A pneumonia epidemic control method and system based on deep learning

Abstract

The application discloses a kind of pneumonia epidemic situation control method and system based on deep learning, comprising the following steps: step S1, based on the historical rehabilitation time series data of a plurality of rehabilitation patients Constructing the stability time series identification model for identifying the disease stability of each time series of patient, and constructing the rehabilitation degree time series identification model for identifying the disease rehabilitation degree of each time series of patient;Step S2, the stability time series identification model and rehabilitation degree time series identification model are combined to construct the risk time series identification model for identifying the disease risk of each time series of patient based on the model weight weighting combination.The application constructs the risk time series identification model for identifying the disease risk of each time series of patient, identifies the real-time disease risk of patient to be controlled based on the real-time pneumonia disease representation data of patient to be controlled, and determines real-time control level according to real-time disease risk, to realize the real-time classification control of patient to be controlled to achieve the real-time reasonable configuration of control resource.

Description

Technical Field

[0001] This invention relates to the field of pneumonia epidemic control technology, specifically to a pneumonia epidemic control method and system based on deep learning. Background Technology

[0002] Pneumonia is a highly contagious disease. Common symptoms include fever, dry cough, and fatigue. A small number of patients may also experience upper respiratory and digestive symptoms such as nasal congestion, runny nose, and diarrhea. Severe cases often develop respiratory distress after one week, and in severe cases, the condition rapidly progresses to acute respiratory distress syndrome, septic shock, uncorrectable metabolic acidosis, coagulation dysfunction, and multiple organ failure. Monitoring lung function is an effective diagnostic tool for assessing a patient's recovery.

[0003] Traditional epidemic dynamics models consider multiple unknowable parameters such as the number of contacts, infection rate, and incubation period, leading to significant uncertainty in predicting epidemic development. In particular, these models cannot accurately incorporate the impact of external interventions on epidemic development, thus failing to effectively quantify the effectiveness of control measures. Huang and Qiao proposed an epidemic dynamics natural growth model based on the widely existing natural growth rate, incorporating only the epidemic transmission rate as a parameter. The epidemic growth rate parameter in this model is derived from statistical data and dynamically changes with the development of the epidemic, characterizing the changes in the dynamic characteristics of the epidemic's development. This change reflects the effectiveness of epidemic control measures, providing a basis for evaluating the effectiveness of epidemic control.

[0004] Based on the above analysis, the problems and shortcomings of the existing technology are: it cannot classify and manage according to the risk of the disease, and uniform management leads to a waste of management resources. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for pneumonia epidemic control based on deep learning, so as to solve the technical problem that the existing technology cannot classify and control according to the risk of the disease, and the unified control leads to the waste of control resources.

[0006] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution:

[0007] A deep learning-based method for pneumonia epidemic control includes the following steps:

[0008] Step S1: Based on the historical rehabilitation time series data of multiple recovered patients, construct a stability time series recognition model to identify the stability of the patient's condition at each time series, and construct a recovery time series recognition model to identify the recovery degree of the patient's condition at each time series. The historical rehabilitation time series data is characterized as the pneumonia condition characterization data of the recovered patients at each time series during the development of pneumonia.

[0009] Step S2: Based on the normalization principle of the model recognition results, construct model weights for the stationarity time series recognition model and the recovery time series recognition model, and construct a risk time series recognition model for recognizing the patient's disease risk at each time series by weighting the stationarity time series recognition model and the recovery time series recognition model based on the model weights;

[0010] Step S3: Using the risk time sequence identification model, identify the real-time disease risk of the patients to be managed based on the real-time pneumonia disease characterization data of the patients to be managed, and determine the real-time management level according to the real-time disease risk, so as to realize the real-time hierarchical management of the patients to be managed and achieve the real-time rational allocation of management resources.

[0011] As a preferred embodiment of the present invention, the step of constructing a stability time-series identification model based on historical rehabilitation time-series data of multiple rehabilitation patients to identify the stability of the patient's condition at each time-series includes:

[0012] The disease inflection points for each recovered patient are identified sequentially from their historical recovery time series data. The proportion of disease inflection points in subsequent time series of each time series is then calculated as the disease stability for each time series. The formula for calculating disease stability is as follows:

[0013]

[0014] In the formula, Characterized as time series t j The stability of the condition at the site, t turn Characterized as time series t j To time series t end The time sequence of disease progression within the interval, ∑ turn∈[j,end] t turn Characterized as time series t j To time series t end The total duration of the disease transition sequence within the interval, t end Represented as the last time series of historical rehabilitation time series data, t j This is represented as the j-th time series, where turn, end, and j are all metric constants.

[0015] The pneumonia disease severity data for each time series is used as the input to the neural network, and the disease stability at each time series is used as the output. The neural network is used to train the model on the input and output to obtain the stability time series recognition model. The model expression of the stability time series recognition model is as follows:

[0016] P t =LSTM({x i,t |i∈[1,k]};

[0017] In the formula, P tCharacterized by the stability of the disease at time t, {x i,t |i∈[1,k]} represents the pneumonia condition data at time t, and LSTM represents the LSTM neural network;

[0018] Preferably, the method for identifying the timing of the disease progression includes:

[0019] Step 101: Calculate the data similarity of historical rehabilitation time series data at adjacent time series points to obtain a set of turning point time series determination curves. The formula for calculating the similarity is:

[0020]

[0021] in, Characterized by historical rehabilitation time series data in adjacent time series t j+1 , t j The data similarity of pneumonia disease symptoms data, and the time series t in the transition time series judgment curve. j+1 Data at the location, These are respectively represented as pneumonia disease severity data in adjacent time series t. j+1 , t j The i-th data component at position t j+1 , t j It is represented as the (j+1)th and jth time series, where i and j are measurement constants and k is the total number of data components;

[0022] Step 102: Plot a similarity threshold interval curve representing the stable state of the disease on the turning point time sequence determination curve, and take all time sequences outside the similarity threshold interval curve on the turning point time sequence determination curve as the turning point time sequence representing the fluctuation of the patient's disease. The threshold interval curve is a time sequence curve composed of the upper similarity limit and the lower similarity limit when the patient's disease is in a stable state.

[0023] As a preferred embodiment of the present invention, the step of constructing a recovery degree time-series recognition model for identifying the patient's recovery degree at each time sequence includes:

[0024] For each recovered patient, the proportion of the time-series distance between each time series and the last time series in the total time-series distance in the historical recovery time-series data is statistically analyzed and used as the recovery degree of each time series. The formula for calculating the recovery degree is as follows:

[0025]

[0026] In the formula, Characterized as time series t j The degree of recovery at the site, t1 represents the initial time series of historical recovery time series data;

[0027] The pneumonia disease severity data at each time series is used as the input to the neural network, and the recovery rate at each time series is used as the output. The neural network is used to train the model on the input and output to obtain the recovery rate time series recognition model. The model expression of the recovery rate time series recognition model is as follows:

[0028] pt = LSTM({x i,t |i∈[1,k]};

[0029] In the formula, p t Characterized by the degree of disease recovery at time t, {x i,t |i∈[1,k]} represents the pneumonia condition data at time t, and LSTM represents the LSTM neural network.

[0030] As a preferred embodiment of the present invention, the step of constructing model weights for the stationarity time-series identification model and the recovery time-series identification model based on the normalization principle of model identification results includes:

[0031] The model weights of the stationarity time series identification model are set, and the formula for calculating the model weights of the stationarity time series identification model is as follows:

[0032]

[0033] The model weights of the recovery time series recognition model are set, and the formula for calculating the model weights of the recovery time series recognition model is as follows:

[0034]

[0035] In the formula, α t β t These are respectively represented as the model weights of the stationarity time series identification model at time t and the recovery time series identification model.

[0036] As a preferred embodiment of the present invention, the step of constructing a risk time series identification model for identifying the patient's disease risk at each time series by weighting the stationarity time series identification model and the recovery time series identification model based on the model weights includes:

[0037] The risk time series identification model is obtained by weighting the model weights of the stationarity time series identification model and the recovery time series identification model, respectively. The functional expression of the risk time series identification model is as follows:

[0038] S t =α t p t +β t P t ;

[0039] In the formula, S t This is represented as the disease risk at time t.

[0040] As a preferred embodiment of the present invention, the step of using the risk time-series identification model to identify the real-time disease risk of the patients to be managed based on the real-time pneumonia disease characterization data of the patients to be managed includes:

[0041] The real-time pneumonia condition data of the patients under management are input into the risk time series identification model S. t =α t pt+β t P t The real-time risk of the patient's condition under control is obtained. t .

[0042] As a preferred embodiment of the present invention, the step of determining the real-time control level based on real-time disease risk includes:

[0043] The real-time disease risk S of the patients to be managed t The real-time control level of the patient to be controlled is determined by matching the risk confidence threshold interval of each control level. The risk confidence threshold interval of the control level includes a threshold interval consisting of the upper limit and lower limit of the disease risk of the corresponding control level.

[0044] As a preferred embodiment of the present invention, the pneumonia disease characterization data, before calculation, includes all data components {x} i |i∈[1,k]} is normalized.

[0045] As a preferred embodiment of the present invention, during model training, the training samples consisting of input and output items are divided into a training set and a test set in a 7:3 ratio.

[0046] As a preferred embodiment of the present invention, the present invention provides a control system based on the deep learning-based pneumonia epidemic control method, comprising:

[0047] The model building unit is used to construct a stability time-series identification model that identifies the stability of a patient's condition at each time sequence based on historical rehabilitation time-series data of multiple rehabilitation patients, and a recovery time-series identification model that identifies the recovery degree of a patient's condition at each time sequence. Based on the normalization principle of the model identification results, model weights are constructed for the stability time-series identification model and the recovery time-series identification model. Based on the model weights, the stability time-series identification model and the recovery time-series identification model are weighted and combined to construct a risk time-series identification model that identifies the risk of a patient's condition at each time sequence.

[0048] The model application unit is used to identify the real-time disease risk of the patients to be managed based on the real-time pneumonia disease characterization data of the patients to be managed using the risk time series identification model, and to determine the real-time management level according to the real-time disease risk, so as to realize the real-time hierarchical management of the patients to be managed and achieve the real-time rational allocation of management resources.

[0049] Compared with the prior art, the present invention has the following advantages:

[0050] This invention constructs a risk time-series identification model to identify the risk of a patient's condition at each time point. Based on the real-time pneumonia condition characterization data of the patients to be managed, the real-time risk of the patients to be managed is identified, and the real-time management level is determined according to the real-time risk of the patients to be managed, so as to achieve real-time hierarchical management of the patients to be managed and achieve real-time rational allocation of management resources. Attached Figure Description

[0051] To more clearly illustrate the embodiments of the present invention or the technical solutions in 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 merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0052] Figure 1 A flowchart of a pneumonia epidemic control method provided in an embodiment of the present invention;

[0053] Figure 2 This is a block diagram of the control system structure provided in an embodiment of the present invention.

[0054] The labels in the diagram represent the following:

[0055] 1-Model building unit; 2-Model application unit. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] like Figure 1 As shown, this invention provides a deep learning-based method for pneumonia epidemic control, comprising the following steps:

[0058] Step S1: Based on the historical rehabilitation time series data of multiple rehabilitation patients, construct a stability time series recognition model to identify the stability of the patient's condition at each time series, and construct a recovery time series recognition model to identify the recovery degree of the patient's condition at each time series. The historical rehabilitation time series data is represented as the pneumonia condition representation data of the rehabilitation patient at each time series during the development of pneumonia.

[0059] Based on historical rehabilitation time-series data from multiple rehabilitation patients, a stability time-series identification model is constructed to identify the stability of the patient's condition at each time series, including:

[0060] The disease inflection points for each recovered patient were identified sequentially from their historical recovery time series data. The proportion of disease inflection points in subsequent time series of each time series was then calculated as the disease stability for each time series. The formula for calculating disease stability is as follows:

[0061]

[0062] In the formula, Characterized as time series t j The stability of the condition at the site, t turn Characterized as time series t j To time series t end The time sequence of disease progression within the interval, ∑ turn∈[j,end] t turn Characterized as time series t j To time series t end The total duration of the disease transition sequence within the interval, t end Represented as the last time series of historical rehabilitation time series data, t j This is represented as the j-th time series, where turn, end, and j are all metric constants.

[0063] The pneumonia disease severity data for each time series is used as the input to the neural network, and the disease stability at each time series is used as the output. The neural network is then used to train the model on the input and output to obtain a stability time series identification model. The model expression for the stability time series identification model is as follows:

[0064] P t =LSTM({x i,t |i∈[1,k]};

[0065] In the formula, P t Characterized by the stability of the disease at time t, {x i,t |i∈[1,k]} represents the pneumonia condition data at time t, and LSTM represents the LSTM neural network;

[0066] Preferred methods for identifying the timing of disease progression include:

[0067] Step 101: Calculate the data similarity of historical rehabilitation time series data at adjacent time series points to obtain a set of turning point time series judgment curves. The formula for calculating the similarity is:

[0068]

[0069] in, Characterized by historical rehabilitation time series data in adjacent time series t j+1 , t j The data similarity of pneumonia disease symptoms data, and the time series t in the transition time series judgment curve. j+1 Data at the location, These are respectively represented as pneumonia disease severity data in adjacent time series t. j+1 , t j The i-th data component at position t j+1 , t j It is represented as the (j+1)th and jth time series, where i and j are measurement constants and k is the total number of data components;

[0070] Step 102: Plot a similarity threshold interval curve representing the stable state of the disease on the turning point time sequence determination curve, and take all time sequences outside the similarity threshold interval curve on the turning point time sequence determination curve as the turning point time sequence representing the fluctuation of the patient's disease. The threshold interval curve is represented by the time sequence curve composed of the upper similarity limit and the lower similarity limit when the patient's disease is in a stable state.

[0071] Construct a recovery time-series recognition model to identify the patient's recovery level at each time point, including:

[0072] For each recovered patient, the proportion of the time-series distance between each time series and the last time series in the total time-series distance in the historical recovery time-series data is statistically analyzed. This proportion is used as the recovery degree for each time series. The formula for calculating the recovery degree is as follows:

[0073]

[0074] In the formula, Characterized as time series t j The degree of recovery at the site, t1 represents the initial time series of historical recovery time series data;

[0075] The pneumonia disease severity data at each time series is used as the input to the neural network, and the recovery rate at each time series is used as the output. The neural network is used to train the model on the input and output to obtain a recovery rate time series recognition model. The model expression of the recovery rate time series recognition model is as follows:

[0076] p t =LSTM({x i,t |i∈[1,k]};

[0077] In the formula, p t Characterized by the degree of disease recovery at time t, {x i,t |i∈[1,k]} represents the pneumonia condition data at time t, and LSTM represents the LSTM neural network.

[0078] The stability of a patient's condition at a given moment indicates the stability of their condition at that time. A higher stability level indicates a more stable condition, allowing for a lower control level. Conversely, a lower stability level indicates a more unstable condition, requiring a higher control level. Similarly, the recovery rate at a given moment indicates the recovery status of a patient's condition. A higher recovery rate indicates that the patient is closer to recovery, allowing for a lower control level. Conversely, a lower recovery rate indicates that the patient is further from recovery, requiring a higher control level. Therefore, by using both stability and recovery rates as indicators to measure the risk of a patient's condition, a control level can be set accordingly.

[0079] Step S2: Based on the normalization principle of the model recognition results, construct model weights for the stationarity time series recognition model and the recovery time series recognition model, and construct a risk time series recognition model for identifying the patient's disease risk at each time series based on the model weights.

[0080] Based on the normalization principle of model recognition results, model weights are constructed for the stationarity time series recognition model and the recovery time series recognition model, including:

[0081] The model weights of the stationarity time series identification model are set, and the formula for calculating the model weights of the stationarity time series identification model is as follows:

[0082]

[0083] The model weights of the recovery time series identification model are set, and the formula for calculating the model weights of the recovery time series identification model is as follows:

[0084]

[0085] In the formula, α t β t These are respectively represented as the model weights of the stationarity time series identification model at time t and the recovery time series identification model.

[0086] Based on model weights, a weighted combination of the stationarity time-series identification model and the recovery time-series identification model is constructed to create a risk time-series identification model that identifies the patient's disease risk at each time series, including:

[0087] The risk time series identification model is obtained by weighting the model weights of the stationarity time series identification model and the recovery time series identification model, respectively. The functional expression of the risk time series identification model is as follows:

[0088] S t =α t p t +β t P t ;

[0089] In the formula, S t This is represented as the disease risk at time t.

[0090] The risk time series identification model is set as a weighted combination of the stationarity time series identification model and the recovery time series identification model. It can comprehensively measure the risk of the disease using stationarity and recovery, thereby increasing the accuracy and comprehensiveness of the risk measurement.

[0091] Step S3: Using a risk time series identification model, identify the real-time disease risk of patients under management based on their real-time pneumonia disease characterization data, and determine the real-time management level according to the real-time disease risk, so as to achieve real-time hierarchical management of patients under management and realize the real-time rational allocation of management resources.

[0092] Using a risk time-series identification model, the real-time disease risk of patients under management is identified based on their real-time pneumonia disease characterization data, including:

[0093] Real-time pneumonia condition data of patients under management are input into the risk time series identification model S. t =α t p t +β t P t The real-time risk of the patient's condition under control is obtained. t .

[0094] The real-time control level is determined based on the real-time risk of the illness, including:

[0095] Real-time disease risk of patients under management t The real-time control level of the patient to be controlled is determined by matching the risk confidence threshold range of each control level. The risk confidence threshold range of the control level includes the threshold range consisting of the upper limit and lower limit of the disease risk of the corresponding control level.

[0096] Before performing calculations, all data components {x} of the pneumonia symptom data were considered. i |i∈[1,k]} is normalized.

[0097] During model training, the training samples, consisting of input and output terms, are divided into a training set and a test set in a 7:3 ratio.

[0098] like Figure 2 As shown, based on the above-mentioned deep learning-based pneumonia epidemic control method, this invention provides a control system, including:

[0099] Model building unit 1 is used to build a stability time series recognition model that identifies the stability of a patient's condition at each time series based on the historical rehabilitation time series data of multiple rehabilitation patients, and a recovery time series recognition model that identifies the recovery degree of a patient's condition at each time series. Based on the normalization principle of the model recognition results, model weights are built for the stability time series recognition model and the recovery time series recognition model. Based on the model weights, the stability time series recognition model and the recovery time series recognition model are weighted and combined to build a risk time series recognition model that identifies the risk of a patient's condition at each time series.

[0100] Model application unit 2 is used to identify the real-time disease risk of patients under management based on the real-time pneumonia disease characterization data of patients under management using the risk time series identification model, and determine the real-time management level according to the real-time disease risk, so as to realize the real-time hierarchical management of patients under management and achieve the real-time rational allocation of management resources.

[0101] This invention constructs a risk time-series identification model to identify the risk of a patient's condition at each time point. Based on the real-time pneumonia condition characterization data of the patients to be managed, the real-time risk of the patients to be managed is identified, and the real-time management level is determined according to the real-time risk of the patients to be managed, so as to achieve real-time hierarchical management of the patients to be managed and achieve real-time rational allocation of management resources.

[0102] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A method for pneumonia epidemic control based on deep learning, characterized in that: Includes the following steps: Step S1: Based on the historical rehabilitation time series data of multiple recovered patients, construct a stability time series recognition model to identify the stability of the patient's condition at each time series, and construct a recovery time series recognition model to identify the recovery degree of the patient's condition at each time series. The historical rehabilitation time series data is characterized as the pneumonia condition characterization data of the recovered patients at each time series during the development of pneumonia. Step S2: Based on the normalization principle of the model recognition results, construct model weights for the stationarity time series recognition model and the recovery time series recognition model, and construct a risk time series recognition model for recognizing the patient's disease risk at each time series by weighting the stationarity time series recognition model and the recovery time series recognition model based on the model weights; Step S3: Using the risk time sequence identification model, identify the real-time disease risk of the patients to be managed based on the real-time pneumonia disease characterization data of the patients to be managed, and determine the real-time management level according to the real-time disease risk, so as to realize the real-time hierarchical management of the patients to be managed and achieve the real-time rational allocation of management resources. Based on historical rehabilitation time-series data from multiple rehabilitation patients, a stability time-series identification model is constructed to identify the stability of the patient's condition at each time series, including: The disease inflection points for each recovered patient were identified sequentially from their historical recovery time series data. The proportion of disease inflection points in subsequent time series of each time series was then calculated as the disease stability for each time series. The formula for calculating disease stability is as follows: In the formula, Characterized as time series The patient's condition is stable. Characterized as time series To time sequence The timing of disease progression within the interval; Characterized as time series To time sequence The total duration of the disease transition sequence within the interval. The last time series of historical rehabilitation time series data is represented. Represented as the j-th time series, , Both j and j are measurement constants; The pneumonia disease severity data for each time series is used as the input to the neural network, and the disease stability at each time series is used as the output. The neural network is then used to train the model on the input and output to obtain a stability time series identification model. The model expression for the stability time series identification model is as follows: In the formula, Characterized by the stability of the disease at time t. The data represents the pneumonia severity at time t. Characterized as an LSTM neural network; Construct a recovery time-series recognition model to identify the patient's recovery level at each time point, including: For each recovered patient, the proportion of the time-series distance between each time series and the last time series in the total time-series distance in the historical recovery time-series data is statistically analyzed. This proportion is used as the recovery degree for each time series. The formula for calculating the recovery degree is as follows: In the formula, Characterized as time series The degree of recovery of the condition at the site The initial time series represented as historical rehabilitation time series data; The pneumonia disease severity data at each time series is used as the input to the neural network, and the recovery rate at each time series is used as the output. The neural network is used to train the model on the input and output to obtain a recovery rate time series recognition model. The model expression of the recovery rate time series recognition model is as follows: In the formula, Characterized by the degree of disease recovery at time t. The data represents the pneumonia severity at time t. It is characterized as an LSTM neural network.

2. The method for pneumonia epidemic control based on deep learning according to claim 1, characterized in that: The method for identifying the timing of the disease progression includes: Step 101: Calculate the data similarity of historical rehabilitation time series data at adjacent time series points to obtain a set of turning point time series determination curves. The formula for calculating the similarity is: in, Characterized by historical rehabilitation time series data in adjacent time series , The data similarity of pneumonia disease symptoms, and the time series in the transition time series judgment curve. Data at the location, , These are respectively represented as pneumonia disease characteristics data in adjacent time series. , The i-th data component at position i, , It is represented as the (j+1)th and jth time series, where i and j are measurement constants and k is the total number of data components; Step 102: Plot a similarity threshold interval curve representing the stable state of the disease on the turning point time sequence determination curve, and take all time sequences outside the similarity threshold interval curve on the turning point time sequence determination curve as the turning point time sequence representing the fluctuation of the patient's disease. The threshold interval curve is a time sequence curve composed of the upper similarity limit and the lower similarity limit when the patient's disease is in a stable state.

3. The method for pneumonia epidemic control based on deep learning according to claim 2, characterized in that: The normalization principle based on the model recognition results is used to construct model weights for the stationarity time series recognition model and the recovery time series recognition model, including: The model weights of the stationarity time series identification model are set, and the formula for calculating the model weights of the stationarity time series identification model is as follows: The model weights of the recovery time series recognition model are set, and the formula for calculating the model weights of the recovery time series recognition model is as follows: In the formula, , These are respectively represented as the model weights of the stationarity time series identification model at time t and the recovery time series identification model.

4. The method for pneumonia epidemic control based on deep learning according to claim 3, characterized in that: The risk-time identification model, which constructs a risk assessment model for identifying the patient's disease risk at each time step by weighting the stationarity time-series identification model and the recovery time-series identification model based on the model weights, includes: The risk time series identification model is obtained by weighting and summing the model weights of the stationarity time series identification model and the recovery time series identification model, respectively. The functional expression of the risk time series identification model is as follows: In the formula, This is represented as the disease risk at time t.

5. The method for pneumonia epidemic control based on deep learning according to claim 4, characterized in that: The process of using the risk time-series identification model to identify the real-time disease risk of patients under management based on their real-time pneumonia disease characterization data includes: The real-time pneumonia condition data of the patients under management are input into the risk time series identification model. The real-time risk of the patient's condition is obtained from the data. .

6. The method for pneumonia epidemic control based on deep learning according to claim 5, characterized in that, The determination of real-time control levels based on real-time disease risk includes: The real-time disease risk of the patients to be managed The real-time control level of the patient to be controlled is determined by matching the risk confidence threshold interval of each control level. The risk confidence threshold interval of the control level includes a threshold interval consisting of the upper limit and lower limit of the disease risk of the corresponding control level.

7. The method for pneumonia epidemic control based on deep learning according to claim 6, characterized in that, The pneumonia symptom data were processed by combining all data components before calculation. Normalization is performed.

8. The method for pneumonia epidemic control based on deep learning according to claim 7, characterized in that, During model training, the training samples, consisting of input and output terms, are divided into a training set and a test set in a 7:3 ratio.

9. A control system for pneumonia epidemic control based on deep learning according to any one of claims 1-8, characterized in that, include: The model building unit (1) is used to build a stability time series recognition model that identifies the stability of the patient's condition at each time series based on the historical rehabilitation time series data of multiple rehabilitation patients, and to build a recovery time series recognition model that identifies the recovery degree of the patient's condition at each time series. Based on the normalization principle of the model recognition results, model weights are built for the stability time series recognition model and the recovery time series recognition model. Based on the model weights, the stability time series recognition model and the recovery time series recognition model are weighted and combined to build a risk time series recognition model that identifies the risk of the patient's condition at each time series. The model application unit (2) is used to identify the real-time disease risk of the patients to be managed based on the real-time pneumonia disease characterization data of the patients to be managed using the risk time sequence identification model, and to determine the real-time management level according to the real-time disease risk, so as to realize the real-time hierarchical management of the patients to be managed and achieve the real-time reasonable allocation of management resources. The stability time-series identification model, which identifies the stability of a patient's condition at each time point based on historical rehabilitation time-series data from multiple rehabilitation patients, includes: The disease inflection points for each recovered patient are identified sequentially from their historical recovery time series data. The proportion of disease inflection points in subsequent time series of each time series is then calculated as the disease stability for each time series. The formula for calculating disease stability is as follows: In the formula, Characterized as time series The patient's condition is stable. Characterized as time series To time sequence The timing of disease progression within the interval; Characterized as time series To time sequence The total duration of the disease transition sequence within the interval. The last time series of historical rehabilitation time series data is represented. Represented as the j-th time series, , Both j and j are measurement constants; The pneumonia disease severity data for each time series is used as the input to the neural network, and the disease stability at each time series is used as the output. The neural network is used to train the model on the input and output to obtain the stability time series recognition model. The model expression of the stability time series recognition model is as follows: In the formula, Characterized by the stability of the disease at time t. The data represents the pneumonia severity at time t. Characterized as an LSTM neural network; The construction of the recovery degree time-series recognition model for identifying the patient's recovery degree at each time sequence includes: For each recovered patient, the proportion of the time-series distance between each time series and the last time series in the total time-series distance in the historical recovery time-series data is statistically analyzed and used as the recovery degree of each time series. The formula for calculating the recovery degree is as follows: In the formula, Characterized as time series The degree of recovery of the condition at the site The initial time series represented as historical rehabilitation time series data; The pneumonia disease severity data at each time series is used as the input to the neural network, and the recovery rate at each time series is used as the output. The neural network is used to train the model on the input and output to obtain the recovery rate time series recognition model. The model expression of the recovery rate time series recognition model is as follows: In the formula, Characterized by the degree of disease recovery at time t. The data represents the pneumonia severity at time t. It is characterized as an LSTM neural network.

Images