CN113724889A - Prediction method, training method, device and equipment of model thereof and storage medium - Google Patents
Prediction method, training method, device and equipment of model thereof and storage medium Download PDFInfo
- Publication number
- CN113724889A CN113724889A CN202111004452.3A CN202111004452A CN113724889A CN 113724889 A CN113724889 A CN 113724889A CN 202111004452 A CN202111004452 A CN 202111004452A CN 113724889 A CN113724889 A CN 113724889A
- Authority
- CN
- China
- Prior art keywords
- time node
- target
- data
- feature
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000012549 training Methods 0.000 title claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims description 29
- 239000011159 matrix material Substances 0.000 claims description 22
- 201000010099 disease Diseases 0.000 claims description 18
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 18
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000004927 fusion Effects 0.000 claims description 5
- 238000010586 diagram Methods 0.000 description 13
- 239000013598 vector Substances 0.000 description 12
- 230000008859 change Effects 0.000 description 8
- 206010022000 influenza Diseases 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 229960005486 vaccine Drugs 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000010410 layer Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 239000002356 single layer Substances 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
- 241000712461 unidentified influenza virus Species 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/80—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for detecting, monitoring or modelling epidemics or pandemics, e.g. flu
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a prediction method, a training device, equipment and a storage medium of a model thereof, wherein the prediction method comprises the following steps: acquiring characteristic data of a target time node about the target event; acquiring the characteristic correlation degree between the target time node and at least one reference time node; and obtaining a prediction result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event and the feature data of the reference time node about the target event. According to the scheme, the target event can be accurately predicted.
Description
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to a prediction method, a training method, an apparatus, a device, and a storage medium for a model thereof.
Background
In recent years, with the development of artificial intelligence technologies such as neural networks and deep learning, the artificial intelligence technology has excellent performances such as high efficiency and stability compared with the traditional manual work, and is beginning to be widely applied to industries such as medical treatment, security protection, household appliances and logistics.
In social life, there are various target events that need to be predicted. Taking epidemics as an example, epidemiology prediction is an important issue, for example, in northern hemisphere temperate regions, the annual winter outbreak of influenza can bring a heavy health burden and economic burden to human society, and the annual vaccine changes due to the variation of influenza virus, so that vaccine manufacturers must produce enough vaccine in a short time to cope with the annual outbreak of influenza. In addition, dynamic allocation of flu beds is also a difficult problem due to the specificity of flu therapy. In view of the above, how to accurately predict a target event becomes an urgent problem to be solved.
Disclosure of Invention
The application provides a prediction method, a training device, equipment and a storage medium of a model thereof.
A first aspect of the present application provides a prediction method, including: acquiring characteristic data of a target time node about the target event; acquiring the characteristic correlation degree between the target time node and at least one reference time node; and obtaining a prediction result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event and the feature data of the reference time node about the target event.
Therefore, by acquiring the feature correlation degree between the target time node and the at least one reference time node, the reference time node similar to the target time node can be obtained based on the feature correlation degree, the feature data of the target time node on the target event and the feature data of the reference time node on the target event, so that the prediction result of the target time node on the target event can be obtained by referring to the change of the number of epidemic diseases of the historically similar reference time node, thereby improving the accuracy and interpretability of the prediction of the target event.
Wherein the obtaining of the feature correlation between the target time node and at least one reference time node comprises: determining a feature correlation between the target time node and each reference time node based on a similarity of the feature data between the target time node and each reference time node.
Therefore, by obtaining the similarity of the feature data between the target time node and each reference time node, the feature correlation between the target time node and each reference time node can be determined, and thus the reference time node similar to the target time node can be obtained, so that the prediction result of the target time node about the target event can be obtained by referring to the change of the number of epidemic diseases of the historically similar reference time node.
Wherein the obtaining of the predicted result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event, and the feature data of the reference time node about the target event comprises: obtaining the weight of each reference time node by using the characteristic correlation degree; fusing the feature data of the target time node and the feature data of the reference time node based on the weight to obtain fused feature data; and obtaining a prediction result of the target time node about the target event based on the fusion characteristic data.
Therefore, the weight of each reference time node can be obtained through the characteristic correlation degree between the target time node and the reference time node, the characteristic data of the target time node and the characteristic data of the reference time node can be fused based on the weight to obtain fused characteristic data, and then the prediction result of the target time node about the target event can be obtained, so that when the result of the target time node about the target event is predicted, historically similar time nodes are referred to, and a more accurate prediction result is obtained.
Wherein after the obtaining of the feature correlation between the target time node and the reference time node, the method further comprises: constructing an adjacency matrix between the target time node and a reference time node based on the feature correlation by using a prediction model, wherein the adjacent relation between the target time node and the reference time node in the adjacency matrix is determined based on the feature correlation; the obtaining the weight of each reference time node by using the feature correlation degree includes: and analyzing the adjacency matrix by using the graph neural network of the prediction model to obtain the weight of each reference time node.
Therefore, an adjacency matrix between the target time node and the reference time node can be constructed through the feature correlation between the target time node and the reference time node, and then the weight of each reference time node can be obtained according to the adjacency matrix between the target time node and the reference time node, so that the feature data of the target time node and the feature data of the reference time node are fused based on the weights, the fused feature data is obtained, and the prediction result of the target time node about the target event can be obtained.
Wherein after the obtaining of the predicted result of the target time node with respect to the target event based on the feature correlation, the feature data of the target time node with respect to the target event, and the feature data of the reference time node with respect to the target event, the method further comprises: and optimizing the prediction result based on the characteristic data of the target time node by using the prediction model.
Therefore, after the prediction result of the target time node about the target event is obtained, the prediction result is optimized by using the prediction model based on the characteristic data of the target time node, and the accuracy of prediction of the target event is further improved.
Wherein the optimizing the prediction result based on the feature data of the target time node is performed by a residual error network of the prediction model; and/or, before the optimizing the prediction result based on the feature data of the target time node, the method further comprises: and converting the characteristic data of the target time node into a second preset dimension for optimizing the prediction result subsequently.
Therefore, the output prediction result and the input feature data of the target time node are learned by using the residual error network, so that the accuracy of the prediction model can be improved, and the accuracy of the optimized prediction result is higher.
Wherein, prior to said obtaining a feature correlation between the target time node and the reference time node, the method comprises: and respectively converting the characteristic data of the target time node and the characteristic data of the reference time node into a first preset dimension by using a prediction model, wherein the preset dimension is higher than the original dimension of the characteristic data.
Therefore, by converting the feature data of the target time node and the feature data of the reference time node into a high-dimensional vector, the obtained high-dimensional vector can be used for representing the similarity degree between the feature data of the target time node and the feature data of the reference time node, and the feature correlation degree between the target time node and the corresponding reference time node can be conveniently obtained.
Wherein the obtaining of the feature data of the target time node about the target event includes: preprocessing original data of a target time node about the target event to obtain characteristic data of the target time node about the target event; wherein the pre-processing comprises at least one of: converting the original data into numerical data under the condition that the original data is non-numerical data; and converting the value range of the original data into a preset range.
Therefore, the non-numerical data is converted into the numerical data, the value range of the original data is converted into the preset range, the prediction model can understand the data more deeply, and the accuracy of the prediction result can be improved.
Wherein the target event is the occurrence of a predetermined disease.
Therefore, the method is beneficial to hospitals, vaccine manufacturers and the like to do treatment and prevention work on diseases by predicting the result of the occurrence of the preset diseases, and can improve the use efficiency of medical resources.
In order to solve the above problem, a second aspect of the present application provides a training method of a predictive model, the training method including: acquiring characteristic data of a sample time node about the target event and the actual result; obtaining a feature correlation between the sample time node and at least one reference time node using a predictive model; obtaining an event prediction result by using the prediction model based on the feature correlation, the feature data of the sample time node about the target event and the feature data of the reference time node about the target event, wherein the event prediction result comprises a prediction result of the sample time node about the target event; adjusting network parameters of the predictive model based on predicted and actual outcomes of the sample time nodes with respect to the target event.
Wherein, prior to said obtaining a feature correlation between the sample time node and at least one reference time node using a predictive model, the method further comprises: inputting feature data of the sample time node and at least one reference time node with respect to the target event to the predictive model; and/or the event prediction result further comprises a prediction result of the reference time node about the target event.
In order to solve the above problem, a third aspect of the present application provides a prediction apparatus comprising: the data acquisition module is used for acquiring characteristic data of a target time node about the target event; a correlation obtaining module, configured to obtain a feature correlation between the target time node and at least one reference time node; and the result prediction module is used for obtaining a prediction result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event and the feature data of the reference time node about the target event.
In order to solve the above problem, a fourth aspect of the present application provides a training apparatus for a prediction model, including: the data acquisition module is used for acquiring characteristic data of the sample time node about the target event and the actual result; a correlation obtaining module, configured to obtain a feature correlation between the sample time node and at least one reference time node by using a prediction model; a result prediction module, configured to obtain an event prediction result based on the feature correlation, the feature data of the sample time node about the target event, and the feature data of the reference time node about the target event by using the prediction model, where the event prediction result includes a prediction result of the sample time node about the target event; and the parameter adjusting module is used for adjusting the network parameters of the prediction model based on the prediction result and the actual result of the sample time node about the target event.
In order to solve the above problem, a fifth aspect of the present application provides an electronic device, which includes a memory and a processor coupled to each other, wherein the processor is configured to execute program instructions stored in the memory to implement the prediction method in the above first aspect or the training method of the prediction model in the above second aspect.
In order to solve the above-mentioned problems, a sixth aspect of the present application provides a computer-readable storage medium on which program instructions are stored, the program instructions, when executed by a processor, implementing the prediction method in the above-mentioned first aspect, or the training method of the prediction model in the above-mentioned second aspect.
According to the scheme, the characteristic correlation degree between the target time node and the at least one reference time node is obtained, the reference time node similar to the target time node can be obtained on the basis of the characteristic correlation degree, the characteristic data of the target time node about the target event and the characteristic data of the reference time node about the target event, so that the prediction result of the target time node about the target event can be obtained by referring to the change of the number of epidemic diseases of the historically similar reference time node, and the accuracy and the interpretability of the prediction of the target event are improved.
Drawings
FIG. 1 is a schematic flow chart diagram illustrating one embodiment of a prediction method of the present application;
FIG. 2 is a flowchart illustrating an embodiment of step S13 in FIG. 1;
FIG. 3 is a schematic flow chart diagram illustrating another embodiment of a prediction method of the present application;
FIG. 4 is a schematic structural diagram of an embodiment of a predictive model of the present application;
FIG. 5 is a schematic flow chart diagram illustrating an embodiment of a method for training a predictive model of the present application;
FIG. 6 is a block diagram of an embodiment of the prediction apparatus of the present application;
FIG. 7 is a block diagram of an embodiment of a device for training a predictive model according to the present application;
FIG. 8 is a block diagram of an embodiment of an electronic device of the present application;
FIG. 9 is a block diagram of an embodiment of a computer-readable storage medium of the present application.
Detailed Description
The following describes in detail the embodiments of the present application with reference to the drawings attached hereto.
In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular system structures, interfaces, techniques, etc. in order to provide a thorough understanding of the present application.
The terms "system" and "network" are often used interchangeably herein. The term "and/or" herein is merely an association describing an associated object, meaning that three relationships may exist, e.g., a and/or B, may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship. Further, the term "plurality" herein means two or more than two.
It will be appreciated that the outcome of the target event is related to the various types of features in the overall process, i.e., the specifics of the various types of features will affect the final trend of the target event. In the present application, the target event may be the occurrence of a predetermined disease, which may be an epidemic occurring only in a certain area, a global epidemic, or a disease without infectivity. Taking the target event as an example of occurrence of the epidemic, the prediction result of the time node to be predicted about the number of infected persons of the epidemic can be obtained by using the prediction model based on the characteristic data related to the occurrence of the epidemic.
Referring to fig. 1, fig. 1 is a schematic flow chart of an embodiment of a prediction method of the present application. Specifically, the method may include the steps of:
step S11: and acquiring characteristic data of the target time node about the target event.
Taking the prediction of the number of infected people with the epidemic disease as an example, the target time node is a time point or a time period to be predicted, and may be a certain day, a certain week, a certain month or the like. The prediction model can be constructed by referring to the infection rate, the death rate, the population base, the sanitary conditions, the climate and other factors of the epidemic diseases, namely, the natural factors and the social factors are characteristic data related to the occurrence of the epidemic diseases.
In one implementation scenario, the step S11 includes: preprocessing original data of a target time node about the target event to obtain characteristic data of the target time node about the target event; wherein the pre-processing comprises at least one of: converting the original data into numerical data under the condition that the original data is non-numerical data; and converting the value range of the original data into a preset range. It can be understood that if the original data is non-numerical data, such as the feature data of seasons like spring, summer, fall, and winter, it is not convenient to learn and predict the prediction model, so that the non-numerical data can be converted into numerical data like 0, 1, etc. by encoding; for another example, if the value range of the original data is relatively wide, the value range of the original data can be converted into a preset range, and the preset range is a range in which the model is easy to train; by preprocessing the original data, the prediction model can be favorable for deeply understanding the data, and the accuracy of the prediction result can be improved.
Step S12: and acquiring the characteristic correlation degree between the target time node and at least one reference time node.
Step S13: and obtaining a prediction result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event and the feature data of the reference time node about the target event.
It can be understood that, when predicting the trend of the epidemic of the target time node, in addition to the feature data related to the target time node, the feature data of the reference time node with high feature correlation with the target time node may be referred to, for example, the development of the epidemic at the same time node in the last year, the trend of the epidemic at a past time, and the like; the characteristic data of the reference time nodes are used for prediction, so that not only can a prediction result be obtained, but also a judgment basis can be obtained, and the prediction result has better interpretability.
In an embodiment, the step S12 specifically includes: determining a feature correlation between the target time node and each reference time node based on a similarity of the feature data between the target time node and each reference time node. It can be understood that by obtaining the similarity of the feature data between the target time node and each reference time node, the feature correlation between the target time node and each reference time node can be determined, and thus a reference time node similar to the target time node can be obtained, so that the prediction result of the target time node about the target event can be obtained by referring to the change of the number of epidemic diseases of the historically similar reference time node.
According to the scheme, the characteristic correlation degree between the target time node and the at least one reference time node is obtained, the reference time node similar to the target time node can be obtained on the basis of the characteristic correlation degree, the characteristic data of the target time node about the target event and the characteristic data of the reference time node about the target event, so that the prediction result of the target time node about the target event can be obtained by referring to the change of the number of epidemic diseases of the historically similar reference time node, and the accuracy and the interpretability of the prediction of the target event are improved.
Referring to fig. 2, fig. 2 is a schematic flowchart illustrating an embodiment of step S13 in fig. 1. In this embodiment, the step S13 may specifically include the following steps:
step S131: and obtaining the weight of each reference time node by using the characteristic correlation.
Step S132: and fusing the characteristic data of the target time node and the characteristic data of the reference time node based on the weight to obtain fused characteristic data.
Step S133: and obtaining a prediction result of the target time node about the target event based on the fusion characteristic data.
It can be understood that the weight of each reference time node can be obtained through the feature correlation between the target time node and the reference time node, the feature data of the target time node and the feature data of the reference time node can be fused based on the weight to obtain fused feature data, and then the prediction result of the target time node about the target event can be obtained, so that when the result of the target time node about the target event is predicted, historically similar time nodes are referred to, and a more accurate prediction result is obtained.
Referring to fig. 3, fig. 3 is a schematic flow chart of another embodiment of the prediction method of the present application. Specifically, the method may include the steps of:
step S31: and acquiring characteristic data of the target time node about the target event.
In this embodiment, step S31 is substantially similar to step S11 of the above embodiments of the present application, and is not repeated here.
Step S32: and respectively converting the characteristic data of the target time node and the characteristic data of the reference time node into a first preset dimension by using a prediction model. Wherein the preset dimension is higher than the original dimension of the feature data.
It can be understood that, by converting the feature data of the target time node and the feature data of the reference time node into a high-dimensional vector, the obtained high-dimensional vector can be used for representing the degree of similarity between the feature data of the target time node and the feature data of the reference time node, so as to facilitate the acquisition of the feature correlation between the target time node and the corresponding reference time node.
Step S33: and acquiring the characteristic correlation degree between the target time node and at least one reference time node.
By calculating the similarity between the high-dimensional vectors, the feature correlation between the target time node and the corresponding reference time node can be obtained, so that the reference time node similar to the target time node can be obtained, and the prediction result of the target time node about the target event can be obtained by referring to the change of the number of epidemic patients of the historically similar reference time node.
Step S34: and constructing an adjacency matrix between the target time node and the reference time node based on the characteristic correlation degree by utilizing a prediction model. Wherein the adjacency relation between the target time node and the reference time node in the adjacency matrix is determined based on the feature correlation.
According to the feature correlation degree between the target time node and each reference time node, an adjacent matrix between the target time node and the reference time node can be constructed, the adjacent matrix comprises vertex data and edge data, the vertex data represents each time node, and the relationship between the vertexes is represented by the edge data, namely the relationship between the time nodes is represented by the feature correlation degree.
Step S35: and analyzing the adjacency matrix by using the graph neural network of the prediction model to obtain the weight of each reference time node.
Step S36: and fusing the characteristic data of the target time node and the characteristic data of the reference time node based on the weight to obtain fused characteristic data.
Step S37: and obtaining a prediction result of the target time node about the target event based on the fusion characteristic data.
Therefore, the weight of each reference time node can be obtained according to the adjacency matrix between the target time node and the reference time node, the feature data of the target time node and the feature data of the reference time node can be fused based on the weight to obtain fused feature data, and further the prediction result of the target time node about the target event can be obtained, so that when the result of the target time node about the target event is predicted, historically similar time nodes are referred to, and a more accurate prediction result is obtained.
Step S38: and optimizing the prediction result based on the characteristic data of the target time node by using the prediction model.
After the prediction result of the target time node about the target event is obtained, the prediction result is optimized by using the prediction model based on the characteristic data of the target time node, so that the accuracy of prediction of the target event is further improved.
In an embodiment, the optimizing the prediction result based on the feature data of the target time node is performed by a residual error network of the prediction model. It can be understood that the residual error network can solve the problem that the classification accuracy rate does not increase and decrease due to the deepening of the network, and the prediction model can achieve the effect of improving the accuracy rate only through simple network deep stacking through the residual error network.
In an embodiment, before the step S38, the method further includes: and converting the characteristic data of the target time node into a second preset dimension for optimizing the prediction result subsequently. Specifically, the prediction result of the target time node about the target event obtained based on the fused feature data is a part responsible for learning features in the network of the prediction model, and the feature data of the target time node is converted into the second preset dimension, which is actually equivalent to an input part of the prediction model, so that the part responsible for learning by the residual error network is actually a difference value between the output prediction result and the input feature data, and thus, the previous output prediction result is optimized by the residual error network, and a better effect can be achieved compared with a direct learning input. Therefore, the output prediction result and the input feature data of the target time node are learned by using the residual error network, so that the accuracy of the prediction model can be improved, and the accuracy of the optimized prediction result is higher.
Referring to fig. 4, fig. 4 is a schematic structural diagram of a prediction model according to an embodiment of the present application. The predictive model 40 may include: a linear neural network 400, a dynamic neural network 402, a graph neural network 404, and a residual network 406.
The linear neural network 400 is configured to convert the feature data of the target time node and the reference time node into a first preset dimension, where the preset dimension is higher than an original dimension of the feature data. The dynamic neural network 402 is configured to obtain a feature correlation degree between a target time node and at least one reference time node, and construct an adjacency matrix between the target time node and the reference time node based on the feature correlation degree to form a virtual graph; wherein the adjacency relation between different time nodes in the adjacency matrix is determined based on the feature correlation. The graph neural network 404 is configured to analyze the adjacency matrix by using a graph neural network of the prediction model to obtain a weight of each time node, fuse feature data of the target time node and the reference time node based on the weights to obtain fused feature data, and obtain a prediction result of the target time node about the target event based on the fused feature data. The residual network 406 is connected to the linear neural network 400 and the graph neural network 404, and is used for optimizing the prediction result based on the feature data of the target time node.
The linear neural network 400 includes a plurality of linear units, two adjacent linear units are connected by a nonlinear activation function, and after the feature data (t1 to tn) of the target time node and the reference time node are input into the linear neural network, the original feature data can be converted into a vector in a high-dimensional space by using the plurality of linear units. The high-dimensional vectors obtained by the linear neural network 400 are input into the dynamic neural network 402, and the similarity of the feature data of each time node and the feature data of other time nodes in the high-dimensional space can be obtained by calculating the similarity of the high-dimensional vectors, so that a virtual graph between a target time node and a reference time node can be constructed, and the virtual graph is specifically an adjacency matrix. Inputting the high-dimensional vector obtained by the linear neural network 400 and the virtual graph obtained by the dynamic neural network 402 into the graph neural network 404, wherein the graph neural network 404 has a plurality of layers, the graph neural network 404 can obtain the weight corresponding to each time node according to the virtual graph, then according to the weight aggregation information, each layer of graph neural network 404 can aggregate the information of the first-order adjacency matrix, and then the result output by the last layer of graph neural network 404 can be one-dimensional data, so that the predicted result, such as the predicted number of people infected by epidemic diseases, can be expressed. In addition, the linear neural network 400 and the graph neural network 404 are connected by the residual error network 406, the high-dimensional vector obtained by the linear neural network 400 is mapped to one-dimensional data by single-layer linear transformation, it can be understood that the single-layer linear transformation can eliminate correlation between components of the high-dimensional vector to reduce the dimension of the high-dimensional vector, and then the output result of the graph neural network 404 is added to the result of the single-layer linear transformation to obtain the final prediction result (y1 to yn) of each time node with respect to the target event.
Referring to fig. 5, fig. 5 is a flowchart illustrating an embodiment of a method for training a prediction model according to the present application. Specifically, the method may include the steps of:
step S51: and acquiring characteristic data of the sample time node about the target event and the actual result.
Step S52: and obtaining the characteristic correlation degree between the sample time node and at least one reference time node by using a prediction model.
Step S53: obtaining an event prediction result by using the prediction model based on the feature correlation, the feature data of the sample time node about the target event and the feature data of the reference time node about the target event, wherein the event prediction result comprises a prediction result of the sample time node about the target event.
Step S54: adjusting network parameters of the predictive model based on predicted and actual outcomes of the sample time nodes with respect to the target event.
In this embodiment, steps S51 to S53 are substantially similar to steps S11 to S13 of the above embodiment of the present application, except that in order to train the prediction model, in addition to obtaining the feature data of the sample time node about the target event to obtain the prediction result by using the prediction model, the actual result of the sample time node about the target event needs to be obtained, so that the network parameters of the prediction model can be adjusted based on the prediction result and the actual result of the sample time node about the target event, and the prediction model is continuously trained.
In an embodiment, before step S52, the method further includes: inputting feature data of the sample time node and at least one reference time node with respect to the target event to the predictive model; and/or the event prediction result further comprises a prediction result of the reference time node about the target event.
It can be understood that, when the prediction model is trained, the characteristic data of the sample time node and the at least one reference time node about the target event needs to be input into the prediction model, and after the trained prediction model is obtained, the characteristic data of the target time node about the target event is directly input, so that the prediction result of the target time node about the target event can be obtained. In addition, in the process of training the prediction model, the prediction result of the reference time node about the target event can be obtained, so that the network parameters of the prediction model can be adjusted based on the prediction result and the actual result of the reference time node about the target event, so that the prediction model is continuously trained.
Referring to fig. 6, fig. 6 is a schematic diagram of an embodiment of a prediction apparatus of the present application. The prediction means 60 includes: a data obtaining module 600, configured to obtain feature data of a target time node about the target event; a correlation obtaining module 602, configured to obtain a feature correlation between the target time node and at least one reference time node; a result predicting module 604, configured to obtain a predicted result of the target time node regarding the target event based on the feature correlation, the feature data of the target time node regarding the target event, and the feature data of the reference time node regarding the target event.
In the above solution, the relevancy obtaining module 602 obtains the feature relevancy between the target time node and at least one reference time node, so that the result predicting module 604 may obtain the reference time node similar to the target time node based on the feature relevancy, the feature data of the target time node about the target event, and the feature data of the reference time node about the target event, so as to obtain the predicted result of the target time node about the target event by referring to the number change of the epidemic disease people of the historically similar reference time node, thereby improving the accuracy and interpretability of the prediction of the target event.
In some embodiments, the correlation obtaining module 602 may be specifically configured to determine the feature correlation between the target time node and each reference time node based on the similarity of the feature data between the target time node and each reference time node.
In some embodiments, the result prediction module 604 may be specifically configured to obtain a weight of each reference time node by using the feature correlation; fusing the feature data of the target time node and the feature data of the reference time node based on the weight to obtain fused feature data; and obtaining a prediction result of the target time node about the target event based on the fusion characteristic data.
In some embodiments, the correlation obtaining module 602 may be further specifically configured to construct an adjacency matrix between the target time node and the reference time node based on the feature correlation by using a prediction model, where an adjacent relationship between the target time node and the reference time node in the adjacency matrix is determined based on the feature correlation. At this time, the result predicting module 604 specifically executes the steps of obtaining the weight of each reference time node by using the feature correlation, including: and analyzing the adjacency matrix by using the graph neural network of the prediction model to obtain the weight of each reference time node.
In some embodiments, the prediction apparatus 60 further includes: a result optimization module 606, where the result optimization module 606 is configured to optimize the prediction result based on the feature data of the target time node by using the prediction model.
In some embodiments, the optimizing the prediction result based on the feature data of the target time node is performed by a residual network of the prediction model. And/or the result optimization module 606 is further configured to convert the feature data of the target time node into a second preset dimension for subsequent optimization of the prediction result.
In some embodiments, the correlation obtaining module 602 is further configured to convert the feature data of the target time node and the feature data of the reference time node into a first preset dimension by using a prediction model, where the preset dimension is higher than an original dimension of the feature data.
In some embodiments, the data obtaining module 600 is further configured to pre-process raw data of a target time node about the target event, so as to obtain feature data of the target time node about the target event; wherein the pre-processing comprises at least one of: converting the original data into numerical data under the condition that the original data is non-numerical data; and converting the value range of the original data into a preset range.
Referring to fig. 7, fig. 7 is a block diagram illustrating an embodiment of a training apparatus for a prediction model according to the present application. The training device 70 for the prediction model includes: a data obtaining module 700, configured to obtain feature data of the sample time node about the target event and the actual result; a correlation obtaining module 702, configured to obtain a feature correlation between the sample time node and at least one reference time node by using a prediction model; a result prediction module 704, configured to obtain an event prediction result based on the feature correlation, the feature data of the sample time node about the target event, and the feature data of the reference time node about the target event by using the prediction model, where the event prediction result includes a prediction result of the sample time node about the target event; a parameter adjusting module 706, configured to adjust a network parameter of the prediction model based on the predicted result and the actual result of the sample time node with respect to the target event.
In the above solution, the correlation obtaining module 702 obtains the feature correlation between the target time node and at least one reference time node, so that the result predicting module 704 may obtain the reference time node similar to the target time node based on the feature correlation, the feature data of the target time node about the target event, and the feature data of the reference time node about the target event, so as to obtain the predicted result of the target time node about the target event by referring to the number change of the epidemic disease people of the historically similar reference time node, thereby improving the accuracy and interpretability of the prediction of the target event.
Referring to fig. 8, fig. 8 is a schematic frame diagram of an embodiment of an electronic device according to the present application. The electronic device 80 comprises a memory 81 and a processor 82 coupled to each other, and the processor 82 is configured to execute program instructions stored in the memory 81 to implement the steps of any of the prediction method embodiments described above, or the steps of any of the prediction model training method embodiments described above. In one particular implementation scenario, the electronic device 80 may include, but is not limited to: microcomputer, server.
In particular, the processor 82 is configured to control itself and the memory 81 to implement the steps of any of the above-described embodiments of the prediction method, or the steps of any of the above-described embodiments of the training method of the prediction model. The processor 82 may also be referred to as a CPU (Central Processing Unit). The processor 82 may be an integrated circuit chip having signal processing capabilities. The Processor 82 may also be a general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. In addition, the processor 82 may be collectively implemented by an integrated circuit chip.
According to the scheme, the processor obtains the feature correlation degree between the target time node and the at least one reference time node, so that the reference time node similar to the target time node can be obtained based on the feature correlation degree, the feature data of the target time node about the target event and the feature data of the reference time node about the target event, the number of epidemic disease people of the historically similar reference time node can be referred to obtain the prediction result of the target time node about the target event, and the accuracy and the interpretability of the prediction of the target event are improved.
Referring to fig. 9, fig. 9 is a block diagram illustrating an embodiment of a computer-readable storage medium according to the present application. The computer readable storage medium 90 stores program instructions 900 capable of being executed by a processor, the program instructions 900 being for implementing the steps of any of the above-described prediction method embodiments, or the steps of any of the above-described training method embodiments of the prediction model.
In the several embodiments provided in the present application, it should be understood that the disclosed method and apparatus may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, a division of a module or a unit is merely one type of logical division, and an actual implementation may have another division, for example, a unit or a component may be combined or integrated with another system, or some features may be omitted, or not implemented. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some interfaces, and may be in an electrical, mechanical or other form.
Units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on network elements. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, a network device, or the like) or a processor (processor) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
Claims (15)
1. A method of prediction, the method comprising:
acquiring characteristic data of a target time node about the target event;
acquiring the characteristic correlation degree between the target time node and at least one reference time node;
and obtaining a prediction result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event and the feature data of the reference time node about the target event.
2. The method of claim 1, wherein obtaining the feature correlation between the target time node and at least one reference time node comprises:
determining a feature correlation between the target time node and each reference time node based on a similarity of the feature data between the target time node and each reference time node.
3. The method according to claim 1 or 2, wherein the obtaining of the predicted result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event, and the feature data of the reference time node about the target event comprises:
obtaining the weight of each reference time node by using the characteristic correlation degree;
fusing the feature data of the target time node and the feature data of the reference time node based on the weight to obtain fused feature data;
and obtaining a prediction result of the target time node about the target event based on the fusion characteristic data.
4. The method of claim 3, wherein after said obtaining a feature correlation between the target time node and the reference time node, the method further comprises:
constructing an adjacency matrix between the target time node and a reference time node based on the feature correlation by using a prediction model, wherein the adjacent relation between the target time node and the reference time node in the adjacency matrix is determined based on the feature correlation;
the obtaining the weight of each reference time node by using the feature correlation degree includes:
and analyzing the adjacency matrix by using the graph neural network of the prediction model to obtain the weight of each reference time node.
5. The method according to any one of claims 1 to 4, wherein after the obtaining of the predicted result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event, and the feature data of the reference time node about the target event, the method further comprises:
and optimizing the prediction result based on the characteristic data of the target time node by using the prediction model.
6. The method of claim 5, wherein the optimizing the prediction result based on the feature data of the target time node is performed by a residual network of the prediction model;
and/or, before the optimizing the prediction result based on the feature data of the target time node, the method further comprises:
and converting the characteristic data of the target time node into a second preset dimension for optimizing the prediction result subsequently.
7. The method according to any of claims 1 to 6, wherein prior to said obtaining a feature correlation between the target time node and the reference time node, the method comprises:
and respectively converting the characteristic data of the target time node and the characteristic data of the reference time node into a first preset dimension by using a prediction model, wherein the preset dimension is higher than the original dimension of the characteristic data.
8. The method according to any one of claims 1 to 7, wherein the obtaining of the characteristic data of the target time node about the target event comprises:
preprocessing original data of a target time node about the target event to obtain characteristic data of the target time node about the target event; wherein the pre-processing comprises at least one of: converting the original data into numerical data under the condition that the original data is non-numerical data; and converting the value range of the original data into a preset range.
9. The method of any one of claims 1 to 8, wherein the target event is the occurrence of a predetermined disease.
10. A method for training a predictive model, the method comprising:
acquiring characteristic data of a sample time node about the target event and the actual result;
obtaining a feature correlation between the sample time node and at least one reference time node using a predictive model;
obtaining an event prediction result by using the prediction model based on the feature correlation, the feature data of the sample time node about the target event and the feature data of the reference time node about the target event, wherein the event prediction result comprises a prediction result of the sample time node about the target event;
adjusting network parameters of the predictive model based on predicted and actual outcomes of the sample time nodes with respect to the target event.
11. The method of claim 10, wherein prior to said using a predictive model to obtain feature correlations between the sample time nodes and at least one reference time node, the method further comprises:
inputting feature data of the sample time node and at least one reference time node with respect to the target event to the predictive model;
and/or the event prediction result further comprises a prediction result of the reference time node about the target event.
12. A prediction apparatus, comprising:
the data acquisition module is used for acquiring characteristic data of a target time node about the target event;
a correlation obtaining module, configured to obtain a feature correlation between the target time node and at least one reference time node;
and the result prediction module is used for obtaining a prediction result of the target time node about the target event based on the feature correlation, the feature data of the target time node about the target event and the feature data of the reference time node about the target event.
13. An apparatus for training a predictive model, comprising:
the data acquisition module is used for acquiring characteristic data of the sample time node about the target event and the actual result;
a correlation obtaining module, configured to obtain a feature correlation between the sample time node and at least one reference time node by using a prediction model;
a result prediction module, configured to obtain an event prediction result based on the feature correlation, the feature data of the sample time node about the target event, and the feature data of the reference time node about the target event by using the prediction model, where the event prediction result includes a prediction result of the sample time node about the target event;
and the parameter adjusting module is used for adjusting the network parameters of the prediction model based on the prediction result and the actual result of the sample time node about the target event.
14. An electronic device comprising a memory and a processor coupled to each other, the processor being configured to execute program instructions stored in the memory to implement the prediction method of any one of claims 1 to 9 or the training method of the prediction model of any one of claims 10 to 11.
15. A computer-readable storage medium having stored thereon program instructions which, when executed by a processor, implement the prediction method of any one of claims 1 to 9 or the training method of the prediction model of any one of claims 10 to 11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111004452.3A CN113724889A (en) | 2021-08-30 | 2021-08-30 | Prediction method, training method, device and equipment of model thereof and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111004452.3A CN113724889A (en) | 2021-08-30 | 2021-08-30 | Prediction method, training method, device and equipment of model thereof and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113724889A true CN113724889A (en) | 2021-11-30 |
Family
ID=78679100
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111004452.3A Withdrawn CN113724889A (en) | 2021-08-30 | 2021-08-30 | Prediction method, training method, device and equipment of model thereof and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113724889A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101645160A (en) * | 2008-08-05 | 2010-02-10 | 上海迪爱斯通信设备有限公司 | Predicting, early warning and intelligent decision making system for social security events |
| CN110111903A (en) * | 2019-04-28 | 2019-08-09 | 中国疾病预防控制中心 | A kind of method and apparatus of public health emergency prediction |
| CN111508609A (en) * | 2020-04-17 | 2020-08-07 | 腾讯科技（深圳）有限公司 | Health condition risk prediction method and device, computer equipment and storage medium |
| CN111684314A (en) * | 2018-02-05 | 2020-09-18 | 株式会社日立制作所 | Weather prediction correction method and weather prediction system |
| CN112988851A (en) * | 2021-04-27 | 2021-06-18 | 腾讯科技（深圳）有限公司 | Counterfactual prediction model data processing method, device, equipment and storage medium |
| CN113139700A (en) * | 2020-11-30 | 2021-07-20 | 中科三清科技有限公司 | River flow prediction method, device, equipment and storage medium |
-
2021
- 2021-08-30 CN CN202111004452.3A patent/CN113724889A/en not_active Withdrawn
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101645160A (en) * | 2008-08-05 | 2010-02-10 | 上海迪爱斯通信设备有限公司 | Predicting, early warning and intelligent decision making system for social security events |
| CN111684314A (en) * | 2018-02-05 | 2020-09-18 | 株式会社日立制作所 | Weather prediction correction method and weather prediction system |
| CN110111903A (en) * | 2019-04-28 | 2019-08-09 | 中国疾病预防控制中心 | A kind of method and apparatus of public health emergency prediction |
| CN111508609A (en) * | 2020-04-17 | 2020-08-07 | 腾讯科技（深圳）有限公司 | Health condition risk prediction method and device, computer equipment and storage medium |
| CN113139700A (en) * | 2020-11-30 | 2021-07-20 | 中科三清科技有限公司 | River flow prediction method, device, equipment and storage medium |
| CN112988851A (en) * | 2021-04-27 | 2021-06-18 | 腾讯科技（深圳）有限公司 | Counterfactual prediction model data processing method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jiwani et al. | Novel healthcare framework for cardiac arrest with the application of AI using ANN | |
| CN107016438B (en) | System based on traditional Chinese medicine syndrome differentiation artificial neural network algorithm model | |
| CN110111888A (en) | A kind of XGBoost disease probability forecasting method, system and storage medium | |
| US11276495B2 (en) | Systems and methods for predicting multiple health care outcomes | |
| Zhou et al. | Credit scoring using support vector machines with direct search for parameters selection | |
| EP3364157A1 (en) | Method and system of outlier detection in energy metering data | |
| WO2018188533A1 (en) | Health model construction method, terminal and storage medium for health assessment | |
| CN113657548A (en) | Medical insurance abnormity detection method and device, computer equipment and storage medium | |
| US20230401466A1 (en) | Method for temporal knowledge graph reasoning based on distributed attention | |
| CN112992375A (en) | Early warning method, early warning device, equipment and medium for diseases | |
| CN116311539B (en) | Sleep motion capturing method, device, equipment and storage medium based on millimeter waves | |
| CN110458355A (en) | Event prediction method, apparatus, equipment and storage medium | |
| CN116611546A (en) | Knowledge-graph-based landslide prediction method and system for target research area | |
| CN113642672A (en) | Feature processing method and device of medical insurance data, computer equipment and storage medium | |
| CN113641659A (en) | Medical characteristic database construction method, device, equipment and storage medium | |
| Xu et al. | Semi-parametric joint modeling of survival and longitudinal data: the r package JSM | |
| CN117292844A (en) | Epidemic situation prediction method and system based on epidemic situation trend and management and control policy association model | |
| JP2021125136A (en) | Optimization device and optimization method | |
| WO2017001885A2 (en) | Method of generating a model of an object | |
| Jahanshahloo et al. | Using the minimum distance of DMUs from the frontier of the PPS for evaluating group performance of DMUs in DEA | |
| CN113299361A (en) | Patient clinical performance-based archive construction method and system | |
| CN111933283B (en) | Health monitoring method and device | |
| KR20220145654A (en) | Time series data processing device configured to process time series data with irregularity | |
| US20230409926A1 (en) | Index for risk of non-adherence in geographic region with patient-level projection | |
| CN113724889A (en) | Prediction method, training method, device and equipment of model thereof and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20211130 |
|
| WW01 | Invention patent application withdrawn after publication |