Data prediction method, device, equipment and storage medium based on diffusion model

Abstract

The present invention discloses a data prediction method, device, equipment and storage medium based on a diffusion model, including obtaining historical and real-time multi-channel signal time series data; constructing a preprocessing model, the preprocessing model is used to perform Min-Max normalization on the input multi-channel signal time series data, converting the normalized data into a two-dimensional tensor image, the horizontal coordinate of the two-dimensional tensor image is the time axis, the vertical coordinate of the two-dimensional tensor image is different signal channels, setting a noise addition starting point on the time axis, and adding Gaussian noise to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis; training a diffusion model based on U-net according to the first noise addition image data to obtain a multi-channel signal data prediction model; inputting the second noise addition image data into the multi-channel signal data prediction model; converting the predicted two-dimensional tensor image into predicted multi-channel signal time series data based on Min-Max normalization. The present invention can realize the prediction of multi-channel signal time series data.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a data prediction method, device, equipment and storage medium based on a diffusion model. Background Art

[0002] Advanced deep learning methods have brought technological innovations to multi-channel signal data analysis. However, there are still some technical challenges in the field of multi-channel signal data analysis, which affect the practical application of deep learning models based on multi-channel signal data. This is mainly reflected in the fact that most technologies and research focus on achieving real-time analysis based on current multi-channel signals and fail to predict future trends; while existing research on future predictions of time series signals can only be achieved for single-channel signals, ignoring the spatiotemporal information of multi-channel signals and failing to predict multi-channel signals simultaneously.

[0003] It can be seen that there is still a research gap in the future trend generation and prediction of multi-channel signals.

[0004] Therefore, a data prediction method, device, equipment and storage medium based on a diffusion model are developed to solve the above problems. Summary of the invention

[0005] The present invention proposes a data prediction method, device, equipment and storage medium based on a diffusion model to solve the problem that the prior art cannot simultaneously predict multi-channel signals.

[0006] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0007] A first aspect of the present invention provides a data prediction method based on a diffusion model, comprising:

[0008] Acquire first information and second information, wherein the first information includes historical multi-channel signal time series data, and the second information includes real-time multi-channel signal time series data;

[0009] Constructing a preprocessing model, the preprocessing model is used to perform Min-Max normalization on the input multi-channel signal time series data, converting the normalized data into a two-dimensional tensor image, the abscissa of the two-dimensional tensor image is the time axis, the ordinate of the two-dimensional tensor image is different signal channels, setting a noise addition starting point on the time axis, and adding Gaussian noise to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis;

[0010] Inputting the first information into the preprocessing model to obtain first noisy image data;

[0011] Inputting the second information into the preprocessing model to obtain second noisy image data;

[0012] According to the first noisy image data, a diffusion model based on U-net is trained to obtain a multi-channel signal data prediction model; the present invention uses the U-net structure to construct a diffusion model, and the diffusion model based on U-net provides a framework for noise addition and denoising. U-net, as the network structure of the diffusion model, performs the task of feature extraction and restoration, so that the diffusion model can gradually denoise and accurately restore the future signal state. The encoding-decoding structure of U-net extracts time series features during the signal noise addition and denoising process, and retains the detailed information in the original signal in the restoration stage through jump connections, so that the model can effectively predict the future signal state.

[0013] Inputting the second noisy image data into the multi-channel signal data prediction model to obtain a predicted two-dimensional tensor image;

[0014] The predicted two-dimensional tensor image is converted into predicted multi-channel signal time series data based on Min-Max normalization.

[0015] Furthermore, the input multi-channel signal time series data is subjected to Min-Max normalization, and the normalized data is converted into a two-dimensional tensor image, including:

[0016] The signal channels of the multi-channel signal time series data are scaled according to the signal channel scaling calculation formula to obtain normalized data. The signal channel scaling calculation formula is:

[0017]

[0018] in, represents the signal channel of the multi-channel signal time series data, represents the multi-channel signal time series data, is the normalized data;

[0019] The normalized data is converted into a two-dimensional tensor image, and the conversion formula is as follows:

[0020] ,

[0021] Among them, j represents the time point, i represents the signal channel, Represents a 2D tensor image.

[0022] Further, the predicted two-dimensional tensor image is converted into predicted multi-channel signal time series data based on Min-Max normalization, including:

[0023] According to the predicted two-dimensional tensor image, based on the conversion formula, calculating the normalized data corresponding to the predicted two-dimensional tensor image;

[0024] According to the normalized data corresponding to the predicted two-dimensional tensor image and based on the signal channel scaling calculation formula, the predicted multi-channel signal time series data corresponding to the predicted two-dimensional tensor image is calculated.

[0025] Further, adding Gaussian noise to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis includes:

[0026] A noise adding formula is constructed based on the Bayesian rule, and the noise adding formula is as follows:

[0027] =

[0028] in, represents the original two-dimensional tensor image, represents the two-dimensional tensor image after noise addition at time t-1, represents the two-dimensional tensor image after adding noise at time t, is the variability control parameter of the noise adding process, represents the posterior probability, represents the likelihood function, represents the prior probability, represents the marginal likelihood;

[0029] Gaussian noise is added to the two-dimensional tensor image from the noise adding starting point toward the bottom of the time axis according to the noise adding formula.

[0030] Furthermore, the formula of the multi-channel signal data prediction model is as follows:

[0031] =

[0032] in, is the scaling factor that controls the denoising process at time step t, To control the scaling factor of the denoising process at time step t-1, represents the noise predicted by the multi-channel signal data prediction model at time step t, θ is the model parameter, represents the parameter that controls the amount of noise, represents the additional noise added at time step t.

[0033] Furthermore, the data prediction method also includes inputting the predicted multi-channel signal time series data into a preset CNN-LSTM model to obtain classified data.

[0034] Furthermore, the multi-channel signal is an electroencephalogram signal.

[0035] The second aspect of the present invention further provides a data prediction device based on a diffusion model, comprising:

[0036] An acquisition module, the acquisition module is used to acquire first information and second information, the first information includes historical multi-channel signal time series data, and the second information includes real-time multi-channel signal time series data;

[0037] A construction module, wherein the construction module is used to construct a preprocessing model, wherein the preprocessing model is used to perform Min-Max normalization on the input multi-channel signal time series data, convert the normalized data into a two-dimensional tensor image, wherein the abscissa of the two-dimensional tensor image is the time axis, the ordinate of the two-dimensional tensor image is different signal channels, a noise addition starting point is set on the time axis, and Gaussian noise is added to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis;

[0038] A first preprocessing module, the first preprocessing module is used to input the first information into the preprocessing model to obtain first noisy image data;

[0039] A second preprocessing module, the second preprocessing module is used to input the second information into the preprocessing model to obtain second noisy image data;

[0040] A training module, the training module is used to train a U-net-based diffusion model according to the first noisy image data to obtain a multi-channel signal data prediction model;

[0041] A prediction module, the prediction module is used to input the second noisy image data into the multi-channel signal data prediction model to obtain a predicted two-dimensional tensor image;

[0042] A conversion module is used to convert the predicted two-dimensional tensor image into predicted multi-channel signal time series data based on Min-Max normalization.

[0043] The third aspect of the present invention further provides a data prediction device based on a diffusion model, comprising:

[0044] Memory for storing computer programs;

[0045] A processor is used to implement the steps of the data prediction method based on the diffusion model when executing the computer program.

[0046] The fourth aspect of the present invention further provides a storage medium, which is a readable storage medium and stores a computer program. When the computer program is executed by a processor, the steps of the data prediction method based on the diffusion model are implemented.

[0047] The beneficial effects of the present invention are:

[0048] The present invention realizes the generation and prediction of multi-channel signal data in the future based on the trained multi-channel signal data prediction model with the currently collected multi-channel signal as input, and the predicted multi-channel signal time series data can be used for early warning of the future multi-channel signal status. The present invention can realize the real-time generation and prediction of personalized multi-channel signal time series data for different multi-channel signal time series data situations, and has a good application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A method step diagram of a data prediction method based on a diffusion model in an embodiment of the present application;

[0050] Figure 2 Schematic diagram of the prediction results in the embodiment of the present application (Figure a shows the situation of adding noise at the beginning of the training stage, and Figure b shows the noise recovery achieved based on the multi-channel signal data prediction model after training). DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0052] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0054] In the description of the present invention, it should be understood that the terms "upper", "lower", "inside", "outside", "left", "right", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, or are directions or positional relationships in which the product of the invention is usually placed when in use, or are directions or positional relationships commonly understood by those skilled in the art. These directions or positional relationships are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0055] Furthermore, the terms “first”, “second”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0056] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms such as "setting" and "connection" should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0057] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0058] like Figure 1 As shown, the data prediction method based on the diffusion model in this embodiment includes:

[0059] S100: Acquire first information and second information, wherein the first information includes historical multi-channel signal time series data, and the second information includes real-time multi-channel signal time series data;

[0060] The multi-channel signal processed in this application refers to data with multiple acquisition channels, such as multi-channel RRG, ECG signals, etc., which reflects the signal strength of different channels at the same time point. Different from IQ signals, IQ signals are two orthogonal components representing complex signals in communication, which represent the phase and amplitude information of radio waves.

[0061] S200: constructing a preprocessing model, the preprocessing model is used to perform Min-Max normalization on the input multi-channel signal time series data, converting the normalized data into a two-dimensional tensor image, the abscissa of the two-dimensional tensor image is the time axis, the ordinate of the two-dimensional tensor image is different signal channels, setting a noise addition starting point on the time axis, and adding Gaussian noise to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis;

[0062] The two-dimensional tensor image in the above-mentioned application is different from the time-frequency diagram. The two-dimensional tensor image in the present application is a matrix representation generated after normalizing the multi-channel time series data on the time axis and the signal channel axis. It represents the state of the signal at different time points and channels, and expresses the relationship between time and channel channels.

[0063] Specifically, the signal channel of the multi-channel signal time series data is scaled according to the signal channel scaling calculation formula to obtain normalized data, and the signal channel scaling calculation formula is:

[0064]

[0065] in, represents the signal channel of the multi-channel signal time series data, represents the multi-channel signal time series data, is the normalized data;

[0066] The normalized data is converted into a two-dimensional tensor image, and the conversion formula is as follows:

[0067] ,

[0068] Among them, j represents the time point, i represents the signal channel, Represents a 2D tensor image.

[0069] Specifically, according to the predicted two-dimensional tensor image, based on the conversion formula, the normalized data corresponding to the predicted two-dimensional tensor image is calculated;

[0070] According to the normalized data corresponding to the predicted two-dimensional tensor image and based on the signal channel scaling calculation formula, the predicted multi-channel signal time series data corresponding to the predicted two-dimensional tensor image is calculated.

[0071] Specifically, a noise adding formula is constructed based on the Bayesian rule, and the noise adding formula is as follows:

[0072] =

[0073] in, represents the original two-dimensional tensor image, represents the two-dimensional tensor image after noise addition at time t-1, represents the two-dimensional tensor image after adding noise at time t, is the variability control parameter of the noise adding process, represents the posterior probability, represents the likelihood function, represents the prior probability, represents the marginal likelihood;

[0074] Among them, each Variables that depend on the next time step and the data at the original time point , the variability of the process is determined by the parameter Control. By combining the posterior probability and the likelihood function, the subsequent U-net-based diffusion model can generate the next step of noise-added data based on the current data, so that the noise gradually accumulates. In this way, when training the U-net-based diffusion model, the U-net-based diffusion model not only sees clear signal data, but also learns the performance characteristics of the signal under noise conditions. Noise processing can improve the robustness of the subsequently generated multi-channel signal data prediction model, so that the multi-channel signal data prediction model can process noisy signal data, so as to better adapt to the uncertainty in future data in actual predictions.

[0075] Gaussian noise is added to the two-dimensional tensor image from the noise adding starting point toward the bottom of the time axis according to the noise adding formula.

[0076] S300: Inputting the first information into the preprocessing model to obtain first noisy image data;

[0077] S400: inputting the second information into the preprocessing model to obtain second noisy image data;

[0078] S500: training a diffusion model based on U-net according to the first noisy image data to obtain a multi-channel signal data prediction model;

[0079] The formula of the multi-channel signal data prediction model is as follows:

[0080] =

[0081] in, is the scaling factor that controls the denoising process at time step t, To control the scaling factor of the denoising process at time step t-1, represents the noise predicted by the multi-channel signal data prediction model at time step t, θ is the model parameter, represents the parameter that controls the amount of noise, represents the additional noise added at time step t.

[0082] The multi-channel signal data prediction model is used to gradually transform the data from a noisy state Predict the noise state at the next time step .

[0083] In the diffusion model based on U-net, the diffusion model provides a framework for adding noise and denoising. U-net, as the network structure of the diffusion model, performs the tasks of feature extraction and restoration, so that the diffusion model can gradually denoise and accurately restore the future signal state. In the forward diffusion process of the diffusion model, the encoder part of U-net receives the signal image after adding noise, extracts key information, and the U-net encoder learns the feature representation of the noisy signal, so that the diffusion model can still effectively learn the underlying pattern of the data in a fake environment. In the reverse diffusion process of the diffusion model, U-net passes the low-detail information in the encoding process to the decoder through jump connections, so as to better restore the signal close to the real one.

[0084] The above training enables the diffusion model based on U-net to learn the signal characteristics under noise disturbance. The model structure is U-net, and its encoding-decoding structure is suitable for capturing the detailed changes in time series data. The multi-channel signal data prediction model can predict and generate the future signal sequence after denoising.

[0085] S600: Inputting the second noisy image data into the multi-channel signal data prediction model to obtain a predicted two-dimensional tensor image;

[0086] S700: Convert the predicted two-dimensional tensor image into predicted multi-channel signal time series data based on Min-Max normalization.

[0087] In some embodiments, the data prediction method further includes inputting the predicted multi-channel signal time series data into a preset CNN-LSTM model to obtain classified data.

[0088] In some embodiments, the multi-channel signal time series data is multi-channel EEG signal time series data.

[0089] This embodiment also provides a data prediction device based on a diffusion model, including:

[0090] An acquisition module, the acquisition module is used to acquire first information and second information, the first information includes historical multi-channel signal time series data, and the second information includes real-time multi-channel signal time series data;

[0091] A construction module, wherein the construction module is used to construct a preprocessing model, wherein the preprocessing model is used to perform Min-Max normalization on the input multi-channel signal time series data, convert the normalized data into a two-dimensional tensor image, wherein the abscissa of the two-dimensional tensor image is the time axis, the ordinate of the two-dimensional tensor image is different signal channels, a noise addition starting point is set on the time axis, and Gaussian noise is added to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis;

[0092] A first preprocessing module, the first preprocessing module is used to input the first information into the preprocessing model to obtain first noisy image data;

[0093] A second preprocessing module, the second preprocessing module is used to input the second information into the preprocessing model to obtain second noisy image data;

[0094] A training module, the training module is used to train a U-net-based diffusion model according to the first noisy image data to obtain a multi-channel signal data prediction model;

[0095] A prediction module, the prediction module is used to input the second noisy image data into the multi-channel signal data prediction model to obtain a predicted two-dimensional tensor image;

[0096] A conversion module is used to convert the predicted two-dimensional tensor image into predicted multi-channel signal time series data based on Min-Max normalization.

[0097] This embodiment also provides a data prediction device based on a diffusion model, including:

[0098] Memory for storing computer programs;

[0099] A processor is used to implement the steps of the data prediction method based on the diffusion model when executing the computer program.

[0100] This embodiment further provides a storage medium, which is a readable storage medium. A computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the data prediction method based on the diffusion model are performed.

[0101] The present invention is further described below by taking the application scenario of the present invention in multi-channel EEG signal data prediction as an example:

[0102] like Figure 2 As shown, the ordinate of the two-dimensional tensor image is the time axis, which represents the time development trend from top to bottom, and the abscissa represents different EEG signal channels, such as Fp1, F3, C3, P3, 1, F7, T3, T5, Fc1, Fc5, Cp1, Cp5, F9, Fz, Cz, and Pz signal channels. This data form can fully characterize the spatiotemporal correlation information of multi-channel EEG signals.

[0103] The size of the two-dimensional tensor image is determined according to the input multi-channel EEG signal data and the length of the prediction time. For example, if 16 channels of EEG signal data are input, a 16*16 two-dimensional tensor image can be constructed.

[0104] Taking the 16*16 signal image as an example, during the noise addition process, Gaussian noise is added to the 8*16 area at the bottom of the image;

[0105] The current 8*16 signal image and 8*16 noise are spliced ​​to obtain a two-dimensional tensor image after adding Gaussian noise, which is then input into the multi-channel signal data prediction model to generate a 8*16 two-dimensional tensor image for a period of time in the future. The two-dimensional tensor image is converted into multi-channel EEG signal data, that is, 16-channel multi-channel EEG signal data for the next 8 moments.

[0106] Among them, the preset CNN-LSTM model can also be replaced with a classification algorithm based on graph convolutional neural network, SVM, and random forest algorithm.

[0107] The prediction results of the multi-channel EEG signal data based on the present invention are as follows: Figure 2 Figure a shows the situation of adding noise at the beginning of training, and Figure b shows the situation of noise recovery achieved by the multi-channel signal data prediction model after training, that is, the multi-channel EEG signals at the previous 8 moments can be used to generate the multi-channel EEG signals at the next 8 moments.

[0108] The present invention realizes the generation and prediction of multi-channel signal data in the future based on the trained multi-channel signal data prediction model with the currently collected multi-channel signal as input, and the predicted multi-channel signal time series data can be used for early warning of the future multi-channel signal status. The present invention can realize the real-time generation and prediction of personalized multi-channel signal time series data for different multi-channel signal time series data situations, and has a good application prospect.

[0109] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A data prediction method based on a diffusion model, characterized in that: include: Acquire first information and second information, wherein the first information includes historical multi-channel signal time series data, and the second information includes real-time multi-channel signal time series data; Constructing a preprocessing model, the preprocessing model is used to perform Min-Max normalization on the input multi-channel signal time series data, converting the normalized data into a two-dimensional tensor image, the abscissa of the two-dimensional tensor image is the time axis, the ordinate of the two-dimensional tensor image is different signal channels, setting a noise addition starting point on the time axis, and adding Gaussian noise to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis; Inputting the first information into the preprocessing model to obtain first noisy image data; Inputting the second information into the preprocessing model to obtain second noisy image data; Training a U-net-based diffusion model according to the first noisy image data to obtain a multi-channel signal data prediction model; Inputting the second noisy image data into the multi-channel signal data prediction model to obtain a predicted two-dimensional tensor image; Converting the predicted two-dimensional tensor image into predicted multi-channel signal time series data based on Min-Max normalization; Wherein, the multi-channel signal time series data is multi-channel electroencephalogram signal time series data.

2. The data prediction method based on the diffusion model according to claim 1, characterized in that: Perform Min-Max normalization on the input multi-channel signal time series data, and convert the normalized data into a two-dimensional tensor image, including: The signal channels of the multi-channel signal time series data are scaled according to the signal channel scaling calculation formula to obtain normalized data. The signal channel scaling calculation formula is: ， in, represents the signal channel of the multi-channel signal time series data, represents the multi-channel signal time series data, is the normalized data; The normalized data is converted into a two-dimensional tensor image, and the conversion formula is as follows: ， Among them, j represents the time point, i represents the signal channel, Represents a 2D tensor image.

3. The data prediction method based on the diffusion model according to claim 2 is characterized in that: The predicted two-dimensional tensor image is converted into predicted multi-channel signal time series data based on Min-Max normalization, including: According to the predicted two-dimensional tensor image, based on the conversion formula, calculating the normalized data corresponding to the predicted two-dimensional tensor image; According to the normalized data corresponding to the predicted two-dimensional tensor image and based on the signal channel scaling calculation formula, the predicted multi-channel signal time series data corresponding to the predicted two-dimensional tensor image is calculated.

4. The data prediction method based on the diffusion model according to claim 1, characterized in that: Adding Gaussian noise to the two-dimensional tensor image from the noise adding starting point toward the bottom of the time axis includes: A noise adding formula is constructed based on the Bayesian rule, and the noise adding formula is as follows: ， in, represents the original two-dimensional tensor image, represents the two-dimensional tensor image after noise addition at time t-1, represents the two-dimensional tensor image after noise addition at time t, σ is the variability control parameter of the noise addition process, represents the posterior probability, represents the likelihood function, represents the prior probability, represents the marginal likelihood function; Gaussian noise is added to the two-dimensional tensor image from the noise adding starting point toward the bottom of the time axis according to the noise adding formula.

5. The data prediction method based on the diffusion model according to claim 4 is characterized in that: The formula of the multi-channel signal data prediction model is as follows: ， in, is the scaling factor that controls the denoising process at time step t, To control the scaling factor of the denoising process at time step t-1, represents the noise predicted by the multi-channel signal data prediction model at time step t, θ is the model parameter, represents the parameter that controls the amount of noise, represents the additional noise added at time step t.

6. The data prediction method based on the diffusion model according to claim 1, characterized in that: The data prediction method also includes inputting the predicted multi-channel signal time series data into a preset CNN-LSTM model to obtain classified data.

7. A data prediction device based on a diffusion model, characterized in that: include: An acquisition module, the acquisition module is used to acquire first information and second information, the first information includes historical multi-channel signal time series data, and the second information includes real-time multi-channel signal time series data; A construction module, wherein the construction module is used to construct a preprocessing model, wherein the preprocessing model is used to perform Min-Max normalization on the input multi-channel signal time series data, convert the normalized data into a two-dimensional tensor image, wherein the abscissa of the two-dimensional tensor image is the time axis, the ordinate of the two-dimensional tensor image is different signal channels, a noise addition starting point is set on the time axis, and Gaussian noise is added to the two-dimensional tensor image from the noise addition starting point toward the bottom of the time axis; A first preprocessing module, the first preprocessing module is used to input the first information into the preprocessing model to obtain first noisy image data; A second preprocessing module, the second preprocessing module is used to input the second information into the preprocessing model to obtain second noisy image data; A training module, the training module is used to train a U-net-based diffusion model according to the first noisy image data to obtain a multi-channel signal data prediction model; A prediction module, the prediction module is used to input the second noisy image data into the multi-channel signal data prediction model to obtain a predicted two-dimensional tensor image; A conversion module, the conversion module is used to convert the predicted two-dimensional tensor image into predicted multi-channel signal time series data based on Min-Max normalization; Wherein, the multi-channel signal time series data is multi-channel electroencephalogram signal time series data.

8. A data prediction device based on a diffusion model, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the data prediction method based on the diffusion model as claimed in any one of claims 1 to 6 when executing the computer program.

9. A storage medium, characterized in that: The storage medium is a readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the data prediction method based on the diffusion model according to any one of claims 1 to 6 are implemented.

Images