Respiratory system disease death number prediction method

Abstract

The invention relates to the technical field of hybrid model prediction, in particular to a respiratory system disease death number prediction method, which comprises the following steps: determining each influence factor related to the respiratory system disease death number, and acquiring a training data set based on each influence factor; constructing a hybrid neural network model in combination with the multi-layer LSTM, and taking the hybrid neural network model as a prediction model; training the prediction model by using the training data set to obtain a trained prediction model; and inputting each influence factor of the current time period into the trained prediction model, and predicting the number of respiratory system disease death in the future. According to the method, the prediction model with higher prediction accuracy is obtained, so that the prediction accuracy and reliability of the respiratory system disease death number are improved.

Description

Technical Field

[0001] This invention relates to the field of hybrid model prediction technology, specifically to a method for predicting the number of deaths from respiratory diseases. Background Technology

[0002] With the advancement of urbanization and industrialization, respiratory diseases caused by air pollution have become a serious public health problem worldwide. Numerous epidemiological studies have shown that air pollutants and meteorological factors have negative health effects on the incidence and mortality of respiratory diseases. Therefore, predicting the number of deaths from respiratory diseases has significant scientific and social value, and plays a positive role in improving environmental quality, promoting public health, and enhancing government governance capabilities.

[0003] Current predictions of respiratory disease mortality typically employ regression models. However, due to limitations in observation time and sampling intervals, regression models may produce spurious trends, resulting in low accuracy in predicting mortality. Furthermore, the diverse range of environmental factors and their complex relationships with respiratory disease mortality make it difficult for traditional regression and statistical models to capture these intricate nonlinear relationships, leading to unreliable predictions of respiratory disease mortality. Summary of the Invention

[0004] To address the aforementioned technical problem of low accuracy in predicting the number of deaths from respiratory diseases, the present invention aims to provide a method for predicting the number of deaths from respiratory diseases, the specific technical solution of which is as follows:

[0005] One embodiment of the present invention provides a method for predicting the number of deaths from respiratory diseases, the method comprising the following steps:

[0006] Identify the various influencing factors related to the number of deaths from respiratory diseases, and obtain a training dataset based on these factors.

[0007] A hybrid neural network model is constructed by combining multi-layer LSTM, and the hybrid neural network model is used as a prediction model;

[0008] The prediction model is trained using the training dataset to obtain a trained prediction model;

[0009] By inputting various influencing factors of the current period into the trained prediction model, the number of future deaths from respiratory diseases can be predicted.

[0010] Furthermore, various influencing factors related to the number of deaths from respiratory diseases were identified, including:

[0011] The influencing factors mentioned belong to air pollutants or meteorological factors;

[0012] Obtain data sequences for each type of air pollutants, each type of meteorological factors, and the number of deaths from respiratory diseases for any city within a set historical period.

[0013] The data sequences corresponding to air pollutants and meteorological factors are used as undetermined data sequences, and the data sequence of deaths from respiratory diseases is used as the baseline data sequence.

[0014] Determine the correlation index between the pending data sequence and the benchmark data sequence. If the correlation index is greater than the set correlation threshold, then the type corresponding to the pending data sequence is taken as an influencing factor.

[0015] Furthermore, determining the correlation index between the data sequence to be determined and the reference data sequence includes:

[0016] In the formula, ρ represents the correlation index between the undetermined data sequence and the baseline data sequence, and x i Let x' represent the i-th data point in the undetermined data sequence, and let x' represent the average value of the data in the undetermined data sequence. i Let represent the i-th data point in the baseline data sequence, y′ represent the average value of the baseline data sequence, and || represent the absolute value function.

[0017] Furthermore, the prediction model includes a preprocessing layer, a neural network recurrent layer, a fully connected layer, and an output processing layer;

[0018] For the preprocessing layer, sliding window technology and CW-Normalization algorithm are used to process the original data sequence sequentially.

[0019] For recurrent layers of neural networks, the Dropout technique is introduced into the recurrent layers of neural networks;

[0020] For fully connected layers, use the Linear function for processing;

[0021] For the output processing layer, add the CW-Denormalization algorithm.

[0022] Furthermore, the expression for the CW-Normalization algorithm is:

[0023] In the formula, x represents the i-th data point in the n-th column of the normalized data. (ni) X represents the i-th data in the n-th column. (n) This represents the data in the nth column, min(X) (n)) represents the minimum value in the nth column of data, max(X) (n) () represents the maximum value in the nth column of data.

[0024] Furthermore, the expression for the Linear function is:

[0025] Output = m × W + b; where Output represents the output data of the fully connected layer, m represents the vector of input data of the fully connected layer, W represents the weight matrix, b represents the bias vector, and × represents matrix multiplication.

[0026] Further, the various training parameters of the prediction model are set;

[0027] The training parameters include neuron activation function, fully connected layer activation function, optimization function, number of neurons, loss function, number of network iterations, batch size, number of model stacking layers, and random inactivation rate.

[0028] Furthermore, the neuron activation function is set to the ReLU activation function, the fully connected layer activation function is set to the linear activation function, the optimization function is set to the Adam function, the number of neurons is set to 200, the loss function is set to the mean squared error function, the number of network iterations is set to 100, the batch size is set to 64, the number of model stacking layers is set to 2, and the random inactivation rate is set to 0.2.

[0029] The present invention has the following beneficial effects:

[0030] This invention provides a method for predicting the number of deaths from respiratory diseases. First, to reduce computational load, various influencing factors related to the number of deaths from respiratory diseases are identified before acquiring the training dataset, thus avoiding the influence of numerous environmental factors and selecting only a subset of influencing factors. A hybrid neural network model constructed using multi-layer LSTM can eliminate the dimensional influence between different features, ensuring they have the same dimensions and similar scales, while possessing strong expressive and generalization abilities. The resulting trained prediction model outperforms the classic GAM model and other deep learning prediction models in terms of prediction performance, thereby improving the accuracy of predicting the number of deaths from respiratory diseases. Attached Figure Description

[0031] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of a method for predicting the number of deaths from respiratory diseases according to an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the MLSCD model structure in an embodiment of the present invention. Detailed Implementation

[0034] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solution proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0036] This invention provides a method for predicting the number of deaths from respiratory diseases, such as... Figure 1 As shown, it includes the following steps:

[0037] S1, identify the various influencing factors related to the number of deaths from respiratory diseases, and obtain a training dataset based on each influencing factor.

[0038] First, it should be noted that the data on influencing factors are of the nature of air pollutants or meteorological factors. When predicting the number of deaths from respiratory diseases, it is necessary to identify predictive factors related to the prediction results and obtain a large amount of historical data based on the predictive factors in order to train the prediction model.

[0039] Step S1 above can be achieved through the following steps S11 to S12:

[0040] S11, identify the various influencing factors related to the number of deaths from respiratory diseases.

[0041] In this embodiment, not all types of air pollutants or meteorological factors are related to the number of deaths from respiratory diseases. In order to reduce the amount of data to a certain extent, it is necessary to find the influencing factors that are strongly correlated with the number of deaths from respiratory diseases. Specifically, the Spearman correlation coefficient method is used to assess the correlation between air pollutants, meteorological factors and the number of deaths from respiratory diseases.

[0042] Air pollutants mainly refer to pollutants in the air, including but not limited to particulate matter, ozone, sulfur dioxide, nitrogen dioxide, and carbon monoxide. They are usually generated by human activities (such as industrial emissions, traffic emissions, and combustion processes) and have a negative impact on the environment and human health. Meteorological factors mainly refer to factors that affect atmospheric movement and climate change, including but not limited to temperature, humidity, wind speed, air pressure, precipitation, and weather type. They are closely related to the dynamic changes in the atmospheric environment and have an important impact on the diffusion, transport, and deposition of air pollutants.

[0043] The above step S11 can be achieved through the following steps S111 to S113:

[0044] S111: Obtain data sequences for each type of air pollutants, each type of meteorological factors, and the number of deaths from respiratory diseases for any city within a set historical period.

[0045] As an example, we obtain the data sequences for a city over a historical 7-day period. All data sequences include data sequences for each type of air pollutants, data sequences for each type of meteorological factors, and data sequences for the number of deaths from respiratory diseases.

[0046] When collecting data on air pollutants and meteorological factors, the data collection interval is 1 hour, meaning data is collected every hour. However, the average value of the same type of data collected within a day should be used as an element in the data sequence. The collection frequency for the number of deaths from respiratory diseases is once a day. Implementers can set the length of the historical period and the data collection frequency according to the specific circumstances, without making specific limitations here.

[0047] S112 uses the data sequences corresponding to air pollutants and meteorological factors as undetermined data sequences, and the data sequence of deaths from respiratory diseases as the baseline data sequence.

[0048] In this embodiment, to facilitate understanding of the scheme and reduce description, all data sequences corresponding to air pollutants and meteorological factors can be collectively referred to as undetermined data sequences, and there are multiple undetermined data sequences; at the same time, in order to compare with the undetermined data sequences, the data sequence of deaths from respiratory diseases can be used as the baseline data sequence, and at this time there is only one baseline data sequence.

[0049] It is worth noting that the number of data points in the undetermined data sequence and the baseline data sequence are the same.

[0050] S113, determine the correlation index between the pending data sequence and the benchmark data sequence. If the correlation index is greater than the set correlation threshold, then the type corresponding to the pending data sequence is taken as an influencing factor.

[0051] In this embodiment, to facilitate understanding of the scheme, taking any undetermined data sequence as an example, we determine whether the type corresponding to the undetermined data sequence is an influencing factor that is strongly correlated with the number of deaths from respiratory diseases.

[0052] Specifically, the Spearman method is used to quantify the correlation between the pending data sequence and the baseline data sequence, determine the correlation index, and apply a threshold to the correlation index. The data sequence type corresponding to the pending data sequence whose correlation index exceeds the set threshold is considered an influencing factor. Of course, implementers can also use other methods to measure the correlation between two sequences, including but not limited to: Pearson correlation coefficient, Kendall correlation coefficient, and cross-entropy.

[0053] The correlation threshold can be set to 0.25. The value of the correlation threshold can be set by the implementer according to the specific actual situation, without any specific limitation. The implementation process of the Spearman method is existing technology and is not within the scope of protection of this invention. It will not be described in detail here.

[0054] As an example, the formula for calculating the correlation index between the undetermined data sequence and the baseline data sequence can be:

[0055] In the formula, ρ represents the correlation index between the undetermined data sequence and the baseline data sequence, and x i Let x' represent the i-th data point in the undetermined data sequence, and let x' represent the average value of the data in the undetermined data sequence. i Let represent the i-th data point in the baseline data sequence, y′ represent the average value of the baseline data sequence, and || represent the absolute value function.

[0056] S12, obtain the training dataset based on various influencing factors.

[0057] In this embodiment, after obtaining the influencing factors related to the number of deaths from respiratory diseases, in order to facilitate the subsequent training of the prediction model, it is necessary to collect historical data and label data corresponding to each influencing factor from the relevant database. The label data is the number of deaths from respiratory diseases. The dataset composed of the collected data is used as the training dataset, and the training dataset is used to train the prediction model.

[0058] The training dataset, which is the sample set input for model training, can contain 1453 samples. The time step can be set to 3. The model input has 9 feature dimensions, representing 9 types of influencing factors: respiratory disease deaths, SO2 (sulfur dioxide), NO2 (nitrogen dioxide), CO (carbon monoxide), and PM2.5. 10 (Particulate matter with a diameter of less than 10 micrometers in the atmosphere) PM 2.5 (Particles with a diameter of less than 2.5 micrometers in the atmosphere, Particulate matter 2.5), maximum temperature, minimum temperature, and average pressure; the ratio of training set to test set can be divided into 8:2.

[0059] Thus, this embodiment has obtained the training dataset for training the model.

[0060] S2 combines a multi-layer LSTM to construct a hybrid neural network model, which is then used as the prediction model.

[0061] First, it's important to clarify that a Recurrent Neural Network (RNN) is a neural network model that constructs non-linear relationships along a time axis. An LSTM (Long Short-Term Memory) model is a variant of an RNN, possessing excellent time-series processing capabilities and broad application prospects in the medical field. It is particularly suitable for processing data with time delays and exhibits good predictive performance. The hybrid neural network model constructed by combining LSTM can overcome the limitations of observation time and sampling intervals, avoid spurious trends, and improve the prediction accuracy of the predictive model. This predictive model can be denoted as an MLSCD (Multilayer LSTM Structure with CW-Normalization and Dropout) model.

[0062] The MLSCD model has a four-layer structure: a preprocessing layer, a recurrent neural network layer, a fully connected layer, and an output processing layer. A schematic diagram of the MLSCD model structure is shown below. Figure 2 As shown, in Figure 2 In the middle, X t X represents the input feature vector at time step t, which contains the data of the current time step and serves as the input to the LSTM unit. t-1 Let X represent the input feature vector at time step t-1.t+1 This represents the input feature vector at time step t+1; LSTM cell (Long Short-Term Memory Cell) represents the Long Short-Term Memory layer. An LSTM cell is the basic computational unit of the LSTM model, containing memory cells, input gates, forget gates, and output gates, used to process and store long and short-term dependencies in sequential data; Dropout represents random deactivation. Dropout is a regularization technique used to prevent overfitting in neural networks. In each training step, a portion of neurons are randomly deactivated, making the network insensitive to certain specific paths, thereby enhancing the model's generalization ability; D t D represents the output of the fully connected layer at time step t. t-1 D represents the output of the fully connected layer at time step t-1. t+1 Y represents the output of the fully connected layer at time step t+1. t Y represents the output of the output layer at time step t. t-1 Y represents the output of the output layer at time step t-1. t+1 This represents the output of the output layer at time step t+1; CW-Denormalization (Column-wise Min-Max Denormalization) represents the inverse normalization algorithm.

[0063] S21. For the preprocessing layer, the sliding window technique and the CW-Normalization algorithm are used to process the input data of the preprocessing layer in sequence.

[0064] In this embodiment, to ensure the prediction accuracy and training effect of the neural network model, the original data sequence is processed in the preprocessing layer using both sliding window technology and column-wise min-max normalization (CW-Normalization) algorithm. The implementation processes of the sliding window technology and the CW-Normalization algorithm are existing technologies and will not be described in detail here.

[0065] It should be noted that the purpose of using the sliding window technique is to transform the original data sequence into data in a format suitable for model input. After the sliding window and data transformation, a data format suitable for direct input into the MLSCD model for training is obtained. The CW-Normalization algorithm can analyze the data processed by the sliding window technique. The core idea of ​​the CW-Normalization algorithm is to regard the first data point of each column of each window as the baseline value of that column, and then scale other data points relative to the baseline value. This allows the data in each column to be scaled to a fixed range while maintaining the relative relationship between the data. The CW-Normalization algorithm eliminates the dimensional differences between different features, making the model easier to converge, and provides an effective means for model training and performance optimization.

[0066] As an example, the expression for the CW-Normalization algorithm can be:

[0067] In the formula, x represents the i-th data point in the n-th column of the normalized data. (ni) X represents the i-th data in the n-th column. (n) This represents the data in the nth column, min(X) (n) ) represents the minimum value in the nth column of data, max(X) (n) () represents the maximum value in the nth column of data.

[0068] S22, For the recurrent layer of the neural network, the Dropout technique is introduced into the recurrent layer of the neural network.

[0069] In this embodiment, to further improve the expressive and generalization capabilities of the MLSCD model, Dropout technology is introduced into the recurrent layers of the LSTM model's neural network. Combining the LSTM model structure with Dropout technology—that is, adding Dropout layers between LSTM layers—allows the multi-layered LSTM and Dropout technology to fully leverage their respective advantages, improving the generalization ability and stability of the deep learning model, effectively reducing overfitting, and increasing the model's robustness.

[0070] S23. For fully connected layers, use the Linear function for processing.

[0071] In this embodiment, the fully connected layer is a basic layer type in neural networks. The MLSCD model uses the Linear function for processing; the Linear function multiplies the input data by the weight matrix and adds a bias vector to achieve a linear combination of features.

[0072] As an example, the expression for the Linear function can be:

[0073] Output = m × W + b; where Output represents the output data of the fully connected layer, m represents the vector of input data of the fully connected layer, W represents the weight matrix, b represents the bias vector, and × represents matrix multiplication.

[0074] S24. For the output processing layer, add the CW-Denormalization algorithm to the output processing layer.

[0075] In this embodiment, to ensure data validity, a CW-Denormalization algorithm is added to the output layer of the model. The purpose is to restore the predicted data output by the model back to the state before the CW-Normalization algorithm processing. The output layer is the output processing layer.

[0076] Thus, this embodiment yields a predictive model for predicting the number of deaths from respiratory diseases.

[0077] S3, use the training dataset to train the prediction model to obtain a trained prediction model.

[0078] In this embodiment, various training parameters of the prediction model are set during the training process. These parameters include neuron activation functions, fully connected layer activation functions, optimization functions, number of neurons, loss functions, number of network iterations, batch size, number of model stacking layers, and random inactivation rate. The configurations of these training parameters are shown in Table 1.

[0079] Table 1

[0080]

[0081] In Table 1, (1) neurons = 200, meaning that the number of hidden neurons within each LSTM neuron is 200; (2) samples = 1453; timestep = 3; input_dim = 9, meaning that the number of samples input for model training is 1453, the time step is 3, and the feature dimension of the model input is 9, which are the number of deaths from respiratory diseases, SO2, NO2, CO, and PM2.5, respectively. 10 PM 2.5(3) The number of stacked layers of the LSTM model is 2, the Dropout is set to 0.2, the activation function of the fully connected layer is the linear activation function, and the activation function of the neuron is the ReLU activation function; (4) The network training optimization function is the Adam function, the loss function is the mean squared error (MSE) to solve the regression problem, the number of model iterations is 100, and 64 samples are used to update the weights in each batch, that is, the batch_size is 64.

[0082] It should be noted that the training process of the prediction model is existing technology and is not within the scope of protection of this invention, and will not be described in detail here.

[0083] Thus, this embodiment has obtained a trained prediction model.

[0084] S4. Input the various influencing factors of the current period into the trained prediction model to predict the number of future deaths from respiratory diseases.

[0085] In this embodiment, the number of respiratory disease deaths in the current 3 days, SO2, NO2, CO, PM10, PM2.5, highest temperature, lowest temperature, and average pressure are used as input data for the prediction model to predict the number of respiratory disease deaths. The current time period can be set to 3 days, and the implementer can adjust the length of the current time period according to specific circumstances; there is no specific limitation.

[0086] Thus, this embodiment achieves a more accurate prediction of the number of deaths from respiratory diseases.

[0087] To verify the implementation effect of the embodiments of the present invention and to better illustrate the advantages of the embodiments, the following experiments are disclosed.

[0088] First, determine the evaluation indicators.

[0089] This embodiment uses the coefficient of determination (R²). 2 The mean absolute error (MAE) and root mean square error (RMSE) are used as evaluation metrics.

[0090] Second, prepare the dataset.

[0091] In this embodiment, the dataset is sourced from the Institute of Environmental and Health Related Product Safety, Chinese Center for Disease Control and Prevention.

[0092] Third, set up the experiment.

[0093] In this embodiment, the experiment was conducted under the Windows 10 operating system, using Python 3.10 as the programming language, TensorFlow 2.5 as the framework and corresponding version, and PyCharm Community Edition 2022.2.1 as the development tool.

[0094] Fourth, comparative experiments.

[0095] In this embodiment, the Generalized Additive Model (GAM) and the Recurrent Neural Network (RNN) are used as comparative experiments. When setting up the comparative experiments, GAM is a traditional regression model, and the model structure and data processing remain unchanged. The RNN model is divided into two comparison experiments: one using the Min-Max normalization method and the other using the CW-Normalization algorithm. Specifically, the experiments use M-RNN, CW-RNN, and MLSCD models for comparison. Here, M-RNN indicates the use of the Min-Max normalization method, and CW-RNN indicates the use of the CW-Normalization algorithm.

[0096] The evaluation results of the comparative experimental model are shown in Table 2:

[0097] Table 2

[0098]

[0099] In Table 2, (1) the MLSCD neural network model has better prediction performance than the traditional GAM model. The RMSE value of the MLSCD model is reduced by 34%, and the corresponding R 2 The value increased by 0.14, and the MAE value decreased by 0.9; (2) Compared with the model using the Min-Max normalization method, the model using the CW-Normalization algorithm has a smaller prediction error, and the R of the CW-RNN model is higher. 2 The value is 0.17 larger than that of the M-RNN model, while the MAE and RMSE values ​​are 0.67 and 1 smaller, respectively; (3) Under the same preprocessing method, the R of the MLSCD model is 0.17 larger than that of the M-RNN model, while the MAE and RMSE values ​​are 0.67 and 1 smaller, respectively; 2 The value is 0.1 larger than that of the CW-RNN model, while the MAE value is 0.5 smaller than that of the CW-RNN model, and the RMSE value is 0.6 smaller than that of the CW-RNN model.

[0100] Therefore, Table 2 shows that the MLSCD model can better capture the complex patterns and regularities of sequence data and has stronger expression and feature extraction capabilities.

[0101] Fifth, ablation experiment.

[0102] It should be noted that, in order to explore the contribution of each component of the MLSCD model to the predictive performance of respiratory disease mortality, an ablation experiment was conducted. The MLSCD model consists of the CW-Normalization algorithm, the multi-layer structure of the LSTM model, and the Dropout technique. These components will be gradually removed to evaluate their impact on model performance.

[0103] (1) Complete MLSCD model: including CW-Normalization algorithm preprocessing, multi-layer LSTM model and Dropout technology.

[0104] (2) Model without CW-Normalization algorithm (No-CWLSTM): The CW-Normalization algorithm processing step is removed from the MLSCD model, and the Min-Max normalization method is used for processing, while the LSTM model and Dropout technology are retained.

[0105] (3) Single-layer LSTM model (S-LSTM): The multi-layer LSTM model in the MLSCD model is changed to a single-layer LSTM structure, while retaining the Dropout technology and CW-Normalization algorithm.

[0106] (4) No Dropout technology (No-DLSTM): The Dropout technology is removed from the MLSCD model, while the CW-Normalization algorithm and multi-layer LSTM model are retained.

[0107] The evaluation results of the ablation experimental model are shown in Table 3:

[0108] Table 3

[0109]

[0110] Table 3 shows that: (1) The MLSCD model without the CW-Normalization algorithm showed a significant decrease in the number of deaths from respiratory diseases, indicating that the preprocessing of the CW-Normalization algorithm is crucial to the model performance; (2) The single-layer LSTM model showed a decrease in prediction accuracy compared to the multi-layer LSTM model, indicating that the multi-layer structure has certain advantages in capturing sequence information; (3) The MLSCD model without Dropout technology showed a higher risk of overfitting, verifying the importance of Dropout technology in the model's generalization ability.

[0111] In summary, this invention first uses air pollutant data, meteorological factor data, and respiratory disease mortality data as a basis, and performs correlation analysis using the Spearman method to select factors with strong correlations. Then, based on a Long Short-Term Memory (LSTM) neural network model, and incorporating column-based Max-Min Normalization (CW-Normalization) and Dropout techniques, a hybrid neural network model combining multi-layer LSTM is proposed. Finally, by training the MLSCD model, the best-performing model is selected for prediction, yielding the prediction results. This invention improves the hybrid neural network model to obtain a more accurate prediction model, thereby enhancing the accuracy and reliability of predicting respiratory disease mortality.

[0112] This concludes the embodiments of the present invention.

[0113] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting the number of deaths from respiratory diseases, characterized in that, Includes the following steps: Identify the various influencing factors related to the number of deaths from respiratory diseases, and obtain a training dataset based on these factors. A hybrid neural network model is constructed by combining multi-layer LSTM, and the hybrid neural network model is used as a prediction model; The prediction model is trained using the training dataset to obtain a trained prediction model; By inputting various influencing factors of the current period into the trained prediction model, the number of future deaths from respiratory diseases can be predicted.

2. The method for predicting the number of deaths from respiratory diseases according to claim 1, characterized in that, The factors identified as being associated with respiratory disease-related deaths include: The influencing factors mentioned belong to air pollutants or meteorological factors; Obtain data sequences for each type of air pollutants, each type of meteorological factors, and the number of deaths from respiratory diseases for any city within a set historical period. The data sequences corresponding to air pollutants and meteorological factors are used as undetermined data sequences, and the data sequence of deaths from respiratory diseases is used as the baseline data sequence. Determine the correlation index between the pending data sequence and the benchmark data sequence. If the correlation index is greater than the set correlation threshold, then the type corresponding to the pending data sequence is taken as an influencing factor.

3. The method for predicting the number of deaths from respiratory diseases according to claim 2, characterized in that, The method for determining the correlation index between the data sequence to be determined and the reference data sequence includes: In the formula, ρ represents the correlation index between the undetermined data sequence and the baseline data sequence, and x i Let x represent the i-th data in the undetermined data sequence. ′ y represents the average value of the undetermined data sequence. i y represents the i-th data point in the baseline data sequence. ′ represents the average value of the baseline data sequence, and || represents the function to remove the absolute value.

4. The method for predicting the number of deaths from respiratory diseases according to claim 2, characterized in that, The prediction model includes a preprocessing layer, a neural network recurrent layer, a fully connected layer, and an output processing layer; For the preprocessing layer, sliding window technology and CW-Normalization algorithm are used to process the original data sequence sequentially. For recurrent layers of neural networks, the Dropout technique is introduced into the recurrent layers of neural networks; For fully connected layers, use the Linear function for processing; For the output processing layer, add the CW-Denormalization algorithm.

5. The method for predicting the number of deaths from respiratory diseases according to claim 4, characterized in that, The expression for the CW-Normalization algorithm is: In the formula, x represents the i-th data point in the n-th column of the normalized data. (ni) X represents the i-th data in the n-th column. (n) This represents the data in the nth column, min(X) (n) ) represents the minimum value in the nth column of data, max(X) (n) () represents the maximum value in the nth column of data.

6. The method for predicting the number of deaths from respiratory diseases according to claim 4, characterized in that, The expression for the Linear function is: Output = m × W + b; where Output represents the output data of the fully connected layer, m represents the vector of input data of the fully connected layer, W represents the weight matrix, b represents the bias vector, and × represents matrix multiplication.

7. The method for predicting the number of deaths from respiratory diseases according to claim 1, characterized in that, Set the various training parameters of the prediction model; The training parameters include neuron activation function, fully connected layer activation function, optimization function, number of neurons, loss function, number of network iterations, batch size, number of model stacking layers, and random inactivation rate.

Images