Multi-source data mapping method and mapping system based on self-supervised learning

By introducing a self-distillation framework in self-supervised pre-training of multi-source timing data, the encoder features and decoder features of visible timing positions are migrated, which solves the problem of insufficient feature representation in the prior art, and improves prediction accuracy and generalization performance.

CN119006843APending Publication Date: 2024-11-22ELECTRIC POWER RES INST OF EAST INNER MONGOLIA ELECTRIC POWER +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411121603.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing multi-source timing data self-supervised pre-training method ignores the influence of visible timing position during model training, resulting in wasting of visible timing feature knowledge in the decoder and affecting the generation of feature representations.

Method used

Based on the original mask reconstruction framework, the self-distillation framework is expanded, and the encoder features with visible timing positions are used as students and the decoder features are used as teachers. Through the self-distillation of features, more advanced knowledge is transferred to the encoder and extracted. Characteristics of stronger representation ability.

Benefits of technology

The prediction accuracy of downstream tasks and the generalization performance of multi-source data mapping features are improved, and more effective feature representations are generated through feature self-distillation and reconstruction of loss optimization models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006843A_ABST
    Figure CN119006843A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of data processing, and provides a multi-source data mapping method and system based on self-supervised learning, and the technical scheme is as follows: a self-distillation framework is expanded on the basis of an original mask reconstruction framework, encoder features at visible time sequence positions are used as students, decoder features at corresponding positions are used as teachers, and the self-distillation framework is used as a self-distillation framework; more advanced knowledge is migrated to the encoder in a feature self-distillation mode, so that the encoder can extract features with stronger representation capability for multi-source data, and the prediction precision of downstream tasks is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of data processing, and in particular relates to a multi-source data mapping method and a mapping system based on self-supervised learning. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, a variety of sensor devices have been deployed in power grid scenarios, generating massive amounts of multi-source data. By analyzing the monitoring data, abnormal conditions can be detected, thereby improving the intelligence level of fault diagnosis and status assessment. For the feature mapping of multi-source data, self-supervised training, as a data pre-training learning method, can learn the characteristic patterns of data, which is conducive to improving the feature representation ability. The current mainstream multi-source time series data self-supervised pre-training method is based on mask modeling. The data samples are encoded into features through the autoencoder and then decoded and reconstructed. The feature reconstruction effect shows the representation ability of the learned features.

[0004] The inventors found that this type of method only imposes constraints on the reconstructed masked multi-source data during model training, but ignores the influence of visible temporal position, resulting in a waste of visible temporal feature knowledge in the decoder, which is not conducive to the generation of more effective feature representation. Summary of the invention

[0005] In order to solve at least one technical problem existing in the above-mentioned background technology, the present invention provides a multi-source data mapping method and mapping system based on self-supervised learning, which expands the self-distillation framework based on the original mask reconstruction framework, and takes the encoder features of the visible time series position as students, and the decoder features of the corresponding position as teachers. More advanced knowledge is transferred to the encoder through feature self-distillation, so that the encoder can extract features with stronger representation capabilities for multi-source data, thereby improving the prediction accuracy of downstream tasks.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A first aspect of the present invention provides a multi-source data mapping method based on self-supervised learning, comprising the following steps:

[0008] Acquire data from multiple sources;

[0009] Mapping features are obtained based on multi-source data and a trained multi-source data mapping model; wherein the construction process of the multi-source data mapping model includes:

[0010] Extract visible time series features based on multi-source data;

[0011] Reconstructing the enhanced mask feature based on the contextual information of the visible temporal features to obtain the enhanced feature;

[0012] Rearrange the enhanced features into enhanced visible features and enhanced mask features;

[0013] Perform feature self-distillation based on visible temporal features and enhanced visible features, reconstruct multi-source data based on enhanced features and enhanced mask features, and optimize the multi-source data mapping model based on the self-distillation loss and reconstruction loss in the feature self-distillation and multi-source data reconstruction stages;

[0014] The prediction task is predicted based on the mapped features to obtain the prediction result.

[0015] Furthermore, the extracting of visible time series features based on multi-source data specifically includes:

[0016] Randomly select a fixed proportion of time point locations;

[0017] Based on the selected time point position, the multi-source data is sampled and masked, thereby obtaining the multi-source data of the unmasked time point;

[0018] Based on the multi-source data of unmasked time points, the encoder is used to extract visible temporal features.

[0019] Further, the reconstructing the enhanced mask feature based on the context information of the visible temporal feature to obtain the enhanced feature includes:

[0020] Get the learnable shared mask feature Token and use the vector copy operation to expand the mask feature at the index position of the masked time point;

[0021] Use the expanded mask features to perform feature filling to generate a feature sequence with the same length as the original multi-source data;

[0022] Based on the feature sequence after mask filling, the decoder completes the prediction of enhanced features.

[0023] Furthermore, feature self-distillation is performed based on the visible temporal features and the enhanced visible features, including:

[0024] Obtain enhanced visible features according to the index of the unmasked time point position and use them as teacher features;

[0025] The visible time series features extracted by the encoder are mapped to the feature space through the feature mapping layer and used as student features; the self-distillation loss is calculated between the teacher features and the student features.

[0026] Furthermore, the multi-source data reconstruction based on the enhanced features and the enhanced mask features includes:

[0027] Obtain enhanced mask features according to the index of the masked time point position;

[0028] The enhanced mask features are used to predict the multi-source data at the mask time point position through a linear reconstruction layer;

[0029] The data reconstruction loss is calculated between the reconstructed data and the original multi-source data.

[0030] Furthermore, the loss function during the training of the multi-source data mapping model is:

[0031] Loss = α*L rec +(1-α)*L dis ,

[0032]

[0033] L rec =MSE(pred,target),

[0034] Where, L dis is the self-distillation loss, L rec is the reconstruction loss, α is the weight to balance the two losses, B is the training batch size, L is the length of the visible feature, D is the dimension of the feature vector, and s ijk Student characteristic value, t ijk is the key mapping feature value, MSE is the mean square error loss, pred and target are the reconstructed value and the original value respectively.

[0035] A second aspect of the present invention provides a multi-source data mapping system based on self-supervised learning, comprising:

[0036] A data acquisition module, which is used to acquire multi-source data;

[0037] A multi-source data mapping model construction module is used to obtain mapping features based on multi-source data and a trained multi-source data mapping model; wherein the construction process of the multi-source data mapping model includes:

[0038] Extract visible time series features based on multi-source data;

[0039] Reconstructing the enhanced mask feature based on the contextual information of the visible temporal features to obtain the enhanced feature;

[0040] Rearrange the enhanced features into enhanced visible features and enhanced mask features;

[0041] Perform feature self-distillation based on visible temporal features and enhanced visible features, reconstruct multi-source data based on enhanced features and enhanced mask features, and optimize the multi-source data mapping model based on the self-distillation loss and reconstruction loss in the feature self-distillation and multi-source data reconstruction stages;

[0042] The prediction module is used to predict the prediction task based on the mapping features to obtain the prediction result.

[0043] A third aspect of the present invention provides a computer-readable storage medium.

[0044] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the multi-source data mapping method based on self-supervised learning as described above.

[0045] A fourth aspect of the present invention provides a computer device.

[0046] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the multi-source data mapping method based on self-supervised learning as described above are implemented.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention expands the self-distillation framework based on the original mask reconstruction framework, performs feature self-distillation based on visible time series features and enhanced visible features, reconstructs multi-source data based on enhanced features and enhanced mask features, and optimizes the multi-source data mapping model based on the self-distillation loss and reconstruction loss in the feature self-distillation and multi-source data reconstruction stages; the encoder features at the visible time series position are used as students, and the decoder features at the corresponding positions are used as teachers, and more advanced knowledge is transferred to the encoder through feature self-distillation, so that the encoder can extract features with stronger representation capabilities for multi-source data, thereby improving the prediction accuracy of downstream tasks.

[0049] 2. Based on the mask reconstruction pre-training paradigm, the present invention also introduces self-distillation loss, which is beneficial to guide the encoder to extract contextual information that is beneficial to downstream prediction tasks and improve the generalization performance of multi-source data mapping features.

[0050] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0052] Figure 1 is a flow chart of a multi-source data mapping method based on self-supervised learning provided by an embodiment of the present invention;

[0053] Figure 2 It is a diagram of the architecture of a multi-source data mapping model based on self-supervised learning provided by an embodiment of the present invention;

[0054] Figure 3 is a visible feature extraction flow chart provided by an embodiment of the present invention;

[0055] Figure 4 is a flowchart of enhanced feature prediction provided by an embodiment of the present invention;

[0056] Figure 5 is a characteristic self-distillation flow chart provided by an embodiment of the present invention;

[0057] Figure 6 This is a multi-source data reconstruction flow chart provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0059] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0060] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0061] As mentioned in the background of the present invention, existing methods only impose constraints on the reconstructed masked multi-source data during model training, while ignoring the influence of visible temporal positions.

[0062] Compared with the visible temporal features in the encoder, the visible features in the encoder have richer knowledge and are conducive to the generation of more effective feature representations.

[0063] Therefore, the present invention proposes a multi-source data mapping method based on self-supervised learning. Based on the original mask reconstruction framework, the self-distillation framework is expanded. The encoder features of visible temporal positions are used as students, and the decoder features of corresponding positions are used as teachers. More advanced knowledge is transferred to the encoder through feature self-distillation, so that the encoder can extract features with stronger representation capabilities for multi-source data, thereby improving the prediction accuracy of downstream tasks.

[0064] Embodiment 1

[0065] like Figure 1-Figure 2 As shown, this embodiment provides a multi-source data mapping method based on self-supervised learning, comprising the following steps:

[0066] Step 1: Obtain multi-source data;

[0067] Step 2: Extract visible time series features based on multi-source data;

[0068] Step 3: Reconstruct the enhanced mask feature based on the context information of the visible temporal features to obtain the enhanced feature;

[0069] Step 4: Rearrange the enhanced features into enhanced visible features and enhanced mask features;

[0070] Step 5: Calculate the self-distillation loss based on the enhanced visible features and visible temporal features;

[0071] Step 6: Based on the enhanced mask features, the multi-source data at the masked time point position is predicted through the linear reconstruction layer, and the data reconstruction loss is calculated with the original multi-source data;

[0072] Step 7: Through the self-distillation loss and the reconstruction loss, the two losses simultaneously optimize the entire model end-to-end.

[0073] In order to more clearly illustrate the technical solution of the present invention, it is described in detail below by way of embodiments.

[0074] Wherein, in step 1, the multi-source data is monitoring data collected by different sensors, such as temperature, gas content, electrical parameters, etc.

[0075] like Figure 3 As shown, the purpose of step 2 is to extract the features of the unmasked time point position, and the extraction of visible time series features based on multi-source data specifically includes:

[0076] First, a fixed proportion of time point positions are randomly selected; then, based on the selected time point positions, the multi-source data is sampled and masked to obtain the multi-source data of the unmasked time points; finally, based on the multi-source data of the unmasked time points, the visible temporal features V are extracted through the encoder.

[0077] like Figure 4 As shown, the purpose of step 3 is to reconstruct the enhanced mask feature using the context information of the visible temporal features, and the enhanced mask feature is reconstructed based on the context information of the visible temporal features, which specifically includes the following steps:

[0078] First, obtain the learnable shared mask feature Token, and use the vector copy operation to expand the mask feature at the index position of the masked time point. The specific operation method is shown in the following formula:

[0079] EToken=Copy(Token,length),

[0080] Among them, Copy is the vector copy function, length is the mask length, and EToken is the expanded mask feature;

[0081] Then, the expanded mask features are used for feature filling to generate a feature sequence with the same length as the original multi-source data;

[0082] Finally, based on the feature sequence after mask filling, the decoder completes the prediction of enhanced features.

[0083] like Figure 5 As shown, the purpose of step 5 is to narrow the gap between upstream and downstream tasks by constraining visible features, including:

[0084] First, the enhanced features are rearranged, and the enhanced visible features are obtained according to the index of the unmasked time point position, which is used as the teacher feature T;

[0085] Then, the visible temporal feature V extracted by the encoder is mapped into the feature space through the feature mapping layer, which is used as the student feature S. The specific calculation method is shown in the following formula:

[0086] S = MLP(V),

[0087] Among them, MLP is a multi-layer perceptron, which contains two linear layers and a ReLU activation function, and the execution order is linear layer-ReLU-linear layer;

[0088] Finally, the self-distillation loss is calculated between the teacher features and the student features. The self-distillation loss function is L dis :

[0089]

[0090] Among them, B is the training batch size, L is the length of the visible feature, D is the dimension of the feature vector, and s ijk Student characteristic value, t ijk is the key mapping feature value.

[0091] like Figure 6 As shown, the purpose of step 6 is to realize the reconstruction of multi-source data at the mask position according to the enhanced mask feature, which specifically includes:

[0092] First, the enhanced features are rearranged and the enhanced mask features M are obtained according to the index of the masked time point position;

[0093] Then, the enhanced mask features are used to predict the multi-source data at the mask time point through the linear reconstruction layer. The specific calculation method is shown in the following formula:

[0094] pred=Linear(M),

[0095] Among them, Linear is a single linear layer, and pred is the reconstructed data;

[0096] Finally, the data reconstruction loss is calculated between the reconstructed data and the original multi-source data. The data reconstruction loss function L rec for:

[0097] L rec =MSE(pred,target),

[0098] Among them, MSE is the mean square error loss, pred and target are the reconstructed value and the original value respectively.

[0099] Therefore, the overall loss function is:

[0100] Loss = α*L rec +(1-α)*L dis ,

[0101] α is the weight to balance the two losses.

[0102] The entire model is optimized end-to-end simultaneously via two losses.

[0103] Embodiment 2

[0104] This embodiment provides a multi-source data mapping system based on self-supervised learning, including:

[0105] A data acquisition module, which is used to acquire multi-source data;

[0106] A multi-source data mapping model construction module is used to obtain mapping features based on multi-source data and a trained multi-source data mapping model; wherein the construction process of the multi-source data mapping model includes:

[0107] Extract visible time series features based on multi-source data;

[0108] Reconstructing the enhanced mask feature based on the contextual information of the visible temporal features to obtain the enhanced feature;

[0109] Rearrange the enhanced features into enhanced visible features and enhanced mask features;

[0110] Perform feature self-distillation based on visible temporal features and enhanced visible features, reconstruct multi-source data based on enhanced features and enhanced mask features, and optimize the multi-source data mapping model based on the self-distillation loss and reconstruction loss in the feature self-distillation and multi-source data reconstruction stages;

[0111] The prediction module is used to predict the prediction task based on the mapping features to obtain the prediction result.

[0112] Furthermore, in the multi-source data mapping model construction module, feature self-distillation is performed based on visible temporal features and enhanced visible features, including:

[0113] Obtain enhanced visible features according to the index of the unmasked time point position and use them as teacher features;

[0114] The visible time series features extracted by the encoder are mapped into the feature space through the feature mapping layer and used as student features; the self-distillation loss is calculated between the teacher features and the student features;

[0115] or,

[0116] The multi-source data reconstruction based on the enhanced features and the enhanced mask features includes:

[0117] Obtain enhanced mask features according to the index of the masked time point position;

[0118] The enhanced mask features are used to predict the multi-source data at the mask time point position through a linear reconstruction layer;

[0119] The data reconstruction loss is calculated between the reconstructed data and the original multi-source data.

[0120] Embodiment 3

[0121] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the multi-source data mapping method based on self-supervised learning as described in the first embodiment are implemented.

[0122] Embodiment 4

[0123] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in a multi-source data mapping method based on self-supervised learning as described in Embodiment 1 are implemented.

[0124] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0125] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0128] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0129] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-source data mapping method based on self-supervised learning, It is characterized in that The steps include: Acquire data from multiple sources; A mapping feature is obtained based on the multi-source data and the trained multi-source data mapping model; wherein the construction process of the multi-source data mapping model includes: Extract visible time series features based on multi-source data; Reconstructing the enhanced mask feature based on the contextual information of the visible temporal features to obtain the enhanced feature; Rearranging the enhanced features into enhanced visible features and enhanced mask features; Perform feature self-distillation based on visible temporal features and enhanced visible features, reconstruct multi-source data based on enhanced features and enhanced mask features, and optimize the multi-source data mapping model based on the self-distillation loss and reconstruction loss in the feature self-distillation and multi-source data reconstruction stages; The prediction task is predicted based on the mapped features to obtain the prediction result.

2. A multi-source data mapping method based on self-supervised learning as claimed in claim 1, It is characterized in that The extracting of visible time series features based on multi-source data specifically includes: Randomly select a fixed proportion of time point locations; Based on the selected time point position, the multi-source data is sampled and masked, thereby obtaining the multi-source data of the unmasked time point; Based on the multi-source data of unmasked time points, the encoder is used to extract visible temporal features.

3. A multi-source data mapping method based on self-supervised learning as claimed in claim 1, It is characterized in that The process of reconstructing the enhanced mask feature based on the context information of the visible temporal feature to obtain the enhanced feature includes: Get the learnable shared mask feature Token and use the vector copy operation to expand the mask feature at the index position of the masked time point; Use the expanded mask features to perform feature filling to generate a feature sequence with the same length as the original multi-source data; Based on the feature sequence after mask filling, the decoder completes the prediction of enhanced features.

4. A multi-source data mapping method based on self-supervised learning as claimed in claim 1, It is characterized in that Feature self-distillation is performed based on visible temporal features and enhanced visible features, including: Obtain enhanced visible features according to the index of the unmasked time point position and use them as teacher features; The visible time series features extracted by the encoder are mapped to the feature space through the feature mapping layer and used as student features; the self-distillation loss is calculated between the teacher features and the student features.

5. A multi-source data mapping method based on self-supervised learning as claimed in claim 1, It is characterized in that The multi-source data reconstruction based on the enhanced features and the enhanced mask features includes: Obtain enhanced mask features according to the index of the masked time point position; The enhanced mask features are used to predict the multi-source data at the masked time point position through a linear reconstruction layer; The data reconstruction loss is calculated between the reconstructed data and the original multi-source data.

6. A multi-source data mapping method based on self-supervised learning as claimed in claim 1, It is characterized in that The loss function during training of the multi-source data mapping model is: Loss=α*L rec +(1-a)*L dis , L rec =MSE(pred,target), Where, L dis is the self-distillation loss, L rec is the reconstruction loss, α is the weight to balance the two losses, B is the training batch size, L is the length of the visible feature, D is the dimension of the feature vector, and s ijk Student characteristic value, t ijk is the key mapping feature value, MSE is the mean square error loss, pred and target are the reconstructed value and the original value respectively.

7. A multi-source data mapping system based on self-supervised learning, It is characterized in that include: A data acquisition module, which is used to acquire multi-source data; A multi-source data mapping model construction module is used to obtain mapping features based on multi-source data and a trained multi-source data mapping model; wherein the construction process of the multi-source data mapping model includes: Extract visible time series features based on multi-source data; Reconstructing the enhanced mask feature based on the contextual information of the visible temporal features to obtain the enhanced feature; Rearranging the enhanced features into enhanced visible features and enhanced mask features; Perform feature self-distillation based on visible temporal features and enhanced visible features, reconstruct multi-source data based on enhanced features and enhanced mask features, and optimize the multi-source data mapping model based on the self-distillation loss and reconstruction loss in the feature self-distillation and multi-source data reconstruction stages; The prediction module is used to predict the prediction task based on the mapping features to obtain the prediction result.

8. A multi-source data mapping system based on self-supervised learning as claimed in claim 7, It is characterized in that In the multi-source data mapping model construction module, feature self-distillation is performed based on visible temporal features and enhanced visible features, including: Obtain enhanced visible features according to the index of the unmasked time point position and use them as teacher features; The visible time series features extracted by the encoder are mapped into the feature space through the feature mapping layer and used as student features; the self-distillation loss is calculated between the teacher features and the student features; or, The multi-source data reconstruction based on the enhanced features and the enhanced mask features includes: Obtain enhanced mask features according to the index of the masked time point position; The enhanced mask features are used to predict the multi-source data at the masked time point position through a linear reconstruction layer; The data reconstruction loss is calculated between the reconstructed data and the original multi-source data.

9. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the steps in a multi-source data mapping method based on self-supervised learning as described in any one of claims 1 to 6 are implemented.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the program, the steps in the multi-source data mapping method based on self-supervised learning as described in any one of claims 1-6 are implemented.