Infectious disease spread prediction method, device and electronic equipment based on data decoupling
Through data decoupling and deep learning technology, the risk spillover characteristics and local evolution characteristics of infectious disease data are extracted and predictions are integrated, which solves the problem of low accuracy in infectious disease prediction in the existing technology and achieves more efficient prediction of infectious disease transmission .
Patent Information
- Application Number
- CN202410927599.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-07-11
AI Technical Summary
Existing methods for infectious disease prediction are limited by simplified assumptions or limited model capabilities, resulting in lower accuracy in infectious disease prediction.
Using a data decoupling method, infectious disease data is decoupled into risk spillover signal data and local evolution signal data. Features are extracted through graph convolution neural networks and long-term memory networks, and feature fusion is performed, and multi-layer perception machines are used for Prediction processing to improve the accuracy of predicting infectious disease transmission.
By combining the transmission nature of infectious diseases and the dynamic evolution laws, the accuracy of infectious disease transmission prediction is significantly improved, and a more accurate prediction method is provided to guide prevention and control policies.
Smart Images

Figure CN118471545B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical care informatics, and in particular to a method, device and electronic equipment for predicting the spread of infectious diseases based on data decoupling. Background Art
[0002] With the development of air travel, globalization, free trade and urbanization, the world is becoming more interconnected, but the global threat of infectious diseases is increasing. There is an urgent need for an accurate and rapid infectious disease prediction method to guide and formulate effective prevention and control policies to minimize the impact of infectious diseases.
[0003] Existing infectious disease prediction methods can be divided into two types: knowledge-driven methods and data-driven methods. Knowledge-driven methods explain the spread dynamics of infectious diseases in the population by combining prior knowledge assumptions and mathematical differential equations, while data-driven methods model the prediction task as a time series prediction problem, using statistical or data mining methods to capture patterns in hidden historical infectious disease data to predict the future development of infectious diseases. These methods are limited by overly simplified assumptions or limited model capabilities, ignoring the complexity and dynamic mechanisms of infectious disease transmission, resulting in low accuracy in infectious disease prediction. Summary of the invention
[0004] The present invention provides a method, device and electronic device for predicting the spread of an infectious disease based on data decoupling, which are used to solve the technical problem that the infectious disease spread prediction method in the prior art is limited by simplified assumptions or limited model capabilities, resulting in low accuracy of infectious disease prediction.
[0005] The present invention provides a method for predicting the spread of infectious diseases based on data decoupling, comprising the following steps:
[0006] Acquiring infectious disease data to be predicted, and encoding the infectious disease data to be predicted to obtain infectious disease encoding data;
[0007] Decoupling the infectious disease coding data to obtain risk spillover signal data and local evolution signal data respectively;
[0008] Performing feature extraction on the risk spillover signal data through a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data, and performing feature extraction on the local evolution signal data through a long short-term memory network to obtain local evolution features of the local evolution signal data;
[0009] The risk spillover feature is fused with the local evolution feature to obtain the infectious disease data fusion feature, and the infectious disease data fusion feature is predicted and processed by a multi-layer perceptron to obtain the propagation prediction result of the infectious disease data to be predicted.
[0010] In some embodiments, the data decoupling of the infectious disease coding data to obtain risk spillover signal data and local evolution signal data respectively includes:
[0011] Acquire prior knowledge of infectious disease characteristics, and determine a ratio weight of risk spillover signals to local evolution signals based on the prior knowledge;
[0012] The infectious disease coding data is weighted by the ratio weight to obtain risk spillover signal data and local evolution signal data respectively.
[0013] In some embodiments, the prior knowledge of infectious disease characteristics includes: epidemic area node codes, epidemic area population flow node codes, and epidemic area infection time codes, and determining the ratio weight of the risk spillover signal and the local evolution signal based on the prior knowledge includes:
[0014] Weighting the epidemic area node code, the epidemic area population flow node code, and the epidemic area infection time code by a first weight matrix to obtain a first weighted result, and performing nonlinear activation processing on the first weighted result to obtain a first linear activation result;
[0015] The first linear activation result is weighted according to a second weight matrix to obtain a second weighted result, and the second weighted result is subjected to nonlinear mapping processing to obtain a ratio weight of a risk spillover signal to a local evolution signal.
[0016] In some embodiments, extracting features from the risk spillover signal data using a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data includes:
[0017] Obtain the origin and destination matrix used to characterize population mobility when an infectious disease occurs;
[0018] The risk spillover signal data and the start-end point matrix are subjected to information aggregation processing through a graph convolutional neural network to obtain the risk spillover characteristics of the risk spillover signal data.
[0019] In some embodiments, extracting features from the local evolving signal data using a long short-term memory network to obtain local evolving features of the local evolving signal data includes:
[0020] Determine the position coding feature of the local evolution signal data, and concatenate the local evolution signal data with the position coding feature to obtain local evolution signal input data;
[0021] Extracting features from the local evolution signal input data through a long short-term memory network to obtain evolution law features;
[0022] The evolution law characteristics are processed by an attention mechanism through a multi-head self-attention network to obtain the local evolution characteristics of the local evolution signal data.
[0023] In some embodiments, the propagation prediction result of the infectious disease data to be predicted is obtained by calling an infectious disease propagation prediction model for prediction, and the infectious disease propagation prediction model is obtained after training with infectious disease data samples. The training process of the infectious disease propagation prediction model includes:
[0024] Calling the infectious disease transmission prediction model to encode the infectious disease data sample to obtain an infectious disease coded data sample, and performing data decoupling on the infectious disease coded data sample to obtain a risk spillover signal sample and a local evolution signal sample respectively;
[0025] By using the infectious disease transmission prediction model, feature extraction is performed on the risk spillover signal samples and the local evolution signal samples respectively to obtain risk spillover sample features of the risk spillover signal samples and local evolution sample features of the local evolution signal samples;
[0026] The risk spillover sample features are integrated with the local evolution sample features to obtain infectious disease data sample integration features, and the infectious disease transmission prediction model is used to predict the infectious disease data sample integration features to obtain a prediction result;
[0027] Determine a mean absolute error between a true result of the infectious disease data sample and the predicted result, and construct a mean absolute error loss value based on the mean absolute error;
[0028] Back propagation is performed in the infectious disease propagation prediction model according to the mean absolute error loss value to update the parameters of the infectious disease propagation prediction model.
[0029] The present invention also provides an infectious disease propagation prediction device based on data decoupling, comprising the following modules:
[0030] An acquisition module is used to acquire the infectious disease data to be predicted, and encode the infectious disease data to be predicted to obtain infectious disease coded data;
[0031] A decoupling module, used for performing data decoupling on the infectious disease coding data to obtain risk spillover signal data and local evolution signal data respectively;
[0032] A feature extraction module is used to perform feature extraction on the risk spillover signal data through a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data, and to perform feature extraction on the local evolution signal data through a long short-term memory network to obtain local evolution features of the local evolution signal data;
[0033] The prediction module is used to fuse the risk spillover characteristics with the local evolution characteristics to obtain the fusion characteristics of infectious disease data, and to perform prediction processing on the fusion characteristics of infectious disease data through a multi-layer perceptron to obtain the propagation prediction results of the infectious disease data to be predicted.
[0034] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an infectious disease transmission prediction method based on data decoupling as described above is implemented.
[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for predicting the spread of infectious diseases based on data decoupling as described above is implemented.
[0036] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the infectious disease transmission prediction methods based on data decoupling as described above.
[0037] The infectious disease propagation prediction method based on data decoupling provided by the present invention utilizes the propagation properties and dynamic evolution laws of infectious diseases to decouple infectious disease data into risk spillover data and local evolution data when predicting infectious disease propagation, thereby extracting risk spillover features and local evolution features of infectious disease data. Finally, the risk spillover features and local evolution features are integrated to predict the spread of infectious diseases, thereby predicting the spread of infectious diseases by combining the propagation properties and evolution laws of infectious diseases, effectively improving the accuracy of infectious disease propagation prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0039] Figure 1 It is a flow chart of the infectious disease propagation prediction method based on data decoupling provided by the present invention.
[0040] Figure 2It is a schematic diagram of the principle of the infectious disease transmission prediction process provided by the present invention.
[0041] Figure 3 It is a structural schematic diagram of the infectious disease propagation prediction device based on data decoupling provided by the present invention.
[0042] Figure 4 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0043] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0044] Combine the following Figure 1 The present invention describes an infectious disease transmission prediction method based on data decoupling.
[0045] Figure 1 is a flow chart of the infectious disease propagation prediction method based on data decoupling provided by the present invention, such as Figure 1 As shown, the method includes the following steps.
[0046] In step 101, the infectious disease data to be predicted is obtained, and the infectious disease data to be predicted is coded to obtain infectious disease coded data.
[0047] First of all, for infectious disease prediction, it is necessary to collect and obtain the data on the infectious diseases to be predicted. These data on the infectious diseases to be predicted can be collected and obtained through various systems and platform channels, specifically the historical daily number of newly confirmed cases (cases) in the epidemic area.
[0048] See also Figure 2 , Figure 2 It is a schematic diagram of the principle of the infectious disease transmission prediction process provided by the present invention, such as Figure 2 As shown in the figure, after obtaining the number of newly confirmed cases per day in history, these data are then input into the encoding layer, which is specifically a linear embedding layer neural network. By inputting the data into the embedding layer neural network for embedded coding processing, the specific data can be mapped into the feature coding space, thereby encoding the corresponding infectious disease coding data x. These coded data are actually the feature representations corresponding to the historical daily newly confirmed data sequence. The formula for the coding process is shown in the following formula (1):
[0049] (1)
[0050] In the above formula (1), X represents the historical daily number of newly confirmed cases input to the encoding layer, and embedding represents the processing function of the neural network in the embedding layer of the encoding layer.
[0051] In some embodiments, Figure 2 As shown in the figure, since the transmission nature of infectious diseases is easily affected by population mobility, when collecting the data on infectious diseases to be predicted based on the epidemic area, it is also necessary to collect inter-regional population mobility data, that is, cross-regional population mobility data. These population mobility data are used for the subsequent auxiliary risk spillover module to extract risk spillover characteristics, which will be explained below.
[0052] Continue to see Figure 1 In step 102, the infectious disease coding data is decoupled to obtain risk spillover signal data and local evolution signal data respectively.
[0053] After encoding the infectious disease data to be predicted, the next step is to extract features from the encoded infectious disease coded data. Here, Figure 2 As shown, the embodiment of the present invention takes into account the propagation nature and evolution law of infectious diseases. Before feature extraction, the infectious disease coding data is first input into the decoupling module, and the infectious disease coding data is decoupled by the decoupling module to obtain risk spillover signal data and local evolution signal data. Among them, the risk spillover signal data is the number of confirmed cases from other regions or epidemic areas to the local area, which is used to characterize the law and characteristics of the spread of infectious diseases. The local evolution signal data is the number of newly confirmed cases that naturally evolve in the local area, which is used to characterize the law and characteristics of the local spread of infectious diseases.
[0054] The specific data decoupling process is expressed as the following formula (2):
[0055] (2)
[0056] In the above formula (2), x represents the infectious disease coding data, x loc represents the local evolving signal data, Represents risk spillover signal data, and Decompose represents the processing function of data decoupling.
[0057] For data decoupling of infectious disease coding data (historical daily newly confirmed data series), considering that there is no prior knowledge to distinguish the proportion of local evolution signals and risk spillover signals in infectious disease coding data, it is necessary to obtain prior knowledge of infectious disease characteristics before decoupling, and estimate the ratio weight of risk spillover signals to local evolution signals based on prior knowledge for data decoupling.
[0058] In the data decoupling module, that is, in the processing of the Decompose data decoupling processing function, we first need to design an estimation gate neural network based on prior knowledge to roughly estimate the proportion of local evolution signals and risk spillover signals. Prior knowledge includes: epidemic area node code (denoted as E u o ), population mobility node code in the epidemic area (denoted as E u d ) and the infection time code of the epidemic area (denoted as T t ), which are used to characterize the propagation characteristics of epidemic area nodes, population mobility characteristics of epidemic areas, and infection time of epidemic areas. Next, these prior knowledge are input into the estimation gate neural network to estimate the ratio weight of risk spillover signal and local evolution signal.
[0059] The design process of the estimated gate neural network here is: weight the input epidemic area node code, epidemic area population flow node code and epidemic area infection time code through the first weight matrix W1 to obtain the first weighted result, and perform nonlinear activation processing on the first weighted result through the nonlinear activation function σ to obtain the first linear activation result. Next, the first linear activation result is weighted according to the second weight matrix W2 to obtain the second weighted result, and the second weighted result is nonlinearly mapped through the sigmoid activation function, so as to map the second weighted result to a value in the range of 0 to 1. This value is the ratio weight of the risk spillover signal to the local evolution signal. The specific calculation formula is as follows:
[0060] (3)
[0061] In the above formula (3), The value range is (0,1), which represents the ratio weight of the risk spillover signal and the local evolution signal, t represents time, and u represents the serial number of the epidemic area. σ represents the nonlinear activation function, E u o 、E u d 、T t They respectively represent the epidemic area node code, the epidemic area population flow node code, and the epidemic area infection time code. W1 represents the first learnable weight matrix, and W2 represents the second learnable weight matrix.
[0062] Through the above steps and calculation formula (3), in the process of processing the Decompose data decoupling processing function, the estimation gate neural network is used to determine the ratio weight of the risk spillover signal and the local evolution signal Then, through the ratio weight Weight the infectious disease coding data x to obtain risk spillover signal data And the local evolution signal data xloc Finally, the risk spillover signal data Send to the risk spillover module and send the local evolution signal data x loc Sent to the local evolution module.
[0063] Continue to see Figure 1 In step 103, feature extraction is performed on the risk spillover signal data through a graph convolutional neural network to obtain the risk spillover features of the risk spillover signal data, and feature extraction is performed on the local evolution signal data through a long short-term memory network to obtain the local evolution features of the local evolution signal data.
[0064] After data decoupling of infectious disease coding data, the next step is feature extraction. In an embodiment of the present invention, feature extraction of risk spillover signal data is performed through a graph convolutional neural network to obtain risk spillover features of risk spillover signal data, and feature extraction of local evolution signal data is performed through a long short-term memory network to obtain local evolution features of local evolution signal data.
[0065] It should be noted that the execution order of the steps of extracting risk spillover features and local evolution features in the embodiment of the present invention is not limited, and can be executed in parallel, that is, the process of extracting risk spillover features and local evolution features is executed at the same time, or there can be a sequence. For the convenience of explanation, the embodiment of the present invention has a sequence, but it should not be limited by this in the specific implementation. The specific process of feature extraction will be described below.
[0066] like Figure 2 As shown, in the risk spillover module, for the extraction of risk spillover features, because the risk spillover features characterize the characteristics of the number of confirmed cases from other epidemic areas to the local epidemic area, the factor of population mobility needs to be considered here. In the embodiment of the present invention, based on the collected inter-regional population mobility data (i.e., cross-regional population mobility data), an origin-destination (OD) matrix for characterizing population mobility when an infectious disease occurs is constructed and obtained. The origin-destination matrix can well characterize the laws and properties of population mobility. Therefore, when extracting risk spillover features in the risk spillover module, the embodiment of the present invention directly uses the risk spillover signal data The origin and destination matrix are input into the graph convolutional neural network, and the risk spillover signal data and the origin and destination matrix are aggregated through the graph convolutional neural network. That is, the risk spillover signal data is convoluted through the graph convolutional neural network, and the origin and destination matrix is input into the convolution calculation process as a weight parameter to aggregate the signal characteristics of areas with high population mobility correlation. Finally, the convolution calculation outputs the feature representation vector, that is, the risk spillover signal data. The risk spillover characteristics of , the information aggregation process is expressed as the following formula (4):
[0067] (4)
[0068] In the above formula (4), GNN represents graph convolutional neural network, and A mobility It represents the matrix of origin and destination.
[0069] In the local evolution module, with regard to the extraction of local evolution features, considering that local evolution features are used to characterize the laws and patterns of the local natural transmission process in the epidemic area, the embodiment of the present invention uses a long short-term memory network to extract local evolution features, and uses a position encoding mechanism and an attention mechanism to assist in the extraction process.
[0070] Specifically, firstly, for the local evolution signal data x loc , it is necessary to determine the position encoding features of the local evolution signal data. Here, the local evolution signal data x loc Its essence is a data sequence. We can first calculate the position code corresponding to the data sequence, denoted as PE, which is used to represent the position information of each code segment in the data sequence. Then, the local evolution signal data x loc It is concatenated with the position encoding feature PE to obtain the local evolution signal input data. Next, the evolution signal input data is input into the Long Short-Term Memory (LSTM) network, and the local evolution signal input data is feature extracted through the LSTM network to obtain the evolution law features and then input into the multi-head self-attention network. The multi-head self-attention network is used to process the evolution law features with the attention mechanism. During the processing, the evolution law features are further extracted to obtain the local evolution features of the local evolution signal data, which are recorded as H. loc , which is specifically expressed as the following formula (5):
[0071] (5)
[0072] In the above formula (5), MultiAttention represents the processing function of the multi-head self-attention network, LSTM represents the processing function of the long short-term memory network, and PE represents the local evolution signal data x loc Position encoding features.
[0073] Continue to see Figure 1 In step 104, the risk spillover feature is fused with the local evolution feature to obtain the fusion feature of infectious disease data, and the fusion feature of infectious disease data is predicted and processed by a multi-layer perceptron to obtain the propagation prediction result of the infectious disease data to be predicted.
[0074] The risk spillover features are extracted through step 103 And the local evolution characteristics H loc Afterwards, these two features will be input into the forward branch and the backward branch. The forward branch will fuse these two features to predict the spread of infectious diseases, and the backward branch will pass these two features back to the decoupling module to update the network parameters of the estimation gate neural network. Figure 2 As shown, on the one hand, the backward branch separates the risk spillover characteristics And the local evolution characteristics H loc The two feature vectors are sent back to the decoupling module for information reconstruction. Here, the two feature vectors are reconstructed through a fully connected layer and then sent back to the decoupling module to adjust the network parameters of the estimation gate neural network in the decoupling module to ensure the accuracy of subsequent decoupling. On the other hand, the forward branch will fuse the risk spillover features with the local evolution features to obtain the fusion features of infectious disease data. The feature fusion process here can be to directly fuse the risk spillover features And the local evolution characteristics H loc After performing vector concatenation operations, the infectious disease data fusion features can be obtained.
[0075] Next, the decoupling module inputs the fusion features of the infectious disease data into the prediction module, which includes a multilayer perceptron (MLP). The multilayer perceptron includes multiple fully connected layer neural networks. By inputting the fusion features of the infectious disease data into the multilayer perceptron for prediction processing, the propagation prediction result of the infectious disease data X to be predicted is finally obtained. The propagation prediction result can be the number of newly confirmed cases per day in the future period of time (time window) (denoted as ). That is, based on the input historical daily number of new cases X, the number of new confirmed cases per day in the future is finally predicted , the prediction process can be expressed as the following formula (6):
[0076] (6)
[0077] In the above formula (6), ⊕ represents the risk spillover characteristics And the local evolution characteristics H loc Perform vector concatenation operations, and MLP represents the processing function of a multi-layer perceptron.
[0078] Through the embodiments of the present invention, the propagation properties and evolution laws of infectious diseases are utilized to decouple infectious disease data into risk data and evolution data when predicting infectious disease propagation, thereby extracting the risk spillover characteristics and local evolution characteristics of infectious disease data, and combining these two characteristics to predict infectious disease propagation. In this way, the propagation properties and dynamic evolution laws of infectious disease propagation are combined when predicting infectious disease propagation, effectively improving the accuracy of infectious disease propagation prediction.
[0079] In some embodiments, the spread prediction result of the infectious disease data X to be predicted It can also be predicted by calling the infectious disease transmission prediction model, which is obtained after training with infectious disease data samples. The infectious disease transmission prediction model can be specifically composed of an embedding layer neural network in the encoding layer, a graph convolutional neural network in the risk spillover module, a long short-term memory network and a multi-head self-attention network in the local evolution module, and a multi-layer perceptron in the prediction module. The following introduces the training process of the infectious disease transmission prediction model.
[0080] First, infectious disease data samples are obtained. These infectious disease data samples are the number of new cases per day in history. At the same time, each infectious data sample records the real result (denoted as y), that is, the actual number of new cases per day in the future time period (time window). Next, the embedded layer neural network of the infectious disease transmission prediction model is called to encode the acquired infectious disease data samples to obtain infectious disease coded data samples, and the infectious disease coded data samples are decoupled to obtain risk spillover signal samples and local evolution signal samples. The process of encoding and data decoupling can refer to the above steps 101 and 102, which will not be repeated here.
[0081] Next, the infectious disease transmission prediction model is continued to be called to perform feature extraction on the risk spillover signal samples and the local evolution signal samples, respectively, to obtain the risk spillover sample features of the risk spillover signal samples and the local evolution sample features of the local evolution signal samples. Specifically, the risk spillover sample features of the risk spillover signal samples are extracted through the graph convolutional neural network in the infectious disease transmission prediction model, and the local evolution sample features of the local evolution signal samples are extracted through the long short-term memory network and the multi-head self-attention network of the infectious disease transmission prediction model. The specific feature extraction process can also refer to the above step 103, which will not be repeated here.
[0082] After the feature extraction is completed, the risk spillover sample features are subsequently fused with the local evolution sample features to obtain the fused features of the infectious disease data samples. The multi-layer perceptron in the infectious disease transmission prediction model is then called to perform the final prediction processing on the fused features of the infectious disease data samples to obtain the prediction result, which is denoted as Y.
[0083] Finally, the average absolute error between the true result y and the predicted result Y of the infectious disease data sample is determined. The true result y of the infectious disease data sample is the actual number of new cases per day in the future time period, and the predicted result is the predicted number of new cases per day in the future time period predicted by the infectious disease transmission prediction model. The average absolute error between the predicted number of new cases per day and the actual number of new cases per day is calculated to measure the prediction error of the infectious disease transmission prediction model. Here, the average absolute error loss value can be constructed based on the average absolute error, denoted as L(Y, y), and the specific calculation formula is as follows: Formula (7):
[0084] (7)
[0085] In the above formula (7), Y represents the prediction result obtained by the infectious disease transmission prediction model, y represents the actual result of the infectious disease data sample, N represents the total number of epidemic areas, j is the ordinal number of the epidemic area (i.e., the jth epidemic area), T p Indicates the size of the prediction window. For example, it can be a time window for predicting the number of cases. It indicates the predicted number of new cases per day in the future time period (time window). i is the ordinal number of the prediction window (i.e., the i-th prediction window).
[0086] After constructing the average error loss value, back propagation is performed in the infectious disease propagation prediction model through the average absolute error loss value to update the parameters of the infectious disease propagation prediction model. During training, the infectious disease data samples can be divided into a training set and a test set according to a ratio of 7 to 3 in the historical time window. The infectious disease propagation prediction model is trained with the training set, and the test set is used to verify the model performance.
[0087] During the training process, the training will be iterated according to the preset iterative training rounds. In each round of iteration, the mean absolute error loss value will be calculated based on the training set samples, and then the mean absolute error loss value will be back-propagated in the infectious disease propagation prediction model, and the parameters of the infectious disease propagation prediction model will be updated. The parameters of the infectious disease propagation prediction model are continuously updated by continuously iterating the loss calculation, back-propagating the gradient calculation, and optimizing the loss through the gradient optimization algorithm.
[0088] When a round of iterations ends, the mean absolute error loss converges (i.e., it no longer decreases with continuous training) or reaches the specified iterative training round, the training ends. Finally, the trained infectious disease transmission prediction model is verified on the test set to obtain the model verification result. Based on the verification result, it can be determined whether the model can be put into actual application scenarios for actual prediction.
[0089] The infectious disease propagation prediction device based on data decoupling provided by the present invention is described below. The infectious disease propagation prediction device based on data decoupling described below and the infectious disease propagation prediction method based on data decoupling described above can be referred to each other.
[0090] Figure 3 is a schematic diagram of the structure of the infectious disease propagation prediction device based on data decoupling provided by the present invention, such as Figure 3 As shown, the present invention provides an infectious disease propagation prediction device based on data decoupling, and the device specifically includes: an acquisition module 301, a decoupling module 302, a feature extraction module 303, and a prediction module 304. Among them, the acquisition module 301 is used to acquire the infectious disease data to be predicted, and encode the infectious disease data to be predicted to obtain the infectious disease encoding data; the decoupling module 302 is used to perform data decoupling on the infectious disease encoding data to obtain risk spillover signal data and local evolution signal data respectively; the feature extraction module 303 is used to extract the risk spillover signal data through a graph convolutional neural network to obtain the risk spillover features of the risk spillover signal data, and extract the local evolution signal data through a long short-term memory network to obtain the local evolution features of the local evolution signal data; the prediction module 304 is used to fuse the risk spillover features with the local evolution features to obtain the fusion features of the infectious disease data, and predict the fusion features of the infectious disease data through a multi-layer perceptron to obtain the propagation prediction result of the infectious disease data to be predicted.
[0091] Specifically, the infectious disease transmission prediction device based on data decoupling provided by the present invention can implement all the method steps implemented in the infectious disease transmission prediction method embodiment based on data decoupling, and can achieve the same technical effects. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.
[0092] Figure 4An example of a physical structure diagram of an electronic device is shown in Figure 4. The electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the infectious disease propagation prediction method based on data decoupling, the method comprising: obtaining the infectious disease data to be predicted, and encoding the infectious disease data to be predicted to obtain infectious disease coding data; performing data decoupling on the infectious disease coding data to obtain risk spillover signal data and local evolution signal data respectively; extracting features of the risk spillover signal data through a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data, and extracting features of the local evolution signal data through a long short-term memory network to obtain local evolution features of the local evolution signal data; performing feature fusion of the risk spillover features and the local evolution features to obtain infectious disease data fusion features, and performing prediction processing on the infectious disease data fusion features through a multi-layer perceptron to obtain the propagation prediction result of the infectious disease data to be predicted.
[0093] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0094] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the infectious disease propagation prediction method based on data decoupling provided by the above methods, which includes: obtaining the infectious disease data to be predicted, and encoding the infectious disease data to be predicted to obtain infectious disease coded data; performing data decoupling on the infectious disease coded data to obtain risk spillover signal data and local evolution signal data respectively; performing feature extraction on the risk spillover signal data through a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data, and performing feature extraction on the local evolution signal data through a long short-term memory network to obtain local evolution features of the local evolution signal data; performing feature fusion on the risk spillover features and the local evolution features to obtain fused features of the infectious disease data, and performing prediction processing on the fused features of the infectious disease data through a multi-layer perceptron to obtain the propagation prediction result of the infectious disease data to be predicted.
[0095] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the infectious disease propagation prediction method based on data decoupling provided by the above-mentioned methods, the method comprising: obtaining infectious disease data to be predicted, and encoding the infectious disease data to be predicted to obtain infectious disease coded data; performing data decoupling on the infectious disease coded data to obtain risk spillover signal data and local evolution signal data respectively; performing feature extraction on the risk spillover signal data through a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data, and performing feature extraction on the local evolution signal data through a long short-term memory network to obtain local evolution features of the local evolution signal data; performing feature fusion on the risk spillover features and the local evolution features to obtain fused features of the infectious disease data, and performing prediction processing on the fused features of the infectious disease data through a multi-layer perceptron to obtain the propagation prediction result of the infectious disease data to be predicted.
[0096] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.
[0097] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting the spread of infectious diseases based on data decoupling, characterized in that: The method comprises: Acquiring infectious disease data to be predicted, and encoding the infectious disease data to be predicted to obtain infectious disease encoding data; Acquire prior knowledge of infectious disease characteristics, and call an estimation gate neural network to determine a ratio weight of a risk spillover signal to a local evolution signal based on the prior knowledge; The infectious disease coding data is weighted by the ratio weight to obtain risk spillover signal data and local evolution signal data, respectively, wherein the risk spillover signal data is the number of confirmed cases from other regions or epidemic areas to the local area, and the local evolution signal data is the number of newly confirmed cases due to natural evolution in the local area; Performing feature extraction on the risk spillover signal data through a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data, and performing feature extraction on the local evolution signal data through a long short-term memory network to obtain local evolution features of the local evolution signal data, wherein the risk spillover features and the local evolution features are used for information reconstruction to adjust network parameters of the estimation gate neural network; The risk spillover feature is fused with the local evolution feature to obtain the infectious disease data fusion feature, and the infectious disease data fusion feature is predicted and processed by a multi-layer perceptron to obtain the propagation prediction result of the infectious disease data to be predicted.
2. The infectious disease propagation prediction method based on data decoupling according to claim 1 is characterized in that: The prior knowledge of the infectious disease characteristics includes the epidemic area node code, the epidemic area population flow node code and the epidemic area infection time code, and the calling of the estimation gate neural network according to the prior knowledge to determine the ratio weight of the risk spillover signal and the local evolution signal includes: Weighting the epidemic area node code, the epidemic area population flow node code, and the epidemic area infection time code by a first weight matrix to obtain a first weighted result, and performing nonlinear activation processing on the first weighted result to obtain a first nonlinear activation result; The first nonlinear activation result is weighted according to a second weight matrix to obtain a second weighted result, and the second weighted result is subjected to nonlinear mapping processing to obtain a ratio weight of a risk spillover signal to a local evolution signal.
3. The infectious disease propagation prediction method based on data decoupling according to claim 1 is characterized in that: The extracting features of the risk spillover signal data by using a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data includes: Obtain the origin and destination matrix used to characterize population mobility when an infectious disease occurs; The risk spillover signal data and the start-end point matrix are subjected to information aggregation processing through a graph convolutional neural network to obtain the risk spillover characteristics of the risk spillover signal data.
4. The infectious disease propagation prediction method based on data decoupling according to claim 1, characterized in that: The extracting features of the local evolving signal data by using a long short-term memory network to obtain local evolving features of the local evolving signal data includes: Determine the position coding feature of the local evolution signal data, and concatenate the local evolution signal data with the position coding feature to obtain local evolution signal input data; Extracting features from the local evolution signal input data through a long short-term memory network to obtain evolution law features; The evolution law characteristics are processed by an attention mechanism through a multi-head self-attention network to obtain the local evolution characteristics of the local evolution signal data.
5. The infectious disease propagation prediction method based on data decoupling according to claim 1, characterized in that: The propagation prediction result of the infectious disease data to be predicted is obtained by calling an infectious disease propagation prediction model for prediction. The infectious disease propagation prediction model is obtained after training with infectious disease data samples. The training process of the infectious disease propagation prediction model includes: Calling the infectious disease transmission prediction model to encode the infectious disease data sample to obtain an infectious disease coded data sample, and performing data decoupling on the infectious disease coded data sample to obtain a risk spillover signal sample and a local evolution signal sample respectively; Calling the infectious disease transmission prediction model to perform feature extraction on the risk spillover signal sample and the local evolution signal sample respectively to obtain the risk spillover sample features of the risk spillover signal sample and the local evolution sample features of the local evolution signal sample; The risk spillover sample feature is integrated with the local evolution sample feature to obtain an infectious disease data sample integration feature, and an infectious disease transmission prediction model is called to perform prediction processing on the infectious disease data sample integration feature to obtain a prediction result; Determine a mean absolute error between a true result of the infectious disease data sample and the predicted result, and construct a mean absolute error loss value based on the mean absolute error; Back propagation is performed in the infectious disease transmission prediction model according to the mean absolute error loss value to update the parameters of the infectious disease transmission prediction model.
6. An infectious disease propagation prediction device based on data decoupling, characterized in that: The device comprises: An acquisition module is used to acquire the infectious disease data to be predicted, and encode the infectious disease data to be predicted to obtain infectious disease coded data; A decoupling module is used to obtain prior knowledge of infectious disease characteristics, and call the estimation gate neural network according to the prior knowledge to determine the ratio weight of the risk spillover signal and the local evolution signal; the infectious disease coding data is weighted by the ratio weight to obtain risk spillover signal data and local evolution signal data, respectively, wherein the risk spillover signal data is the number of confirmed cases that have spilled over from other regions or epidemic areas to the local area, and the local evolution signal data is the number of newly confirmed cases that have evolved naturally in the local area; a feature extraction module, configured to perform feature extraction on the risk spillover signal data through a graph convolutional neural network to obtain risk spillover features of the risk spillover signal data, and perform feature extraction on the local evolution signal data through a long short-term memory network to obtain local evolution features of the local evolution signal data, wherein the risk spillover features and the local evolution features are used for information reconstruction to adjust network parameters of the estimation gate neural network; The prediction module is used to fuse the risk spillover characteristics with the local evolution characteristics to obtain the fusion characteristics of infectious disease data, and to perform prediction processing on the fusion characteristics of infectious disease data through a multi-layer perceptron to obtain the propagation prediction results of the infectious disease data to be predicted.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the infectious disease spread prediction method based on data decoupling as described in any one of claims 1 to 5 is implemented.
8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the infectious disease propagation prediction method based on data decoupling as described in any one of claims 1 to 5 is implemented.
9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the infectious disease propagation prediction method based on data decoupling as described in any one of claims 1 to 5 is implemented.
