CN116720158A - Time sequence regression prediction method and system with uncertainty estimation - Google Patents

Abstract

The present invention provides a time series regression prediction method and system with uncertainty estimation. The method is based on a deep learning model of the attention mechanism and can perform time series regression prediction and modeling of its uncertainty. The system includes: data acquisition and preprocessing module, neural network module, training module, and testing module. The time series regression prediction method with uncertainty estimation according to the embodiment of the present invention, by constructing a time series prediction and uncertainty model, based on the attention mechanism, extracts features of the input time series and decodes the output time series at the same time and its prediction uncertainty, which can effectively improve the actual reference value of time series prediction results. Compared with existing uncertainty estimation methods, it has higher execution efficiency and lower time and space cost. The invention is also applied in the discharge modeling of Tokamak 0-dimensional diagnostic physical quantities, making the modeling results more reliable and of practical value.

Description

A time series regression prediction method and system with uncertainty estimation

Technical field

The invention belongs to the field of deep learning and uncertainty measurement, and specifically relates to a time series regression prediction method and system with uncertainty estimation.

Background technique

Deep neural networks are powerful machine learning tools that can be used to handle time series forecasting problems. However, deep neural networks often face many challenges in practical applications, such as insufficient sample size, skewed data set distribution, unknown inputs, etc. These problems will lead to uncertainty in the model. This uncertainty will have a negative impact on deep neural networks. It will have a negative impact on the prediction results and reduce the accuracy and reliability of its predictions. For example, in the financial field, uncertainty may cause models to fail to accurately predict stock prices or exchange rate fluctuations, causing losses to investors; in the transportation field, uncertainty may cause traffic forecasts to be inaccurate, causing transportation planners and It causes inconvenience to passengers; in the field of controllable nuclear fusion, uncertainty may lead to inaccurate discharge predictions, thereby affecting the safe and stable operation of the device, and may even damage the device.

Therefore, the uncertainty measurement of deep neural networks is very important. How to effectively measure and manage the uncertainty in deep learning models is a current research hotspot in the field of deep learning.

A search of existing patents found that there are almost no uncertainty measurement methods for time series. Most of the existing literature uses methods such as Monte Carlo dropout and model integration for uncertainty estimation.

The shortcomings of existing uncertainty measurement methods mainly include the following aspects:

1. High computational cost: These methods require running multiple models or sampling, so the computational cost is relatively high, especially when high-precision predictions or large amounts of data are required;

2. Poor interpretability: The uncertainty measures produced by these methods are usually difficult to interpret. For the application scenarios using these measures, it is difficult to clearly explain the source of prediction uncertainty. This is difficult for some key application scenarios, such as controllable nuclear fusion, Medical care, finance, etc. may not be suitable.

In short, uncertainty measurement is a very important research field in deep learning. Its research results can help us better understand the performance and behavior of deep learning models, thereby improving its effectiveness and reliability in practical applications. Efficient and low-cost modeling of uncertainty estimates for time series remains a challenge.

Contents of the invention

In order to solve the problem of uncertainty in deep neural networks and overcome the problem that existing uncertainty measurement methods have high computational overhead and may affect the accuracy of prediction results, the present invention provides a time series regression prediction method with uncertainty estimation, which The attention mechanism is used to model time series and their prediction uncertainty. At the same time, the model is trained using tokamak discharge experimental data to achieve high-fidelity and rapid diagnostic signal time series and prediction uncertainty modeling, and to realize the verification of 0-dimensional diagnostic physical quantities of the tokamak experimental proposal.

In order to achieve the above objects, the technical solutions adopted by the present invention are:

The deep learning model based on the attention mechanism constructs a time series regression prediction method with uncertainty estimation, which specifically includes the following steps:

S1. Data acquisition and preprocessing: Obtain time series data, including input time series and output time series, and perform resampling and standardized preprocessing operations on the data to establish a data set for model training and testing;

S2. Construct an attention-based time series prediction and uncertainty estimation model: Pass the input time series into the attention-based time series prediction and uncertainty estimation model. The model will extract features of the input time series and Map to the latent space, and obtain the output time series and its uncertainty through the time series output module and uncertainty estimation module respectively;

S3. Training model: First, use the target output time series and the predicted output of the time series output module to calculate the loss according to the loss function L1, and use the error back propagation algorithm to optimize the model parameters; then, based on the existing model parameters, according to The loss function L2 calculates the loss of the prediction output of the actual deviation and uncertainty estimation module, and uses the error back propagation algorithm to optimize the parameters of the uncertainty estimation network in the model, and finally obtains the optimal time series prediction regression with uncertainty estimation. Model;

S4. Test and verify the validity of the model: Input the test set data into the trained time series prediction and uncertainty model, and output the model prediction time series and its prediction uncertainty.

Furthermore, the time series prediction and uncertainty model based on the attention mechanism constructed in step 2 includes a position encoder, an input encoder, a time series output module, a direct uncertainty estimation module, and a linear output layer.

Position encoder: Adds timing information to the data to help the model learn the relative and absolute position information of the data;

Input encoder: perform feature extraction on the input time series, compress it into a semantic vector of specified length, and map it into the latent space;

Time series output module: Decode according to the feature vector mapped to the latent space by the input encoder to obtain the global features of the output time series, and obtain the final expression of the output time series through the linear fully connected layer;

Direct uncertainty estimation module: Decode according to the feature vector mapped by the input encoder into the latent space to obtain the global characteristics of the time series uncertainty, and obtain the final expression of the uncertainty time series through the linear fully connected layer.

Linear output layer: A simple linear mapping of the outputs of the timing output module and the direct uncertainty estimation module to the target output dimension.

Furthermore, the loss functions L1 and L2 use a mask mechanism to calculate effective mean square error loss and weighted mean square error loss based on the effective length of the time series.

Further, the position encoder uses a periodic function to encode the position information of the original tensor to obtain a time series tensor, and combines the time series tensor with the original input tensor, so that the model has the ability to learn time series information.

Further, the direct uncertainty estimation module includes an uncertainty estimation decoder and a linear output layer, where the uncertainty estimation decoder is used to decode and obtain the global uncertainty of the time series according to the mapping of the input features in the latent space. Features; the linear output layer maps the uncertainty global features to the uncertainty output vector space of the sample by adjusting the weight of the uncertainty global features.

Further, the uncertainty estimation decoder includes a multi-head attention mechanism and a fully connected neural network. The multi-head attention mechanism uses linear layers to establish multiple attentions. Each attention focuses on different parts of the input information and then splices them. , can enhance the expression ability of the model; the fully connected neural network is a series connection of multiple linear layers and activation function Relu. Through the composite mapping of simple linear and nonlinear processing units, relatively complex nonlinear processing capabilities can be obtained.

On the other hand, the present invention applies for a time series regression prediction system with uncertainty estimation, which builds a time series prediction and uncertainty model based on the attention mechanism, specifically including the following:

Data acquisition and preprocessing module: used to obtain time series data, including input time series and output time series, and perform resampling and standardized preprocessing operations on the data to establish a data set for model training and testing;

Neural network module: The neural network includes a position encoder, an input encoder, a timing output module, a direct uncertainty estimation module, and a linear output layer; used to obtain the output time series and its uncertainty based on the input time series;

Training module: used to train the neural network using the training data set of the task to obtain a trained neural network;

Test module: used to input the input time series data of the test set into the trained neural network model to obtain the predicted time series and its uncertainty.

In an embodiment of the present invention, an electronic device is provided including a readable storage medium, a central processing unit, and a graphics processor. The readable storage medium stores a computer program that can be run on the central processor and the graphics processor. When the central processor and the graphics processor execute the computer program, the time series prediction is implemented. Method steps.

In an embodiment of the present invention, a readable storage medium is provided. A computer program is stored on the readable storage medium. The computer program executes the time series regression prediction with uncertainty estimation when run by a processor. Steps of the model approach.

The beneficial effects of the present invention are:

1. This deep learning model based on the attention mechanism constructs a time series regression prediction model with uncertainty estimation. By constructing a time series prediction and uncertainty model, based on the attention mechanism, the characteristics of the input time series are evaluated Extract and innovatively decode the output time series and its prediction uncertainty at the same time, which can effectively improve the actual reference value of the time series prediction results. Compared with the existing uncertainty estimation method, it has higher execution efficiency and lower cost. time and space costs.

2. This deep learning model based on the attention mechanism constructs a time series regression prediction model with uncertainty estimation, and is innovatively applied in the discharge modeling of Tokamak's 0-dimensional diagnostic physical quantities, making the modeling results more reliable. , more practical value.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only They are used for the purpose of illustrating preferred embodiments and are not to be construed as limitations of the invention.

Figure 1 is a schematic flow chart of a time series regression prediction method with uncertainty estimation according to an embodiment of the present invention;

Figure 2 is a basic flow chart for obtaining time series data according to the embodiment of the present invention;

Figure 3 is an architecture diagram of an attention-based time series prediction and uncertainty estimation neural network model according to an embodiment of the present invention;

Figure 4 is a basic flow chart of training a neural network to obtain a trained neural network according to an embodiment of the present invention;

Figure 5 is a schematic diagram of the modeling effect of the attention-based time series prediction and uncertainty estimation model according to the embodiment of the present invention;

Figure 6 is a schematic structural diagram of a time series prediction device according to an embodiment of the present invention;

FIG. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

The present invention uses a time series prediction and uncertainty estimation model based on the attention mechanism to model the time series regression problem, so that the model outputs the time series prediction results and its uncertainty at the same time. Therefore, the main contribution of the present invention is the construction of time series prediction and uncertainty estimation models based on the attention mechanism. In addition, in the embodiment, discharge modeling and uncertainty estimation modeling are performed for Tokamak 0-dimensional diagnostic physical quantities.

As shown in Figure 1, the present invention provides a time series regression prediction method with uncertainty, which includes the following steps:

S1. Data acquisition and preprocessing: Obtain time series data, including input time series and output time series, and perform resampling and standardized preprocessing operations on the data to establish a data set for model training and testing;

S2. Construct an attention-based time series prediction and uncertainty estimation model: Pass the input time series into the attention-based time series prediction and uncertainty estimation model. The model will extract features of the input time series and Map to the latent space, and obtain the output time series and its uncertainty through the time series output module and uncertainty estimation module respectively;

S3. Training model: First, use the target output time series and the predicted output of the time series output module to calculate the loss according to the loss function L1, and use the error back propagation algorithm to optimize the model parameters; then, based on the existing model parameters, according to The loss function L2 calculates the loss of the prediction output of the actual deviation and uncertainty estimation module, and uses the error back propagation algorithm to optimize the parameters of the uncertainty estimation network in the model, and finally obtains the optimal time series prediction regression with uncertainty estimation. Model;

S4. Test and verify the validity of the model: Input the test set data into the trained time series prediction and uncertainty model, and output the model prediction time series and its prediction uncertainty.

Specifically, as shown in Figure 2, the data acquisition and preprocessing process described in step S1 includes the following steps:

S11. Obtain original time series data. Taking the tokamak dynamics system as an example, the original time series data is read from the MDSplus database and stored in different HDF5 files according to the experimental order for subsequent use;

S12. Use a fixed sampling rate to resample the original time series data to obtain a resampled time series, thereby reducing computing costs while retaining sufficient data information;

S13. Use the Z-Score standardization method to standardize the resampled time series data. After data processing, it conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1;

S14. Based on the above preprocessed time series data, divide it into two training sets and one test set according to the ratio of 4:4:2.

Specifically, as shown in Figure 3, the time series regression prediction model with uncertainty estimation includes a position encoder, an input encoder, a time series output module, a direct uncertainty estimation module, and a linear output layer.

The position encoder uses a periodic function to encode the position information of the original tensor to obtain a time series tensor, and combines the time series tensor with the original input tensor, so that the model has the ability to learn time series relative position information and absolute position information. .

The input encoder performs feature extraction on the input time series, compresses it into a semantic vector of specified length, and maps it into the latent space.

The time series output module decodes according to the feature vector mapped to the latent space by the input encoder to obtain the global features of the output diagnostic signal, and obtains the final expression of the time series of the diagnostic signal through the linear fully connected layer.

The direct uncertainty estimation module includes an uncertainty estimation decoder and a regression output layer.

Among them, the uncertainty estimation decoder based on the attention mechanism is composed of a multi-head attention mechanism and a fully connected neural network, and is used to decode and obtain the global characteristics of the time series uncertainty according to the mapping of input features in the latent space. The multi-head attention mechanism refers to splitting the input vector into multiple heads, performing attention calculations on each head, and finally splicing the results of all heads together to obtain the final output vector. Through this method Extract task-related information, increase the model's attention to different aspects of information, and improve the model's generalization ability and effect. Each head is calculated using the following function,

Among them, Q, K, and V are the inputs of the multi-head attention mechanism. In the multi-head self-attention mechanism, Q, K, and V are the same; Softmax is an activation function that can normalize a numerical vector into a probability distribution vector. And the sum of each probability is 1; T represents the transposition operation; d _model is the dimension of the position vector, which is the same as the hidden state dimension value of the entire model; h represents the number of heads in the multi-head attention mechanism, i∈[1, h]; W _i ^Q , W _i ^K , and W _i ^V are the weight matrices of Q, K, and V respectively.

The fully connected neural network uses the following function to perform relatively complex nonlinear processing on the output of the attention mechanism,

Among them, x is the input of the fully connected neural network; Relu is the linear rectification function in the activation function; W ₁ and W ₂ are the weight parameters of the two linear layers in the fully connected neural network; T represents the transpose operation; b ₁ and b ₂ is the bias of two linear layers in a fully connected neural network.

The linear output layer is a simple one-dimensional linear layer. By adjusting the weight of the uncertainty global feature, it is mapped to the uncertainty output vector space of the sample, and a simple linear mapping is performed to align the model output tensor and 0-dimensional diagnostic data tensor.

Specifically, as shown in Figure 4, the model training process in step S3 includes the following steps:

S31. Randomly initialize the weights and bias parameters of the neural network;

S32. Initialize the neural network hyperparameters and the stochastic gradient descent optimizer. The neural network hyperparameters include batch size, learning rate, and number of iterations;

S33. Use the target output time series and the predicted output of the time series output module to calculate the loss according to the loss function L1, and use the error back propagation algorithm to optimize the model parameters;

S34. Based on the existing model parameters, calculate the loss of the actual deviation and uncertainty estimation module prediction output according to the loss function L2, and use the error back propagation algorithm to optimize the parameters of the uncertainty estimation network in the model;

The loss functions L1 and L2 use a mask mechanism to calculate the effective mean square error loss VMSE and the weighted mean square error loss VWMSE according to the effective length of the time series;

The calculation formula of the effective mean square error loss function VMSE is as follows:

Among them, n _V is the effective length of the time series; W _V is a matrix containing only 0 and 1, used for time series

The mean square error of ^ extracts the effective part; y is the real experimental data; y is the model prediction output.

The calculation formula of the effective weighted mean square error loss VWMSE is as follows:

Among them, n _{V_in} is the number of time slices that are valid and within the prediction uncertainty coverage; n _{V_out} is the number of time slices that are valid and outside the prediction uncertainty coverage; y _{V_in} is within the prediction uncertainty coverage real experimental data; is the model prediction output within the coverage range of prediction uncertainty; y _{V_out} is the real experimental data within the coverage range of prediction uncertainty;/> is the model prediction output within the prediction uncertainty coverage; λ is a parameter used to balance the importance of coverage and uncertainty width; μ is the predefined target uncertainty coverage; PUCP is the prediction uncertainty Coverage, the PUCP calculation formula is as follows:

Among them, N is the time series length, a _i is a binary value, and the calculation formula is as follows:

where _yi is the target value, is the point prediction value, and uncertainty is the uncertainty.

S35. Finally, the optimal time series prediction regression model with uncertainty estimation is obtained.

The time series prediction method provided in this embodiment can be applied to a variety of tasks. Below we will discuss the tokamak discharge modeling task as an example.

The discharge modeling and uncertainty estimation modeling model for tokamak 0-dimensional diagnostic physical quantities are implemented as follows: the input data of the discharge modeling and uncertainty estimation modeling model for tokamak 0-dimensional diagnostic physical quantities It is a time series of 92 control signals, including plasma current feedforward, poloidal magnetic field coil current, toroidal magnetic field, power of low-clutter current drive and heating system, neutral beam injection system, ion cyclotron resonance heating system, and electron cyclotron resonance. Heating/current drive system, gas blowing system, supersonic molecular beam injection, particle injection system, plasma shape feedforward; the above-mentioned discharge modeling and uncertainty estimation modeling model for Tokamak 0-dimensional diagnostic physical quantities The modeling target is 11 diagnostic signals, including the actual plasma current I _p , the average electron density of the tokamak magnetic axis plasma n _e , the plasma energy storage W _mhd , the tokamak ring voltage V _loop , and the tokamak normalization The specific magnetic pressure β _n , the toroidal specific pressure β _t , the poloidal specific pressure β _p , the elongation ratio k, the internal inductance li, the safety factor q ₀ , and the safety factor q ₉₅ on the 95% flux surface. As shown in Figure 2, the stacked layer sizes are: D0, D1, and D2 are all 6, and the hidden layer size d _model is 512.

Under the premise that the target uncertainty coverage μ is set to 0.9 and the parameter λ is 4, the discharge modeling and uncertainty estimation modeling results for the tokamak 0-dimensional diagnostic physical quantities are shown in Figure 5. Gun #73873 was selected as test data. The discharge time of this gun exceeds 70s and the sequence length exceeds 7×10 ⁴ . The specific process is: first, set the input amount of the tokamak control system. In this embodiment, the control system source file corresponding to the #73873 gun is directly called, and then the actual actuator input signal of the #73873 gun is used as the input of the system. Next, convert the data into Tensor type and load it into the GPU, and load the trained deep learning model into the GPU. Use the trained deep learning model to calculate the input data to obtain 11 0-dimensional diagnostic physical quantities and their prediction uncertainties. Modeling results, and finally visualizing the data.

In the Hefei "Eastern Super Ring (EAST)" tokamak device, a total of 932 shots were used as a test set to test the performance of the model on the overall test set. In this invention, the target uncertainty coverage μ is set to 0.9. Under the premise that λ is set to 4, the average prediction uncertainty coverage PUCP in 0-dimensional diagnostic physical quantity discharge modeling reaches 90.891%, which is in line with expectations. The good performance on the test set can illustrate that the modeling results of the present invention are accurate and reliable and have practical value.

On the other hand, as shown in Figure 6, the present invention is a time series regression prediction system with uncertainty estimation. The device includes the following modules:

Data acquisition and preprocessing module: used to obtain time series data, including input time series and output time series, and perform resampling and standardized preprocessing operations on the data to establish a data set for model training and testing;

Neural network module: The neural network includes a position encoder, an input encoder, a timing output module, a direct uncertainty estimation module, and a linear output layer; used to obtain the output time series and its uncertainty based on the input time series;

Training module: used to train the neural network using the training data set of the task to obtain a trained neural network;

Test module: used to input the input time series data of the test set into the trained neural network model to obtain the predicted time series and its uncertainty.

As shown in Figure 7, the present invention also provides an electronic device, including a readable storage medium, a core processor, and a graphics processor. A computer program stored on the readable storage medium and executable on the core processor and graphics processor. When the core processor and graphics processor execute the computer program, they will implement what is described in this embodiment. The steps of a time series regression forecasting method with uncertainty estimation.

Specifically, the readable storage medium is a general storage medium, such as a removable disk, a hard disk, an optical disk, etc. A computer program is stored on the storage medium. When the computer program is executed by the processor, the above-mentioned time estimation with uncertainty is realized. Each process of the sequence prediction method embodiment can achieve the same technical effect. To avoid duplication, it will not be described again here.

Although preferred embodiments of the embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once the basic inventive concepts are understood. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

Claims (17)

1. A time series regression prediction method with uncertainty estimation, which is characterized by constructing a time series prediction and uncertainty model based on the attention mechanism, specifically including the following steps: S1. Data acquisition and preprocessing: Obtain time series data, including input time series and output time series, and perform resampling, standardization, and discharge category label preprocessing operations on the data to establish a data set for model training and testing; S2. Construct an attention-based time series prediction and uncertainty estimation model: Pass the input time series into the attention-based time series prediction and uncertainty estimation model. The model will extract features of the input time series and Map to the latent space, and obtain the output time series and its uncertainty through the time series output module and uncertainty estimation module respectively; S3. Training model: First, use the target output time series and the predicted output of the time series output module to calculate the loss according to the loss function L1, and use the error back propagation algorithm to optimize the model parameters; then, based on the existing model parameters, according to The loss function L2 calculates the loss of the prediction output of the actual deviation and uncertainty estimation module, and uses the error back propagation algorithm to optimize the parameters of the uncertainty estimation network in the model, and finally obtains the optimal time series prediction regression with uncertainty estimation. Model; S4. Test and verify the validity of the model: Input the test set data into the trained time series prediction and uncertainty model, and output the model prediction time series and its prediction uncertainty. 2. The time series regression prediction method with uncertainty estimation according to claim 1, characterized in that the data acquisition and preprocessing process in S1 includes the following steps: S11. Obtain the original time series data. The original time series data is read from the MDSplus database and stored in different HDF5 files according to the experimental order for subsequent use; S12. Use a fixed sampling rate to resample the original time series data to obtain a resampled time series, thereby reducing computing costs while retaining sufficient data information; S13. Use the Z-Score standardization method to standardize the resampled time series data. After data processing, it conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1; S14. Based on the above preprocessed time series data, divide it into two training sets and one test set according to the ratio of 4:4:2. 3. The time series regression prediction method with uncertainty estimation according to claim 1, characterized in that the attention mechanism-based time series prediction and uncertainty model constructed in S2 includes: The position encoder will use a periodic function to encode the position information of the original tensor to obtain a time series tensor, and combine the time series tensor with the original input tensor, so that the model has the ability to learn the relative position information and absolute position information of the time series; The input encoder performs feature extraction on the input time series, compresses it into a semantic vector of specified length, and maps it into the latent space; The time series output module decodes the feature vector mapped to the latent space by the input encoder to obtain the global features of the output time series, and obtains the final expression of the output time series through the linear fully connected layer; The direct uncertainty estimation module decodes the feature vector mapped to the latent space by the input encoder to obtain the global characteristics of the time series uncertainty, and obtains the final expression of the uncertainty time series through the linear fully connected layer; The linear output layer performs a simple linear mapping of the outputs of the timing output module and the direct uncertainty estimation module to the target output dimension. 4. The time series regression prediction method with uncertainty estimation according to claim 1, characterized in that the loss functions L1 and L2 in the S3 use a mask mechanism to calculate an effective mean according to the effective length of the time series. square error loss and weighted mean square error loss. 5. The time series regression prediction method with uncertainty estimation according to claim 2, characterized in that the position encoder uses a periodic function to encode the position information of the original tensor to obtain the time series tensor, and The time series tensor is combined with the original input tensor to give the model the ability to learn time series information. 6. The time series regression prediction method with uncertainty estimation according to claim 2, characterized in that the direct uncertainty estimation module includes an uncertainty estimation decoder and a linear output layer, wherein the uncertainty estimation The decoder is used to decode and obtain the global features of the time series uncertainty based on the mapping of the input features in the latent space; the linear output layer maps the global features of the uncertainty into the uncertainty output vector space of the sample by adjusting the weight of the global features of the uncertainty. . 7. The time series regression prediction method with uncertainty estimation according to claim 5, characterized in that the uncertainty estimation decoder is composed of a multi-head attention mechanism and a fully connected neural network, and the multi-head attention mechanism uses The linear layer establishes multiple attentions, each paying attention to different parts of the input information, and then splicing them together, which can enhance the expression ability of the model; the fully connected neural network is a series connection of multiple linear layers and the activation function Relu. Through simple linear Composite mapping with the nonlinear processing unit can obtain relatively complex nonlinear processing capabilities. 8. The time series regression prediction method with uncertainty estimation according to claim 3, characterized in that the position encoder uses sine and cosine functions to add timing information to the original vector to help the model learn the relative relationship of the data. and absolute location information. 9. The time series regression prediction method with uncertainty estimation according to claim 2, characterized in that, The uncertainty estimation decoder based on the attention mechanism is composed of a multi-head attention mechanism and a fully connected neural network; the multi-head attention mechanism refers to splitting the input vector into multiple heads and analyzing each head. Attention calculation is performed on the head, and finally the results of all heads are spliced together to obtain the final output vector. Through this method, task-related information is extracted, the model's attention to different aspects of information is increased, and the generalization ability and effect of the model are improved. , where each head is calculated using the following function, Among them, Q, K, and V are the inputs of the multi-head attention mechanism. In the multi-head self-attention mechanism, Q, K, and V are the same; Softmax is an activation function that can normalize a numerical vector into a probability distribution vector. And the sum of each probability is 1; T represents the transposition operation; d _model is the dimension of the position vector, which is the same as the hidden state dimension value of the entire model; h represents the number of heads in the multi-head attention mechanism, i∈[1, h]; W _i ^Q , W _i ^K , and W _i ^V are the weight matrices of Q, K, and V respectively. 10. The time series regression prediction method with uncertainty estimation according to claim 9, characterized in that the fully connected neural network uses the following function to perform relatively complex nonlinear processing on the output of the attention mechanism, Among them, x is the input of the fully connected neural network; Relu is the linear rectification function in the activation function; W ₁ and W ₂ are the weight parameters of the two linear layers in the fully connected neural network; T represents the transpose operation; b ₁ and b ₂ is the bias of two linear layers in a fully connected neural network. 11. The time series regression prediction method with uncertainty estimation according to claim 10, characterized in that the linear output layer is a simple one-dimensional linear layer, and a simple linear mapping is performed to align the model output tensor and 0-dimensional diagnostic data tensor. 12. The time series regression prediction method with uncertainty estimation according to claim 1, characterized in that the model training process in step S3 includes the following steps: S31. Randomly initialize the weights and bias parameters of the neural network; S32. Initialize the neural network hyperparameters and the stochastic gradient descent optimizer. The neural network hyperparameters include batch size, learning rate, and number of iterations; S33. Use the target output time series and the predicted output of the time series output module to calculate the loss according to the loss function L1, and use the error back propagation algorithm to optimize the model parameters; S34. Based on the existing model parameters, calculate the loss of the actual deviation and uncertainty estimation module prediction output according to the loss function L2, and use the error back propagation algorithm to optimize the parameters of the uncertainty estimation network in the model; S35. Finally, the optimal time series prediction regression model with uncertainty estimation is obtained. 13. The time series regression prediction method with uncertainty estimation according to claim 4, characterized in that the loss functions L1 and L2 use a mask mechanism to calculate the effective mean square error loss VMSE according to the effective length of the time series. And the weighted mean square error loss VWMSE specifically includes: The calculation formula of the effective mean square error loss function VMSE is as follows: Among them, n _V is the effective length of the time series; W _V is a matrix containing only 0 and 1, used to extract the effective part for the mean square error of the time series; y is the real experimental data; is the model prediction output; The calculation formula of the effective weighted mean square error loss VWMSE is as follows: Among them, n _{V_in} is the number of time slices that are valid and within the prediction uncertainty coverage; n _{V_out} is the number of time slices that are valid and outside the prediction uncertainty coverage; y _{V_in} is within the prediction uncertainty coverage real experimental data; is the model prediction output within the coverage range of prediction uncertainty; y _{V_out} is the real experimental data within the coverage range of prediction uncertainty;/> is the model prediction output within the prediction uncertainty coverage; λ is a parameter used to balance the importance of coverage and uncertainty width; μ is the predefined target uncertainty coverage; PUCP is the prediction uncertainty Coverage. 14. The time series regression prediction method with uncertainty estimation according to claim 13, characterized in that the PUCP calculation formula is as follows: Among them, N is the time series length, ai is a binary value, and the calculation formula is as follows: Among them, _yi is the target value, is the point prediction value, and uncertainty is the uncertainty. 15. A time series regression prediction system with uncertainty estimation, characterized in that the system includes the following modules: Data acquisition and preprocessing module: used to obtain time series data, including input time series and output time series, and perform resampling and standardized preprocessing operations on the data to establish a data set for model training and testing; Neural network module: The neural network includes a position encoder, an input encoder, a timing output module, a direct uncertainty estimation module, and a linear output layer; used to obtain the output time series and its uncertainty based on the input time series; Training module: used to train the neural network using the training data set of the task to obtain a trained neural network; Test module: used to input the input time series data of the test set into the trained neural network model to obtain the predicted time series and its uncertainty. 16. An electronic device, characterized by comprising a readable storage medium, a central processing unit, and a graphics processor. Wherein, the readable storage medium is used to store one or more computer programs. When the processor executes the computer program, the time series regression with uncertainty estimation as described in any one of claims 1 to 14 is implemented. Steps in the forecasting method. 17. A readable storage medium, characterized in that a computer program is stored on the readable storage medium, and when the computer program is run by a processor, the computer program executes the method described in any one of claims 1 to 14 with uncertainty. Steps in the time series regression forecasting method of degree estimation.