Training method and device of epidemic situation prediction model, computer equipment and storage medium
By performing Hilbert yellow transformation on epidemic training data and using long and short-term memory networks for model training, the problem of low accuracy in existing epidemic prediction models when processing nonlinear and non-stationary time series data is solved, and higher prediction accuracy is achieved.
Patent Information
- Application Number
- CN202510183681.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-19
- Publication Date
- 2025-07-01
AI Technical Summary
The existing epidemic prediction model has the problem of low accuracy when processing time series data, and it is difficult to effectively capture nonlinear and non-stationary dynamic changes.
The epidemic training data is preprocessed by Hilbert yellow transformation, the data is decomposed into intrinsic modal functions and their instantaneous frequency, and the model is trained using a long and short-term memory network, and the loss value is calculated based on the mean square error.
Through the combination of Hilbert yellow transformation and long and short-term memory networks, the processing effect and prediction accuracy of the epidemic prediction model on time series data is significantly improved.
Smart Images
Figure CN120236783A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of epidemic prediction. Specifically, this application relates to a method, device, computer device, and storage medium for training an epidemic prediction model. Background Art
[0002] In the event of an epidemic spread, reasonable epidemic prediction has important reference significance for epidemic prevention and control. For example, predicting when the epidemic will enter the high outbreak period, when it will reach the peak, and when it can enter the stable and controllable stage. The epidemic prediction model is a key tool in dealing with the epidemic spread event, helping government departments and health organizations understand the development trend of the epidemic, so as to better allocate resources and formulate prevention and control strategies.
[0003] In current epidemic predictions, epidemic data is often time-series data, which often exhibits non-linearity and non-stationarity. Traditional epidemic prediction models have limitations in capturing these complex dynamic changes, ultimately resulting in low accuracy of epidemic prediction by the epidemic prediction model. Summary of the Invention
[0004] The main purpose of this application is to provide a method, device, computer device, and storage medium for training an epidemic prediction model, so as to improve the processing effect of the epidemic prediction model on time-series data, thereby improving the accuracy of the model's epidemic prediction.
[0005] To achieve the above invention purpose, this application provides a method for training an epidemic prediction model, including:
[0006] Obtain epidemic training data;
[0007] Preprocess the epidemic training data to obtain standard epidemic training data;
[0008] Perform Hilbert-Huang transform on the standard epidemic training data to obtain transformed data;
[0009] Use the transformed data to train a preset long short-term memory network, and calculate the loss value of the long short-term memory network during the training process based on the mean square error calculation formula;
[0010] When it is determined that the loss value meets the requirements, generate an epidemic prediction model.
[0011] Preferably, the performing Hilbert-Huang transform on the standard epidemic training data to obtain transformed data includes:
[0012] Screen the original signal corresponding to the standard epidemic training data, find all local extreme points of the original signal, and obtain local maximum values and local minimum values;
[0013] Connect the local maxima and local minima respectively according to the cubic spline interpolation method to obtain the upper envelope and the lower envelope;
[0014] Calculate the mean of the upper envelope and the lower envelope, and subtract the mean from the original signal corresponding to the standard epidemic training data to obtain the detail component;
[0015] Determine whether the detail component satisfies the preset constraint conditions;
[0016] If it is satisfied, perform Hilbert transform on the detail component to obtain the instantaneous frequency and amplitude of the detail component, and use the detail component and the corresponding instantaneous frequency as the transformed data;
[0017] If it is not satisfied, use the detail component as the new original signal, and return to execute the step of finding all local extreme points of the original signal to obtain local maxima and local minima until the detail component satisfies the preset constraint conditions.
[0018] Preferably, the preset constraint conditions include:
[0019]
[0020] wherein, the x(t) is the original signal corresponding to the standard epidemic training data, the IMF i (t) is the i-th intrinsic mode function, the r(t) is the detail component, and the n is the number of the detail components;
[0021] Or at any point of the original signal, the average value of the upper envelope formed by local maxima and the lower envelope formed by local minima is zero, and the number of local maxima and local minima must be equal or differ by at most one.
[0022] Further, after performing Hilbert transform on the detail component to obtain the instantaneous frequency and amplitude of the detail component, it further includes:
[0023] Generate a Hilbert spectrum according to the instantaneous frequency and amplitude of the detail component, and the Hilbert spectrum is used to characterize the time-frequency distribution of the standard epidemic training data.
[0024] Preferably, the using the transformed data to train a preset long short-term memory network includes:
[0025] Extract the detail component and the corresponding instantaneous frequency of each piece of the transformed data;
[0026] Input each detail component and the corresponding instantaneous frequency into the preset long short-term memory network, and use each detail component to train the long short-term memory network.
[0027] Further, after generating the epidemic prediction model when it is determined that the loss value meets the requirements, the following steps are further included:
[0028] Obtain epidemic data;
[0029] Preprocess the epidemic data to obtain standard epidemic data;
[0030] Perform Hilbert-Huang transform on the standard epidemic data and then input it into the epidemic prediction model to generate an epidemic prediction result.
[0031] Preferably, the preprocessing of the epidemic training data to obtain standard epidemic training data includes:
[0032] Remove outliers, fill in missing values, and perform standardization processing on the epidemic training data to obtain standard epidemic training data.
[0033] This application also provides a training model for an epidemic prediction model, including:
[0034] An acquisition module for acquiring epidemic training data;
[0035] A preprocessing module for preprocessing the epidemic training data to obtain standard epidemic training data;
[0036] A transformation module for performing Hilbert-Huang transform on the standard epidemic training data to obtain transformed data;
[0037] A training module for training a preset long short-term memory network using the transformed data and calculating the loss value of the long short-term memory network during the training process based on the mean square error calculation formula;
[0038] A generation module for generating an epidemic prediction model when it is determined that the loss value meets the requirements.
[0039] This application also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the training method for the epidemic prediction model described in any one of the above are implemented.
[0040] This application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the training method for the epidemic prediction model described in any one of the above are implemented.
[0041] A training method, device, computer device and storage medium for an epidemic prediction model provided by this application can remove noise and outliers by preprocessing epidemic training data, improve data quality, and thus enhance the prediction accuracy of the model. The Hilbert-Huang transform is an adaptive non-linear and non-stationary signal analysis method that can process non-linear and non-stationary time series data. By combining empirical mode decomposition and Hilbert transform, it decomposes complex signals into a series of intrinsic mode functions and their instantaneous frequencies, reducing the complexity of the data. Therefore, through this transform, the inherent period and trend in epidemic data can be extracted, providing more effective features for the long short-term memory network and further enhancing the prediction ability and efficiency of the model. At the same time, by using the long short-term memory network for epidemic prediction, the long-term and short-term dependencies in time series data can be captured, which helps to process data such as epidemics that have obvious time correlations. In addition, based on the calculation of the loss function of mean square error, the difference between the model prediction value and the actual value can be effectively measured, guiding the parameter optimization in the model training process and accelerating the model convergence speed. Therefore, combining the Hilbert-Huang transform with the long short-term memory network in this application can significantly improve the accuracy of epidemic prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a schematic flowchart of the training method of the epidemic prediction model according to an embodiment of this application;
[0043] Figure 2 It is a schematic block diagram of the structure of the training device of the epidemic prediction model according to an embodiment of this application;
[0044] Figure 3 It is a schematic block diagram of the structure of the computer device according to an embodiment of this application.
[0045] The realization, functional features and advantages of the purpose of this application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0046] In order to make the purpose, technical solution and advantages of this application clearer, the following further details this application with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.
[0047] A training method for an epidemic prediction model proposed in this application, the execution subject is a computer device, and the computer device can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (such as tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device can also be a mobile or stationary server. The components of the computing device include but are not limited to a memory and a processor. The processor is connected to the memory through a bus, and the database is used to store data. The computing device also includes an access device, and the access device enables the computing device to communicate via one or more networks. Examples of these networks include the Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device can include one or more of any type of wired or wireless network interface (such as a Network Interface Card (NIC)), such as an IEEE802.11 Wireless Local Area Network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.
[0048] Reference Figure 1 , in one embodiment, this application provides a training method for an epidemic prediction model, and the method includes:
[0049] S11. Obtain epidemic training data;
[0050] S12. Preprocess the epidemic training data to obtain standard epidemic training data;
[0051] S13. Perform Hilbert-Huang transform on the standard epidemic training data to obtain transformed data;
[0052] S14. Use the transformed data to train a preset long short-term memory network, and calculate the loss value of the long short-term memory network during the training process based on the mean square error calculation formula;
[0053] S15. When it is determined that the loss value meets the requirements, generate an epidemic prediction model.
[0054] As described in step S11 above, the computer device can first obtain historical epidemic data and use it as epidemic training data, which may include information such as the number of daily confirmed cases, cumulative confirmed cases, the number of cured and discharged patients, the case fatality rate, the infection time, and the infection location. For example, the historical epidemic data can be collected from the Internet through public health department data, datasets released by research institutions, or web crawler technology to obtain the epidemic training data. High-quality epidemic training data is the basis for establishing an effective epidemic prediction model.
[0055] As described in step S12 above, this step can preprocess the epidemic training data to obtain standard epidemic training data. The preprocessing methods include data cleaning (removing useless information), missing value handling, outlier handling, data transformation (such as normalization or standardization), etc. Preprocessing can reduce the influence of noise and outliers, improve the generalization ability and prediction accuracy of the model, and thus provide more accurate input for model training.
[0056] As described in step S13 above, this step can perform Hilbert-Huang transform on the standard epidemic training data to convert non-linear and non-stationary time series data into linear and stationary data, thus facilitating data analysis. Among them, the Hilbert-Huang transform includes empirical mode decomposition and Hilbert transform to decompose the data into a series of intrinsic mode functions, so as to reveal the internal period and trend of the data and provide richer features for subsequent model training.
[0057] Specifically, the Hilbert-Huang Transform (HHT) is a method for analyzing non-linear and non-stationary time series. The HHT method combines Empirical Mode Decomposition (EMD) and Hilbert Spectral Analysis (HSA), and is particularly suitable for processing complex, non-linear and non-stationary signals. EMD is the first step of HHT, and its purpose is to decompose a complex data set into a series of simple waveforms called Intrinsic Mode Functions (IMFs). The decomposition process of EMD is recursive. Starting from the original data, through local extreme value identification, envelope creation and mean removal, IMFs are gradually extracted. Each IMF extracted at each step is the fastest-changing part of the data. After extraction, this part is removed from the data, and the same process is continued for the remaining data until the remaining data can no longer be decomposed.
[0058] Once the IMFs are obtained, the next step is to perform the Hilbert transform on each IMF to obtain the instantaneous frequency and instantaneous amplitude of the signal. The Hilbert transform is a mathematical operation that can convert a real-valued signal into a complex signal, thereby allowing the calculation of the instantaneous frequency and phase of the signal. For each IMF, the Hilbert transform provides a method to determine the time-varying frequency and amplitude characteristics of the signal. The HHT result reflects the time-frequency characteristics of the signal, that is, the law of change of the frequency-domain characteristics of the signal over time. HHT can reflect local characteristics, which mainly benefits from the role of EMD. EMD can perform time-frequency localization analysis adaptively and effectively extract the characteristic information of the original signal.
[0059] As described in step S14 above, this step uses the transformed data to train a preset long short-term memory network and uses the mean square error calculation formula as the loss function to optimize the model parameters. Among them, the long short-term memory network is a special recurrent neural network that can capture long-term dependencies in time series data, so it can effectively process time series data. The loss function of the mean square error calculation formula can quantify the accuracy of model prediction and guide model training.
[0060] Among them, the long short-term memory network (Long Short-Term Memory, LSTM) is a special recurrent neural network (RNN) that can learn long-term dependence information, so it can effectively capture long-term and short-term dependencies in time series data.
[0061] LSTM solves the problems of traditional RNN by introducing "gate" structures and "cell states". The gate structures can control the inflow and outflow of information, and the cell state can store the state for a long time. This enables LSTM to better capture dependencies in long sequences. The main components of LSTM include:
[0062] Forget Gate: Determines which information to forget from the cell state.
[0063] Input Gate: Determines how much of the new input information should be retained.
[0064] Output Gate: Controls how much of the information in the current cell state is available for output.
[0065] The cell state of LSTM is updated at each time step, and this state can transmit information across time steps. The calculation at each time step includes:
[0066] The forget gate uses the sigmoid activation function to determine whether to retain or forget the information in the cell state.
[0067] The input gate determines how much of the new input information should be added to the cell state.
[0068] The cell state is updated by combining the information from the forget gate and the input gate.
[0069] The output gate determines how much of the information in the cell state is available for output.
[0070] In the LSTM model, the Mean Squared Error (MSE) calculation formula is a loss function used to measure the degree of difference between the model's predicted values and the true values. For sequence data prediction tasks, it is desired that the model can accurately predict future values. Therefore, a reliable loss function is needed to evaluate the model's performance. The MSE calculation formula can obtain the loss value by squaring the difference between the model's predicted values and the true values and then taking the average.
[0071] As described in step S15 above, monitor the loss value during the training process. When the loss value is stable or reaches the expected target, stop the training, consider the epidemic prediction model to be trained, save the parameters of the epidemic prediction model, and this epidemic prediction model can be used for actual prediction, thus ensuring that the model has good performance on the training data and providing a reliable model for actual prediction.
[0072] For example, assume that it is necessary to predict the number of newly confirmed COVID-19 cases per day in a certain region. First, obtain the number of newly confirmed cases per day in the region for the past year from public data sources. Clean the obtained data, remove holidays or days with missing data, and normalize the data so that its range is between 0 and 1. Apply HHT to the preprocessed data to decompose it into several IMFs, and these IMFs can represent different frequency components of the data. Use the decomposed IMFs as input features to train an LSTM model. During the training process, use MSE as the loss function to calculate the difference between the model's predicted values and the actual values, and continuously optimize the model parameters. When the training loss value drops to an acceptable range, stop the training, save the model parameters, and generate the final epidemic prediction model. Use this model to predict the number of newly confirmed cases per day for a period of time in the future to provide support for public health decision-making.
[0073] A training method for an epidemic prediction model provided by this application can remove noise and outliers by preprocessing epidemic training data, improve data quality, and thus enhance the prediction accuracy of the model. The Hilbert-Huang transform is an adaptive non-linear and non-stationary signal analysis method that can process non-linear and non-stationary time series data. By combining empirical mode decomposition and Hilbert transform, it decomposes complex signals into a series of intrinsic mode functions and their instantaneous frequencies, reducing the complexity of the data. Therefore, through this transform, the inherent period and trend in epidemic data can be extracted, providing more effective features for the long short-term memory network and further enhancing the prediction ability and efficiency of the model. At the same time, epidemic prediction through the long short-term memory network can capture long-term and short-term dependencies in time series data, which helps to process data such as epidemics with obvious time correlations. In addition, based on the calculation of the loss function of mean square error, the difference between the model prediction value and the actual value can be effectively measured, guiding the parameter optimization in the model training process and accelerating the model convergence speed. Therefore, combining the Hilbert-Huang transform with the long short-term memory network in this application can significantly improve the accuracy of epidemic prediction.
[0074] In one embodiment, performing the Hilbert-Huang transform on the standard epidemic training data to obtain transformed data includes:
[0075] Screen the original signal corresponding to the standard epidemic training data to find all local extreme points of the original signal, obtaining local maximum values and local minimum values;
[0076] Connect the local maximum values and local minimum values respectively according to the cubic spline interpolation method to obtain an upper envelope and a lower envelope;
[0077] Calculate the mean of the upper envelope and the lower envelope, and subtract the mean from the original signal corresponding to the standard epidemic training data to obtain a detail component;
[0078] Judge whether the detail component meets a preset constraint condition;
[0079] If it meets the condition, perform the Hilbert transform on the detail component to obtain the instantaneous frequency and amplitude of the detail component, and use the detail component and the corresponding instantaneous frequency as the transformed data;
[0080] If it does not meet the condition, use the detail component as a new original signal, and return to execute the step of finding all local extreme points of the original signal to obtain local maximum values and local minimum values until the detail component meets the preset constraint condition.
[0081] This embodiment can identify all local maxima and minima points in the original signal, which are the basis for constructing the signal envelope. For example, the local maxima and minima points of the original signal can be identified by the difference method.
[0082] At the same time, by connecting the local extreme points to construct the upper and lower envelopes of the signal, it helps to extract the detailed components from the original signal. For example, the cubic spline interpolation method (or other interpolation methods) is used to connect the local maxima points and local minima points to form the upper envelope and the lower envelope. Among them, the cubic spline interpolation method is an interpolation method in numerical analysis. It approximates the original data points by constructing a series of cubic polynomials, and is particularly suitable for the case where a smooth curve needs to pass through all data points.
[0083] In addition, by calculating the mean of the upper and lower envelopes and subtracting this mean from the original signal, the detailed components of the signal are obtained. Determine whether the detailed components meet the preset constraint conditions to ensure that the extracted detailed components are valid and meet the conditions for being an Intrinsic Mode Function (IMF). Perform the Hilbert transform on the detailed components that meet the preset constraint conditions to obtain the instantaneous frequency and amplitude information. If the detailed components do not meet the preset constraint conditions, they need to be reprocessed to extract valid IMFs, that is, the detailed components that do not meet the conditions are used as the new original signal, and return to the step of finding local extreme points, repeating the above process until the conditions are met.
[0084] For example, assume there is a non-stationary epidemic time series data, and it is desired to use the Hilbert-Huang transform to analyze its inherent periodicity and trend. Then collect the number of newly confirmed cases per day over a period of time. By analyzing the data, find all local maxima and minima points, use the cubic spline interpolation method to connect these extreme points to construct the upper and lower envelopes. Then calculate the mean of the upper and lower envelopes and subtract this mean from the original signal to obtain the first detailed component, and check whether this detailed component meets the IMF conditions. If not, use this detailed component as the new original signal and repeat the above steps. Once the detailed components that meet the conditions are obtained, perform the Hilbert transform on them to obtain the instantaneous frequency and amplitude information. By analyzing the instantaneous frequency and amplitude of all detailed components, the inherent periodicity and trend of the epidemic development can be accurately understood, providing a scientific basis for epidemic prevention and control.
[0085] In one embodiment, the preset constraint conditions include:
[0086]
[0087] Among them, the x(t) is the original signal corresponding to the standard epidemic training data, the IMF i (t) is the i-th intrinsic mode function, the r(t) is the detailed component, and the n is the number of the detailed components;
[0088] Or at any point of the original signal, the average value of the upper envelope formed by local maximum points and the lower envelope formed by local minimum points is zero, and the number of local maximum points and local minimum points must be equal or differ by at most one.
[0089] In one embodiment, after performing Hilbert transform on the detail component to obtain the instantaneous frequency and amplitude of the detail component, it further includes:
[0090] Generating a Hilbert spectrum according to the instantaneous frequency and amplitude of the detail component, and the Hilbert spectrum is used to characterize the time-frequency distribution of the standard epidemic training data.
[0091] In this embodiment, the Hilbert spectrum is a time-frequency analysis tool used to characterize the time-frequency distribution of a signal. Through Hilbert transform, instantaneous frequency and amplitude information can be extracted from the signal, and then the Hilbert spectrum is generated.
[0092] Specifically, first perform Hilbert transform on the signal to obtain an analytic signal. Then, extract the instantaneous frequency and amplitude from the analytic signal. By showing these time-frequency characteristics, the Hilbert spectrum helps to analyze the local characteristics and variation laws of the signal.
[0093] The Hilbert spectrum is one of the results of the Hilbert-Huang transform, which intuitively reflects the relationship between time, instantaneous frequency and amplitude of the signal. It can be used to analyze the variation laws of each component in the mixed-component signal over time and identify local characteristics. Therefore, the Hilbert spectrum can be used to analyze signals containing mixed components, especially non-linear and non-stationary signals, and can reveal the time-varying characteristics of each component in the signal, which is very useful for identifying and analyzing local characteristics in the signal.
[0094] For example, assume there is a standard epidemic training data set containing the number of newly confirmed cases per day over a period of time. If the Hilbert spectrum is used to analyze the time-frequency distribution of these data, then by observing the Hilbert spectrum, periodic changes and trends in the epidemic data can be identified. For example, certain seasonal patterns can be found, or the acceleration and deceleration stages of the epidemic development. According to the analysis results of the Hilbert spectrum, decision-making support can be provided for epidemic prevention and control, such as predicting the development trend of the epidemic and formulating corresponding prevention and control measures.
[0095] In this embodiment, by generating the Hilbert spectrum, not only can the overall trend of the epidemic data be understood, but also the local characteristics and time-varying characteristics of the data can be deeply analyzed, providing more comprehensive information support for epidemic prevention and control.
[0096] In one embodiment, the using the transformed data to train a preset long short-term memory network includes:
[0097] Extract the detailed components and corresponding instantaneous frequencies of each of the transformation data;
[0098] Input each of the detailed components and the corresponding instantaneous frequencies into a preset long short-term memory network, and train the long short-term memory network using each of the detailed components.
[0099] In this embodiment, the instantaneous frequency of each IMF is used as a time series label, and the corresponding detailed component is used as an input feature, which is input into a preset long short-term memory network for training, so that the long short-term memory network can learn the dynamic change law of the signal.
[0100] Through training, the long short-term memory network can capture the long-term dependence relationship of the signal and predict future trends. During the training process, the parameters of the long short-term memory network are adjusted to minimize the difference between the predicted value and the actual value. For example, the mean square error (MSE) is used as the loss function, and finally a trained epidemic prediction model is obtained. The trained epidemic prediction model can quickly respond to new epidemic data, update the prediction results in real time, and provide support for decision-making.
[0101] In one embodiment, after determining that the loss value meets the requirements and generating the epidemic prediction model, it further includes:
[0102] Obtain epidemic data;
[0103] Preprocess the epidemic data to obtain standard epidemic data;
[0104] Perform Hilbert-Huang transform on the standard epidemic data and then input it into the epidemic prediction model to generate an epidemic prediction result.
[0105] In this embodiment, relevant data on the epidemic are collected, which usually include the number of newly confirmed cases, recoveries, deaths, etc. per day. This data can be obtained from public health departments, scientific research institutions or international organizations such as the World Health Organization (WHO), etc.
[0106] By preprocessing the epidemic data, the consistency and accuracy of the data are ensured, providing a high-quality data basis for subsequent analysis. The preprocessing methods include data cleaning (processing outliers and missing values), normalization or standardization (comparing data on the same scale), etc.
[0107] In addition, the Hilbert-Huang transform is used to analyze the time-frequency characteristics of the epidemic data, and this information is input into the epidemic prediction model to generate an epidemic prediction result. Thus, by combining HHT and LSTM, the model can capture the time-frequency characteristics and long-term dependence relationship of the epidemic data, improving the accuracy of the prediction.
[0108] In one embodiment, preprocessing the epidemic training data to obtain standard epidemic training data includes:
[0109] Removing outliers, filling in missing values, and normalizing the epidemic training data to obtain standard epidemic training data.
[0110] In this embodiment, outliers may have an adverse impact on the training and prediction results of the model. Removing outliers can improve the data quality and the accuracy of the model. For example, statistical methods (such as the IQR method) or visualization methods (such as box plots) can be used to identify and remove outliers.
[0111] Missing values affect the integrity of the data. Filling in missing values can ensure the integrity of the data set and avoid problems during model training. Filling methods can include mean filling, median filling, mode filling, K-nearest neighbor filling, regression filling, etc.
[0112] Normalization processing can convert data with different dimensions into the same dimension, improving the training efficiency and prediction accuracy of the model. Normalization processing methods can include Min-Max normalization, Z-Score normalization, decimal scaling normalization, etc. Min-Max normalization scales the data into the range [0, 1], Z-Score normalization converts the data into a distribution with a mean of 0 and a standard deviation of 1, and decimal scaling normalization scales the data by moving the decimal point.
[0113] Referring to Figure 2 , an apparatus for training an epidemic prediction model is also provided in an embodiment of the present application. The apparatus includes:
[0114] An acquisition module 11 for acquiring epidemic training data;
[0115] A preprocessing module 12 for preprocessing the epidemic training data to obtain standard epidemic training data;
[0116] A transformation module 13 for performing Hilbert-Huang transform on the standard epidemic training data to obtain transformed data;
[0117] A training module 14 for training a preset long short-term memory network using the transformed data and calculating the loss value of the long short-term memory network during the training process based on the mean square error calculation formula;
[0118] A generation module 15 for generating an epidemic prediction model when it is determined that the loss value meets the requirements.
[0119] The training device for an epidemic prediction model provided by this application can remove noise and outliers by preprocessing epidemic training data, improve data quality, and thus enhance the prediction accuracy of the model. The Hilbert-Huang transform is an adaptive non-linear and non-stationary signal analysis method that can process non-linear and non-stationary time series data. By combining empirical mode decomposition and Hilbert transform, it decomposes complex signals into a series of intrinsic mode functions and their instantaneous frequencies, reducing the complexity of the data. Therefore, through this transform, the inherent period and trend in epidemic data can be extracted, providing more effective features for the long short-term memory network and further enhancing the prediction ability and efficiency of the model. At the same time, epidemic prediction by the long short-term memory network can capture long-term and short-term dependencies in time series data, which helps to process data such as epidemics that have obvious time correlations. In addition, based on the calculation of the loss function of mean square error, the difference between the model prediction value and the actual value can be effectively measured, guiding the parameter optimization in the model training process and accelerating the model convergence speed. Therefore, combining the Hilbert-Huang transform with the long short-term memory network in this application can significantly improve the accuracy of epidemic prediction.
[0120] As described above, it can be understood that each component of the training device for the epidemic prediction model proposed in this application can implement the functions of any one of the training methods for the epidemic prediction model described above, and the specific structure will not be elaborated.
[0121] Referring to Figure 3 , an embodiment of this application also provides a computer device, and its internal structure can be as Figure 3 shown. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a storage medium and an internal memory. The storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the storage medium. The database of the computer device is used to store relevant data of the training method for the epidemic prediction model. The network interface of the computer device is used to communicate with external computer devices through a network connection. When the computer program is executed by the processor, it implements a training method for an epidemic prediction model.
[0122] An embodiment of this application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a training method for an epidemic prediction model.
[0123] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided in this application and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.
[0124] It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, device, article, or method including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, device, article, or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, device, article, or method including that element.
[0125] The above are only the preferred embodiments of this application, and do not limit the patent scope of this application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of this application, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of this application by the same token.
Claims
1. A training method for an epidemic prediction model, characterized in that: include: Obtain epidemic training data; Preprocessing the epidemic training data to obtain standard epidemic training data; Performing Hilbert-Huang transform on the standard epidemic training data to obtain transformed data; Using the transformed data to train a preset long short-term memory network, and calculating the loss value of the long short-term memory network during the training process based on a mean square error calculation formula; When it is determined that the loss value meets the requirements, an epidemic prediction model is generated.
2. The method according to claim 1, characterized in that The Hilbert-Huang transform is performed on the standard epidemic training data to obtain transformed data, including: Screening the original signal corresponding to the standard epidemic training data, finding all local extreme value points of the original signal, and obtaining local maximum and local minimum values; The local maximum and the local minimum are connected respectively according to the cubic spline interpolation method to obtain an upper envelope and a lower envelope; Calculate the mean of the upper envelope and the lower envelope, and subtract the mean from the original signal corresponding to the standard epidemic training data to obtain the detail component; Determining whether the detail component satisfies a preset constraint condition; If the conditions are met, performing Hilbert transform on the detail component to obtain the instantaneous frequency and amplitude of the detail component, and using the detail component and the corresponding instantaneous frequency as transformation data; If not, the detail component is used as a new original signal, and the step of finding all local extreme points of the original signal to obtain local maximum and local minimum is returned to execute until the detail component meets the preset constraint conditions.
3. The method according to claim 2, characterized in that The preset constraints include: Wherein, x(t) is the original signal corresponding to the standard epidemic training data, and the IMF i (t) is the i-th intrinsic mode function, r(t) is the detail component, and n is the number of the detail components; Or at any point of the original signal, the average values of the upper envelope formed by the local maximum points and the lower envelope formed by the local minimum points are zero, and the number of the local maximum points and the local minimum points must be equal or differ by at most one.
4. The method according to claim 2, characterized in that: After performing Hilbert transform on the detail component to obtain the instantaneous frequency and amplitude of the detail component, the method further includes: A Hilbert spectrum is generated according to the instantaneous frequency and amplitude of the detail component, and the Hilbert spectrum is used to characterize the time-frequency distribution of the standard epidemic training data.
5. The method according to claim 2, characterized in that: The method of using the transformed data to train a preset long short-term memory network includes: Extracting detail components and corresponding instantaneous frequencies of each of the transformed data; Each of the detail components and the corresponding instantaneous frequency are input into a preset long short-term memory network, and each of the detail components is used to train the long short-term memory network.
6. The method according to claim 1, characterized in that When it is determined that the loss value meets the requirement, after generating the epidemic prediction model, the method further includes: Obtain epidemic data; Preprocessing the epidemic data to obtain standard epidemic data; The standard epidemic data is subjected to Hilbert-Huang transform and then input into the epidemic prediction model to generate an epidemic prediction result.
7. The method according to claim 1, characterized in that The preprocessing of the epidemic training data to obtain standard epidemic training data includes: The epidemic training data is subjected to outlier removal, missing value filling and standardization processing to obtain standard epidemic training data.
8. A training model for an epidemic prediction model, characterized in that: include: Acquisition module, used to obtain epidemic training data; A preprocessing module, used to preprocess the epidemic training data to obtain standard epidemic training data; A transformation module, used for performing Hilbert-Huang transformation on the standard epidemic training data to obtain transformed data; A training module, used to train a preset long short-term memory network using the transformed data, and calculate the loss value of the long short-term memory network during the training process based on a mean square error calculation formula; A generation module is used to generate an epidemic prediction model when it is determined that the loss value meets the requirements.
9. A computer device, characterized in that: include: processor; Memory; Wherein, the memory stores a computer program, and when the processor executes the computer program, it implements the training method of the epidemic prediction model described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the training method for the epidemic prediction model described in any one of claims 1 to 7.