A geomagnetic disturbance index intelligent prediction and uncertainty evaluation method and system

Abstract

The present application relates to geomagnetic disturbance index prediction technology field, provide a kind of geomagnetic disturbance index intelligent prediction and uncertainty evaluation method and system.The method includes, obtaining space weather observation data during geomagnetic storm event;Based on space weather observation data, using iTransformer model, at random inactivation layer is in non-activated state, obtain the SYM-H index point prediction result of future multiple time steps;Based on space weather observation data, using iTransformer model, at random inactivation layer is in activated state, carry out multiple random forward propagation, obtain multiple prediction results;According to the empirical quantile of multiple prediction results, construct original prediction interval;Calibration set is constructed, according to calibration set, select the lower bound correction and upper bound correction corresponding to empirical quantile, respectively, the lower boundary and upper boundary of original prediction interval are modified, to obtain the calibration prediction interval.

Description

Technical Field

[0001] This invention relates to the field of geomagnetic disturbance index prediction technology, and in particular to a method and system for intelligent prediction and uncertainty assessment of geomagnetic disturbance index. Background Technology

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

[0003] As solar activity intensifies, solar eruptions such as solar flares and coronal mass ejections cause dramatic changes in the solar wind and interplanetary magnetic field, which interact with the Earth's magnetosphere, triggering geomagnetic storms. Geomagnetic storms cause significant disturbances to the Earth's magnetosphere and ionosphere, potentially impacting satellite operation, communication and navigation systems, and the safety of power systems. Particularly during strong geomagnetic storms, rapid changes in the geomagnetic field can induce geomagnetic currents in long-distance conductors such as power transmission networks, affecting the safe and stable operation of high-voltage power transmission equipment. Therefore, accurate prediction of geomagnetic storm evolution is crucial for space weather monitoring and early warning, and risk prevention and control of related engineering systems.

[0004] Geomagnetic indices can be used to describe the intensity and variation of geomagnetic activity. Among them, the SYM-H index has a high temporal resolution and can reflect the changes in loop current and geomagnetic disturbances during geomagnetic storms with relatively fine detail, making it suitable for predicting geomagnetic disturbances on short timescales. However, the variation of the SYM-H index is influenced by multiple factors, including solar wind parameters, interplanetary magnetic field parameters, and the Earth's magnetosphere response. Complex nonlinear relationships exist between these different physical variables, making accurate prediction of the SYM-H index still quite challenging.

[0005] In recent years, machine learning and deep learning methods have been increasingly applied to geomagnetic index prediction. Existing methods typically use solar wind parameters, interplanetary magnetic field parameters, and historical geomagnetic indices as inputs, and employ techniques such as Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Gradient Boosting Models (GBM) to predict future geomagnetic indices. These data-driven methods can learn the nonlinear relationship between input parameters and geomagnetic response from historical observation data, and have found some application in short-term prediction of geomagnetic disturbance indices.

[0006] However, existing methods for predicting geomagnetic disturbance indices still have shortcomings. Some data-driven methods focus on time-series feature extraction, failing to adequately consider the correlation between solar wind parameters, interplanetary magnetic field components, and historical geomagnetic indices, and thus struggling to fully utilize the dependencies between different physical variables. Furthermore, existing methods often employ deterministic prediction approaches, typically providing only a single prediction result without describing the uncertainty of the prediction. When space weather conditions change rapidly, a single prediction value is insufficient to reflect the possible range of changes in the prediction result, thus failing to meet the information requirements for risk assessment and early warning decisions. Summary of the Invention

[0007] To address the shortcomings of existing geomagnetic disturbance index prediction methods in terms of insufficient multivariate dependency information mining and inadequate uncertainty assessment capabilities, this invention provides an intelligent prediction and uncertainty assessment method and system for geomagnetic disturbance index. By extracting the dependency features between solar wind parameters, interplanetary magnetic field parameters, and historical SYM-H indexes, the method achieves short-term prediction of the SYM-H index and obtains the uncertainty prediction interval corresponding to the prediction results.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for intelligent prediction and uncertainty assessment of geomagnetic disturbance index.

[0009] A method for intelligent prediction and uncertainty assessment of geomagnetic disturbance index, comprising: Acquire space weather observation data during geomagnetic storm events, including: interplanetary magnetic field parameters, solar wind speed, solar wind proton density, solar wind dynamic pressure, twilight electric field, and historical SYM-H index; Based on space weather observation data during geomagnetic storm events, the iTransformer model was used to obtain SYM-H index point prediction results for multiple future time steps in the inactive state of the random deactivation layer. Based on space weather observation data during geomagnetic storm events, the iTransformer model is used, which is activated in the random deactivation layer and undergoes multiple random forward propagations to obtain multiple prediction results. The original prediction interval is constructed based on the empirical quantiles of the multiple prediction results. A calibration set is constructed, and the lower and upper bound corrections corresponding to the empirical quantiles are selected based on the calibration set to correct the lower and upper boundaries of the original prediction interval, respectively, to obtain the calibrated prediction interval.

[0010] Furthermore, after acquiring space weather observation data during geomagnetic storm events, the process also includes: time-aligning the space weather observation data during geomagnetic storm events, using linear interpolation to fill in missing observations, and then standardizing the data to obtain a multivariate time series input matrix.

[0011] Furthermore, after obtaining the multivariate time series input matrix, the process also includes: taking the multivariate time series input matrix, obtaining the variable center representation through matrix transpose, and then inputting it into the iTransformer model.

[0012] Furthermore, the process of adopting the iTransformer model includes: A variable embedding module is used to linearly map each spatial weather variable, converting the input sequences of different physical variables into a hidden feature representation of a unified dimension; Based on hidden feature representation, a variable attention encoding module is used to extract observation features of multiple space weather variables and the dependency features between different space weather variables, and then fuse them to output fused features; Based on the fusion features, a prediction output module is used to output the prediction results of the SYM-H index points for multiple future time steps.

[0013] Further, the step of selecting the lower bound correction and upper bound correction corresponding to the empirical quantile based on the calibration set includes: The calibration set includes a lower bound set of deviations and an upper bound set of deviations; Let the number of deviation samples contained in each deviation set be... Then, conservative empirical quantiles for interval calibration are determined based on the target coverage level. :

[0014] in, This represents the floor function; From the lower bound deviation set respectively and upper bound deviation set Selecting from the empirical quantiles The empirical quantiles are used as lower bound corrections. and upper bound correction :

[0015]

[0016] in, This represents the function for calculating empirical quantiles. Indicates empirical quantiles. This represents the lower bound correction amount determined by the empirical distribution of the lower bound deviation. This represents the upper bound correction amount determined by the empirical distribution of the upper bound deviation.

[0017] Furthermore, the correction of the lower and upper boundaries of the original prediction interval includes: The lower bound of the calibrated prediction interval is expressed as:

[0018] The upper bound of the calibrated prediction interval is expressed as:

[0019] No. The input sample at the th th The final calibration prediction interval over each prediction step size is represented as follows:

[0020] in, and These represent the lower and upper bounds of the calibrated prediction interval, respectively. This represents the lower boundary of the original prediction interval. This represents the upper boundary of the original prediction interval. This represents the lower bound correction amount determined by the empirical distribution of the lower bound deviation. This represents the upper bound correction amount determined by the empirical distribution of the upper bound deviation.

[0021] A second aspect of the present invention provides an intelligent prediction and uncertainty assessment system for geomagnetic disturbance index.

[0022] A smart prediction and uncertainty assessment system for geomagnetic disturbance index includes: The data acquisition module is configured to acquire space weather observation data during geomagnetic storm events, including: interplanetary magnetic field parameters, solar wind speed, solar wind proton density, solar wind dynamic pressure, twilight electric field, and historical SYM-H index. The SYM-H index point prediction module is configured to: based on space weather observation data during geomagnetic storm events, use the iTransformer model, and be in an inactive state in the random deactivation layer to obtain SYM-H index point prediction results for multiple future time steps; The uncertainty assessment module is configured to: use the iTransformer model based on space weather observation data during geomagnetic storm events, activate the model in the random deactivation layer, perform multiple random forward propagations to obtain multiple prediction results; construct the original prediction interval based on the empirical quantiles of the multiple prediction results; construct a calibration set, select the lower bound correction amount and upper bound correction amount corresponding to the empirical quantiles based on the calibration set, and correct the lower boundary and upper boundary of the original prediction interval respectively to obtain the calibrated prediction interval.

[0023] A third aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index as described in the first aspect above.

[0024] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and to execute steps in the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index as described in the first aspect above.

[0025] The fifth aspect of the present invention provides a computer program product or computer program.

[0026] This invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index as described in the first aspect above.

[0027] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves short-term prediction of the SYM-H index by extracting the dependency features between solar wind parameters, interplanetary magnetic field parameters and historical SYM-H index, and obtains the uncertainty prediction range corresponding to the prediction results.

[0028] This invention directly utilizes the real calibration set to statistically analyze the deviations of the lower and upper bounds from the true values, obtaining two correction values ​​to correct the actual deviations at each boundary, thus yielding an asymmetric prediction interval. This eliminates the need for manually setting the search range and step size, and the upper and lower boundaries can be adjusted independently, better reflecting the potential asymmetry in actual prediction errors. Attached Figure Description

[0029] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0030] Figure 1 This is a flowchart illustrating the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index according to an embodiment of the present invention; Figure 2 This is a flowchart of another embodiment of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index shown in the embodiments of the present invention; Figure 3This is a comparison diagram of the time-centered and variable-centered representations of multivariate time series according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the iTransformer prediction network structure shown in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the uncertainty prediction interval of the geomagnetic storm event 34 in the 1-hour and 2-hour prediction tasks, as shown in an embodiment of the present invention. Figure 6 This is a structural diagram of the intelligent prediction and uncertainty assessment system for geomagnetic disturbance index shown in an embodiment of the present invention; Figure 7 This is a structural diagram of a computer device shown in an embodiment of the present invention. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0033] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0034] Figure 1 This is a flowchart illustrating the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index according to an embodiment of the present invention; see reference. Figure 1 The method includes: Acquire space weather observation data during geomagnetic storm events, including: interplanetary magnetic field parameters, solar wind speed, solar wind proton density, solar wind dynamic pressure, twilight electric field, and historical SYM-H index; Based on space weather observation data during geomagnetic storm events, the iTransformer model was used to obtain SYM-H index point prediction results for multiple future time steps in the inactive state of the random deactivation layer. Based on space weather observation data during geomagnetic storm events, the iTransformer model is used, which is activated in the random deactivation layer and undergoes multiple random forward propagations to obtain multiple prediction results. The original prediction interval is constructed based on the empirical quantiles of the multiple prediction results. A calibration set is constructed, and the lower and upper bound corrections corresponding to the empirical quantiles are selected based on the calibration set to correct the lower and upper boundaries of the original prediction interval, respectively, to obtain the calibrated prediction interval.

[0035] This invention achieves short-term prediction of the SYM-H index by extracting the dependency features between solar wind parameters, interplanetary magnetic field parameters and historical SYM-H index, and obtains the uncertainty prediction range corresponding to the prediction results.

[0036] Figure 2 This is a flowchart of another embodiment of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index shown in this invention; the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index includes the following steps: Step 1: Data Input and Sample Construction, focusing on the interplanetary magnetic field component B. x B y B z， Solar wind speed V proton density N Solar wind dynamic pressure Pdyn Electric field E at dawn and dusk y Preprocessing was performed on multi-parameter space weather observation data, including historical SYM-H index data, and a sliding window was used to construct prediction samples. Specifically, this invention acquires space weather observation data during geomagnetic storm events and constructs a multivariate time series input for predicting the geomagnetic disturbance index. For example... Figure 2 As shown in the input section on the left, the input data includes: interplanetary magnetic field parameter B. x B y and B z Solar wind speed V Solar wind proton density N and solar wind dynamics Pdyn Electric field E at dawn and dusk y And the historical SYM-H index. The above eight input variables respectively characterize the solar wind and interplanetary magnetic field driving conditions and the Earth's magnetosphere response state during geomagnetic storm processes.

[0037] The acquired space weather observation data were time-aligned, and missing observations were filled in using linear interpolation. Subsequently, the input variables were standardized to reduce the impact of differences in the units and numerical ranges of different physical variables on model training.

[0038] A sliding window approach is used to construct prediction samples. For any prediction time...t Selected previously L Multi-parameter spatial weather observation data at each time step are used as model input to form a multivariate time series input matrix:

[0039] in, Indicates the number of input variables. Indicates the length of the historical time window. Indicates the first The input variables at the prediction time Before The historical sequence within a time step.

[0040] Predicting time After the future The SYM-H index at each time step is used as the prediction target to construct the target sequence:

[0041] in, This indicates the prediction step size. Therefore, using... As model input, As the prediction target, input-output sample pairs are formed for model training and prediction.

[0042] Step 2: Variable center representation and feature extraction. The complete sequence of each input variable within the historical time window is labeled as an independent variable. After variable embedding, it is input into the iTransformer encoder. The dependency features between solar wind parameters, interplanetary magnetic field parameters and historical SYM-H index are extracted through the multi-head self-attention mechanism in the variable dimension. like Figure 3 As shown, this invention reorganizes multivariate time series inputs using a variable-centric representation. Unlike the traditional Transformer method, which combines multiple variables at the same point in time to form a time stamp and performs attention calculations along the time dimension, this invention treats the complete historical sequence of each physical variable as an independent variable token. Therefore, the object of self-attention calculation is transformed from different time locations into different spatial weather physical variables, enabling the model to model the dependencies between solar wind parameters, interplanetary magnetic field parameters, and historical geomagnetic responses along the variable dimension.

[0043] Specifically, the original input matrix:

[0044] The central representation of the variables is obtained by matrix transpose:

[0045] in,L Indicates the number of historical time steps; M This represents the number of spatial weather input variables. After transformation, B... x B y B z , V , N , Pdyn E y The historical SYM-H values ​​correspond to independent variables, tokens. This variable-centric representation enables the model to extract the dependency features between different spatial weather variables along the variable dimension.

[0046] Step 3: Deterministic prediction. Input the fused features output by the encoder into the prediction mapping layer, and output the SYM-H exponent point prediction results for multiple future time steps in deterministic prediction mode. like Figure 4 As shown, the iTransformer prediction network constructed in this invention includes a variable embedding module, an encoder module, and a prediction output module.

[0047] (1) Variable Embedding Module First, a linear mapping is performed on each variable token to transform the historical sequences of different physical variables into a unified-dimensional hidden feature representation:

[0048] in, Indicates the first A historical sequence of spatial weather variables, This represents the characteristic vector of the corresponding variable. and These represent the weight matrix and bias term of the linear mapping, respectively. The variable embedding module maps different physical variables to a unified feature space, providing a foundation for subsequent extraction of inter-variable dependency features.

[0049] (2) Variable attention encoding module The results of embedding all variables are input into the encoder module. The encoder module includes a multi-head self-attention layer, a residual connection structure, a normalization layer, a feedforward neural network layer, and a random deactivation layer. Unlike traditional Transformers that perform attention calculations along the time dimension, this invention uses the labels of each space weather variable as the objects of attention calculation and extracts the dependency features between different space weather variables along the variable dimension. Through this structure, the nonlinear coupling relationship between solar wind parameters, interplanetary magnetic field parameters, and historical SYM-H indices can be modeled, forming a fused feature representation for predicting geomagnetic disturbance indices.

[0050] (3) Prediction output module After encoding by the encoder, the resulting fused features are input into the prediction mapping layer. The prediction mapping layer outputs a SYM-H exponent prediction sequence over a future time range:

[0051] in, H This indicates the prediction step size.

[0052] Step 4: Uncertainty Assessment and Prediction Interval Calibration. In the uncertainty assessment mode, the random deactivation layer remains active. Multiple random forward propagations are performed on the same input sample, and the original prediction interval is constructed based on the empirical quantiles of the multiple prediction results. In this embodiment, a validation set is selected as the calibration set. The validation set contains geomagnetic storm events of different intensities, covering the SYM-H index variation characteristics under different geomagnetic disturbance states. Therefore, it can be used to statistically analyze the deviation between the original prediction interval and the actual observations, and thereby determine the calibration correction amount for the prediction interval. Further using the validation set as calibration data, the deviations between the actual observations and the lower and upper boundaries of the original prediction interval are calculated respectively, determining the lower and upper boundary correction amounts. The lower and upper boundaries of the original prediction interval are then corrected to obtain the calibrated prediction interval. Finally, the SYM-H index point prediction results and their corresponding calibrated prediction intervals are output.

[0053] Specifically, to quantify the uncertainty of the SYM-H index prediction results, this invention introduces an uncertainty assessment and prediction interval calibration mechanism based on Monte Carlo Dropout (MC Dropout) on the basis of deterministic prediction. This mechanism includes four stages: MC Dropout random prediction, construction of the original prediction interval, statistical analysis of upper and lower boundary deviations, and prediction interval calibration. First, MC Dropout is used to perform multiple random forward propagations on the same sample, constructing the original prediction interval based on the empirical distribution of the predicted values. Second, using the validation set as the calibration set, the deviations between the true values ​​and the upper and lower boundaries of the original prediction interval are statistically analyzed. Then, the corresponding correction amounts are determined based on the empirical deviation distributions of the upper and lower boundaries. Finally, the correction amounts are used to independently calibrate the two boundaries of the original prediction interval, outputting a calibrated prediction interval corresponding to the SYM-H index point prediction results.

[0054] In deterministic prediction mode, the random deactivation layer in the model is turned off, and the sample to be predicted is input into the trained iTransformer model. The SYM-H exponent prediction result for the next H prediction steps is obtained through one forward propagation. This point prediction result is the SYM-H exponent prediction sequence generated by the aforementioned prediction output module.

[0055] In the uncertainty assessment mode, the randomly deactivated layers in the model are kept in an active state. Since the random deactivated state is different in each forward propagation, performing multiple random forward propagations on the same input sample can yield different prediction results. For the ... The nth input sample, the nth The predicted sequence obtained by the random forward propagation is represented as follows:

[0056] in, Indicates the first One input sample, This represents the model parameters after training. Indicates the first The random inactivation state corresponding to the next forward propagation. Indicates the number of random forward propagations. Indicates the first The predicted sequence of the future SYM-H index obtained by the random forward propagation is used to construct the uncertainty interval of the prediction results.

[0057] For the The input sample, at the _ , in the _ Sum of prediction steps The predicted values ​​obtained from each random forward propagation form the set of random predicted samples corresponding to that prediction step size:

[0058] in, Indicates the first In the second random forward propagation, the first The predicted SYM-H index value corresponding to each prediction step size.

[0059] Based on the random prediction sample set Based on the empirical distribution, the lower and upper bounds of the original prediction interval are determined. For the target coverage level... The prediction interval, the lower bound of the original prediction interval is represented as:

[0060] The upper bound of the original prediction interval is represented as:

[0061] Therefore, the first The input sample at the th ... The original prediction interval over each prediction step size is represented as follows:

[0062] in, This represents the function for calculating empirical quantiles. and These represent the lower and upper bounds of the original prediction interval, respectively. This indicates the significance level corresponding to the target coverage level. When the target coverage level is set to 90%, Take 0.1, and use respectively The 5% and 95% empirical quantiles of the random prediction results are used as the lower and upper bounds of the original prediction interval.

[0063] The original prediction interval is obtained directly from the empirical distribution of multiple random prediction results of MC Dropout. Since the original prediction interval may have a lower actual coverage than the target coverage level on unknown samples, this invention further uses the validation set as a calibration dataset to calibrate the original prediction interval.

[0064] For each calibration sample in the validation set, the randomly deactivated layer is kept in an active state, and the same input sample is then processed. The process involves a random forward propagation. Following the original prediction interval construction method described above, the original prediction intervals for each sample in the calibration set at each prediction step size are obtained.

[0065] Let the calibration set be the first The sample at the th The true SYM-H index at each prediction step size is: The corresponding lower bound of the original prediction interval is The upper bound of the original prediction interval is To reflect the deviations of the two interval boundaries from the actual observed values, this invention calculates the lower bound deviation and the upper bound deviation separately.

[0066] The lower bound deviation is expressed as:

[0067] The upper bound deviation is expressed as:

[0068] in, This represents the deviation of the actual observed value from the lower bound of the original prediction interval. This indicates the deviation of the upper bound of the original prediction interval from the actual observed value.

[0069] The lower bound biases of all samples and all prediction steps in the calibration dataset are summarized to form a lower bound bias set:

[0070] Simultaneously, the upper bound deviations of all samples and all prediction step sizes in the calibration dataset are summarized to form an upper bound deviation set:

[0071] in, This indicates the number of samples in the calibration dataset. Indicates the sample number in the calibration dataset. Indicates the prediction step size number. This represents the total prediction step size. Through the above processing, the lower bound bias and upper bound bias each form two independent empirical bias distributions.

[0072] Let the number of deviation samples contained in each deviation set be... Then, conservative empirical quantiles for interval calibration are determined based on the target coverage level. :

[0073] in, This represents the function for rounding up.

[0074] From the lower bound deviation set respectively and upper bound deviation set Selecting from the empirical quantiles The empirical quantiles are used as lower bound corrections. and upper bound correction :

[0075]

[0076] in, This represents the function for calculating empirical quantiles. Indicates empirical quantiles. This represents the lower bound correction amount determined by the empirical distribution of the lower bound deviation. This represents the upper bound correction amount determined by the empirical distribution of the upper bound deviation. The lower bound correction amount and the upper bound correction amount are determined based on the calibration deviation of the corresponding boundaries, and therefore they can have different values, thus enabling independent correction for different deviation characteristics of the two boundaries of the original prediction interval.

[0077] For new samples to be predicted, MC Dropout is first used. A second random forward propagation is performed, and the original prediction interval is obtained based on the empirical quantiles of the random prediction results. Subsequently, a lower bound correction amount predetermined from the calibration data is applied. and upper bound correction The lower and upper bounds of the original prediction interval are corrected respectively.

[0078] The lower bound of the calibrated prediction interval is expressed as:

[0079] The upper bound of the calibrated prediction interval is expressed as:

[0080] Therefore, the first The input sample at the th ... The final calibration prediction interval over each prediction step size is represented as follows:

[0081] in, and These represent the lower and upper bounds of the calibrated prediction interval, respectively. The lower bound correction amount... and upper bound correction The validation set is predetermined and remains fixed during model testing and actual prediction phases.

[0082] Through the above steps, this invention constructs the original prediction interval based on the MC Dropout method, and independently determines the upper and lower boundary correction amounts using the validation set bias, thereby achieving asymmetric calibration of the two sides of the prediction interval.

[0083] The following examples use the short-term prediction of the SYM-H index during a typical geomagnetic storm as an example to illustrate the specific implementation details and verification of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index proposed in this invention.

[0084] (1) Data sources and prediction task construction This embodiment uses 42 typical geomagnetic storm events from 1998 to 2018 as the research object. The SYM-H index, solar wind parameters, and interplanetary magnetic field parameters were all obtained from OMNI data provided by the Space Physics Data Facility (SPDF). All input variables were standardized to a 5-minute time resolution, and missing observations were filled using linear interpolation. The historical input window length was set to... L =96, corresponding to the spatial weather historical feature information of the past 8 hours, with the step size set to... H =12 or 24, corresponding to the SYM-H index prediction tasks for the next 1 hour and 2 hours, respectively.

[0085] The 42 geomagnetic storm events were divided into training, validation, and test sets at the event level. The specific division is shown in Table 1: the training set contains 20 geomagnetic storm events, which are used for model parameter learning; the validation set contains 5 geomagnetic storm events, which are used for model parameter adjustment, model selection, and prediction interval calibration; and the test set contains 17 geomagnetic storm events, which are used to evaluate the model's ability to predict unknown geomagnetic storm events.

[0086] Table 1. Dataset partitioning for geomagnetic storm events

[0087] (2) Introduction to performance evaluation indicators To evaluate the predictive performance of the method of this invention, the root mean square error (RMSE) and the coefficient of determination (R²) are used.

[0088] The RMSE calculation formula is as follows:

[0089] In the formula, N Indicates the number of predicted samples ， Indicates the first i One actual observation value, This represents the model's predicted value. The smaller the RMSE value, the smaller the prediction error.

[0090] The formula for calculating R² is as follows:

[0091] In the formula, R² represents the average of the actual observations. The closer R² is to 1, the higher the agreement between the predicted and actual observations.

[0092] (3) Network training and testing To verify the effectiveness of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index proposed in this invention, this embodiment trains and tests the model based on a real geomagnetic storm event dataset, and conducts comparative experiments with several representative geomagnetic disturbance index prediction methods.

[0093] During the training phase, the Huber loss function is used as the loss function, and the Adam optimizer is used for parameter updates. The learning rate is set to 0.0001, the batch size to 256, the feature dimension to 512, the number of encoder layers to 2, the hidden dimension of the feedforward neural network to 2048, the random inactivation rate to 0.05, and the number of training epochs to 100. In this embodiment, the target coverage level is set to 90%, therefore the 5% and 95% empirical quantiles are used as the lower and upper bounds of the original prediction interval.

[0094] To comprehensively evaluate the geomagnetic disturbance index prediction performance of this invention, representative prediction methods such as LSTM, CNN-LSTM, GBM, physical-information fusion model (Hybrid model), Transformer, DLinear, and TimeXer were selected for comparison, covering different modeling methods including recurrent neural networks, traditional machine learning, physical-information fusion, attention mechanisms, and linear time series prediction. Specifically, Transformer, DLinear, and TimeXer were trained and tested using the same dataset partitioning, input variables, preprocessing methods, and prediction time scales as this invention. LSTM, CNN-LSTM, GBM, and the physical-information fusion model used publicly reported prediction results from existing studies as references for comparison, where "—" indicates that the relevant published studies did not report this index.

[0095] Table 2. Comparison of average performance of different prediction methods in 1-hour and 2-hour SYM-H prediction tasks.

[0096] Table 2 shows that iTransformer achieves the lowest average RMSE and the highest R-value in the 1-hour prediction task. 2 In the 2-hour prediction task, its average RMSE was 9.35 nT, lower than other comparison methods in the table, while R... 2 It is 0.91.

[0097] Geomagnetic storm event 34 was selected, and the original prediction intervals and calibrated prediction intervals for 1-hour and 2-hour prediction tasks are presented. The results are as follows: Figure 5 As shown. Figure 5 (a) and Figure 5 (c) The original prediction intervals for the 1-hour and 2-hour prediction tasks, respectively. Figure 5 (b) and Figure 5 (d) represents the corresponding calibrated prediction intervals.

[0098] from Figure 5 (a) and Figure 5 (c) It can be seen that the original prediction interval obtained by MC Dropout is relatively narrow. During periods of rapid decline or large fluctuations in SYM-H, some observations fall outside the prediction interval. Therefore, based on the original prediction interval, the lower and upper bounds of the interval are corrected using calibration data. After calibration, the prediction interval is expanded, and some observations that were originally outside the interval are now covered, such as... Figure 5 (b) and Figure 5As shown in (d), interval calibration only adjusts the upper and lower boundaries of the prediction interval, without changing the deterministic prediction value. Therefore, interval calibration can improve the problem of the original MC Dropout prediction interval being too narrow, and enhance the prediction interval's coverage of actual SYM-H changes.

[0099] The above combination Figure 1 The intelligent prediction and uncertainty assessment method for geomagnetic disturbance index provided in the embodiments of the present invention has been described in detail. Next, the intelligent prediction and uncertainty assessment system for geomagnetic disturbance index provided in the embodiments of the present invention will be described in conjunction with the accompanying drawings.

[0100] Figure 6 This is a schematic diagram of the intelligent prediction and uncertainty assessment system for geomagnetic disturbance index shown in an embodiment of the present invention, with reference to... Figure 6 The system described in this invention includes: The data acquisition module is configured to acquire space weather observation data during geomagnetic storm events, including: interplanetary magnetic field parameters, solar wind speed, solar wind proton density, solar wind dynamic pressure, twilight electric field, and historical SYM-H index. The SYM-H index point prediction module is configured to: based on space weather observation data during geomagnetic storm events, use the iTransformer model, and be in an inactive state in the random deactivation layer to obtain SYM-H index point prediction results for multiple future time steps; The uncertainty assessment module is configured to: use the iTransformer model, which is activated in the random deactivation layer, based on space weather observation data during geomagnetic storm events, and perform multiple random forward propagations to obtain multiple prediction results; construct the original prediction interval based on the empirical quantiles of the multiple prediction results; construct a calibration set, select the lower bound correction amount and upper bound correction amount corresponding to the empirical quantiles based on the calibration set, and correct the lower boundary and upper boundary of the original prediction interval respectively to obtain the calibrated prediction interval.

[0101] In some embodiments, after acquiring space weather observation data during a geomagnetic storm event, the method further includes: aligning the space weather observation data during the geomagnetic storm event in time, using linear interpolation to fill in missing observations, and then performing standardization to obtain a multivariate time series input matrix.

[0102] In some embodiments, after obtaining the multivariate time series input matrix, the method further includes: taking the multivariate time series input matrix, obtaining the variable center representation by matrix transpose, and then inputting it into the iTransformer model.

[0103] In some embodiments, the process of employing the iTransformer model includes: using a variable embedding module to linearly map each space weather variable, converting the input sequences of different physical variables into a hidden feature representation of a unified dimension; based on the hidden feature representation, using a variable attention encoding module to extract observation features of multiple space weather variables and dependency features between different space weather variables, and fusing them to output fused features; based on the fused features, using a prediction output module to output the SYM-H index point prediction results for multiple future time steps.

[0104] In some embodiments, selecting the lower bound correction and upper bound correction corresponding to the empirical quantile based on the calibration set includes: The calibration set includes a lower bound set of deviations and an upper bound set of deviations; Let the number of deviation samples contained in each deviation set be... Then, conservative empirical quantiles for interval calibration are determined based on the target coverage level. :

[0105] in, This represents the floor function; From the lower bound deviation set respectively and upper bound deviation set Selecting from the empirical quantiles The empirical quantiles are used as lower bound corrections. and upper bound correction :

[0106]

[0107] in, This represents the function for calculating empirical quantiles. Indicates empirical quantiles. This represents the lower bound correction amount determined by the empirical distribution of the lower bound deviation. This represents the upper bound correction amount determined by the empirical distribution of the upper bound deviation.

[0108] In some embodiments, the correction of the lower and upper boundaries of the original prediction interval includes: The lower bound of the calibrated prediction interval is expressed as:

[0109] The upper bound of the calibrated prediction interval is expressed as:

[0110] No. The input sample at the th th The final calibration prediction interval over each prediction step size is represented as follows:

[0111] in, and These represent the lower and upper bounds of the calibrated prediction interval, respectively. This represents the lower boundary of the original prediction interval. This represents the upper boundary of the original prediction interval. This represents the lower bound correction amount determined by the empirical distribution of the lower bound deviation. This represents the upper bound correction amount determined by the empirical distribution of the upper bound deviation.

[0112] According to embodiments of the present invention, the intelligent prediction and uncertainty assessment system for geomagnetic disturbance index can correspond to the execution of the methods described in the embodiments of the present invention, and the above and other operations and / or functions of each module of the intelligent prediction and uncertainty assessment system for geomagnetic disturbance index are respectively for the purpose of implementing... Figure 1 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0113] See Figure 7 The diagram shows the structure of a computer device, which includes a processor, a communication interface, and a computer-readable storage medium. The processor, communication interface, and computer-readable storage medium are connected via a bus or other means. The communication interface is used to receive and send data. The computer-readable storage medium can be stored in the computer device's memory. The computer-readable storage medium stores computer programs, including program instructions, and the processor executes the program instructions stored in the computer-readable storage medium. The processor (or CPU, Central Processing Unit) is the computing and control core of the computer device, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve the corresponding steps in the embodiment of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index.

[0114] This embodiment provides a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the processing system of the computer device.

[0115] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM memory or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0116] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above-described embodiment of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index.

[0117] This embodiment provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding steps in the above-described embodiment of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index.

[0118] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0119] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0122] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0123] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent prediction and uncertainty assessment of geomagnetic disturbance index, characterized in that, include: Acquire space weather observation data during geomagnetic storm events, including: interplanetary magnetic field parameters, solar wind speed, solar wind proton density, solar wind dynamic pressure, twilight electric field, and historical SYM-H index; Based on space weather observation data during geomagnetic storm events, the iTransformer model was used to obtain SYM-H index point prediction results for multiple future time steps in the inactive state of the random deactivation layer. Based on space weather observation data during geomagnetic storm events, the iTransformer model is used, which is activated in the random deactivation layer and undergoes multiple random forward propagations to obtain multiple prediction results. The original prediction interval is constructed based on the empirical quantiles of the multiple prediction results. A calibration set is constructed, and the lower and upper bound corrections corresponding to the empirical quantiles are selected based on the calibration set to correct the lower and upper boundaries of the original prediction interval, respectively, to obtain the calibrated prediction interval.

2. The intelligent prediction and uncertainty assessment method for geomagnetic disturbance index according to claim 1, characterized in that, After acquiring space weather observation data during geomagnetic storm events, the process also includes: time alignment of the space weather observation data during geomagnetic storm events, completion of missing observations using linear interpolation, and standardization to obtain a multivariate time series input matrix.

3. The intelligent prediction and uncertainty assessment method for geomagnetic disturbance index according to claim 2, characterized in that, After obtaining the multivariate time series input matrix, the process also includes: taking the multivariate time series input matrix, obtaining the variable center representation through matrix transpose, and then inputting it into the iTransformer model.

4. The intelligent prediction and uncertainty assessment method for geomagnetic disturbance index according to claim 1, characterized in that, The process of using the iTransformer model includes: A variable embedding module is used to linearly map each spatial weather variable, converting the input sequences of different physical variables into a hidden feature representation of a unified dimension; Based on hidden feature representation, a variable attention encoding module is used to extract observation features of multiple space weather variables and the dependency features between different space weather variables, and then fuse them to output fused features; Based on the fusion features, a prediction output module is used to output the prediction results of the SYM-H index points for multiple future time steps.

5. The intelligent prediction and uncertainty assessment method for geomagnetic disturbance index according to claim 1, characterized in that, The step of selecting the lower and upper bound correction values ​​corresponding to the empirical quantiles based on the calibration set includes: The calibration set includes a lower bound set of deviations and an upper bound set of deviations; Let the number of deviation samples contained in each deviation set be... Then, conservative empirical quantiles for interval calibration are determined based on the target coverage level. : in, This represents the floor function; From the lower bound deviation set respectively and upper bound deviation set Selecting from the empirical quantiles The empirical quantiles are used as lower bound corrections. and upper bound correction : in, This represents the function for calculating empirical quantiles. Indicates empirical quantiles. This represents the lower bound correction amount determined by the empirical distribution of the lower bound deviation. This represents the upper bound correction amount determined by the empirical distribution of the upper bound deviation.

6. The intelligent prediction and uncertainty assessment method for geomagnetic disturbance index according to claim 1, characterized in that, The correction of the lower and upper boundaries of the original prediction interval includes: The lower bound of the calibrated prediction interval is expressed as: The upper bound of the calibrated prediction interval is expressed as: No. The input sample at the th ... The final calibration prediction interval over each prediction step size is represented as follows: in, and These represent the lower and upper bounds of the calibrated prediction interval, respectively. This represents the lower boundary of the original prediction interval. This represents the upper boundary of the original prediction interval. This represents the lower bound correction amount determined by the empirical distribution of the lower bound deviation. This represents the upper bound correction amount determined by the empirical distribution of the upper bound deviation.

7. A system for intelligent prediction and uncertainty assessment of geomagnetic disturbance index, characterized in that, include: The data acquisition module is configured to acquire space weather observation data during geomagnetic storm events, including: interplanetary magnetic field parameters, solar wind speed, solar wind proton density, solar wind dynamic pressure, twilight electric field, and historical SYM-H index. The SYM-H index point prediction module is configured to: based on space weather observation data during geomagnetic storm events, use the iTransformer model, and be in an inactive state in the random deactivation layer to obtain SYM-H index point prediction results for multiple future time steps; The uncertainty assessment module is configured to: use the iTransformer model based on space weather observation data during geomagnetic storm events, activate the model in the random deactivation layer, perform multiple random forward propagations to obtain multiple prediction results; construct the original prediction interval based on the empirical quantiles of the multiple prediction results; construct a calibration set, select the lower bound correction amount and upper bound correction amount corresponding to the empirical quantiles based on the calibration set, and correct the lower boundary and upper boundary of the original prediction interval respectively to obtain the calibrated prediction interval.

8. A computer device, characterized in that, A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps of the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and execute the steps in the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps in the intelligent prediction and uncertainty assessment method for geomagnetic disturbance index as described in any one of claims 1-6.

Images