CN117334334B - A health risk prediction method, device, equipment and medium - Google Patents

Abstract

The present invention discloses a health risk prediction method, device, equipment and medium. The method obtains current environmental data; inputs the current environmental data into a prediction model, wherein the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST model includes a VARMA model and a BI-LSTM model, and weightedly fuses the prediction results of the VARLST model and the prediction results of the AdaBoost model to obtain a final prediction value; based on the final prediction value predicted under the current environmental data and the number of disease patients and non-disease patients under the dangerous reference baseline value in the past time period, calculate the relative risk value of the predicted health risk; and perform health risk warning based on the relative risk value. The present invention can prepare for early prevention by rapid prediction and calculation of the relative risk value of health risks when the environmental quality index is abnormal.

Description

A health risk prediction method, device, equipment and medium

Technical Field

The present invention relates to the technical field of health risk prediction, and in particular to a health risk prediction method, device, equipment and medium.

Background technique

In recent years, with the continuous development of urbanization, the population density of cities has also been increasing. Urban areas with dense populations are also high-risk areas for the spread of pollutants and mass infection. When abnormal pollutant values appear near the city, if there is no effective prediction and warning and the public ignores the impact of the objective factor of the environment on physical health, it will cause irreversible consequences for the health of local people. At present, the scope of smart city popularization is becoming wider and wider, and there are many technologies that can integrate various service resources of the city to provide more accurate and convenient services for the city. Therefore, a model that combines and analyzes environmental data and medical data is needed to predict the impact of environmental changes on physical health, aiming to solve the above problems.

The commonly used algorithms for predicting health risks with multiple time series parameters can be roughly divided into three categories, including the use of modeling mathematical principles such as autoregressive mean moving average, vector autoregression, and ARIMA models for environmental time series prediction, including machine learning models such as AdaBoost model and random forest, and deep learning models such as LSTM suitable for processing sequence data, which can handle multiple time series predictions to a certain extent. However, the existing technologies for multi-parameter health risk prediction have problems such as low model generalization ability, poor data fitting effect for nonlinear relationships, and high model calculation complexity. Therefore, it is difficult to model complex environmental data, and many algorithms are difficult to put into practical use. The prediction algorithm based on deep learning shows great advantages in time series prediction tasks due to its unique network structure, but it also has high requirements for data. The network application to multivariate data requires a large number of samples, otherwise the prediction accuracy will drop sharply, and sufficient training data and computing resources are required, and there is a significant lag in the prediction results.

Summary of the invention

In order to solve the above technical problems, the present invention proposes a health risk prediction method, device, equipment and medium, which can make early prevention preparations through rapid prediction and calculation of relative danger values of health risks when the environmental quality index is abnormal.

In order to achieve the above object, the technical solution of the present invention is as follows:

A health risk prediction method comprises the following steps:

Obtain current environmental data;

The current environment data is input into a prediction model, wherein the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST model includes a VARMA model and a BI-LSTM model, and the VARMA model is used for multi-dimensional data prediction, and the prediction result of the VARMA prediction model is obtained and the prediction result is compared with the actual value to obtain a fitting residual, and the fitting residual is used as an input of the BI-LSTM model to obtain the prediction result of the BI-LSTM model, and the prediction result of the VARMA prediction model is superimposed with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST model; the prediction result of the VARLST model and the prediction result of the AdaBoost model are weighted and fused to obtain a final prediction value;

Based on the final predicted value predicted under the current environmental data and the number of disease patients and non-disease patients under the risk reference benchmark value in the past time period, the relative risk value of the predicted health risk is calculated; health risk warning is carried out based on the relative risk value.

Preferably, the training process of the VARLST model is as follows:

Obtain historical environmental data and medical statistics data and preprocess them, and divide the preprocessed data set into a training set and a test set according to a preset ratio;

The data set is subjected to difference smoothing, and the autocorrelation function and partial autocorrelation function analysis are performed on the difference smoothed data. Based on the analysis results, the parameters of the VARMA prediction model and the order of the difference are estimated. The VARMA prediction model is:

Where y _n,t is the number of medical visits in n departments on the tth day; y _n,tp is the p-order lag of the number of medical visits in n departments on the tth day; θ _n,t is the time series composition structure of the number of medical visits in n departments on the tth day; A is the coefficient corresponding to the composition structure; B _p is the coefficient corresponding to the p-order lag; C is the coefficient of the exogenous variable; x _m,t is the environmental index data of m items on the tth day, ε _n is the residual term,

The training set is input into a VARLST model for training, wherein the VARLST model includes a VARMA model and a BI-LSTM model, wherein the VARMA model is used for multi-dimensional data prediction, and the prediction result of the obtained VARMA prediction model is compared with the actual value to obtain a fitting residual;

Input the fitted residual into the BI-LSTM model for training until the preset requirements or number of iterations are met, and output the trained BI-LSTM model. In each iteration, the BI-LSTM model is backward-transferred, and the error of the output layer at each moment and the parameter derivatives of the forward LSTM and backward LSTM are calculated through the output of the backward transfer. After obtaining the parameter derivatives, the network parameters of the BI-LSTM model are updated;

Input the test set into the trained BI-LSTM model to obtain the prediction results of the BI-LSTM model;

The prediction results of the VARMA prediction model are superimposed on the prediction results of the BI-LSTM model to obtain the prediction results of the VARLST model.

Preferably, the training process of the AdaBoost model is as follows:

Train the base learner based on the data set and the initial weight distribution, and optimize the base learner;

Calculate the prediction error, proportional error and connection weight of each training sample in the tth iteration;

Adjust the weight distribution of each training sample in the t+1 iteration, and continue training until the maximum number of iterations is reached to obtain K models;

The prediction results of the K models are added together as the prediction result of the AdaBoost model.

Preferably, the historical environmental data includes air quality data, urban drinking water quality data, waste gas, wastewater and other total emission data.

Preferably, obtaining historical environmental data and medical statistics and preprocessing them includes the following steps:

Perform data cleaning on environmental data and medical statistics, including processing missing values, outliers and duplicate values;

Perform correlation analysis on the cleaned data, calculate the correlation coefficients between each feature using the Pearson correlation coefficient, and finally obtain the correlation coefficient matrix C;

According to the correlation coefficient matrix C, environmental indices with high correlation with medical statistics data are selected as input variables of the model.

Preferably, MAE, MAPE, and RMSE are selected as evaluation indicators to evaluate the performance of the prediction model.

Preferably, the relative risk value RR of the predicted health risk is calculated by the following formula:

Among them, IE is the predicted number of people suffering from the disease and the predicted number of people with the disease, IN is the predicted number of people without the disease, CE is the number of people suffering from the disease under the risk reference benchmark value, and CN is the number of people without the disease under the risk reference benchmark.

Based on the above content, the present invention also discloses a health risk prediction device, including: a data acquisition module, a mixed multivariate time series prediction module and a health risk prediction module, wherein:

The data acquisition module is used to acquire current environment data;

The hybrid multivariate time series prediction module is used to input the current environmental data into a prediction model, wherein the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST model includes a VARMA model and a BI-LSTM model, wherein the VARMA model is used for multi-dimensional data prediction, and the prediction result of the VARMA prediction model is obtained and the prediction result is compared with the actual value to obtain a fitting residual, and the fitting residual is used as an input of the BI-LSTM model to obtain the prediction result of the BI-LSTM model, and the prediction result of the VARMA prediction model is superimposed with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST model; the prediction result of the VARLST model and the prediction result of the AdaBoost model are weighted and fused to obtain a final prediction value;

The health risk prediction module is used to calculate the relative risk value of the predicted health risk based on the final predicted value predicted under the current environmental data and the number of people with the disease and the number of people without the disease under the risk reference baseline value in the past time period; and to issue a health risk warning based on the relative risk value.

Based on the above content, the present invention further discloses a computer device, including: a memory for storing a computer program; and a processor for implementing any of the above methods when executing the computer program.

Based on the above content, the present invention further discloses a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, any of the methods described above is implemented.

Based on the above technical scheme, the beneficial effects of the present invention are as follows: the present invention first obtains the data of various air quality indexes, urban drinking water quality data, waste gas, wastewater and other total emission data, and medical population statistics, performs data cleaning, data correlation analysis, and feature selection, so that the data is more complete and accurate. The selected feature data is trained by VARLST and AdaBoost models respectively. First, after ACF and PACF analysis of the pre-processed data, the parameters and difference order of the VARMA prediction model are estimated, a multidimensional data prediction network is established for data analysis and prediction, the obtained prediction results are compared with the actual values, the obtained fitting residuals are trained by the BI-LSTM network, and the output prediction data is analyzed and processed by the neural network, and the prediction results of the two are calculated and corrected to obtain the prediction results of the VARLST model. On the other hand, the pre-processed data is initialized with weight distribution, iteratively trained with weak regressors, and the weights of weak regressors are adjusted and combined for prediction, and the final prediction results of the AdaBoost model are obtained. The prediction results of the two models are averaged and fused to obtain the final prediction results. The final predicted number of sick people and non-sick people are compared and calculated with the risk reference benchmark value to obtain the relative risk value, judge the severity of health risks, and achieve the purpose of predicting health risks in advance. The present invention uses the Pearson correlation coefficient to analyze the data correlation, solves the problem of requiring a large amount of data for prediction, improves prediction efficiency, and strengthens the connection between environmental factors and the number of medical treatments; through the VARLST model, combined with the advantages of deep learning and traditional statistical prediction algorithms, the prediction model has higher execution efficiency and generalization ability to improve the accuracy and practicality of the health risk prediction model; the AdaBoost model is integrated, combined with the integrated learning advantages of the AdaBoost model, and important features are automatically selected to improve the prediction accuracy and stability of the model.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a diagram of an application environment of a health risk prediction method in an embodiment;

FIG2 is a flow chart of a health risk prediction method in one embodiment;

FIG3 is a schematic diagram of the architecture of a VARLST-AdaBoost hybrid model in a health risk prediction method in one embodiment;

FIG4 is a schematic structural diagram of a health risk prediction device in one embodiment;

FIG. 5 is a diagram showing the internal structure of a computer device in an embodiment.

Detailed ways

The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

A health risk prediction method provided in an embodiment of the present application can be applied in an application environment as shown in FIG1 . As shown in FIG1 , the application environment includes a computer device 110 . The computer device 110 can obtain current environmental data; the computer device 110 can input the current environmental data into a prediction model (VARLST-AdaBoost hybrid model), which includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST model includes a VARMA model and a BI-LSTM model, the VARMA model is used for multi-dimensional data prediction, and the prediction result of the VARMA prediction model is obtained and the prediction result is compared with the actual value to obtain the fitting residual, and the fitting residual is used as the input of the BI-LSTM model to obtain the prediction result of the BI-LSTM model, and the prediction result of the VARMA prediction model is superimposed with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST model; the prediction result of the VARLST model and the prediction result of the AdaBoost model are weighted and fused to obtain a final prediction value; the computer device 110 can calculate the relative risk value of the predicted health risk based on the final prediction value predicted under the current environmental data and the number of people with the disease and the number of people without the disease under the risk reference benchmark value in the past time period; and health risk warning is performed based on the relative risk value. The computer device 110 may be, but is not limited to, various personal computers, laptop computers, robots, tablet computers, and other devices.

In one embodiment, as shown in FIGS. 2 and 3 , a health risk prediction method is provided, comprising the following steps:

Step 1: Obtain historical data on air quality data such as AQI, PM2.5, PM10, CO, NO2, SO2, O3, urban drinking water quality data, total amount of waste gas, wastewater and other emissions, and medical treatment statistics.

Step 2: Preprocess all the historical data obtained. Specifically, it includes the following steps:

Step 2-1: Clean all historical data obtained, including processing missing values, outliers and duplicate values, to ensure the integrity and accuracy of the data;

Step 2-2: Perform correlation analysis on the cleaned data, calculate the correlation coefficient between each feature through the Pearson correlation coefficient, and finally obtain a correlation coefficient matrix C. If the correlation coefficient is close to 1 or -1, it means that there is a strong linear correlation between the two. If the correlation coefficient is close to 0, it means that there is no linear correlation between the two.

Step 2-3: According to the correlation coefficient matrix C, select environmental indices with high correlation with the statistical data of the number of medical visits as the input variables of the model. The size and significance of the correlation coefficient determine which environmental indices can be used as input variables for a specific disease. Selecting index dimensions with high correlation with the disease and significant correlation coefficient as input variables can more effectively improve the prediction performance of the model.

Step 3: Take the highly correlated environmental index data and the medical population statistics as input data, first determine the autocorrelation and partial autocorrelation of the above time series through the autocorrelation function (ACF) and partial autocorrelation function (PACF), and build a VARLST hybrid model based on the analysis to obtain the final prediction results. The specific steps include the following:

Step 3-1: The multi-parameter time series data of environmental index and medical statistics are non-stationary series. It is necessary to smooth the time series through difference processing and ensure that the conditions for applying the prediction model are met. Then, the prediction results are restored through difference reduction.

Step 3-2: Analyze the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the difference-smoothed data to determine the autocorrelation and partial autocorrelation of the time series and better understand the characteristics and trends of the data;

Step 3-3: Based on the autocorrelation function and partial autocorrelation function of the multivariate environmental time series, estimate the parameters and order of the difference of the vector autoregression moving average (VARMA) forecasting model to ensure the practicality of the forecasting model and the stability of the time series. The VARMA model adopts the form of a multi-equation system, which is usually used to predict the system of interconnected time series and analyze the dynamic impact of random disturbances on the variable system. Assuming that the optimal order is p, the total number of days in the sample data set is T, the number of medical visits for each disease is an endogenous variable, and the environmental index data is an exogenous variable, the vector autoregression model is constructed as follows:

Where y _n,t is the number of medical visits in n departments on the tth day; y _n,tp is the p-order lag of the number of medical visits in n departments on the tth day; θ _n,t is the time series composition structure of the number of medical visits in n departments on the tth day; A is the coefficient corresponding to the composition structure; B _p is the coefficient corresponding to the p-order lag; C is the coefficient of the exogenous variable; x _m,t is the environmental index data of m items on the tth day, and ε _n is the residual term;

The above equation can be simplified to:

Y _t =Aθ _t +B ₁ Y _t-1 +···+B _p Y _tp +Cx _t +ε _t ,t=1,2,···,T (2)

Step 3-4: Compare the fitted value predicted by the model with the true value to obtain the fitted residual. The calculation formula for the data fitting residual is as follows:

E _t = Z _t - X _t (3)

Where _Xt is the output fitting result, _Zt is the real data, and _Et is the data fitting residual;

Step 3-5: Build a BI-LSTM neural network model, forget and retain the fitted residual data, use the residual data to train the BI-LSTM neural network model, and test the BI-LSTM neural network model to reduce the error that the VARMA model cannot accurately analyze. Specifically, the following steps are included:

Step 3-5-1: Through a point-by-point multiplication and sigmoid neural layer operation, the information is selective when passing through, and a forget gate is constructed. The formula is as follows:

_ft = σ( _Wf ·[ht _-1 , xt _] + _bf ) (4)

Step 3-5-2: Construct the input gate structure, the formula is as follows:

_it =σ( _Wi ·[ _ht-1 , _xt ]+ _bi ) (5)

c _t = tanh(W _C ·[h _t-1 ,x _t ]+b _C ) (6)

y _t = f _t * y _{t - 1} + i _t * c _t (7)

Where W _C is the weight coefficient of c _t , W _i is the input weight coefficient, and tanh is the activation function;

Step 3-5-3: Construct the output gate structure, the formula is as follows:

o _t =σ(W _o [h _t-1 ,x _t ]+b _o ) (8)

h _t = o _t *tanh(y _t ) (9)

After steps 3-5-1 and 3-5-2, the cell information y _t is updated. The cell information is processed by the tanh activation function, and a value in the interval [-1,1] is output after processing. The output value is then multiplied by the output of the sigmoid gate to finally obtain the determined output part;

Step 3-5-4: Calculate the total output value of the BI-LSTM structure. The total output value of the BI-LSTM structure at time t is the sum of the forward LSTM and backward LSTM outputs. The formula is as follows:

in Take and operation for vector;

Step 3-5-5: Iteratively train the BI-LSTM network parameters. In each iteration, first perform a backward pass on the BI-LSTM network, and then calculate the error of the output layer at each moment and the parameter derivatives of the forward LSTM and backward LSTM through the output of the backward pass;

After obtaining the parameter derivatives, the network parameters of BI-LSTM are finally updated.

Step 3-6: The trained BI-LSTM model is used to predict the residual of the test set data, and the predicted residual result is recorded as E. Finally, the corrected test set prediction result is calculated, expressed as Y _VARLST :

_YVARLST ＝ _Xt +E (13)

Step 4: Take the highly correlated environmental index data and medical population statistics obtained in step 2 as input data, iterate through the AdaBoost model, and finally combine all weak regressors according to the weights to obtain the final prediction result. Specifically, the following steps are included:

Step 4-1: Let a sample set be U = {(x _i , y _i |i = 1, 2, ..., N)}, and the weight distribution of the sample at the tth iteration is represented by the symbol D _t(i) , t = 1, 2, ..., K. When t = 1 during the first iteration, the weight distribution is initialized

Step 4-2: Calculate the maximum error E _t = max|y _i -G _k ( _xi )| on the training set, i = 1, 2, ..., N, calculate the expected output of f _t ( _xi ) according to the weight distribution, and calculate the prediction error. The calculation formula is as follows:

Step 4-3: Calculate the proportional error. The calculation formula is as follows:

Step 4-4: Calculate the connection weight Where/>

Step 4-5: Adjust weighted distribution and/>

Step 4-6: After K iterations, the final prediction result can be obtained by

Step 4-7: The prediction result Y obtained by the VARLST model The prediction result Y obtained by _{the VARLST} and AdaBoost models _AdaBoost is fused according to their respective weights to obtain the final prediction result Y. The calculation formula is as follows:

Y＝W ₁ Y _VARLST +W ₂ Y _AdaBoost (17)

In order to reduce the risk of overfitting, reduce the variance of the model, and improve the stability and robustness of the model, average fusion is used in this embodiment, so W ₁ =W ₂ =0.5.

Step 5: Use the mean square error (RMSE), absolute percentage error (MAPE), and mean absolute error (MAE) to test the fitting efficiency of the nonlinear model training data after the time series distribution and evaluate the prediction results. The calculation formulas for the above model evaluation indicators are as follows:

Where Pi _is the predicted data value, _Xi is the true data value, is the mean of the true data, and n is the number of datasets.

Step 6: The number of people with and without diseases predicted by the current environmental factor data is predicted, and the median of the disease rate and non-disease rate in the past period is used as the risk reference benchmark value to calculate the relative risk value of health risk. The calculation formula is as follows:

Among them, RR is the relative risk value of the predicted health risk under the multi-parameter environmental index, IE is the predicted number of people with the disease, IN is the predicted number of non-disease people, CE is the number of people with the disease under the risk reference benchmark value, and CN is the number of non-disease people under the risk reference benchmark.

The health risk warning level is divided according to the relative risk value of the predicted health risk under the multi-parameter environmental index. When the relative risk value of the predicted health risk is less than 1.4, it is low risk; when the relative risk value of the predicted health risk is 1.4 to 2.9, it is medium risk; when the relative risk value of the predicted health risk is greater than 2.9, it is high risk. The purpose of predicting health risks in advance is achieved through the final defined risk level.

It should be understood that, although the various steps in the above-mentioned flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flow chart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

Simulation example:

In order to verify the effectiveness of the algorithm proposed in this example, the monthly air quality index data, monthly data on total emissions of waste gas, wastewater and other emissions, and monthly data on the number of medical treatment in Haikou City, Hainan Province from 2014 to June 2023 were selected. The VARMA model, AdaBoost model and VARLST model were used to compare the predictions with the prediction model (VARLST-AdaBoost hybrid model) in this application. Finally, the fitting performance of the model was compared according to the model evaluation index.

In this embodiment, the mean square error (RMSE), absolute percentage error (MAPE), and mean absolute error (MAE) coefficients are used to test the fitting efficiency of the nonlinear model training data after the time series distribution, and the final indicators are shown in the following table.

Table 1 Model comparison experimental results

It can be seen from Table 1 that when the total amount of data is not large, the conventional VARMA and AdaBoost prediction models have insignificant prediction effects, while the VARLST model improves the prediction accuracy to a certain extent, but there is a risk of overfitting and instability. The VARLST-AdaBoost hybrid model proposed in the present invention improves the original model, adjusts the sample weight according to the classification error rate, reduces the overfitting problem of the model, and has a more excellent performance in the accuracy of the prediction results, thereby improving the prediction performance of the model.

As shown in FIG4 , in one embodiment, a health risk prediction device 400 is provided, including: a data acquisition module 410, a mixed multivariate time series prediction module 420 and a health risk prediction module 430, wherein:

The data acquisition module 410 is used to acquire current environment data;

A hybrid multivariate time series prediction module 420 is used to input the current environmental data into a prediction model, wherein the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST model includes a VARMA model and a BI-LSTM model, wherein the VARMA model is used for multi-dimensional data prediction, obtain the prediction result of the VARMA prediction model and compare the prediction result with the actual value to obtain the fitting residual, use the fitting residual as the input of the BI-LSTM model to obtain the prediction result of the BI-LSTM model, superimpose the prediction result of the VARMA prediction model with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST model; perform weighted fusion on the prediction result of the VARLST model and the prediction result of the AdaBoost model to obtain the final prediction value;

The health risk prediction module 430 is used to calculate the relative risk value of the predicted health risk based on the final predicted value predicted under the current environmental data and the number of people with the disease and the number of people without the disease under the risk reference baseline value in the past time period; and to issue a health risk warning based on the relative risk value.

In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG5. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a health risk prediction method is implemented. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

Those skilled in the art will understand that the structure shown in FIG. 5 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

Obtain current environmental data;

The current environment data is input into a prediction model, wherein the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST model includes a VARMA model and a BI-LSTM model, wherein the VARMA model is used for multi-dimensional data prediction, and the prediction result of the VARMA prediction model is obtained and the prediction result is compared with the actual value to obtain a fitting residual, and the fitting residual is used as an input of the BI-LSTM model to obtain the prediction result of the BI-LSTM model, and the prediction result of the VARMA prediction model is superimposed with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST model; and the prediction result of the VARLST model and the prediction result of the AdaBoost model are weightedly fused to obtain a final prediction value;

Based on the final predicted value predicted under the current environmental data and the number of disease patients and non-disease patients under the risk reference benchmark value in the past time period, the relative risk value of the predicted health risk is calculated; health risk warning is carried out based on the relative risk value.

In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

Obtain current environmental data;

The current environment data is input into a prediction model, wherein the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST model includes a VARMA model and a BI-LSTM model, and the VARMA model is used for multi-dimensional data prediction, and the prediction result of the VARMA prediction model is obtained and the prediction result is compared with the actual value to obtain a fitting residual, and the fitting residual is used as an input of the BI-LSTM model to obtain the prediction result of the BI-LSTM model, and the prediction result of the VARMA prediction model is superimposed with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST model; the prediction result of the VARLST model and the prediction result of the AdaBoost model are weighted and fused to obtain a final prediction value;

Based on the final predicted value predicted under the current environmental data and the number of disease patients and non-disease patients under the risk reference benchmark value in the past time period, the relative risk value of the predicted health risk is calculated; health risk warning is carried out based on the relative risk value.

Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims (7)

1. A health risk prediction method, characterized in that it comprises the following steps: Obtain current environmental data; Input the current environment data into a prediction model, wherein the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST hybrid model includes a VARMA model and a BI-LSTM model, and the VARMA model is used for multi-dimensional data prediction, obtain the prediction result of the VARMA prediction model, and compare the prediction result with the actual value to obtain the fitting residual, use the fitting residual as the input of the BI-LSTM model to obtain the prediction result of the BI-LSTM model, superimpose the prediction result of the VARMA prediction model with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST hybrid model; perform weighted fusion on the prediction result of the VARLST hybrid model and the prediction result of the AdaBoost model to obtain the final prediction value; Based on the final predicted value predicted under the current environmental data and the number of people with the disease and the number of people without the disease under the risk reference baseline value in the past time period, the relative risk value of the predicted health risk is calculated; based on the relative risk value, a health risk warning is performed, and the relative risk value RR of the predicted health risk is calculated as follows: Among them, IE is the predicted number of people with the disease and the predicted number of people with the disease, IN is the predicted number of people without the disease, CE is the number of people with the disease under the risk reference benchmark value, and CN is the number of people without the disease under the risk reference benchmark. The training process of the VARLST hybrid model is specifically as follows: historical environmental data and medical statistics are obtained and preprocessed, and the preprocessed data set is divided into a training set and a test set according to a preset ratio; the data set is subjected to differential smoothing, and the autocorrelation function and partial autocorrelation function analysis are performed on the differential smoothed data, and the VARMA prediction model parameters and the order of the difference are estimated based on the analysis results. The VARMA prediction model is: Where y _n,t is the number of medical visits in n departments on the tth day; y _n,tp is the p-order lag of the number of medical visits in n departments on the tth day; θ _n,t is the time series composition structure of the number of medical visits in n departments on the tth day; A is the coefficient corresponding to the composition structure; B _p is the coefficient corresponding to the p-order lag; C is the coefficient of the exogenous variable; x _m,t is the environmental index data of m items on the tth day, ε _n is a residual term, and the training set is input into a VARLST hybrid model for training. The VARLST hybrid model includes a VARMA model and a BI-LSTM model. The VARMA model is used for multi-dimensional data prediction, and the prediction result of the obtained VARMA prediction model is compared with the actual value to obtain a fitting residual; the fitting residual is input into a BI-LSTM model for training until the preset requirement or the number of iterations is reached, and the trained BI-LSTM model is output, wherein, in each iteration, the BI-LSTM model is backward-transferred, and the error of the output layer at each moment and the parameter derivatives of the forward LSTM and the backward LSTM are calculated through the output of the backward transfer, and after the parameter derivatives are obtained, the network parameters of the BI-LSTM model are updated; the test set is input into the trained BI-LSTM model to obtain the prediction result of the BI-LSTM model; the prediction result of the VARMA prediction model is superimposed with the prediction result of the BI-LSTM model to obtain the prediction result of the VARLST hybrid model; The training process of the AdaBoost model is specifically as follows: training a base learner based on a data set and an initialization weight distribution to optimize the base learner; calculating the prediction error, proportional error, and connection weight of each training sample in the t-th iteration; adjusting the weight distribution of each training sample in the t+1 iteration, and continuing training until the maximum number of iterations is reached to obtain K models; and adding the prediction results of the K models as the prediction result of the AdaBoost model. 2. A health risk prediction method according to claim 1, characterized in that the historical environmental data includes air quality data, urban drinking water quality data, waste gas, wastewater and other total emission data. 3. A health risk prediction method according to claim 1, characterized in that the acquisition of historical environmental data and medical statistics data and preprocessing specifically comprises the following steps: Perform data cleaning on environmental data and medical statistics, including processing missing values, outliers and duplicate values; Perform correlation analysis on the cleaned data, calculate the correlation coefficients between each feature using the Pearson correlation coefficient, and finally obtain the correlation coefficient matrix C; According to the correlation coefficient matrix C, environmental indices with high correlation with medical statistics are selected as input variables of the model. 4. A health risk prediction method according to claim 1, characterized in that MAE, MAPE, and RMSE are selected as evaluation indicators to evaluate the performance of the prediction model. 5. A health risk prediction device, characterized by comprising: a data acquisition module, a mixed multivariate time series prediction module and a health risk prediction module, wherein: The data acquisition module is used to acquire current environment data; The hybrid multivariate time series prediction module is used to input the current environmental data into the prediction model, the prediction model includes a VARLST hybrid model and an AdaBoost model, wherein the VARLST hybrid model includes a VARMA model and a BI-LSTM model, the VARMA model is used for multi-dimensional data prediction, the prediction result of the VARMA prediction model is obtained and the prediction result is compared with the actual value to obtain the fitting residual, the fitting residual is used as the input of the BI-LSTM model, the prediction result of the BI-LSTM model is obtained, and the prediction result of the VARMA prediction model is compared with the prediction result of the BI-LSTM model. The prediction results of the VARLST hybrid model are obtained by superimposing the prediction results of the VARLST hybrid model and the prediction results of the AdaBoost model, and the final prediction value is obtained. The training process of the VARLST hybrid model is specifically as follows: historical environmental data and medical statistics data are obtained and preprocessed, and the preprocessed data set is divided into a training set and a test set according to a preset ratio; the data set is subjected to differential smoothing, and the autocorrelation function and partial autocorrelation function analysis are performed on the differential smoothed data, and the parameters of the VARMA prediction model and the order of the difference are estimated based on the analysis results. The VARMA prediction model is: Where y _n,t is the number of medical visits in n departments on the tth day; y _n,tp is the p-order lag of the number of medical visits in n departments on the tth day; θ _n,t is the time series composition structure of the number of medical visits in n departments on the tth day; A is the coefficient corresponding to the composition structure; B _p is the coefficient corresponding to the p-order lag; C is the coefficient of the exogenous variable; x _m,t is the environmental index data of m items on the tth day, ε _n is a residual term. The training set is input into a VARLST hybrid model for training. The VARLST hybrid model includes a VARMA model and a BI-LSTM model. The VARMA model is used for multi-dimensional data prediction. The prediction result of the obtained VARMA prediction model is compared with the actual value to obtain a fitting residual. The fitting residual is input into a BI-LSTM model for training until the preset requirement or the number of iterations is met, and a trained BI-LSTM model is output. In each iteration, the BI-LSTM model is backward-transferred. Through the output of the backward transfer, the error of the output layer at each moment and the parameter derivatives of the forward LSTM and the backward LSTM are calculated. After the parameter derivatives are obtained, the BI-LSTM model is updated. network parameters; input the test set into the trained BI-LSTM model to obtain the prediction results of the BI-LSTM model; superimpose the prediction results of the VARMA prediction model with the prediction results of the BI-LSTM model to obtain the prediction results of the VARLST hybrid model; the training process of the AdaBoost model is specifically as follows: training the base learner based on the data set and the initialization weight distribution to optimize the base learner; calculating the prediction error, proportional error and connection weight of each training sample in the tth iteration; adjusting the weight distribution of each training sample in the t+1th iteration, and continuing training until the maximum number of iterations is reached to obtain K models; adding the prediction results of the K models as the prediction results of the AdaBoost model; The health risk prediction module is used to calculate the relative risk value of the predicted health risk based on the final predicted value predicted under the current environmental data and the number of people with the disease and the number of people without the disease under the risk reference baseline value in the past time period; and to perform health risk warning based on the relative risk value. The relative risk value RR of the predicted health risk is calculated by the following formula: Among them, IE is the predicted number of people suffering from the disease and the predicted number of people with the disease, IN is the predicted number of people without the disease, CE is the number of people suffering from the disease under the risk reference benchmark value, and CN is the number of people without the disease under the risk reference benchmark. 6. A computer device, characterized by comprising: a memory for storing a computer program; and a processor for implementing the method according to any one of claims 1 to 4 when executing the computer program. 7. A readable storage medium, characterized in that a computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.