Large model infectious disease prediction and early warning method and system based on multi-stage learning

Abstract

The invention discloses a large model infectious disease prediction and early warning method and system based on multi-stage learning, and mainly aims at constructing a multi-stage learning model for data, learning basic features, time sequence features and expansion features in the data through different stages, and then predicting and early warning the infectious diseases through a re-defined attention mechanism model. According to the scheme, different information and a self-defined attention fusion mechanism are respectively learned according to learning models which are constructed in sequence and aim at different features and are in different stages, so that the learning efficiency is improved. The results of different stages are fully fused through attention traversal pairwise calculation, different weight factors are given to the results of each stage, and then weighted fusion is performed, so that multi-stage, comprehensive and accurate prediction and early warning of the infectious disease epidemic situation are realized through fusion of multi-source data.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and specifically relates to a large-model infectious disease prediction and early warning method and system based on multi-stage learning. Background Art

[0002] Infectious diseases have always been a major threat to human health. For example, early warning systems for hand, foot and mouth disease and hepatitis B virus are conducive to the management of infectious disease risks. With the intensification of globalization and population mobility, the speed and scope of the spread of infectious diseases are constantly expanding, posing huge challenges to public health security and social economy. Prediction technology is the basis of early warning systems, which helps make decisions on medical care and intervention strategies. Traditional infectious disease prediction methods mainly rely on statistical analysis and empirical judgment, and have problems such as low prediction accuracy and poor timeliness.

[0003] In recent years, with the rapid development of technologies such as big data and artificial intelligence, infectious disease prediction methods based on machine learning and deep learning have gradually become a research hotspot. Due to the success of attention mechanisms in time series prediction, in order to improve prediction technology, some prediction models have introduced them into the prediction of the number of infectious disease cases. These attention mechanisms focus on some important time steps of the input data, so the prediction of the number of outpatient cases of infectious diseases is often regarded as a time series prediction problem. The previous attention mechanisms mainly focused on the impact of time series at different times on the prediction. However, this method mostly focuses on the prediction of a single model or a single stage, ignoring the complexity and dynamics of the infectious disease transmission process, and cannot integrate different data information to comprehensively judge the transmission growth rate of infectious diseases, and thus cannot give accurate infectious disease data warnings. Summary of the invention

[0004] In response to the above-mentioned deficiencies in the prior art, the present application provides a large-model infectious disease prediction and early warning method and system based on multi-stage learning.

[0005] In the first aspect, the present application proposes a large-model infectious disease prediction and early warning method based on multi-stage learning, comprising the following steps:

[0006] Acquire characteristic description information of infectious disease data, and input the characteristic description information into a pre-built autoencoder model to obtain a basic characteristic representation of the infectious disease data;

[0007] A gated recurrent unit model is used to capture the temporal features of the characteristic description information, wherein the feature values ​​in all the temporal features are sorted in time order and then learned and extracted to obtain a temporal feature representation;

[0008] The characteristic description information is transformed and organized into a natural language description, the natural language description is expanded using a large model, and the expanded text is transformed into a vector representation to obtain an expanded characteristic representation;

[0009] The basic feature representation, the temporal feature representation and the expanded feature representation are integrated through an attention mechanism model to obtain a comprehensive feature representation;

[0010] Based on the comprehensive feature representation, a regression model is used to predict the trend of infectious disease transmission, and early warnings are issued according to set thresholds.

[0011] In some optional implementations of some embodiments, the step of obtaining characteristic description information of infectious disease data and inputting the characteristic description information into a pre-built autoencoder model to obtain a basic characteristic representation of the infectious disease data includes:

[0012] Assume that the characteristic description information of the input autoencoder model is , represents an n-dimensional data set, in which each element has n-dimensional features;

[0013] The embedding layer accepts input x and maps x to a vector representation of a specified dimension. The calculation of the embedding layer is:

[0014]

[0015] in, and Represent the weight matrix and bias matrix of the embedding layer respectively;

[0016] The encoder consists of a linear layer and an activation function. The output of the embedding layer is Enter the encoder to calculate:

[0017]

[0018] in, and denote the weight matrix and bias matrix of the hidden layer respectively, The basic feature representation indicating that the basic feature learning is completed, is a nonlinear activation function, and the calculation of the nonlinear activation function is:

[0019]

[0020] in, is a nonlinear activation function Received input.

[0021] In some optional implementations of some embodiments, the use of a gated recurrent unit model to capture the temporal features of the characteristic description information, wherein the feature values ​​in all the temporal features are sorted in chronological order and then learned and extracted to obtain a temporal feature representation, includes:

[0022] The learning and extraction of time series features adopts the model structure of the GRU network, which is composed of multiple gated networks, including a reset gate, a candidate gate, and an update gate. The final hidden state calculated by the multiple gated networks is the time series feature representation.

[0023] In some optional implementations of some embodiments, the learning and extraction of the temporal features adopts a model structure of a GRU network, which is composed of multiple gated networks, including a reset gate, a candidate gate, and an update gate. The final hidden state calculated by the multiple gated networks is the temporal feature representation, and the specific calculation is:

[0024] Reset the gate:

[0025]

[0026] in, and are the weight matrices of the reset gate, is the bias matrix of the reset gate, and The hidden state representations of the current input character and the previous character respectively represent characteristic description information;

[0027] The calculation of candidate gates is divided into two cases: receiving the previous hidden state and not receiving the previous hidden state;

[0028] When receiving the previous hidden state:

[0029]

[0030] in, and They are the weight matrix and bias matrix of the current network layer respectively;

[0031] When not receiving the previous hidden state:

[0032]

[0033] in, and They are the weight matrix and bias matrix of the current network layer respectively;

[0034] Update Gate:

[0035] ( )

[0036] in, and are the weight matrices of the update gates, is the bias matrix of the update gate;

[0037] Get the final hidden state:

[0038]

[0039] in, is the timing characteristic representation.

[0040] In some optional implementations of some embodiments, the converting and organizing the characteristic description information into a natural language description, using a large model to expand the natural language description, and converting the expanded text into a vector representation to obtain an expanded characteristic representation includes:

[0041] Construct prompt word instructions for inputting into the big model: construct the feature name and feature content corresponding to each line of data of the acquired characteristic description information into nested list format data, record the constructed data as content, and then splice it with the task instruction prompt for the big model, and use the format after splicing content and prompt as input, and input it into the big model for feature expansion;

[0042] Use the fully connected layer to convert the feature text expanded by the large model into a vector representation:

[0043]

[0044] in, and are the weight matrix and bias matrix of the fully connected layer, is the text after the received large model features are expanded, It is a vector representation of the text transformation after the large model feature is expanded, that is, the expanded feature representation.

[0045] In some optional implementations of some embodiments, the basic feature representation, the temporal feature representation and the expanded feature representation are fused through an attention mechanism model to obtain a comprehensive feature representation, including:

[0046] Will , and After three first linear layers respectively, three Query vectors are obtained: , and , the weight matrices and bias matrices of the three first linear layers use the same weights:

[0047]

[0048] in, and Represent the weight matrix and bias matrix in each first linear layer, , and There is a set, denoted as ;

[0049] Will , and After three second linear layers respectively, three Key vectors are obtained , and , the weight matrices and bias matrices of the three second linear layers also use the same weights:

[0050]

[0051] in, and Represent the weight matrix and bias matrix in each second linear layer, , and There is a set, denoted as ;

[0052] Will , and After three third linear layers respectively, the Value vector , and , the weight matrices and bias matrices of the three third linear layers also use the same weights:

[0053]

[0054] in, and Represent the weight matrix and bias matrix in each second linear layer, , and There is a set, denoted as ;

[0055] right , , The Query vector, Key vector and Value vector saved in the Query vector are combined in pairs to obtain vector combinations, and all the combinations of vector combinations are traversed and the Att results of the Query vector, Key vector and Value vector in each vector combination are calculated respectively;

[0056] Assign corresponding weight influence factors to the Query vector, Key vector and Value vector in each of the Att results, sum all the Att results according to the assigned weight influence factors, multiply the sum result and each of the Att results to obtain the attention score of the fused weight information, and use the attention score of the fused weight information as the comprehensive feature representation for infectious disease prediction.

[0057] In some optional implementations of some embodiments, the prediction of the infectious disease transmission trend using a regression model based on the comprehensive feature representation and the early warning according to a set threshold include:

[0058] The comprehensive feature representation is input into the regression model. The regression model is divided into a hidden layer and a fully connected layer. The hidden layer calculation process is:

[0059]

[0060] in, and are the weight matrix and bias matrix of the hidden layer, respectively. It is a comprehensive feature representation;

[0061] The fully connected layer is calculated as:

[0062]

[0063] in, and are the weight matrix and bias matrix of the fully connected layer respectively, The final predicted number of infected people at the next time point;

[0064] Said Used to calculate the growth rate of the number of infections at the current time point:

[0065]

[0066] Set the warning threshold. When the rate exceeds the threshold, the warning mechanism is automatically triggered, and an infection out-of-control warning is output. The model prediction results and warning information are sent to relevant departments for early warning prompts.

[0067] The second aspect is a large-model infectious disease prediction and early warning system based on multi-stage learning, including a basic feature representation acquisition module, a time series feature representation acquisition module, an expanded feature representation acquisition module, a feature fusion module, and an infectious disease transmission trend prediction and early warning module;

[0068] The basic feature representation acquisition module is used to obtain characteristic description information of infectious disease data, input the characteristic description information into a pre-built autoencoder model, and obtain the basic feature representation of infectious disease data;

[0069] The temporal feature representation acquisition module is used to capture the temporal features of the characteristic description information using a gated recurrent unit model, wherein the feature values ​​in all the temporal features are sorted in chronological order and then learned and extracted to obtain the temporal feature representation;

[0070] The expanded feature representation acquisition module is used to transform and organize the characteristic description information into a natural language description, expand the natural language description using a large model, and transform the expanded text into a vector representation to obtain an expanded feature representation;

[0071] The feature fusion module is used to fuse the basic feature representation, the temporal feature representation and the expanded feature representation through an attention mechanism model to obtain a comprehensive feature representation;

[0072] The infectious disease transmission trend prediction and early warning module is used to predict the infectious disease transmission trend based on the comprehensive feature representation using a regression model, and to issue an early warning according to a set threshold.

[0073] In a third aspect, the present application proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0074] In a fourth aspect, the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0075] Beneficial effects of the present invention:

[0076] The present invention mainly constructs a multi-stage learning model for the data, and learns the basic features, time series features, and extended features in the data through different stages. Then, through our customized attention mechanism model, the results learned in the previous stage are fused and calculated to form a comprehensive feature representation. The learning models for different stages of different features constructed in sequence in this scheme learn different information respectively. The customized attention fusion mechanism fully integrates the results of different stages through attention traversal and pairwise calculations and assigns different weight factors to the results of each stage and then performs weighted fusion. By fusing multi-source data, multi-stage, comprehensive, and accurate prediction and early warning of infectious disease epidemics can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1It is the overall flow chart of the present invention.

[0078] Figure 2 It is a system principle block diagram of the present invention. DETAILED DESCRIPTION

[0079] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein; on the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0080] On the one hand, this application proposes a large-scale infectious disease prediction and early warning method based on multi-stage learning, such as Figure 1 As shown, the following steps are included:

[0081] S100: Acquire characteristic description information of infectious disease data, input the characteristic description information into a pre-built autoencoder model, and obtain a basic characteristic representation of the infectious disease data;

[0082] Among them, characteristic descriptive information includes basic characteristics such as the number of cases, transmission routes, population distribution, and climate environment;

[0083] In some optional implementations of some embodiments, the step of obtaining characteristic description information of infectious disease data and inputting the characteristic description information into a pre-built autoencoder model to obtain a basic characteristic representation of the infectious disease data includes:

[0084] Assume that the characteristic description information of the input autoencoder model is , represents an n-dimensional data set, in which each element has n-dimensional features;

[0085] The embedding layer accepts input x and maps x to a vector representation of a specified dimension. The calculation of the embedding layer is:

[0086]

[0087] in, and Represent the weight matrix and bias matrix of the embedding layer respectively;

[0088] The encoder consists of a linear layer and an activation function. The output of the embedding layer is Enter the encoder to calculate:

[0089]

[0090] in, and denote the weight matrix and bias matrix of the hidden layer respectively, The basic feature representation indicating that the basic feature learning is completed, is a nonlinear activation function, and the calculation of the nonlinear activation function is:

[0091]

[0092] in, is a nonlinear activation function Received input.

[0093] S200: using a gated recurrent unit model to capture the temporal features of the characteristic description information, wherein the feature values ​​in all the temporal features are sorted in chronological order and then learned and extracted to obtain a temporal feature representation;

[0094] In some optional implementations of some embodiments, the use of a gated recurrent unit model to capture the temporal features of the characteristic description information, wherein the feature values ​​in all the temporal features are sorted in chronological order and then learned and extracted to obtain a temporal feature representation, includes:

[0095] The learning and extraction of time series features adopts the model structure of the GRU network, which is composed of multiple gated networks, including a reset gate, a candidate gate, and an update gate. The final hidden state calculated by the multiple gated networks is the time series feature representation.

[0096] Among them, time series feature learning: using the gated recurrent unit model to capture the time series features of infectious disease data. Time series features include feature information that is statistically analyzed in time, such as statistical time, vaccination rate, and population mobility. The feature values ​​of all features are sorted in chronological order, and the time series features are learned and extracted;

[0097] In some optional implementations of some embodiments, the learning and extraction of the temporal features adopts a model structure of a GRU network, which is composed of multiple gated networks, including a reset gate, a candidate gate, and an update gate. The final hidden state calculated by the multiple gated networks is the temporal feature representation, and the specific calculation is:

[0098] Reset the gate:

[0099]

[0100] in, and are the weight matrices of the reset gate, is the bias matrix of the reset gate, and The hidden state representations of the current input character and the previous character respectively represent characteristic description information;

[0101] The calculation of candidate gates is divided into two cases: receiving the previous hidden state and not receiving the previous hidden state;

[0102] When receiving the previous hidden state:

[0103]

[0104] in, and They are the weight matrix and bias matrix of the current network layer respectively;

[0105] When not receiving the previous hidden state:

[0106]

[0107] in, and They are the weight matrix and bias matrix of the current network layer respectively;

[0108] Update Gate:

[0109] ( )

[0110] in, and are the weight matrices of the update gates, is the bias matrix of the update gate;

[0111] Get the final hidden state:

[0112]

[0113] in, is the timing characteristic representation.

[0114] S300: transforming and organizing the characteristic description information into a natural language description, expanding the natural language description using a large model, and transforming the expanded text into a vector representation to obtain an expanded characteristic representation;

[0115] In some optional implementations of some embodiments, the converting and organizing the characteristic description information into a natural language description, using a large model to expand the natural language description, and converting the expanded text into a vector representation to obtain an expanded characteristic representation includes:

[0116] Construct prompt word instructions for inputting into the big model: construct the feature name and feature content corresponding to each line of data of the acquired characteristic description information into nested list format data, record the constructed data as content, and then splice it with the task instruction prompt for the big model, and use the format after splicing content and prompt as input, and input it into the big model for feature expansion;

[0117] The specific instructions for constructing the prompt word for the input large model are:

[0118] The feature name and feature content corresponding to each row of data of the existing characteristic description information are constructed into nested list format data, such as: [['time', '2023-04-01'], ['new cases', 50], ['transmission route', 'mainly droplet transmission, supplemented by contact transmission'], ['population distribution', 'mainly young and middle-aged people, followed by the elderly'], ['population mobility', 'after the Qingming Festival holiday, the number of people returning home increases'], ['climate environment', 'spring'], ['prevention and control measures', 'strengthen disinfection of public places and remind citizens to wear masks'], ['vaccination rate', 50], ['medical resource utilization', 'hospital beds are tight, and temporary medical points need to be added in some areas']], and the constructed data is recorded as content.

[0119] Then add the task instruction prompt to the model: "Please rewrite and expand the data based on the current data and combine it with medical knowledge of infectious diseases to enrich the information. For example, the "spring" in it can be combined with the impact of seasons and temperature on current infectious diseases and supplemented as: "In spring, the temperature rises and the humidity is moderate, which is conducive to the spread of the virus." Splice the content and prompt and input them into the large model for feature expansion.

[0120] Use the fully connected layer to convert the feature text expanded by the large model into a vector representation:

[0121]

[0122] in, and are the weight matrix and bias matrix of the fully connected layer, is the text after the received large model features are expanded, It is a vector representation of the text transformation after the large model feature is expanded, that is, the expanded feature representation.

[0123] S400: fusing the basic feature representation, the temporal feature representation and the expanded feature representation through an attention mechanism model to obtain a comprehensive feature representation;

[0124] In some optional implementations of some embodiments, the basic feature representation, the temporal feature representation and the expanded feature representation are fused through an attention mechanism model to obtain a comprehensive feature representation, including:

[0125] Will , and After three first linear layers respectively, three Query vectors are obtained: , and , the weight matrices and bias matrices of the three first linear layers use the same weights:

[0126]

[0127] in, and Represent the weight matrix and bias matrix in each first linear layer, , and There is a set, denoted as ;

[0128] Will , and After three second linear layers respectively, three Key vectors are obtained , and , the weight matrices and bias matrices of the three second linear layers also use the same weights:

[0129]

[0130] in, and Represent the weight matrix and bias matrix in each second linear layer, , and There is a set, denoted as ;

[0131] Will , and After three third linear layers respectively, the Value vector , and , the weight matrices and bias matrices of the three third linear layers also use the same weights:

[0132]

[0133] in, and Represent the weight matrix and bias matrix in each second linear layer, , and There is a set, denoted as ;

[0134] right , , The Query vector, Key vector and Value vector saved in the Query vector are combined in pairs to obtain vector combinations, and all the combinations of vector combinations are traversed and the Att results of the Query vector, Key vector and Value vector in each vector combination are calculated respectively;

[0135] Specifically, they are:

[0136] right , , The Query vector, Key vector, and Value vector saved in between are combined and calculated. The combination methods are as follows: [Q1, K1, V1], [Q1, K2, V1], etc. After traversing all the combination methods, there are 3*3*3=27 kinds in total.

[0137] Calculate Q, K, and V in each combination separately. For Q, K, and V:

[0138]

[0139] And so on, there are 27 calculated Att results;

[0140] Assign corresponding weight influence factors to the Query vector, Key vector and Value vector in each of the Att results, sum all the Att results according to the assigned weight influence factors, multiply the sum result and each of the Att results to obtain the attention score of the fused weight information, and use the attention score of the fused weight information as the comprehensive feature representation for infectious disease prediction.

[0141] Among them, for Q1, Q2, and Q3, each is manually assigned a weighted impact factor 1. 2. 3, among which 1+ 2+ 3=1, similarly, for V1, V2, V3 and , 2, 3. The same weight influence factors of Q1, Q2, and Q3 are also used to distinguish the feature weights of learning results at different stages.

[0142] For each Att result, add up the weighted influence factors corresponding to Q, K, and V used in it, and record the sum as , such as: Att2 uses Q1, K2, V1, then the corresponding Calculated as:

[0143]

[0144] Finally, the 27 calculated Att results correspond to 27 Value, continue to calculate, the attention Att and the corresponding Multiply them to get the attention score of the fused weight information, then sum them up, divide them by 27 for average, and divide them by The cumulative sum of is calculated and restored:

[0145]

[0146] The Att result calculated from multi-source information is output as a comprehensive feature representation to calculate the number of infected people in the next stage;

[0147] S500: Based on the comprehensive feature representation, a regression model is used to predict the spread trend of the infectious disease, and an early warning is issued according to a set threshold.

[0148] Among them, based on the comprehensive feature representation of the fusion multi-stage model learning information, the final stage of prediction is carried out through the regression model to predict the number of infections at the next time point. According to the prediction results, an early warning is issued, and the early warning information and analysis report are submitted to the relevant disease control department.

[0149] In some optional implementations of some embodiments, the prediction of the infectious disease transmission trend using a regression model based on the comprehensive feature representation and the early warning according to a set threshold include:

[0150] The comprehensive feature representation is input into the regression model. The regression model is divided into a hidden layer and a fully connected layer. The hidden layer calculation process is:

[0151]

[0152] in, and are the weight matrix and bias matrix of the hidden layer, respectively. It is a comprehensive feature representation;

[0153] The fully connected layer is calculated as:

[0154]

[0155] in, and are the weight matrix and bias matrix of the fully connected layer respectively, The final predicted number of infected people at the next time point;

[0156] Said Used to calculate the growth rate of the number of infections at the current time point:

[0157]

[0158] Set the warning threshold. When the rate exceeds the threshold, the warning mechanism is automatically triggered, and an infection out-of-control warning is output. The model prediction results and warning information are sent to relevant departments for early warning prompts.

[0159] The second aspect is the large-scale infectious disease prediction and early warning system based on multi-stage learning, such as Figure 2 As shown, it includes a basic feature representation acquisition module, a time series feature representation acquisition module, an extended feature representation acquisition module, a feature fusion module and an infectious disease transmission trend prediction and warning module;

[0160] The basic feature representation acquisition module is used to obtain characteristic description information of infectious disease data, input the characteristic description information into a pre-built autoencoder model, and obtain the basic feature representation of infectious disease data;

[0161] The temporal feature representation acquisition module is used to capture the temporal features of the characteristic description information using a gated recurrent unit model, wherein the feature values ​​in all the temporal features are sorted in chronological order and then learned and extracted to obtain the temporal feature representation;

[0162] The expanded feature representation acquisition module is used to transform and organize the characteristic description information into a natural language description, expand the natural language description using a large model, and transform the expanded text into a vector representation to obtain an expanded feature representation;

[0163] The feature fusion module is used to fuse the basic feature representation, the temporal feature representation and the expanded feature representation through an attention mechanism model to obtain a comprehensive feature representation;

[0164] The infectious disease transmission trend prediction and early warning module is used to predict the infectious disease transmission trend based on the comprehensive feature representation using a regression model, and to issue an early warning according to a set threshold.

[0165] In a third aspect, the present application proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0166] In a fourth aspect, the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0167] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0168] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0169] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0170] In the embodiments provided in the present disclosure, it should be understood that the disclosed apparatus / computer equipment and methods can be implemented in other ways. For example, the apparatus / computer equipment embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of the apparatus or unit, which may be electrical, mechanical or other forms.

[0171] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0173] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0174] The above are only preferred implementations of the present invention. It should be pointed out that a number of modifications and improved technical solutions made by those skilled in the art without departing from the technical solution should also be regarded as falling within the scope of protection required by the claims.

Claims

1. A large-model infectious disease prediction and early warning method based on multi-stage learning, characterized by: The following steps are involved: Acquire characteristic description information of infectious disease data, and input the characteristic description information into a pre-built autoencoder model to obtain a basic characteristic representation of the infectious disease data; A gated recurrent unit model is used to capture the temporal features of the characteristic description information, wherein the feature values ​​in all the temporal features are sorted in time order and then learned and extracted to obtain a temporal feature representation; The characteristic description information is transformed and organized into a natural language description, the natural language description is expanded using a large model, and the expanded text is transformed into a vector representation to obtain an expanded characteristic representation; The basic feature representation, the temporal feature representation and the expanded feature representation are integrated through an attention mechanism model to obtain a comprehensive feature representation; Based on the comprehensive feature representation, a regression model is used to predict the trend of infectious disease transmission, and early warnings are issued according to set thresholds.

2. The method according to claim 1, characterized in that: The step of obtaining characteristic description information of infectious disease data and inputting the characteristic description information into a pre-built autoencoder model to obtain a basic characteristic representation of the infectious disease data includes: Assume that the characteristic description information of the input autoencoder model is , represents an n-dimensional data set, in which each element has n-dimensional features; The embedding layer accepts input x and maps x to a vector representation of a specified dimension. The calculation of the embedding layer is: in, and Represent the weight matrix and bias matrix of the embedding layer respectively; The encoder consists of a linear layer and an activation function. The output of the embedding layer is Enter the encoder to calculate: in, and denote the weight matrix and bias matrix of the hidden layer respectively, The basic feature representation indicating that the basic feature learning is completed, is a nonlinear activation function, and the calculation of the nonlinear activation function is: in, is a nonlinear activation function Received input.

3. The method according to claim 2, characterized in that: The method of using a gated recurrent unit model to capture the temporal features of the characteristic description information, wherein the feature values ​​in all the temporal features are sorted in chronological order and then learned and extracted to obtain a temporal feature representation, includes: The learning and extraction of time series features adopts the model structure of the GRU network, which is composed of multiple gated networks, including a reset gate, a candidate gate, and an update gate. The final hidden state calculated by the multiple gated networks is the time series feature representation.

4. The method according to claim 3, characterized in that: The learning and extraction of the temporal features adopts the model structure of the GRU network, which is composed of multiple gated networks, including a reset gate, a candidate gate, and an update gate. The final hidden state calculated by the multiple gated networks is the temporal feature representation, and the specific calculation is: Reset the gate: in, and They are the weight matrices of the reset gate, is the bias matrix of the reset gate, and The hidden state representations of the current input character and the previous character respectively represent characteristic description information; The calculation of candidate gates is divided into two cases: receiving the previous hidden state and not receiving the previous hidden state; When receiving the previous hidden state: in, and They are the weight matrix and bias matrix of the current network layer respectively; When not receiving the previous hidden state: in, and They are the weight matrix and bias matrix of the current network layer respectively; Update Gate: ( ) in, and are the weight matrices of the update gates, is the bias matrix of the update gate; Get the final hidden state: in, is the timing characteristic representation.

5. The method according to claim 4, characterized in that: The step of converting and organizing the characteristic description information into a natural language description, expanding the natural language description using a large model, and converting the expanded text into a vector representation to obtain an expanded characteristic representation includes: Construct prompt word instructions for inputting into the big model: construct the feature name and feature content corresponding to each line of data of the acquired characteristic description information into nested list format data, record the constructed data as content, and then splice it with the task instruction prompt for the big model, and use the format after splicing content and prompt as input, and input it into the big model for feature expansion; Use the fully connected layer to convert the feature text expanded by the large model into a vector representation: in, and are the weight matrix and bias matrix of the fully connected layer, is the text after the received large model features are expanded, It is a vector representation of the text transformation after the large model feature is expanded, that is, the expanded feature representation.

6. The method according to claim 5, characterized in that: The basic feature representation, the temporal feature representation and the expanded feature representation are fused through the attention mechanism model to obtain a comprehensive feature representation, including: Will , and After three first linear layers respectively, three Query vectors are obtained: , and , the weight matrices and bias matrices of the three first linear layers use the same weights: in, and Represent the weight matrix and bias matrix in each first linear layer, , and There is a set, denoted as ; Will , and After three second linear layers respectively, three Key vectors are obtained , and , the weight matrices and bias matrices of the three second linear layers also use the same weights: in, and Represent the weight matrix and bias matrix in each second linear layer, , and There is a set, denoted as ; Will , and After three third linear layers respectively, the Value vector , and , the weight matrices and bias matrices of the three third linear layers also use the same weights: in, and Represent the weight matrix and bias matrix in each second linear layer, , and There is a set, denoted as ; right , , The Query vector, Key vector and Value vector saved in the Query vector are combined in pairs to obtain vector combinations, and all the combinations of vector combinations are traversed and the Att results of the Query vector, Key vector and Value vector in each vector combination are calculated respectively; Assign corresponding weight influence factors to the Query vector, Key vector and Value vector in each of the Att results, sum all the Att results according to the assigned weight influence factors, multiply the sum result and each of the Att results to obtain the attention score of the fused weight information, and use the attention score of the fused weight information as the comprehensive feature representation for infectious disease prediction.

7. The method according to claim 6, characterized in that: Based on the comprehensive feature representation, the regression model is used to predict the spread trend of infectious diseases, and early warning is issued according to the set threshold, including: The comprehensive feature representation is input into the regression model. The regression model is divided into a hidden layer and a fully connected layer. The hidden layer calculation process is: in, and are the weight matrix and bias matrix of the hidden layer, respectively. It is a comprehensive feature representation; The fully connected layer is calculated as: in, and are the weight matrix and bias matrix of the fully connected layer respectively, The final predicted number of infected people at the next time point; Said Used to calculate the growth rate of the number of infections at the current time point: Set the warning threshold. When the rate exceeds the threshold, the warning mechanism is automatically triggered, and an infection out-of-control warning is output. The model prediction results and warning information are sent to relevant departments for early warning prompts.

8. A large-scale infectious disease prediction and early warning system based on multi-stage learning, characterized by: It includes basic feature representation acquisition module, time series feature representation acquisition module, extended feature representation acquisition module, feature fusion module and infectious disease transmission trend prediction and early warning module; The basic feature representation acquisition module is used to obtain characteristic description information of infectious disease data, input the characteristic description information into a pre-built autoencoder model, and obtain the basic feature representation of infectious disease data; The temporal feature representation acquisition module is used to capture the temporal features of the characteristic description information using a gated recurrent unit model, wherein the feature values ​​in all the temporal features are sorted in chronological order and then learned and extracted to obtain the temporal feature representation; The expanded feature representation acquisition module is used to transform and organize the characteristic description information into a natural language description, expand the natural language description using a large model, and transform the expanded text into a vector representation to obtain an expanded feature representation; The feature fusion module is used to fuse the basic feature representation, the temporal feature representation and the expanded feature representation through an attention mechanism model to obtain a comprehensive feature representation; The infectious disease transmission trend prediction and early warning module is used to predict the infectious disease transmission trend based on the comprehensive feature representation using a regression model, and to issue an early warning according to a set threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images