input layer nodes = 1; the number of hidden layer nodes = 64; the number of hidden layer layers = 3; the number of output layer nodes = 1; the learning rate = 0.

001.

6. The method according to claim 3, wherein The construction steps of the DLNM model specifically include: Step S221: Convert the data into a format suitable for DLNM analysis; the model emphasizes the impact of HCHO lag exposure, with a lag of 7 days, and uses natural cubic spline cross-basis functions to generate cross-bases; Step S222: Use the generalized additive model GAM to fit the model, considering various environmental factors and their non-linear effects.

7. The method according to claim 3, wherein When predicting the atmospheric environment data of the next time series based on the LSTM model, since the 5 environmental variables have their own time variation laws, separately predict the specified number of days X for the 5 environmental variables, where X ≤ 14.

8. The method according to claim 7, wherein X=4。 9. An early warning device for children's respiratory diseases based on atmospheric environment monitoring and deep learning models, characterized in that It includes a processor, a memory, and a display. The processor reads the computer program in the memory to enable the warning device to execute the warning method according to any one of claims 1-8, and the result is displayed on the display.

Images