WO2021159739A1 - 疫情趋势预测方法、装置、电子设备及存储介质 - Google Patents

疫情趋势预测方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2021159739A1
WO2021159739A1 PCT/CN2020/124608 CN2020124608W WO2021159739A1 WO 2021159739 A1 WO2021159739 A1 WO 2021159739A1 CN 2020124608 W CN2020124608 W CN 2020124608W WO 2021159739 A1 WO2021159739 A1 WO 2021159739A1
Authority
WO
WIPO (PCT)
Prior art keywords
epidemic
feature matrix
sequence data
feature
data
Prior art date
Application number
PCT/CN2020/124608
Other languages
English (en)
French (fr)
Inventor
张渊
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2021159739A1 publication Critical patent/WO2021159739A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/80ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for detecting, monitoring or modelling epidemics or pandemics, e.g. flu
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/044Recurrent networks, e.g. Hopfield networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks

Definitions

  • This application relates to the field of computer technology, and in particular to an epidemic trend prediction method, device, electronic equipment, and storage medium.
  • the embodiments of the present application provide an epidemic trend prediction method, device, electronic equipment, and storage medium, which combine multi-dimensional features to predict the epidemic trend, which is more referable.
  • an embodiment of the present application provides an electronic device, including a processor and a memory, the processor and the memory are connected to each other, wherein the memory is used to store a computer program, and the computer program includes program instructions ,
  • the processor is configured to call the program instructions to execute the following steps:
  • the epidemic sequence data includes the first disease characteristic data, the second disease characteristic data, and the meteorological characteristic data on each date within the first preset time range;
  • the first epidemic trend prediction result includes the addition of each date within the predicted second preset date range The number of cases and/or the number of new deaths; the second preset date range is after the first preset date range;
  • the embodiments of the present application provide a method for predicting epidemic trends, including:
  • the epidemic sequence data includes the first disease characteristic data, the second disease characteristic data, and the meteorological characteristic data on each date within the first preset time range;
  • the first epidemic trend prediction result includes the addition of each date within the predicted second preset date range The number of cases and/or the number of new deaths; the second preset date range is after the first preset date range;
  • an embodiment of the present application provides an epidemic trend prediction device, including:
  • the acquiring module is used to acquire the epidemic sequence data of the target area, the epidemic sequence data including the first disease characteristic data, the second disease characteristic data, and the meteorological characteristic data of each date within the first preset time range;
  • the processing module is used to call the pre-trained time series model to predict the epidemic trend according to the target feature matrix to obtain the first epidemic trend prediction result, and the first epidemic trend prediction result includes the predicted second preset date range The number of new cases and/or the number of new deaths on each date; the second preset date range is after the first preset date range;
  • the processing module is also used to display the first epidemic trend prediction result through a terminal device.
  • an embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the following method:
  • the epidemic sequence data includes the first disease characteristic data, the second disease characteristic data, and the meteorological characteristic data on each date within the first preset time range;
  • the first epidemic trend prediction result includes the addition of each date within the predicted second preset date range The number of cases and/or the number of new deaths; the second preset date range is after the first preset date range;
  • This application can be used for epidemic trend prediction based on combining multiple disease features and meteorological features, which is highly referable.
  • FIG. 1 is a schematic flowchart of an epidemic trend prediction method provided by an embodiment of the present application
  • FIG. 2A is a schematic flowchart of another epidemic trend prediction method provided by an embodiment of the present application.
  • 2B is a schematic diagram of an epidemic trend prediction process provided by an embodiment of the present application.
  • FIG. 3 is a schematic structural diagram of an epidemic trend prediction system provided by an embodiment of the present application.
  • FIG. 4 is a schematic structural diagram of an epidemic trend prediction device provided by an embodiment of the present application.
  • Fig. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • the technical solution of this application can be applied to the fields of artificial intelligence, digital medical care, smart city, blockchain and/or big data technology to realize epidemic trend prediction.
  • the data involved in this application such as epidemic sequence data and/or prediction results, can be stored in a database, or can be stored in a blockchain, which is not limited in this application.
  • FIG. 1 is a schematic flowchart of an epidemic trend prediction method provided by an embodiment of this application.
  • This method can be applied to electronic devices.
  • the electronic device can be a terminal device or a server.
  • Terminal devices include but are not limited to smart terminals such as notebook computers and desktop computers.
  • the server can be a server or a server cluster. Specifically, the method may include the following steps:
  • S101 Acquire epidemic sequence data of a target area, where the epidemic sequence data includes first disease feature data, second disease feature data, and meteorological feature data on each date within a first preset time range.
  • the first disease characteristic data may include disease characteristic data of the first disease.
  • the second disease characteristic data includes disease characteristic data of the second disease.
  • the first disease is different from the second disease.
  • the first disease may be new coronary pneumonia.
  • the first disease characteristic data may be new coronary pneumonia characteristic data.
  • the second disease can be influenza.
  • the second disease characteristic data may be influenza characteristic data.
  • New coronary pneumonia characteristic data can include the number of cases and deaths of new coronary pneumonia.
  • the first disease characteristic data may include the number of people who have suffered onset of the first disease and the number of people who have died.
  • the second disease characteristic data may include the number of people who have suffered from the second disease and the number of people who have died.
  • Meteorological feature data include, but are not limited to, temperature, humidity, atmospheric pressure and other meteorological feature data.
  • the process for the electronic device to obtain the epidemic sequence data of the target area may be: when the timed task arrives, the electronic device obtains the epidemic sequence data of the target area crawled from the designated platform.
  • the process for the electronic device to obtain the epidemic sequence data of the target area may be: the electronic device obtains the epidemic sequence data of the target area submitted by the terminal device.
  • the electronic device can perform feature extraction on each feature data included in the epidemic sequence data to obtain the feature vector of each feature data, and according to the feature vector of each feature data, splicing to obtain the corresponding epidemic sequence data Target feature matrix.
  • the process of the electronic device splicing to obtain the target feature matrix corresponding to the epidemic sequence data according to the feature vector of each feature data may be: the electronic device splicing the feature vector of each feature data to the same feature matrix, and The feature matrix to which each feature vector is spliced is determined as the target feature matrix corresponding to the epidemic sequence data.
  • the process of the electronic device splicing to obtain the target feature matrix corresponding to the epidemic sequence data according to the feature vector of each feature data may be: the electronic device according to the features of each first disease feature data included in the epidemic sequence data The vector and the feature vector of each second disease feature data are spliced to obtain the first feature matrix, and according to the feature vector of each meteorological feature data included in the epidemic sequence data, the second feature matrix is spliced; the electronic device uses the first feature matrix And the second feature matrix are determined as the target feature matrix corresponding to the epidemic sequence data.
  • the first feature matrix can be a 2*100*300 feature matrix
  • the second feature matrix can also be a 2*100*300 feature matrix.
  • the difference between the two may be small, while the difference between the disease feature and the meteorological feature may be large, so the first disease feature can be spliced according to the feature vector of the disease feature data.
  • a feature matrix is spliced according to the feature vector of the meteorological feature data to obtain a second feature matrix, which is input to the pre-trained time series model, so that the model can predict the epidemic trend based on the two feature matrices and has higher prediction accuracy.
  • the first epidemic trend prediction result includes the predicted results of each date within the second preset date range. The number of new cases and/or the number of new deaths.
  • the time series model may be an autoregressive model, a moving average model, a moving average model, a differential autoregressive moving average model, or a recurrent neural network (RNN) model.
  • the second preset date range is after the first preset date range.
  • the first date range may be a system date and a date range before the system date.
  • the second date range can be a date range after the system date.
  • the number of dates corresponding to the second preset date range may be the same as or different from the number of dates corresponding to the first preset date range, which is not limited in this application.
  • the pre-trained time series model can be obtained in the following way: the electronic device obtains historical epidemic series data of the target area, and the historical epidemic series data includes the data of each date within the third preset date range.
  • the sequence model is trained to obtain a pre-trained time sequence model.
  • the way that the electronic device constructs the feature matrix corresponding to the historical epidemic sequence data according to the historical epidemic sequence data can be referred to the aforementioned method for the electronic device to construct the target feature matrix corresponding to the epidemic sequence data according to the epidemic sequence data.
  • the application examples are not repeated here.
  • the electronic device uses the feature matrix corresponding to the historical epidemic sequence data to train the original time series model, and the process of obtaining the pre-trained time series model may be: electronic equipment
  • the hidden layer in the original RNN model is processed according to the feature matrix corresponding to the historical epidemic sequence data to obtain the high-dimensional feature matrix corresponding to the feature matrix; the electronic device is based on the high-dimensional feature matrix and the original RNN model.
  • the output layer obtains the second epidemic trend prediction result.
  • the second epidemic trend prediction result includes the number of new cases and/or the number of new deaths on each date within the predicted fourth preset date range; the electronic device uses the first 2.
  • the prediction result of the epidemic trend and the actual result of the corresponding epidemic trend are used to construct a loss function, and use the loss function to train the original RNN model to obtain a pre-trained RNN model.
  • the real result of the epidemic trend includes the real fourth preset date range The number of new cases and/or the number of new deaths as of the date.
  • the prediction result of the second epidemic trend here is used to distinguish it from the prediction result of the first epidemic trend, and does not indicate a sequential relationship.
  • the fourth preset date range is after the third preset date range.
  • the number of dates corresponding to the fourth preset date range may be the same or different from the number of dates corresponding to the third preset date range, and this application is not limited.
  • the electronic device calls the pre-trained time series model to predict the epidemic trend according to the target feature matrix
  • the process of obtaining the first epidemic trend prediction result may be: electronic equipment
  • the hidden layer in the pre-trained RNN model is processed according to the target feature matrix to obtain the target high-dimensional feature matrix corresponding to the target feature matrix; the electronic device is based on the target high-dimensional feature matrix and the output in the pre-trained RNN model Layer, obtain the first epidemic trend forecast result.
  • the hidden layer here may include a hidden layer, for example, the hidden layer may be a 5-layer RNN, including 5-layer hidden units.
  • the output layer here can be a linear regression layer, for example.
  • the electronic device obtains the first epidemic trend prediction result according to the target high-dimensional feature matrix and the output layer in the pre-trained RNN model: the electronic device inputs the target high-dimensional feature matrix The output layer in the pre-trained RNN model is processed to obtain the first epidemic trend prediction result.
  • the target high-dimensional feature matrix may include the high-dimensional feature matrix of each of the two feature matrices, and the electronic device can compare the two feature matrices. The high-dimensional feature matrix of each matrix is fused and input to the output layer for processing to obtain the first epidemic trend prediction result.
  • the server may send the first epidemic trend prediction result to the terminal device, and the terminal device may display the first epidemic trend prediction result.
  • the terminal device can display the prediction result of the first epidemic trend.
  • the electronic device can obtain the epidemic sequence data of the target area.
  • the epidemic sequence data includes the first disease feature data, the second disease feature data, and the meteorological data of each date within the first preset time range.
  • Feature data the electronic device constructs the target feature matrix corresponding to the epidemic sequence data according to the epidemic sequence data, and calls the pre-trained time series model to predict the epidemic trend according to the target feature matrix, and obtains the first epidemic trend prediction result to pass the terminal
  • the device displays the prediction result of the first epidemic trend.
  • this application can be used to predict the epidemic trend based on combining multiple disease features and meteorological features, which is highly referable .
  • FIG. 2A is a schematic flowchart of another epidemic trend prediction method provided by an embodiment of this application.
  • This method can be applied to the aforementioned electronic devices. Specifically, the method may include the following steps:
  • step S201 may refer to step S101 in the embodiment of FIG.
  • S202 Perform feature extraction on each feature data included in the epidemic sequence data to obtain a feature vector of each feature data.
  • S203 According to the feature vector of each first disease feature data and the feature vector of each second disease feature data included in the epidemic sequence data, splicing to obtain a first feature matrix.
  • the electronic device may perform feature extraction on each feature data included in the epidemic sequence data to obtain the feature vector of each feature data, and based on the features of each first disease feature data included in the epidemic sequence data
  • the vector and the feature vector of each second disease feature data are spliced to obtain the first feature matrix
  • the second feature matrix is spliced to obtain the second feature matrix, so that the first feature matrix and The second feature matrix is determined as the target feature matrix corresponding to the epidemic sequence data.
  • the hidden layer in the RNN model may include a first hidden layer and a second hidden layer, and the structure of the first hidden layer may be the same as or different from the structure of the second hidden layer.
  • the pre-trained RNN model can be obtained in the following manner:
  • the electronic device can splice the feature vectors of each first disease feature data included in the historical epidemic sequence data and the feature vectors of each second disease feature data to obtain a third feature matrix.
  • the feature vector of the feature data is spliced to obtain a fourth feature matrix, and the third feature matrix and the fourth feature matrix are determined as feature matrices corresponding to the historical epidemic sequence data.
  • the third feature matrix can be a 2*100*300 feature matrix
  • the fourth feature matrix can also be a 2*100*300 feature matrix.
  • the electronic device processes the feature matrix corresponding to the historical epidemic sequence data through the hidden layer in the original RNN model to obtain the high-dimensional feature matrix corresponding to the feature matrix.
  • the electronic device can process the third feature matrix through the first hidden layer in the pre-trained RNN model to obtain the third high-dimensional feature matrix corresponding to the third feature matrix, and pass the data in the pre-trained RNN model
  • the second hidden layer processes the fourth feature matrix to obtain a fourth high-dimensional feature matrix corresponding to the fourth feature matrix.
  • the third high-dimensional feature matrix here refers to the high-dimensional feature matrix corresponding to the third feature matrix
  • the fourth high-dimensional feature matrix refers to the high-dimensional feature matrix corresponding to the fourth feature matrix.
  • the characteristics of the first disease such as new coronary pneumonia
  • the characteristics of the second disease such as influenza
  • the two models can be better modeled based on the shared network parameters of the first hidden layer. climate characteristics do not participate in parameter sharing, but will be combined with disease characteristics in the future.
  • the electronic device obtains a second epidemic trend prediction result according to the high-dimensional feature matrix and the output layer in the original RNN model, and the second epidemic trend prediction result includes each date within the predicted fourth preset date range The number of new cases and/or the number of new deaths.
  • the electronic device may perform fusion processing on the third high-dimensional feature matrix and the fourth high-dimensional feature matrix to obtain a fused feature matrix, and output the second epidemic trend prediction result according to the fused feature matrix through the output layer in the RNN model.
  • the electronic device performs fusion processing on the third high-dimensional feature matrix and the fourth high-dimensional feature matrix
  • the process of obtaining the fused feature matrix may be: the electronic device determines the attention corresponding to the fourth high-dimensional feature matrix And use the attention weight to perform weighting processing on the fourth high-dimensional feature matrix to obtain a weighted feature matrix; the electronic device performs splicing processing on the third high-dimensional feature matrix and the weighted feature matrix to obtain a fused feature matrix.
  • the dimension of the fusion feature matrix is the same as the dimension of the third feature matrix.
  • the process of the electronic device determining the attention weight corresponding to the fourth high-dimensional feature matrix may be: the electronic device performs an attention operation according to the third high-dimensional feature matrix and the fourth high-dimensional feature matrix to obtain the first The attention weight corresponding to the four-dimensional feature matrix.
  • the electronic device uses the second epidemic trend prediction result and the corresponding actual epidemic trend result to construct a loss function, and uses the loss function to train the original RNN model to obtain a pre-trained RNN model.
  • the actual epidemic trend result includes The number of new cases and/or the number of new deaths on each date within the true fourth preset date range.
  • the fourth preset date range here is after the third preset date range.
  • the number of dates corresponding to the fourth preset date range may be the same as or different from the number of dates corresponding to the third preset date range.
  • S208 Perform fusion processing on the first high-dimensional feature matrix and the second high-dimensional feature matrix to obtain a fused feature matrix.
  • the output layer in the pre-trained RNN model is processed according to the fusion feature matrix to obtain a first epidemic trend prediction result.
  • the first epidemic trend prediction result includes the predicted second preset date range. The number of new cases and/or the number of new deaths as of the date.
  • the electronic device can process the second feature matrix through the second hidden layer in the pre-trained RNN model to obtain the second high-dimensional feature matrix corresponding to the second feature matrix, and can The first high-dimensional feature matrix and the second high-dimensional feature matrix are fused to obtain a fused feature matrix, and then the output layer in the pre-trained RNN model is processed according to the fused feature matrix to obtain the first epidemic trend forecast result.
  • the first high-dimensional feature matrix may be a 1*4096 feature matrix
  • the second high-dimensional feature matrix may be a 1*4096 feature matrix.
  • the electronic device performs fusion processing on the first high-dimensional feature matrix and the second high-dimensional feature matrix
  • the process of obtaining the fused feature matrix may be: the electronic device determines the attention corresponding to the second high-dimensional feature matrix Weight, and use the attention weight to perform weighting processing on the second high-dimensional feature matrix to obtain a weighted feature matrix; the electronic device performs splicing processing on the first high-dimensional feature matrix and the weighted feature matrix to obtain a fused feature matrix.
  • the dimension of the fusion feature matrix here is the same as the dimension of the first feature matrix.
  • the process of the electronic device determining the attention weight corresponding to the second high-dimensional feature matrix may be: the electronic device performs an attention operation according to the first high-dimensional feature matrix and the second high-dimensional feature matrix to obtain the first high-dimensional feature matrix.
  • the attention weight corresponding to the two-dimensional feature matrix may be: the electronic device performs an attention operation according to the first high-dimensional feature matrix and the second high-dimensional feature matrix to obtain the first high-dimensional feature matrix.
  • the electronic device can obtain the first feature matrix according to the new coronary pneumonia feature data and the flu feature data to input the first hidden layer in the pre-trained RNN model for processing to obtain the first high-dimensional feature matrix, and can be based on the first feature matrix.
  • the climate feature data obtains the second feature matrix and inputs it to the second hidden layer in the pre-trained RNN model for processing to obtain the second high-dimensional feature matrix.
  • the electronic device can use the attention weight corresponding to the second high-dimensional feature matrix to weight the second high-dimensional feature matrix to obtain a weighted feature matrix, and perform splicing processing on the first high-dimensional feature matrix and the weighted feature matrix to obtain the fused feature
  • the matrix is then processed according to the fusion feature matrix through the output layer in the pre-trained RNN model, and the first epidemic trend prediction result is output.
  • step S210 may refer to step S104 in the embodiment of FIG.
  • the electronic device can splice the first feature matrix according to the feature vector of each first disease feature data included in the epidemic sequence data and the feature vector of each second disease feature data, and according to the epidemic sequence
  • the feature vector of each meteorological feature data included in the data is spliced to obtain a second feature matrix;
  • the electronic device processes the first feature matrix through the first hidden layer in the pre-trained RNN model to obtain the corresponding
  • the first high-dimensional feature matrix is processed by the second hidden layer in the pre-trained RNN model to obtain the second high-dimensional feature matrix corresponding to the second feature matrix;
  • the high-dimensional feature matrix and the second high-dimensional feature matrix are fused to obtain a fused feature matrix, and the output layer in the pre-trained RNN model is processed according to the fused feature matrix to obtain the first epidemic trend prediction result, and Different hidden layers process different features separately and then merge them for prediction, which can improve the accuracy of model prediction.
  • This application can be used in the field of medical technology and involves blockchain technology.
  • the prediction result of the first epidemic situation can be written into the blockchain, or the compressed data of the prediction data of the first epidemic situation can be written into the blockchain.
  • the epidemic trend prediction system shown in FIG. 3 includes a server 10 and a terminal device 20. in:
  • the server 10 can obtain the first epidemic trend prediction result by performing step S101-step S103, and can make the terminal device 20 display the first epidemic trend prediction result by performing step S104.
  • This process combines multi-dimensional features to predict the epidemic trend, which can be referred to Higher sex.
  • FIG. 4 is a schematic structural diagram of an epidemic trend prediction device provided by an embodiment of this application.
  • the epidemic trend prediction device can be applied to the aforementioned electronic equipment.
  • the epidemic trend prediction device may include:
  • the acquiring module 401 is configured to acquire epidemic sequence data in a target area, where the epidemic sequence data includes first disease characteristic data, second disease characteristic data, and meteorological characteristic data on each date within a first preset time range.
  • the construction module 402 is configured to construct a target feature matrix corresponding to the epidemic sequence data according to the epidemic sequence data.
  • the processing module 403 is configured to call a pre-trained time series model to predict the epidemic trend according to the target feature matrix to obtain a first epidemic trend prediction result, and the first epidemic trend prediction result includes the predicted second preset date range The number of new cases and/or the number of new deaths on each date within the period; the second preset date range is after the first preset date range,
  • the processing module 403 is further configured to display the first epidemic trend prediction result through a terminal device.
  • the epidemic trend prediction device shown in FIG. 4 may further include an output module (not shown in the figure).
  • the processing module 403 may send the first epidemic trend prediction result to the terminal device through the output module, so that the terminal device can display the first epidemic trend prediction result.
  • the processing module 403 may display the first epidemic trend prediction result through the output module.
  • the processing module 403 calls a pre-trained time series model to predict the epidemic trend according to the target feature matrix, and obtains the first epidemic trend prediction result, specifically the pre-trained RNN model
  • the hidden layer is processed according to the target feature matrix to obtain the target high-dimensional feature matrix corresponding to the target feature matrix; according to the target high-dimensional feature matrix and the output layer in the pre-trained RNN model, the first 1. Prediction results of epidemic trend.
  • the construction module 402 constructs a target feature matrix corresponding to the epidemic sequence data according to the epidemic sequence data, specifically, extracting features of each feature data included in the epidemic sequence data to obtain the The feature vector of each feature data; according to the feature vector of each feature data, splicing to obtain the target feature matrix corresponding to the epidemic sequence data.
  • the construction module 402 splices the target feature matrix corresponding to the epidemic sequence data according to the feature vector of each feature data, specifically according to each first disease included in the epidemic sequence data
  • the feature vector of the feature data and the feature vector of each second disease feature data are spliced to obtain the first feature matrix; according to the feature vectors of the meteorological feature data included in the epidemic sequence data, the second feature matrix is spliced;
  • a feature matrix and the second feature matrix are determined as target feature matrices corresponding to the epidemic sequence data.
  • the processing module 403 performs processing according to the target feature matrix through the hidden layer in the pre-trained RNN model to obtain the target high-dimensional feature matrix corresponding to the target feature matrix, which is specifically
  • the first feature matrix is processed through the first hidden layer in the pre-trained RNN model to obtain the first high-dimensional feature matrix corresponding to the first feature matrix; through the pre-trained RNN model
  • the second hidden layer processes the second feature matrix to obtain a second high-dimensional feature matrix corresponding to the second feature matrix.
  • the processing module 403 obtains the first epidemic trend prediction result according to the target high-dimensional feature matrix and the output layer in the pre-trained RNN model, specifically the prediction result of the first epidemic trend
  • the fusion process of the dimensional feature matrix and the second high-dimensional feature matrix is performed to obtain a fusion feature matrix; the output layer in the pre-trained RNN model is processed according to the fusion feature matrix to obtain the first epidemic trend prediction result.
  • the processing module 403 is further configured to obtain historical epidemic sequence data of the target area, where the historical epidemic sequence data includes the first disease characteristic data of each date within the third preset date range , The second disease feature data and meteorological feature data; construct the feature matrix corresponding to the historical epidemic sequence data according to the historical epidemic sequence data; use the feature matrix corresponding to the historical epidemic sequence data to train the original time series model, Get the pre-trained time series model.
  • the time series model is an RNN model
  • the processing module 403 uses the feature matrix corresponding to the historical epidemic sequence data to train the original time series model to obtain a pre-trained time series model
  • the hidden layer in the original RNN model is processed according to the feature matrix corresponding to the historical epidemic sequence data to obtain the high-dimensional feature matrix corresponding to the feature matrix; according to the high-dimensional feature matrix and the original
  • the second epidemic trend prediction result is obtained, and the second epidemic trend prediction result includes the number of new cases and/or the number of new deaths on each date within the predicted fourth preset date range
  • the fourth preset date range is after the third preset date range; the second epidemic trend prediction result and the corresponding actual epidemic trend result are used to construct a loss function, and the loss function is used to train the original
  • the RNN model obtains a pre-trained RNN model, and the actual result of the epidemic trend includes the number of new cases and/or the number of new deaths on each date within the true fourth preset date range.
  • the epidemic trend prediction device can obtain the epidemic sequence data of the target area, and the epidemic sequence data includes the first disease characteristic data and the second disease characteristic data of each date within the first preset time range. And meteorological feature data; the epidemic trend prediction device constructs the target feature matrix corresponding to the epidemic sequence data according to the epidemic sequence data, and calls the pre-trained time series model to predict the epidemic trend based on the target feature matrix, and obtain the first epidemic trend prediction As a result, the first epidemic trend prediction result is displayed through the terminal device. Compared with the process of predicting the epidemic trend based on a single factor in the prior art, this application can be used for epidemic trend prediction based on combining multiple disease characteristics and meteorological features. The reference is high.
  • FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of this application.
  • the electronic device described in this embodiment may include: one or more processors 1000 and a memory 2000.
  • the processor 1000 and the memory 2000 may be connected by a bus.
  • the processor 1000 may be a central processing module (Central Processing Unit, CPU), the processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC) , Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
  • the memory 2000 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as a magnetic disk memory. Wherein, the memory 2000 is used to store a computer program, and the computer program includes program instructions.
  • the processor 1000 is configured to call the program instructions and execute the following steps:
  • the epidemic sequence data includes the first disease characteristic data, the second disease characteristic data, and the meteorological characteristic data on each date within the first preset time range;
  • the first epidemic trend prediction result includes the addition of each date within the predicted second preset date range The number of cases and/or the number of new deaths; the second preset date range is after the first preset date range;
  • the electronic device shown in FIG. 5 may further include an output device (not shown).
  • the processor 1000 may send the first epidemic trend prediction result to the terminal device through the output device, so that the terminal device can display the first epidemic trend prediction result.
  • the processor 1000 may display the processing result of the first epidemic trend through an output device.
  • the output device may be a standard wired/wireless interface, or may be a display screen, touch display screen or other devices.
  • the time series model is a recurrent neural network RNN model.
  • the processor 1000 Is configured to call the program instructions to perform the following steps:
  • the first epidemic trend prediction result is obtained.
  • the processor 1000 when constructing the target feature matrix corresponding to the epidemic sequence data according to the epidemic sequence data, the processor 1000 is configured to call the program instructions to perform the following steps:
  • the target feature matrix corresponding to the epidemic sequence data is obtained by splicing.
  • the processor 1000 when the target feature matrix corresponding to the epidemic sequence data is spliced according to the feature vectors of the feature data, the processor 1000 is configured to call the program instructions to perform the following steps:
  • the first feature matrix is spliced together;
  • the first feature matrix and the second feature matrix are determined as target feature matrices corresponding to the epidemic sequence data.
  • the processor 1000 when the hidden layer in the pre-trained RNN model is processed according to the target feature matrix to obtain the target high-dimensional feature matrix corresponding to the target feature matrix, the processor 1000 is configured to Call the program instructions and perform the following steps:
  • the processor 1000 is configured to call the program instructions to perform the following steps:
  • the output layer in the pre-trained RNN model is processed according to the fusion feature matrix to obtain the first epidemic trend prediction result.
  • the processor 1000 is configured to call the program instructions and further execute the following steps:
  • the historical epidemic sequence data including the first disease characteristic data, the second disease characteristic data, and the meteorological characteristic data of each date within the third preset date range;
  • the original time series model is trained by using the feature matrix corresponding to the historical epidemic sequence data to obtain a pre-trained time series model.
  • the time series model is an RNN model.
  • the processor 1000 is configured Used to call the program instructions to perform the following steps:
  • a second epidemic trend prediction result is obtained, and the second epidemic trend prediction result includes the addition of each date within the predicted fourth preset date range The number of cases and/or the number of new deaths; the fourth preset date range is after the third preset date range;
  • the second epidemic trend prediction result and the corresponding actual result of the epidemic trend to construct a loss function
  • the actual result of the epidemic trend includes the real first 4. The number of new cases and/or the number of new deaths on each date within the preset date range.
  • the processor 1000 described in the embodiment of the present application can perform the implementation described in the embodiment of FIG. 1 and the embodiment of FIG. 2A, and can also perform the implementation described in the embodiment of the present application, which will not be repeated here. .
  • the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method of the above-mentioned embodiment can be implemented, which will not be repeated here.
  • the storage medium involved in this application such as a computer-readable storage medium, may be non-volatile or volatile.
  • the functional modules in each embodiment of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module.
  • the above-mentioned integrated modules can be implemented in the form of sampling hardware, and can also be implemented in the form of sampling software functional modules.
  • the program can be stored in a computer readable storage medium. During execution, it may include the procedures of the above-mentioned method embodiments.
  • the computer-readable storage medium may be volatile or non-volatile.
  • the computer storage medium may be a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM).
  • the computer-readable storage medium may mainly include a storage program area and a storage data area, where the storage program area may store an operating system, an application program required by at least one function, etc.; the storage data area may store information based on the blockchain node Use the created data, etc.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种疫情趋势预测方法、装置、电子设备及存储介质,应用于医疗科技领域,该电子设备包括处理器和存储器,存储器用于存储计算机程序,计算机程序包括程序指令,处理器被配置用于调用程序指令,执行以下步骤:获取目标地区的疫情序列数据(S101);根据疫情序列数据构建疫情序列数据对应的目标特征矩阵(S102);调用预训练的时间序列模型以根据目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数(S103)。结合多维度特征来进行疫情趋势预测,可参考性更高,涉及区块链技术,可将第一疫情趋势预测结果写入区块链中。

Description

疫情趋势预测方法、装置、电子设备及存储介质
本申请要求于2020年9月28日提交中国专利局、申请号为202011043913.3,发明名称为“疫情趋势预测方法、装置、电子设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域,尤其涉及一种疫情趋势预测方法、装置、电子设备及存储介质。
背景技术
疫情的爆发和蔓延会对各个地方的经济以及人民生活带来严重的影响。近段时间来,COVID-19引起了世界范围内的疫情爆发,造成了极大的生命损失和经济损失。发明人意识到,现有的流行病学预测模型大多是针对单一疾病在人群中的演化进行建模预测,存在一定的局限性:1、只考虑单一疾病的演化,没有兼顾同时流行的多种疾病的影响。2、只采用了单一模态的数据,未能考虑多种因素的协同作用。可见,现有的疫情趋势预测方法可参考性较低。
发明内容
本申请实施例提供了一种疫情趋势预测方法、装置、电子设备及存储介质,结合多维度的特征进行疫情趋势预测,可参考性更高。
第一方面,本申请实施例提供了一种电子设备,包括处理器和存储器,所述处理器和所述存储器相互连接,其中,所述存储器用于存储计算机程序,所述计算机程序包括程序指令,所述处理器被配置用于调用所述程序指令,执行以下步骤:
获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
通过终端设备展示所述第一疫情趋势预测结果。
第二方面,本申请实施例提供了一种疫情趋势预测方法,包括:
获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
通过终端设备展示所述第一疫情趋势预测结果。
第三方面,本申请实施例提供了一种疫情趋势预测装置,包括:
获取模块,用于获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
构建模块,用于根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
处理模块,用于调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
所述处理模块,还用于通过终端设备展示所述第一疫情趋势预测结果。
第四方面,本申请实施例提供了一种计算机可读存储介质,所述计算机可读存储介质存储有计算机程序,所述计算机程序被处理器执行以实现以下方法:
获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
通过终端设备展示所述第一疫情趋势预测结果。
本申请可以基于结合多个疾病特征以及气象特征以用于疫情趋势预测,可参考性较高。
附图说明
图1是本申请实施例提供的一种疫情趋势预测方法的流程示意图;
图2A是本申请实施例提供的另一种疫情趋势预测方法的流程示意图;
图2B是本申请实施例提供的一种疫情趋势预测过程示意图;
图3是本申请实施例提供的一种疫情趋势预测系统的结构示意图;
图4是本申请实施例提供的一种疫情趋势预测装置的结构示意图;
图5是本申请实施例提供的一种电子设备的结构示意图。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行描述。
本申请的技术方案可应用于人工智能、数字医疗、智慧城市、区块链和/或大数据技术领域,以实现疫情趋势预测。可选的,本申请涉及的数据如疫情序列数据和/或预测结果等可存储于数据库中,或者可以存储于区块链中,本申请不做限定。
请参阅图1,为本申请实施例提供的一种疫情趋势预测方法的流程示意图。该方法可以应用于电子设备。电子设备可以为终端设备或服务器。终端设备包括但不限于笔记本电脑、台式电脑等智能终端。服务器可以为一个服务器或服务器集群。具体地,该方法可以包括以下步骤:
S101、获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据。
其中,第一疾病特征数据可以包括第一疾病的疾病特征数据。第二疾病特征数据包括第二疾病的疾病特征数据。第一疾病与第二疾病不同。例如,第一疾病可以为新冠肺炎。相应地,第一疾病特征数据可以为新冠肺炎特征数据。第二疾病可以为流感。相应地,第二疾病特征数据可以为流感特征数据。新冠肺炎特征数据可以包括关于新冠肺炎的发病人数及死亡人数。第一疾病特征数据可以包括关于第一疾病发病的人数和死亡的人数。第二疾病特征数据可以包括关于第二疾病发病的人数及死亡的人数。气象特征数据包括但不限于气温、湿度、气压等气象特征数据。
在一个实施例中,电子设备获取目标地区的疫情序列数据的过程可以为:电子设备在定时任务到达时,获取从指定平台爬取的目标地区的疫情序列数据。
在一个实施例中,电子设备获取目标地区的疫情序列数据的过程可以为:电子设备获取终端设备提交的目标地区的疫情序列数据。
S102、根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵。
本申请实施例中,电子设备可以对该疫情序列数据包括的各特征数据进行特征提取,得到该各特征数据的特征向量,并根据该各特征数据的特征向量,拼接得到该疫情序列数据对应的目标特征矩阵。
在一个实施例中,电子设备根据该各特征数据的特征向量,拼接得到该疫情序列数据 对应的目标特征矩阵的过程可以为:电子设备将该各特征数据的特征向量拼接至同一特征矩阵,并将各特征向量拼接至的特征矩阵确定为该疫情序列数据对应的目标特征矩阵。
在一个实施例中,电子设备根据该各特征数据的特征向量,拼接得到该疫情序列数据对应的目标特征矩阵的过程可以为:电子设备根据该疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵,并根据该疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵;电子设备将该第一特征矩阵和该第二特征矩阵确定为该疫情序列数据对应的目标特征矩阵。例如,第一特征矩阵可以为2*100*300的特征矩阵,第二特征矩阵也可以为2*100*300的特征矩阵。由于第一疾病特征与第二疾病特征同属疾病特征,两者之间的差异可能较小,而疾病特征与气象特征之间的差异可能较大,因此可以根据疾病特征数据的特征向量拼接得到第一特征矩阵,并根据气象特征数据的特征向量拼接得到第二特征矩阵,以输入预训练的时间序列模型,使得模型可以根据两个特征矩阵进行疫情趋势预测,具有更高的预测精度。
具体地,电子设备根据该疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵的过程可以为:电子设备将该疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量拼接至同一特征矩阵;电子设备将该各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量拼接至的特征矩阵,确定为第一特征矩阵。
S103、调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
其中,该时间序列模型可以为自回归模型、移动平均模型、移动平均模型、差分自回归移动平均模型或循环神经网络(Recurrent Neural Network,RNN)模型。该第二预设日期范围在该第一预设日期范围之后。在一个实施例,第一日期范围可以为系统日期及系统日期之前的日期范围。第二日期范围可以为系统日期之后的日期范围。该第二预设日期范围对应的日期数量可以与第一预设日期范围对应的日期数量相同或不同,本申请不做限制。
在一个实施例中,所述的预训练的时间序列模型可以通过如下方式得到:电子设备获取目标地区的历史疫情序列数据,该历史疫情序列数据包括在第三预设日期范围内的各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;电子设备根据该历史疫情序列数据构建该历史疫情序列数据对应的特征矩阵,并利用该历史疫情序列数据对应的特征矩阵对待训练的时间序列模型进行训练,得到预训练的时间序列模型。其中,电子设备根据该历史疫情序列数据构建该历史疫情序列数据对应的特征矩阵的方式,可以参见前述提及的电子设备根据该疫情序列数据构建该疫情序列数据对应的目标特征矩阵的方式,本申请实施例在此不做赘述。
在一个实施例中,当时间序列模型为RNN模型时,电子设备利用该历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型的过程可以为:电子设备通过该原始的RNN模型中的隐藏层根据该历史疫情序列数据对应的特征矩阵进行处理,得到该特征矩阵对应的高维特征矩阵;电子设备根据该高维特征矩阵以及该原始的RNN模型中的输出层,获得第二疫情趋势预测结果,该第二疫情趋势预测结果包括预测的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;电子设备利用该第二疫情趋势预测结果以及对应的疫情趋势真实结果构建损失函数,利用该损失函数训练该原始的RNN模型,得到预训练的RNN模型,该疫情趋势真实结果包括真实的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。此处的第二疫情趋势预测结果用于与第一疫情趋势预测结果区分开来描述,不表示顺序关系。该第四预设日期范围在该第三预设日期范围之后。该第四预设日期范围对应的日期数量可以与第三预设日期范围对 应的日期数量相同或不同,本申请不做限制。
在一个实施例中,当时间序列模型为RNN模型时,电子设备调用该预训练的时间序列模型以根据该目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果的过程可以为:电子设备通过预训练的RNN模型中的隐藏层根据该目标特征矩阵进行处理,得到该目标特征矩阵对应的目标高维特征矩阵;电子设备根据该目标高维特征矩阵以及该预训练的RNN模型中的输出层,获得第一疫情趋势预测结果。此处的隐藏层可以包括一个隐藏层,该隐藏层如可以为5层RNN,包括5层隐含单元。此处的输出层如可以为线性回归linear regression层。
在一个实施例中,电子设备根据该目标高维特征矩阵以及该预训练的RNN模型中的输出层,获得第一疫情趋势预测结果的过程可以为:电子设备将对该目标高维特征矩阵输入该预训练的RNN模型中的输出层进行处理,得到第一疫情趋势预测结果。在一个实施例中,当目标特征矩阵包括第一特征矩阵和第二特征矩阵时,目标高维特征矩阵可以包括这两个特征矩阵各自的高维特征矩阵,电子设备可以在对这两个特征矩阵各自的高维特征矩阵进行融合处理后输入到输出层进行处理,得到第一疫情趋势预测结果。
S104、通过终端设备展示所述第一疫情趋势预测结果。
本申请实施例中,当电子设备为服务器时,服务器可以将第一疫情趋势预测结果发送至终端设备,终端设备可以展示该第一疫情趋势预测结果。当电子设备为终端设备时,终端设备可以展示该第一疫情趋势预测结果。
可见,图1所示的实施例中,电子设备可以获取目标地区的疫情序列数据,该疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;电子设备根据该疫情序列数据构建该疫情序列数据对应的目标特征矩阵,并调用预训练的时间序列模型以根据该目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果以通过终端设备展示该第一疫情趋势预测结果,相较于现有技术基于单一因素进行疫情趋势预测的过程,本申请可以基于结合多个疾病特征以及气象特征以用于疫情趋势预测,可参考性较高。
请参阅图2A,为本申请实施例提供的另一种疫情趋势预测方法的流程示意图。该方法可以应用于前述提及的电子设备。具体地,该方法可以包括以下步骤:
S201、获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据。
其中,步骤S201可以参见图1实施例中的步骤S101,本申请实施例在此不做赘述。
S202、对所述疫情序列数据包括的各特征数据进行特征提取,得到所述各特征数据的特征向量。
S203、根据所述疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵。
S204、根据所述疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵。
S205、将所述第一特征矩阵和所述第二特征矩阵确定为所述疫情序列数据对应的目标特征矩阵。
在步骤S202-步骤S205中,电子设备可以对该疫情序列数据包括的各特征数据进行特征提取,得到该各特征数据的特征向量,并根据该疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵,并根据该疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵,从而将该第一特征矩阵和该第二特征矩阵确定为该疫情序列数据对应的目标特征矩阵。
S206、通过所述预训练的RNN模型中的第一隐藏层对所述第一特征矩阵进行处理,得 到所述第一特征矩阵对应的第一高维特征矩阵。
本申请实施例中,所述的RNN模型中的隐藏层可以包括第一隐藏层和第二隐藏层,第一隐藏层的结构可以与第二隐藏层的结构相同,也可以不同。
本申请实施例中,所述的预训练的RNN模型可以通过以下方式得到:
①根据所述历史疫情序列数据构建所述历史疫情序列数据对应的特征矩阵。
电子设备可以对该历史疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量进行拼接处理,得到第三特征矩阵,对该历史疫情序列数据包括的各气象特征数据的特征向量进行拼接处理,得到第四特征矩阵,并将该第三特征矩阵和该第四特征矩阵确定为该历史疫情序列数据对应的特征矩阵。例如,第三特征矩阵可以为2*100*300的特征矩阵,第四特征矩阵也可以为2*100*300的特征矩阵。
②电子设备通过所述原始的RNN模型中的隐藏层根据所述历史疫情序列数据对应的特征矩阵进行处理,得到所述特征矩阵对应的高维特征矩阵。
电子设备可以通过该预训练的RNN模型中的第一隐藏层对该第三特征矩阵进行处理,得到该第三特征矩阵对应的第三高维特征矩阵,并通过该预训练的RNN模型中的第二隐藏层对该第四特征矩阵进行处理,得到该第四特征矩阵对应的第四高维特征矩阵。此处的第三高维特征矩阵是指第三特征矩阵对应的高维特征矩阵,第四高维特征矩阵是指第四特征矩阵对应的高维特征矩阵。第一疾病(如新冠肺炎)的特征和第二疾病(如流感)的特征基于第一隐藏层共享网络参数,互相学习,这是因为在一定情况下两种疾病的传播方式有一定的相似性,两者基于第一隐藏层共享网络参数可以更好的进行建模。气候特征不参与参数共享,而是在后续会与疾病特征相结合。
③电子设备根据所述高维特征矩阵以及所述原始的RNN模型中的输出层,获得第二疫情趋势预测结果,所述第二疫情趋势预测结果包括预测的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
电子设备可以对该第三高维特征矩阵和该第四高维特征矩阵进行融合处理,得到融合特征矩阵,并通过该RNN模型中的输出层根据该融合特征矩阵输出第二疫情趋势预测结果。
在一个实施例中,电子设备对该第三高维特征矩阵和该第四高维特征矩阵进行融合处理,得到融合特征矩阵的过程可以为:电子设备确定该第四高维特征矩阵对应的attention权重,并利用该attention权重对该第四高维特征矩阵进行加权处理,得到加权特征矩阵;电子设备对该第三高维特征矩阵以及该加权特征矩阵进行拼接处理,得到融合特征矩阵。融合特征矩阵的维度与第三特征矩阵的维度相同。在一个实施例中,电子设备确定该第四高维特征矩阵对应的attention权重的过程可以为:电子设备根据该第三高维特征矩阵以及该第四高维特征矩阵执行attention操作,得到该第四高维特征矩阵对应的attention权重。
④电子设备利用所述第二疫情趋势预测结果以及对应的疫情趋势真实结果构建损失函数,利用所述损失函数训练所述原始的RNN模型,得到预训练的RNN模型,所述疫情趋势真实结果包括真实的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
此处的第四预设日期范围在第三预设日期范围之后。第四预设日期范围对应的日期数量可以与第三预设日期范围对应的日期数量相同,也可以不同。
S207、通过所述预训练的RNN模型中的第二隐藏层对所述第二特征矩阵进行处理,得到所述第二特征矩阵对应的第二高维特征矩阵。
S208、对所述第一高维特征矩阵和所述第二高维特征矩阵进行融合处理,得到融合特征矩阵。
S209、通过所述预训练的RNN模型中的输出层根据所述融合特征矩阵进行处理,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日 期的新增病例的数量和/或新增死亡的人数。
在步骤S207-步骤S209中,电子设备可以通过该预训练的RNN模型中的第二隐藏层对该第二特征矩阵进行处理,得到该第二特征矩阵对应的第二高维特征矩阵,并可以对该第一高维特征矩阵和该第二高维特征矩阵进行融合处理,得到融合特征矩阵,然后通过该预训练的RNN模型中的输出层根据该融合特征矩阵进行处理,得到第一疫情趋势预测结果。此处的第一高维特征矩阵如可以为1*4096的特征矩阵,第二高维特征矩阵如可以为1*4096的特征矩阵。
在一个实施例中,电子设备对该第一高维特征矩阵和该第二高维特征矩阵进行融合处理,得到融合特征矩阵的过程可以为:电子设备确定该第二高维特征矩阵对应的attention权重,并利用该attention权重对该第二高维特征矩阵进行加权处理,得到加权特征矩阵;电子设备对第一高维特征矩阵以及加权特征矩阵进行拼接处理,得到融合特征矩阵。此处的融合特征矩阵的维度与第一特征矩阵的维度相同。在一个实施例中,电子设备确定该第二高维特征矩阵对应的attention权重的过程可以为:电子设备根据该第一高维特征矩阵以及该第二高维特征矩阵执行attention操作,得到该第二高维特征矩阵对应的attention权重。
假设第一疾病特征数据为新冠肺炎特征数据,第二疾病特征数据为流感特征数据。参见图2B,电子设备可以根据新冠肺炎特征数据以及流感特征数据获得第一特征矩阵以输入预训练的RNN模型中的第一隐藏层进行处理,得到第一高维特征矩阵,并可以根据第一气候特征数据获得第二特征矩阵以输入至预训练的RNN模型中的第二隐藏层进行处理,得到第二高维特征矩阵。电子设备可以用第二高维特征矩阵对应的attention权重对该第二高维特征矩阵进行加权处理,得到加权特征矩阵,并对第一高维特征矩阵以及加权特征矩阵进行拼接处理,得到融合特征矩阵,然后通过预训练的RNN模型中的输出层根据该融合特征矩阵进行处理,输出第一疫情趋势预测结果。
S210、通过终端设备展示所述第一疫情趋势预测结果。
其中,步骤S210可以参见图1实施例中的步骤S104,本申请实施例在此不做赘述。
可见,图2A的实施例中,电子设备可以根据疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵,并根据该疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵;电子设备通过该预训练的RNN模型中的第一隐藏层对该第一特征矩阵进行处理,得到该第一特征矩阵对应的第一高维特征矩阵,通过该预训练的RNN模型中的第二隐藏层对该第二特征矩阵进行处理,得到该第二特征矩阵对应的第二高维特征矩阵;电子设备对该第一高维特征矩阵和该第二高维特征矩阵进行融合处理,得到融合特征矩阵,并通过该预训练的RNN模型中的输出层根据该融合特征矩阵进行处理,得到第一疫情趋势预测结果,通过不同隐藏层对不同特征分别进行处理再融合以用于预测,能够提升模型预测的准确度。
本申请可以用于医疗科技领域,涉及区块链技术,如可以将第一疫情预测结果写入区块链中,或可以将第一疫情预测数据的压缩数据写入区块链中。
下面以电子设备为服务器来阐述下本申请实施例的疫情趋势预测系统。参见图3,图3所示的疫情趋势预测系统包括服务器10和终端设备20。其中:
服务器10可以通过执行步骤S101-步骤S103得到第一疫情趋势预测结果,并可以通过执行步骤S104使终端设备20展示该第一疫情趋势预测结果,该过程结合多维度特征进行疫情趋势预测,可参考性更高。
请参阅图4,为本申请实施例提供的一种疫情趋势预测装置的结构示意图。该疫情趋势预测装置可以应用于前述提及的电子设备。具体的,该疫情趋势预测装置可以包括:
获取模块401,用于获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据。
构建模块402,用于根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵。
处理模块403,用于调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后,
处理模块403,还用于通过终端设备展示所述第一疫情趋势预测结果。
在一种可选的实施方式中,图4所示的疫情趋势预测装置还可以包括输出模块(图未示)。在电子设备为服务器的情况下,处理模块403可以通过输出模块将该第一疫情趋势预测结果发送至终端设备,以便终端设备展示该第一疫情趋势预测结果。在电子设备为终端设备的情况下,处理模块403可以通过输出模块展示该第一疫情趋势预测结果。
在一种可选的实施方式中,处理模块403调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,具体为通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵;根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果。
在一种可选的实施方式中,构建模块402根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵,具体为对所述疫情序列数据包括的各特征数据进行特征提取,得到所述各特征数据的特征向量;根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵。
在一种可选的实施方式中,构建模块402根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵,具体为根据所述疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵;根据所述疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵;将所述第一特征矩阵和所述第二特征矩阵确定为所述疫情序列数据对应的目标特征矩阵。
在一种可选的实施方式中,处理模块403通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵,具体为通过所述预训练的RNN模型中的第一隐藏层对所述第一特征矩阵进行处理,得到所述第一特征矩阵对应的第一高维特征矩阵;通过所述预训练的RNN模型中的第二隐藏层对所述第二特征矩阵进行处理,得到所述第二特征矩阵对应的第二高维特征矩阵。
在一种可选的实施方式中,处理模块403根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果,具体为对所述第一高维特征矩阵和所述第二高维特征矩阵进行融合处理,得到融合特征矩阵;通过所述预训练的RNN模型中的输出层根据所述融合特征矩阵进行处理,得到第一疫情趋势预测结果。
在一种可选的实施方式中,处理模块403,还用于获取目标地区的历史疫情序列数据,所述历史疫情序列数据包括在第三预设日期范围内的各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;根据所述历史疫情序列数据构建所述历史疫情序列数据对应的特征矩阵;利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型。
在一种可选的实施方式中,所述时间序列模型为RNN模型,处理模块403利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型,具体为通过所述原始的RNN模型中的隐藏层根据所述历史疫情序列数据对应的特征矩阵进行处理,得到所述特征矩阵对应的高维特征矩阵;根据所述高维特征矩阵以及所述原始的RNN模型中的输出层,获得第二疫情趋势预测结果,所述第二疫情趋势预测结果包 括预测的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第四预设日期范围在所述第三预设日期范围之后;利用所述第二疫情趋势预测结果以及对应的疫情趋势真实结果构建损失函数,利用所述损失函数训练所述原始的RNN模型,得到预训练的RNN模型,所述疫情趋势真实结果包括真实的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
可见,图4所示的实施例中,疫情趋势预测装置可以获取目标地区的疫情序列数据,该疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;疫情趋势预测装置根据该疫情序列数据构建该疫情序列数据对应的目标特征矩阵,并调用预训练的时间序列模型以根据该目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果以通过终端设备展示该第一疫情趋势预测结果,相较于现有技术基于单一因素进行疫情趋势预测的过程,本申请可以基于结合多个疾病特征以及气象特征以用于疫情趋势预测,可参考性较高。
请参阅图5,为本申请实施例提供的一种电子设备的结构示意图。本实施例中所描述的电子设备可以包括:一个或多个处理器1000和存储器2000。处理器1000和存储器2000可以通过总线连接。
处理器1000可以是中央处理模块(Central Processing Unit,CPU),该处理器还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现成可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
存储器2000可以是高速RAM存储器,也可为非不稳定的存储器(non-volatile memory),例如磁盘存储器。其中,存储器2000用于存储计算机程序,所述计算机程序包括程序指令。处理器1000被配置用于调用所述程序指令,执行以下步骤:
获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
通过终端设备展示所述第一疫情趋势预测结果。
在一个实施例中,图5所示的电子设备还可以包括输出装置(图未示)。在电子设备为服务器的情况下,处理器1000可以通过输出装置将该第一疫情趋势预测结果发送至终端设备,以便终端设备展示该第一疫情趋势预测结果。在电子设备为终端设备的情况下,处理器1000可以通过输出装置展示该第一疫情趋势处理结果。所述的输出装置如可以为标准的有线/无线接口,或可以为显示屏、触摸显示屏等装置。
在一个实施例中,所述时间序列模型为循环神经网络RNN模型,在调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果时,处理器1000被配置用于调用所述程序指令,执行以下步骤:
通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵;
根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果。
在一个实施例中,在根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵时,处理器1000被配置用于调用所述程序指令,执行以下步骤:
对所述疫情序列数据包括的各特征数据进行特征提取,得到所述各特征数据的特征向量;
根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵。
在一个实施例中,在根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵时,处理器1000被配置用于调用所述程序指令,执行以下步骤:
根据所述疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵;
根据所述疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵;
将所述第一特征矩阵和所述第二特征矩阵确定为所述疫情序列数据对应的目标特征矩阵。
在一个实施例中,在通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵时,处理器1000被配置用于调用所述程序指令,执行以下步骤:
通过所述预训练的RNN模型中的第一隐藏层对所述第一特征矩阵进行处理,得到所述第一特征矩阵对应的第一高维特征矩阵;
通过所述预训练的RNN模型中的第二隐藏层对所述第二特征矩阵进行处理,得到所述第二特征矩阵对应的第二高维特征矩阵;
在根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果时,处理器1000被配置用于调用所述程序指令,执行以下步骤:
对所述第一高维特征矩阵和所述第二高维特征矩阵进行融合处理,得到融合特征矩阵;
通过所述预训练的RNN模型中的输出层根据所述融合特征矩阵进行处理,得到第一疫情趋势预测结果。
在一个实施例中,处理器1000被配置用于调用所述程序指令,还执行以下步骤:
获取目标地区的历史疫情序列数据,所述历史疫情序列数据包括在第三预设日期范围内的各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
根据所述历史疫情序列数据构建所述历史疫情序列数据对应的特征矩阵;
利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型。
在一个实施例中,所述时间序列模型为RNN模型,在利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型时,处理器1000被配置用于调用所述程序指令,执行以下步骤:
通过所述原始的RNN模型中的隐藏层根据所述历史疫情序列数据对应的特征矩阵进行处理,得到所述特征矩阵对应的高维特征矩阵;
根据所述高维特征矩阵以及所述原始的RNN模型中的输出层,获得第二疫情趋势预测结果,所述第二疫情趋势预测结果包括预测的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第四预设日期范围在所述第三预设日期范围之后;
利用所述第二疫情趋势预测结果以及对应的疫情趋势真实结果构建损失函数,利用所述损失函数训练所述原始的RNN模型,得到预训练的RNN模型,所述疫情趋势真实结果包括真实的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
具体实现中,本申请实施例中所描述的处理器1000可执行图1实施例、图2A实施例所描述的实现方式,也可执行本申请实施例所描述的实现方式,在此不再赘述。
本申请实施例还提供了一种计算机可读存储介质,该计算机可读存储介质存储有计算机程序,计算机程序被处理器执行时可实现上述实施例的方法,这里不再赘述。可选的,本申请涉及的存储介质如计算机可读存储介质可以是非易失性的,也可以是易失性的。
在本申请各个实施例中的各功能模块可以集成在一个处理模块中,也可以是各个模块单独物理存在,也可以是两个或两个以上模块集成在一个模块中。上述集成的模块既可以采样硬件的形式实现,也可以采样软件功能模块的形式实现。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,所述的程序可存储于一计算机可读取存储介质中,该程序在执行时,可包括如上述各方法的实施例的流程。其中,所述的计算机可读存储介质可为易失性的或非易失性的。例如,该计算机存储介质可以为磁碟、光盘、只读存储记忆体(Read-Only Memory,ROM)或随机存储记忆体(Random Access Memory,RAM)等。所述的计算机可读存储介质可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需的应用程序等;存储数据区可存储根据区块链节点的使用所创建的数据等。
其中,本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain),本质上是一个去中心化的数据库,是一串使用密码学方法相关联产生的数据块,每一个数据块中包含了一批次网络交易的信息,用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
以上所揭露的仅为本申请一种较佳实施例而已,当然不能以此来限定本申请之权利范围,本领域普通技术人员可以理解实现上述实施例的全部或部分流程,并依本申请权利要求所作的等同变化,仍属于本申请所涵盖的范围。

Claims (20)

  1. 一种电子设备,其中,包括处理器和存储器,所述处理器和所述存储器相互连接,其中,所述存储器用于存储计算机程序,所述计算机程序包括程序指令,所述处理器被配置用于调用所述程序指令,执行以下步骤:
    获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
    根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
    调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
    通过终端设备展示所述第一疫情趋势预测结果。
  2. 根据权利要求1所述的电子设备,其中,所述时间序列模型为循环神经网络RNN模型,在调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果时,所述处理器被配置用于调用所述程序指令,执行以下步骤:
    通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵;
    根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果。
  3. 根据权利要求2所述的电子设备,其中,在根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵时,所述处理器被配置用于调用所述程序指令,执行以下步骤:
    对所述疫情序列数据包括的各特征数据进行特征提取,得到所述各特征数据的特征向量;
    根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵。
  4. 根据权利要求3所述的电子设备,其中,在根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵时,所述处理器被配置用于调用所述程序指令,执行以下步骤:
    根据所述疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵;
    根据所述疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵;
    将所述第一特征矩阵和所述第二特征矩阵确定为所述疫情序列数据对应的目标特征矩阵。
  5. 根据权利要求4所述的电子设备,其中,
    在通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵时,所述处理器被配置用于调用所述程序指令,执行以下步骤:
    通过所述预训练的RNN模型中的第一隐藏层对所述第一特征矩阵进行处理,得到所述第一特征矩阵对应的第一高维特征矩阵;
    通过所述预训练的RNN模型中的第二隐藏层对所述第二特征矩阵进行处理,得到所述第二特征矩阵对应的第二高维特征矩阵;
    在根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果时,所述处理器被配置用于调用所述程序指令,执行以下步骤:
    对所述第一高维特征矩阵和所述第二高维特征矩阵进行融合处理,得到融合特征矩阵;
    通过所述预训练的RNN模型中的输出层根据所述融合特征矩阵进行处理,得到第一疫情趋势预测结果。
  6. 根据权利要求1-5任一项所述的电子设备,其中,所述处理器被配置用于调用所述程序指令,还执行以下步骤:
    获取目标地区的历史疫情序列数据,所述历史疫情序列数据包括在第三预设日期范围内的各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
    根据所述历史疫情序列数据构建所述历史疫情序列数据对应的特征矩阵;
    利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型。
  7. 根据权利要求6所述的电子设备,其中,所述时间序列模型为RNN模型,在利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型时,所述处理器被配置用于调用所述程序指令,执行以下步骤:
    通过所述原始的RNN模型中的隐藏层根据所述历史疫情序列数据对应的特征矩阵进行处理,得到所述特征矩阵对应的高维特征矩阵;
    根据所述高维特征矩阵以及所述原始的RNN模型中的输出层,获得第二疫情趋势预测结果,所述第二疫情趋势预测结果包括预测的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第四预设日期范围在所述第三预设日期范围之后;
    利用所述第二疫情趋势预测结果以及对应的疫情趋势真实结果构建损失函数,利用所述损失函数训练所述原始的RNN模型,得到预训练的RNN模型,所述疫情趋势真实结果包括真实的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
  8. 一种疫情趋势预测方法,其中,包括:
    获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
    根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
    调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
    通过终端设备展示所述第一疫情趋势预测结果。
  9. 根据权利要求8所述的方法,其中,所述时间序列模型为循环神经网络RNN模型,所述调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,包括:
    通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵;
    根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果。
  10. 根据权利要求9所述的方法,其中,所述根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵,包括:
    对所述疫情序列数据包括的各特征数据进行特征提取,得到所述各特征数据的特征向量;
    根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵。
  11. 根据权利要求10所述的方法,其中,所述根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵,包括:
    根据所述疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵;
    根据所述疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵;
    将所述第一特征矩阵和所述第二特征矩阵确定为所述疫情序列数据对应的目标特征矩 阵。
  12. 根据权利要求8-11任一项所述的方法,其中,所述方法还包括:
    获取目标地区的历史疫情序列数据,所述历史疫情序列数据包括在第三预设日期范围内的各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
    根据所述历史疫情序列数据构建所述历史疫情序列数据对应的特征矩阵;
    利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型。
  13. 根据权利要求12所述的方法,其中,所述时间序列模型为RNN模型,所述利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型,包括:
    通过所述原始的RNN模型中的隐藏层根据所述历史疫情序列数据对应的特征矩阵进行处理,得到所述特征矩阵对应的高维特征矩阵;
    根据所述高维特征矩阵以及所述原始的RNN模型中的输出层,获得第二疫情趋势预测结果,所述第二疫情趋势预测结果包括预测的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第四预设日期范围在所述第三预设日期范围之后;
    利用所述第二疫情趋势预测结果以及对应的疫情趋势真实结果构建损失函数,利用所述损失函数训练所述原始的RNN模型,得到预训练的RNN模型,所述疫情趋势真实结果包括真实的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
  14. 一种疫情趋势预测装置,其中,包括:
    获取模块,用于获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
    构建模块,用于根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
    处理模块,用于调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
    所述处理模块,还用于通过终端设备展示所述第一疫情趋势预测结果。
  15. 一种计算机可读存储介质,其中,所述计算机可读存储介质存储有计算机程序,所述计算机程序被处理器执行以实现以下方法:
    获取目标地区的疫情序列数据,所述疫情序列数据包括第一预设时间范围内各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
    根据所述疫情序列数据构建所述疫情序列数据对应的目标特征矩阵;
    调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果,所述第一疫情趋势预测结果包括预测的第二预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第二预设日期范围在所述第一预设日期范围之后;
    通过终端设备展示所述第一疫情趋势预测结果。
  16. 根据权利要求15所述的计算机可读存储介质,其中,所述时间序列模型为循环神经网络RNN模型,所述调用预训练的时间序列模型以根据所述目标特征矩阵进行疫情趋势预测,得到第一疫情趋势预测结果时,具体实现:
    通过所述预训练的RNN模型中的隐藏层根据所述目标特征矩阵进行处理,得到所述目标特征矩阵对应的目标高维特征矩阵;
    根据所述目标高维特征矩阵以及所述预训练的RNN模型中的输出层,获得第一疫情趋势预测结果。
  17. 根据权利要求16所述的计算机可读存储介质,其中,所述根据所述疫情序列数据 构建所述疫情序列数据对应的目标特征矩阵时,具体实现:
    对所述疫情序列数据包括的各特征数据进行特征提取,得到所述各特征数据的特征向量;
    根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵。
  18. 根据权利要求17所述的计算机可读存储介质,其中,所述根据所述各特征数据的特征向量,拼接得到所述疫情序列数据对应的目标特征矩阵时,具体实现:
    根据所述疫情序列数据包括的各第一疾病特征数据的特征向量以及各第二疾病特征数据的特征向量,拼接得到第一特征矩阵;
    根据所述疫情序列数据包括的各气象特征数据的特征向量,拼接得到第二特征矩阵;
    将所述第一特征矩阵和所述第二特征矩阵确定为所述疫情序列数据对应的目标特征矩阵。
  19. 根据权利要求12-18任一项所述的计算机可读存储介质,其中,所述计算机程序被处理器执行时还用于实现:
    获取目标地区的历史疫情序列数据,所述历史疫情序列数据包括在第三预设日期范围内的各日期的第一疾病特征数据、第二疾病特征数据以及气象特征数据;
    根据所述历史疫情序列数据构建所述历史疫情序列数据对应的特征矩阵;
    利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型。
  20. 根据权利要求19所述的计算机可读存储介质,其中,所述时间序列模型为RNN模型,所述利用所述历史疫情序列数据对应的特征矩阵对原始的时间序列模型进行训练,得到预训练的时间序列模型时,具体实现:
    通过所述原始的RNN模型中的隐藏层根据所述历史疫情序列数据对应的特征矩阵进行处理,得到所述特征矩阵对应的高维特征矩阵;
    根据所述高维特征矩阵以及所述原始的RNN模型中的输出层,获得第二疫情趋势预测结果,所述第二疫情趋势预测结果包括预测的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数;所述第四预设日期范围在所述第三预设日期范围之后;
    利用所述第二疫情趋势预测结果以及对应的疫情趋势真实结果构建损失函数,利用所述损失函数训练所述原始的RNN模型,得到预训练的RNN模型,所述疫情趋势真实结果包括真实的第四预设日期范围内各日期的新增病例的数量和/或新增死亡的人数。
PCT/CN2020/124608 2020-09-28 2020-10-29 疫情趋势预测方法、装置、电子设备及存储介质 WO2021159739A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011043913.3 2020-09-28
CN202011043913.3A CN111883262B (zh) 2020-09-28 2020-09-28 疫情趋势预测方法、装置、电子设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021159739A1 true WO2021159739A1 (zh) 2021-08-19

Family

ID=73199240

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/124608 WO2021159739A1 (zh) 2020-09-28 2020-10-29 疫情趋势预测方法、装置、电子设备及存储介质

Country Status (2)

Country Link
CN (1) CN111883262B (zh)
WO (1) WO2021159739A1 (zh)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112561121B (zh) * 2020-11-12 2021-07-20 北京大学 基于手机信令数据的复工趋势预测方法及系统
CN112820414B (zh) * 2021-01-29 2021-11-09 南威软件股份有限公司 基于改进的三次指数平滑模型及lstm模型在新冠疫情的预警方法
CN112988851B (zh) * 2021-04-27 2021-07-27 腾讯科技(深圳)有限公司 反事实预测模型数据处理方法、装置、设备及存储介质
CN113707337B (zh) * 2021-08-30 2024-05-10 平安科技(深圳)有限公司 基于多源数据的疾病预警方法、装置、设备及存储介质
CN114049968A (zh) * 2021-10-21 2022-02-15 北京北明数科信息技术有限公司 传染病发展趋势预测方法、系统、装置及存储介质
CN115840895B (zh) * 2021-12-31 2024-05-03 江苏常胜电器(淮安)有限公司 一种电子器件温度保护系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108648829A (zh) * 2018-04-11 2018-10-12 平安科技(深圳)有限公司 疾病预测方法及装置、计算机装置及可读存储介质
WO2018195401A1 (en) * 2017-04-21 2018-10-25 The Board Of Trustees Of The Leland Stanford Junior University A collector device of environmental exposure for biotic and abiotic agents
CN109448781A (zh) * 2018-11-06 2019-03-08 云南大学 一种流感病毒抗原变化的预测方法
CN109656918A (zh) * 2019-01-04 2019-04-19 平安科技(深圳)有限公司 流行病发病指数的预测方法、装置、设备及可读存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2014183100A2 (en) * 2013-05-10 2014-11-13 Wello, Inc. System and process for notification of contagion risk conditions and mitigation thereof
CN106778014B (zh) * 2016-12-29 2020-06-16 浙江大学 一种基于循环神经网络的患病风险预测建模方法
CN107194508B (zh) * 2017-05-18 2020-08-18 成都数联易康科技有限公司 一种基于时间序列动态预测中短期区域儿科门诊人次的方法
CN109859854A (zh) * 2018-12-17 2019-06-07 中国科学院深圳先进技术研究院 传染病预测方法、装置、电子设备及计算机可读介质
KR20200029425A (ko) * 2020-03-11 2020-03-18 네이버 주식회사 딥 어텐션 네트워크를 이용하여 환자 의료 기록으로부터 질병 예후를 예측하는 방법 및 시스템
CN111524611B (zh) * 2020-04-24 2023-03-03 腾讯科技(深圳)有限公司 构建传染病趋势预测模型的方法、预测方法、装置及设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018195401A1 (en) * 2017-04-21 2018-10-25 The Board Of Trustees Of The Leland Stanford Junior University A collector device of environmental exposure for biotic and abiotic agents
CN108648829A (zh) * 2018-04-11 2018-10-12 平安科技(深圳)有限公司 疾病预测方法及装置、计算机装置及可读存储介质
CN109448781A (zh) * 2018-11-06 2019-03-08 云南大学 一种流感病毒抗原变化的预测方法
CN109656918A (zh) * 2019-01-04 2019-04-19 平安科技(深圳)有限公司 流行病发病指数的预测方法、装置、设备及可读存储介质

Also Published As

Publication number Publication date
CN111883262A (zh) 2020-11-03
CN111883262B (zh) 2021-01-15

Similar Documents

Publication Publication Date Title
WO2021159739A1 (zh) 疫情趋势预测方法、装置、电子设备及存储介质
WO2022022274A1 (zh) 一种模型训练方法及装置
WO2021047593A1 (zh) 推荐模型的训练方法、预测选择概率的方法及装置
WO2019196286A1 (zh) 疾病预测方法及装置、计算机装置及可读存储介质
WO2021120677A1 (zh) 一种仓储模型训练方法、装置、计算机设备及存储介质
WO2019196280A1 (zh) 疾病预测方法及装置、计算机装置及可读存储介质
CN112836064B (zh) 知识图谱补全方法、装置、存储介质及电子设备
EP3913545A2 (en) Method and apparatus for updating parameter of multi-task model, and electronic device
JP7206419B2 (ja) 人工知能推奨モデルの特徴処理方法、装置、電子機器、及びコンピュータプログラム
WO2022141968A1 (zh) 对象推荐方法及装置、计算机设备和介质
WO2022001724A1 (zh) 一种数据处理方法及装置
CN113963110B (zh) 纹理图生成方法、装置、电子设备及存储介质
CN112861662B (zh) 基于人脸和交互文本的目标对象行为预测方法及相关设备
CN111933300A (zh) 疫情防控效果预测方法、装置、服务器及存储介质
WO2021174881A1 (zh) 多维度信息的组合预测方法、装置、计算机设备及介质
WO2021238513A1 (zh) 区块链节点扩容方法、装置、计算机设备及存储介质
WO2021218037A1 (zh) 目标检测方法、装置、计算机设备和存储介质
WO2024067373A1 (zh) 一种数据处理方法及相关装置
WO2021217937A1 (zh) 姿态识别模型的训练方法及设备、姿态识别方法及其设备
WO2021036397A1 (zh) 目标神经网络模型的生成方法和装置
CN114493376A (zh) 一种基于工单数据的任务调度管理方法及系统
WO2022179603A1 (zh) 一种增强现实方法及其相关设备
Xue et al. Modeling the impact of social distancing on the COVID-19 pandemic in a low transmission setting
CN114912972A (zh) 基于区块链的网络推广营销平台及其方法
CN111933303B (zh) 事件预测方法、装置、电子设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20918179

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20918179

Country of ref document: EP

Kind code of ref document: A1