Geomagnetic single station and regional magnetic field strength prediction method and system based on lstm network

Abstract

This invention belongs to the field of geomagnetic data processing technology. Addressing the problems of data gaps and low detection accuracy in existing geomagnetic field prediction methods, this invention proposes a method and system for predicting the magnetic field intensity of a single geomagnetic station and a geomagnetic region based on an LSTM network. The method includes: acquiring and preprocessing historical observation data from a geomagnetic observatory; constructing an LSTM prediction model; training and predicting the LSTM prediction model using the preprocessed historical data to obtain an optimized LSTM prediction model; and using the optimized LSTM prediction model to predict the magnetic field intensity of a single geomagnetic station and a geomagnetic region. This invention utilizes geomagnetic constants prediction from different components of the LSTM prediction model, enabling the processing of geomagnetic data with complex time dependencies and effectively improving prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of geomagnetic data processing technology, and in particular to a method and system for predicting the magnetic field intensity of a single geomagnetic station and a region based on an LSTM network. Background Technology

[0002] Geomagnetic field prediction is crucial for geophysics, space science, and engineering applications. In geophysics, sensors based on geomagnetic properties can measure the magnetism of underground minerals. Anomalies in geomagnetic data are often closely related to geological disasters, and the accuracy of geomagnetic data analysis determines the timeliness of early warnings for natural disasters such as earthquakes, volcanic activity, auroras, and solar magnetic storms. The high-frequency and low-frequency components of geomagnetic data play different roles in geomagnetic research, reflecting different time scales and physical phenomena. Low-frequency signals are mainly used to analyze long-term changes and trends in the Earth's magnetic field, involving dynamic processes within the Earth, the historical evolution of the Earth's magnetic field, and earthquake prediction. However, despite the significant application value of geomagnetism, its acquisition and analysis processes face challenges such as noise, missing data, and outliers. Furthermore, traditional methods (such as spherical harmonic analysis) are limited by truncation levels, making it difficult to meet the high-precision requirements of applications like detection and navigation.

[0003] In recent years, Long Short-Term Memory (LSTM) networks have been gradually applied in geomagnetic field modeling and prediction due to their unique gating mechanism and strong nonlinear modeling capabilities. LSTM networks are a special type of recurrent neural network (RNN). By introducing gating mechanisms and cell states, they can effectively capture long-term dependencies in sequences, aiming to solve the gradient vanishing and gradient explosion problems faced by traditional RNN networks when processing long sequence data.

[0004] To this end, a geomagnetic intensity prediction method based on LSTM network was developed, which is suitable for processing dynamic geomagnetic field data with complex time dependencies and provides strong support for the prediction of short-term and long-term geomagnetic field intensity. Summary of the Invention

[0005] To address the problems mentioned above in the background section, this invention proposes a method and system for predicting the magnetic field intensity of a single geomagnetic station and a region based on LSTM networks.

[0006] The present invention achieves the above objectives by employing the following technical solutions: In a first aspect, this invention discloses a method for predicting the magnetic field intensity of a single geomagnetic station and a region based on an LSTM network, comprising the following steps: Acquire historical data from geomagnetic observatories and perform preprocessing; Build an LSTM prediction model, train and predict using preprocessed historical data to obtain an optimized LSTM prediction model; The optimized LSTM prediction model is used to predict the magnetic field intensity of a single geomagnetic station and a geomagnetic region.

[0007] As a further optimized technical solution: the historical data obtained from the geomagnetic observatory observations include multidimensional time series data of magnetic field intensity D component, H component and Z component, and Savitzky-Golay filtering is used to smooth the time series data of each dimension.

[0008] A further preferred technical solution is that the steps to obtain the optimized LSTM prediction model include: An LSTM prediction model is established based on an LSTM network, wherein the LSTM prediction model includes an input layer, at least one LSTM layer, a batch normalization layer, a dropout layer, a fully connected layer, and an output layer. A sliding window is used to further process the smoothed time series data to form input and output sample pairs. The smoothed time series data is divided into training and test sets, and the time series data in the training set is normalized. The LSTM prediction model is trained using the normalized training set, and then tested using the test set to obtain the optimized LSTM prediction model.

[0009] As a preferred implementation method, when predicting the magnetic field intensity of a single geomagnetic station, a single-input single-output approach is adopted. The D, H, or Z components of the geomagnetic intensity of a single geomagnetic station in the previous stage are input into the optimized LSTM prediction model to obtain the corresponding D, H, or Z components of the geomagnetic intensity in the next stage.

[0010] As a preferred implementation method, when predicting the magnetic field strength of a geomagnetic region, an input-output mapping relationship is established between the geomagnetic observation data of multiple geomagnetic observatories and the prior data of the target geomagnetic observatory. The geomagnetic observation data of multiple geomagnetic observatories are input into the optimized LSTM prediction model, and the D component, H component and Z component of the magnetic field strength of the target geomagnetic observatory are output.

[0011] Secondly, this invention also discloses a geomagnetic single-station and regional magnetic field intensity prediction system based on LSTM networks, comprising: The data acquisition module is used to acquire historical data from geomagnetic observatories. The data preprocessing module is used to preprocess historical data; The model building and optimization module is used to build an LSTM prediction model and train and predict based on preprocessed historical data to obtain an optimized LSTM prediction model. The prediction module uses an optimized LSTM prediction model to predict the magnetic field strength of a single geomagnetic station and a geomagnetic region.

[0012] As a further preferred option, the prediction module includes a single-station prediction submodule and a regional prediction submodule, wherein: The single-station prediction submodule is used to predict the geomagnetic intensity components of a single geomagnetic station. The regional prediction submodule is used to predict the geomagnetic intensity of a target geomagnetic station using geomagnetic observation data from multiple input geomagnetic observatories.

[0013] In addition, the present invention also discloses a computer device, comprising: one or more processors, and a memory storing one or more computer programs, wherein the processor invokes the computer programs to implement the steps of the prediction method described above.

[0014] In addition, the present invention also discloses a computer-readable storage medium storing a computer program that is invoked by a processor to implement the steps of the prediction method described above.

[0015] Compared with the prior art, the present invention can produce the following effects: 1. By constructing an LSTM prediction model, we conducted an in-depth study on geomagnetic field modeling and prediction for single geomagnetic stations and geomagnetic regions. The experimental results fully demonstrate the advantages of LSTM in processing geomagnetic data with complex time dependencies, proving its improvement in prediction accuracy compared to traditional methods. In particular, the analysis of the differences in prediction performance for different time scales and different geomagnetic components provides an important reference for further optimizing geomagnetic prediction models.

[0016] 2. Regarding the prediction method of this invention for a single geomagnetic station, the prediction accuracy of different components varies through the LSTM prediction model. In long-term prediction, the Z component has the highest accuracy, followed by D, and the H component has the lowest accuracy. In short-term prediction, the Z component has the highest accuracy, followed by H, and the D component has the lowest accuracy. In terms of geomagnetic regions, data with a time span of one year has the highest accuracy for predicting the H component and the lowest for Z; data with a time span of one month has the highest accuracy for predicting the Z component and the lowest for D. Overall, short-term prediction is better than long-term prediction, providing an effective method for geomagnetic field prediction, and is especially suitable for processing geomagnetic data with complex time dependencies. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0018] Figure 1 This is a structural diagram of the LSTM network of the present invention; Figure 2 This is a logic diagram of the LSTM network of the present invention for predicting the magnetic field strength of a single geomagnetic station using a single input; Figure 3 The LSTM network of this invention uses a multi-input magnetic field strength prediction logic diagram for geomagnetic regions. Figure 4 This is a schematic diagram showing the locations of various magnetostations in Embodiment 2 of the present invention; Figure 5 This is the short-term prediction result of the D component observed at the Chengdu Geomagnetic Observatory (data from January 1994) in Embodiment 2 of the present invention; Figure 6 This is the short-term prediction result of the H component observed at the Sheshan Geomagnetic Observatory (data from January 1994) in Embodiment 2 of the present invention; Figure 7 This is the short-term prediction result of the Z component observed at the Qianling Geomagnetic Observatory (data from January 1994) in Embodiment 2 of the present invention; Figure 8 This is the long-term prediction result of the D component observed at the Qiongzhong Geomagnetic Observatory (1996 data) in Embodiment 2 of the present invention; Figure 9 This is the long-term prediction result of the Z component observed at the Chengdu Geomagnetic Observatory (1996 data) in Embodiment 2 of the present invention; Figure 10 This is the long-term prediction result of the H component observed at the Lanzhou Geomagnetic Observatory (1996 data) in Embodiment 2 of the present invention; Figure 11 The long-term (one-year) prediction results and relative errors of the Z component of geomagnetic regional observation in Embodiment 2 of the present invention are shown. Figure 12 The long-term prediction results and relative errors of the H component observed in the geomagnetic region in Embodiment 2 of the present invention; Figure 13 The long-term prediction results and relative errors of the D component of geomagnetic regional observation in Embodiment 2 of the present invention; Figure 14 The short-term (one-month) prediction results and relative errors of the Z component of geomagnetic region observation in Embodiment 2 of the present invention are shown. Figure 15 The short-term (one-month) prediction results and relative errors of the H component of geomagnetic region observation in Embodiment 2 of the present invention are shown. Figure 16 This is the short-term (one-month) prediction result and relative error of the D component of the geomagnetic region observation in Embodiment 2 of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0020] Example 1: Reference Figures 1-3 This invention proposes a method for predicting the magnetic field intensity of a single geomagnetic station and a region based on an LSTM network, comprising the following steps: Acquire historical data from geomagnetic observatories and perform preprocessing; Build an LSTM prediction model, train and predict using preprocessed historical data to obtain an optimized LSTM prediction model; The optimized LSTM prediction model is used to predict the magnetic field intensity of a single geomagnetic station and a geomagnetic region.

[0021] Specifically, the historical data obtained from the geomagnetic observatory includes multidimensional time series data of magnetic field intensity D, H and Z components. The original geomagnetic data usually contains noise from sensor errors, environmental interference or other external factors, which can mask the true patterns and trends in the geomagnetic data. Savitzky-Golay filtering is used to smooth the time series data of each dimension.

[0022] Savitzky-Golay filtering is a smoothing technique that uses polynomial fitting to effectively remove noise while preserving the main trend of the data. During the filtering process, the window size is selected based on the characteristics of the geomagnetic data. Smaller windows can retain more detailed information, while larger windows are more suitable for smoothing larger noise. In this embodiment, a window size of 11 and a second-order polynomial are selected to carry out the filtering work. The second-order polynomial can smooth most time series data and effectively capture the local trends in the data. Through the filtering operation, high-frequency noise in the geomagnetic data is removed, while its main trend is preserved.

[0023] Next, an LSTM prediction model is constructed. This invention's LSTM prediction model is based on an LSTM network. The LSTM network, through its gating mechanism, can dynamically capture the temporal variation of the geomagnetic field and automatically extract key features. The LSTM network contains unique memory units and a gating mechanism. Each Long Short-Term Memory (LSTM) unit covers three main gates: an input gate, a forget gate, and an output gate, such as... Figure 1As shown, it can effectively control the flow of information, avoid long-term dependence problems, and improve the accuracy of prediction. Its powerful nonlinear modeling capability can capture the complex nonlinear variation law in the geomagnetic field. With multivariate time series processing, it can capture the interrelationship between different dimensions and gain a more comprehensive understanding of the changes in the geomagnetic field.

[0024] Based on the diagram, the memory units and gating mechanism of the LSTM network structure are as follows: (1) Input gate: ; (2) Gate of Oblivion: ; (3) Output gate: ; (4) Candidate values ​​for cell state: ; (5) Cell state renewal: ; (6) Hidden state update ; In the formula: , , , These are the weights of the forget gate, input gate, output gate, and memory unit, respectively. , , and These are the bias values ​​for the forget gate, input gate, output gate, and memory unit, respectively. This represents a vector consisting of the hidden state of the previous time unit and the input data at the current time unit; The forget gate at time t determines the information to be deleted from the memory unit; The input gate at time t determines the information added to the memory cell; At time t, the output gate determines which information to select from the memory cell for output. It is the final state of the memory unit at time t; It represents the hidden state of the cell at time t.

[0025] The principle is as follows: For data input into the LSTM network, the forget gate first controls the amount of historical information to be discarded in the current cell state. Then, the input gate controls how much information to retain in the current cell state. Next, the cell state is updated by multiplying the result of the forget gate by the cell state from the previous time step, and then adding the result of the input gate. This process achieves the change of cell state from one time step to the next. Updated to Finally, the hidden state of the current unit is obtained through the output gate. .

[0026] In this embodiment, the LSTM prediction model includes an input layer, three LSTM layers, a batch normalization layer, a dropout layer, a fully connected layer, and an output layer. The input layer receives single feature data, which is then processed through three LSTM layers (containing 300, 200, and 100 hidden units respectively, with the output mode being sequences), a batch normalization layer, and a dropout layer to extract sequence features and prevent overfitting. Subsequently, nonlinearity is introduced through a fully connected layer containing 128 neurons and a ReLU activation layer. Finally, the prediction result is obtained through the Dropout layer and the output layer. The output layer contains a single node for regression prediction.

[0027] For the LSTM prediction model constructed in this embodiment, it is trained and tested using denoised geomagnetic data. Specifically, the total length of the denoised geomagnetic data is calculated, and the training set and test set are divided in an 8:2 ratio. The training set is used for the training process of the LSTM prediction model, allowing the LSTM prediction model to learn the patterns and rules of geomagnetic data. The test set is used to verify the generalization ability of the LSTM prediction model and to predict the performance of the LSTM prediction model on new geomagnetic data.

[0028] Furthermore, the time series data in the training set is normalized to map the time series data to the [0,1] interval. Normalization can eliminate the dimensional differences between different features, enabling the LSTM prediction model to converge faster during training. The aim is to improve the training efficiency and prediction accuracy of the LSTM prediction model. The mean mu and standard deviation sig of the time series data in the training set are calculated, and these parameters are used to normalize the training set and the test set to ensure data consistency. This normalization process not only accelerates the training of the LSTM prediction model but also reduces the problems of gradient vanishing or exploding, thus improving the stability of the LSTM prediction model.

[0029] In this embodiment, the LSTM prediction model is trained using a normalized training set, and then tested using a test set to obtain an optimized LSTM prediction model. Before training and testing the LSTM prediction model, both the training set and the test set need to be converted into a format that matches the input of the LSTM network, forming a three-dimensional tensor of the number of samples, time step, and number of features. After the format conversion of the geomagnetic data, the LSTM network can accurately process and extract features, thereby improving the prediction performance of the LSTM prediction model.

[0030] In this embodiment, a single-input, single-output method is used to predict the magnetic field strength of a single geomagnetic station, such as... Figure 2As shown, the D, H, or Z components of the geomagnetic intensity of a single geomagnetic station in the previous stage are input into the optimized LSTM prediction model to obtain the corresponding D, H, or Z components of the geomagnetic intensity in the next stage.

[0031] In this embodiment, when predicting the magnetic field strength of a geomagnetic region, an input-output mapping relationship is established between geomagnetic observation data from multiple geomagnetic observatories and prior data from the target geomagnetic observatory. Figure 3 As shown, geomagnetic observation data from multiple geomagnetic observatories are input into the optimized LSTM prediction model, and the D, H, and Z components of the magnetic field intensity of the target geomagnetic observatory are output.

[0032] Furthermore, an error function system is established to evaluate the prediction effect, assuming... For the actual observed signal, The output signal predicted by the LSTM prediction model. Given the number of samples, the following four evaluation metrics are used to evaluate and analyze the LSTM prediction model: (1) Root Mean Square Error (RMSE): This measures the difference between the predicted and actual values. It is calculated by first taking the square of the difference between the predicted and actual values, then taking the average of these squares, and finally taking the square root of the average. The formula is as follows: ; The smaller the RMSE value, the higher the prediction accuracy of the model.

[0033] (2) Mean Squared Error (MSE): MSE is the square of RMSE, and its formula is: ; Mean squared error also reflects the difference between the predicted and the true values. The smaller the value, the better the model's overall predictive performance.

[0034] (3) Mean Absolute Error (MAE): refers to the average of the absolute differences between the predicted value and the actual value. Its formula is: ; MAE directly measures the average absolute difference between predicted and actual values. The smaller the value, the higher the accuracy of the model's prediction.

[0035] (4) Mean Relative Percentage Error (MAPE): MAPE is the average of the relative differences between the predicted and actual values, and its formula is: ; MAPE measures the percentage of prediction error relative to the true value; the smaller the value, the higher the prediction accuracy of the model.

[0036] Example 2: The data used in this embodiment are the D, H, and Z components of the geomagnetic field observed by six geomagnetic stations in China from 1994 to 1996. The six geomagnetic stations are referenced from... Figure 4 As shown in Table 1, geomagnetic data with different time resolutions and geomagnetic data from different geomagnetic activity periods were selected for comparative analysis. The specific names and location parameters of each geomagnetic station are shown in Table 1.

[0037] Table 1. Names and location parameters of various magnetostations ; (1) The following data is for a period of one month, which can well reflect the short-term changes in the geomagnetic field: Taking the D-component data from the Chengdu Geomagnetic Observatory in January 1994 as an example, the results are analyzed, and the predicted results are as follows: Figure 5 As shown, from Figure 5 As can be seen from Table 5, the root mean square error is 0.26834. It can be seen that the prediction error curve is basically a noise curve, with an absolute error of 0.32032% and a root mean square error of 0.22306.

[0038] Table 2. Modeling and prediction of the short-time geomagnetic D component by the LSTM prediction model. ; Taking the H-component data from the Sheshan Geomagnetic Observatory in January 1994 as an example, the results are analyzed, and the predicted results are as follows: Figure 6 As shown, from Figure 6 As can be seen from Table 6, the root mean square error is 8.605, and the prediction error curve is basically a noise curve, with an absolute error of 0.02050% and a root mean square error of 6.8774.

[0039] Table 3. Modeling and prediction of the short-time geomagnetic H component by the LSTM prediction model. ; Taking the Z-component data from the Qianling Geomagnetic Observatory in January 1994 as an example, the results are analyzed, and the predicted results are as follows: Figure 7 As shown, from Figure 7 As can be seen from Table 7, the root mean square error is 3.2005. It can be seen that the prediction error curve is basically a noise curve, with an absolute error of 0.00677% and a root mean square error of 2.3676.

[0040] Table 4. Modeling and prediction of the short-time geomagnetic Z component by the LSTM prediction model. ; For short-term variation analysis, the LSTM prediction model achieved a MAE ranging from 0.22306 to 0.5165, an RMSE between 0.26834 and 0.62316, and a MAEP from 0.11920% to 0.34706% for element D. For element Z, the MAE ranged from 1.4698 to 5.1249, the RMSE varied from 2.1133 to 6.3469, and the MAEP ranged from 0.00419% to 0.01486%. For element H, the MAE ranged from 4.7604 to 6.8774, the RMSE ranged from 5.9317 to 8.605, and the MAEP ranged from 0.01381% to 0.02050%. The results indicate that the LSTM prediction model achieves the highest prediction accuracy for element Z in short-term forecasting, followed by element H, while exhibiting the worst prediction accuracy for element D.

[0041] (2) The following data represents a year's worth of data, which can well reflect the long-term changes in the geomagnetic field: Taking the D-component data from the Qiongzhong Geomagnetic Observatory in 1996 as an example, the results are analyzed, and the predicted results are as follows: Figure 8 As shown, from Figure 8 As can be seen from Table 2, each predicted value is basically consistent with the true value. The root mean square error is 0.34519, and the prediction error curve is basically a noise curve. The absolute error is 0.455706%, and the root mean square error is 0.2653, indicating that the prediction effect is very good.

[0042] Table 5. Modeling and prediction of the geomagnetic D component by the LSTM prediction model. ; Taking the Z-component data from the Chengdu Geomagnetic Observatory in 1994 as an example, the results are analyzed, and the predicted results are as follows: Figure 9 As shown, from Figure 9 As can be seen from Table 3, the root mean square error is 11.2035, indicating that the prediction error curve is basically a noise curve. The absolute error is 0.020092% and the root mean square error is 6.992, which shows that the prediction effect is also quite good.

[0043] Table 6. Modeling and prediction of the geomagnetic Z component by the LSTM prediction model. ; Taking the H-component data from the Lanzhou Geomagnetic Observatory in 1996 as an example, the results are analyzed, and the predicted results are as follows: Figure 10 As shown, from Figure 10As can be seen from Table 4, the root mean square error is 5.8558, indicating that the prediction error curve is basically a noise curve. The absolute error is 0.13241%, and the root mean square error is 4.1283, which shows that the prediction effect is still very good.

[0044] Table 7. Modeling and prediction of the H component of the geomagnetic field by the LSTM prediction model. ; In the long-term variation analysis, the prediction accuracy of the LSTM deep learning model can be quantitatively evaluated by analyzing the evaluation metrics of the LSTM model on the validation sets of D, H, and Z elements at different geomagnetic stations. Specifically, when predicting the D element, the MAE of the LSTM model ranges from 0.2653 to 0.43102, the RMSE ranges from 0.34519 to 0.55599, and the MAEP varies from 0.12188% to 0.61842%. For the Z element, the MAE ranges from 2.9815 to 6.922, the RMSE ranges from 3.2996 to 11.2035, and the MAEP ranges from 0.0069164% to 0.020092%. In the prediction of the H element, the MAE varied from 4.1283 to 7.5569, the RMSE ranged from 5.8598 to 11.8679, and the MAEP increased from 0.012878% to 0.021933%. Comprehensive analysis shows that the LSTM model achieved the highest accuracy in predicting the Z element, followed by the D element, while the prediction accuracy for the H element was relatively low.

[0045] Comparative analysis revealed that the LSTM prediction model performs better in predicting long-term changes in geomagnetic stations than in predicting short-term changes.

[0046] (3) Prediction of geomagnetic regions based on LSTM prediction model: Combination Figure 4 As shown, the data uses the D, H, and Z components of the Chengdu, Sheshan, Qianling, and Wuhan geomagnetic stations in 1994 as input. The specific information of the geomagnetic stations is shown in Table 1.

[0047] The input data comes from magnetic field observations from multiple geomagnetic stations, presented in time series format. Each geomagnetic station's data is considered an independent time series, and all the data collectively constitute a multidimensional time series input. The LSTM prediction model, through its memory units and gating mechanism, effectively captures the temporal dependencies in the data from each geomagnetic station, extracts key features, and uses prior data from the target geomagnetic station (such as historical observations) as a supervisory signal to train the network. By linking the prior data of the target geomagnetic station with the input data, the network can learn the mapping relationship between the input (i.e., data from multiple geomagnetic stations) and the output (i.e., data from the target geomagnetic station). This mapping relationship reflects the spatial correlation of regional magnetic field changes. Some geomagnetic stations' data may contribute more to the prediction of the target geomagnetic station, and the LSTM network automatically learns these weight relationships. LSTM networks are continuously trained and their weights adjusted to optimize prediction performance. The network extracts features from the input data, capturing dynamic features in the time series. Based on the weight relationships between input data, it constructs the spatial correlation of magnetic field changes in a geomagnetic region. It leverages multi-source data to capture the complex spatiotemporal characteristics of magnetic field changes in the geomagnetic region, using prior data from the target geomagnetic station as a supervisory signal to enhance the predictive power of the LSTM network. Furthermore, LSTM networks possess powerful time series modeling capabilities, handling nonlinear relationships and are suitable for dynamic modeling of magnetic field changes. By incorporating prior data from the target geomagnetic station, the LSTM network can gain a better understanding of the station's spatiotemporal characteristics, improving prediction accuracy.

[0048] Tables 8 and 9 below show the long-term and short-term predictions of geomagnetic intensity in the geomagnetic region, respectively: Table 8 Analysis of the long-term geomagnetic intensity prediction results of the LSTM prediction model for the geomagnetic region. ; Table 9. Analysis of the short-term geomagnetic intensity prediction results of the LSTM prediction model for the geomagnetic region. ; Based on the above analysis of long-term and short-term geomagnetic intensity predictions, we can conclude that: Reference Figures 11-16 When dealing with data over a period of one year, the LSTM prediction model has the highest prediction accuracy for the H component and the lowest for the Z component; when dealing with data over a period of one month, it has the highest prediction accuracy for the Z component and the lowest for the D component.

[0049] like Figures 11-13As shown, when dealing with data with a time span of one year, the LSTM prediction model has the highest prediction accuracy for the H component. This is because the H component exhibits a relatively stable and regular trend over the long term, with strong spatial correlation, which is conducive to the LSTM prediction model capturing long-term trends. In contrast, the Z component is affected by various factors such as underground geological structure and rock magnetism, and the accumulation of short-term fluctuations affects the grasp of long-term trends, resulting in the lowest prediction accuracy.

[0050] like Figures 14-16 As shown, for data with a time span of one month, the Z component has the highest prediction accuracy because the changes are small due to short-term geomagnetic activity such as geological structure and rock magnetism, and the LSTM prediction model can easily extract effective features. The D component, on the other hand, has the lowest accuracy because its changes are complex and random, affected by a combination of factors such as geomagnetic storms and ionospheric currents, and it lacks long-term trend guidance, resulting in high uncertainty in short-term predictions.

[0051] Overall, short-term predictions outperform long-term predictions because short-term data are highly correlated and continuous, making it easier for LSTM prediction models to capture short-term dependencies and dynamic changes. In contrast, long-term data is complex and variable, prone to error accumulation and model drift, and the sample distribution may change, affecting the model's ability to grasp the true patterns.

[0052] Example 3: A geomagnetic field intensity prediction system based on LSTM networks for single geomagnetic stations and regions includes: The data acquisition module is used to acquire historical data from geomagnetic observatories. The data preprocessing module is used to preprocess historical data; The model building and optimization module is used to build an LSTM prediction model and train and predict based on preprocessed historical data to obtain an optimized LSTM prediction model. The prediction module uses an optimized LSTM prediction model to predict the magnetic field strength of a single geomagnetic station and a geomagnetic region.

[0053] As a further preferred option, the prediction module includes a single-station prediction submodule and a regional prediction submodule, wherein: The single-station prediction submodule is used to predict the geomagnetic intensity components of a single geomagnetic station. The regional prediction submodule is used to predict the geomagnetic intensity of a target geomagnetic station using geomagnetic observation data from multiple input geomagnetic observatories.

[0054] In addition, the present invention also discloses a computer device, including one or more processors and a memory communicatively connected to the one or more processors; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method described in the above embodiments.

[0055] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0056] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0057] Furthermore, this invention also discloses a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the methods described in the above embodiments. That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or a processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0058] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the magnetic field intensity of a single geomagnetic station and a region based on LSTM networks, characterized in that, Includes the following steps: Acquire historical data from geomagnetic observatories and perform preprocessing; The preprocessed historical data is input into the LSTM prediction model for training and prediction, resulting in an optimized LSTM prediction model. The optimized LSTM prediction model is used to predict the magnetic field intensity of a single geomagnetic station and a geomagnetic region.

2. The method for predicting the magnetic field intensity of a single geomagnetic station and a region based on LSTM networks according to claim 1, characterized in that, The acquired historical data from the geomagnetic observatory includes multidimensional time series data of magnetic field intensity D, H and Z components. Savitzky-Golay filtering was used to smooth the time series data of each dimension.

3. The method for predicting the magnetic field intensity of a single geomagnetic station and a region based on an LSTM network according to claim 2, characterized in that, The steps to obtain the optimized LSTM prediction model include: An LSTM prediction model is established based on an LSTM network, wherein the LSTM prediction model includes an input layer, at least one LSTM layer, a batch normalization layer, a dropout layer, a fully connected layer, and an output layer. A sliding window is used to further process the smoothed time series data to form input and output sample pairs. The smoothed time series data is divided into training and test sets, and the time series data in the training set is normalized. The LSTM prediction model is trained using the normalized training set, and then tested using the test set to obtain the optimized LSTM prediction model.

4. The method for predicting the magnetic field intensity of a single geomagnetic station and a region based on an LSTM network according to claim 3, characterized in that, When predicting the magnetic field intensity of a single geomagnetic station, a single-input single-output method is adopted. The D, H, or Z components of the geomagnetic intensity of a single geomagnetic station in the previous stage are input into the optimized LSTM prediction model to obtain the corresponding D, H, or Z components of the geomagnetic intensity in the next stage.

5. The method for predicting the magnetic field intensity of a single geomagnetic station and a region based on an LSTM network according to claim 3, characterized in that, When predicting the magnetic field strength of a geomagnetic region, an input-output mapping relationship is established between geomagnetic observation data from multiple geomagnetic observatories and prior data from the target geomagnetic observatory. The geomagnetic observation data from multiple geomagnetic observatories are input into the optimized LSTM prediction model, and the D, H, and Z components of the magnetic field strength of the target geomagnetic observatory are output.

6. A geomagnetic field intensity prediction system for single stations and regions based on LSTM networks, characterized in that, include: The data acquisition module is used to acquire historical data from geomagnetic observatories. The data preprocessing module is used to preprocess historical data; The model building and optimization module is used to build an LSTM prediction model and train and predict based on preprocessed historical data to obtain an optimized LSTM prediction model. The prediction module uses an optimized LSTM prediction model to predict the magnetic field strength of a single geomagnetic station and a geomagnetic region.

7. The geomagnetic field intensity prediction system for a single geomagnetic station and region based on LSTM networks according to claim 6, characterized in that, The prediction module includes a single-station prediction submodule and a regional prediction submodule, wherein: The single-station prediction submodule is used to predict the geomagnetic intensity components of a single geomagnetic station. The regional prediction submodule is used to predict the geomagnetic intensity of a target geomagnetic station using geomagnetic observation data from multiple input geomagnetic observatories.

8. A computer device, characterized in that: include: One or more processors, and a memory storing one or more computer programs, wherein the processors invoke the computer programs to implement the steps of the prediction method according to any one of claims 1-5.

9. A computer-readable storage medium, characterized in that: A computer program is stored, which is invoked by a processor to implement the steps of the prediction method according to any one of claims 1-5.

Images