Geomagnetic modeling method and system based on long short-term memory network optimization

Abstract

The invention relates to a geomagnetic modeling method and system based on long short-term memory network optimization, and the method comprises the steps: building a plurality of long short-term memory networks, and enabling the plurality of long short-term memory networks to be connected in series to form a first network module; and then, establishing communication connection between the Colmogorov-Arnold network after modification of the activation function and the first network module, and finally, forming an optimization mechanism for jointly optimizing a long-short-term memory network by using a sparrow search algorithm and the Colmogorov-Arnold network, and constructing a high-precision geomagnetic modeling system. The problems that a traditional geomagnetic model is low in precision, gradient disappears or explodes in geomagnetic modeling based on a traditional long-short-term memory network, and the nonlinear fitting capacity is weak are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer models and systems, and in particular to a geomagnetic modeling method and system based on long short-term memory network optimization. Background Art

[0002] The distribution of the geomagnetic field carries information about the distribution of the magnetic field in the space inside and outside the earth, which affects the progress of geophysical exploration, remote sensing positioning, description of the laws of planetary dynamics and communication. The conventional method of geomagnetic field modeling is spherical harmonic analysis, which uses the sum of a finite number of spherical harmonic functions (truncated spherical harmonic series) with increasing orders (increasing spatial resolution) to approximate the geomagnetic field scalar potential and obtain the best fit for the geomagnetic survey data.

[0003] However, the spherical harmonic analysis method is limited by the truncation level of the spherical harmonic series, and the spatial resolution is generally not high. When the truncation order increases, the number of spherical harmonic coefficients to be determined increases sharply, and the coefficients of the high-order terms are much smaller than the coefficients of the low-order terms (the difference is as large as 6 orders of magnitude), which makes the statistical significance of this method questionable. Therefore, the conventional spherical harmonic analysis method is far from meeting the actual needs of prospecting and navigation.

[0004] In this regard, the existing technology uses a long short-term memory (LSTM) network model to complete the modeling of the geomagnetic field distribution. This model is a special recurrent neural network (RNN). The linear combination + activation function mode is used between the layers of this neural network to achieve the processing effect of first implementing linear combination and then mapping through the activation function.

[0005] However, the model is prone to gradient vanishing and gradient exploding problems during back propagation and gradient calculation; at the same time, the activation function between the network layers of the model cannot be changed after selection, so only the weight parameters of the linear combination of the model can be optimized, which makes the nonlinear ability of the model weak, and thus makes it difficult for the model to characterize the nonlinear relationship between the geomagnetic field and the spatial and temporal factors in the geomagnetic field distribution. If its nonlinear ability is to be increased, it is necessary to add additional nodes and layers, which increases the consumption of computing resources. Summary of the invention

[0006] The technical problem to be solved by the present invention is how to overcome the technical defects of the prior art, that is, the low accuracy of the geomagnetic model and the long updating time of the spherical harmonic parameters. In order to overcome the defects of the prior art, the present invention provides a geomagnetic modeling method and system based on long short-term memory network optimization, including a geomagnetic modeling method based on long short-term memory network optimization and a geomagnetic modeling system based on long short-term memory network optimization.

[0007] The present invention provides a geomagnetic modeling method based on long short-term memory network optimization, comprising the following steps: S1: acquiring satellite magnetic survey data, performing data preprocessing on the satellite magnetic survey data to obtain preprocessing results, and dividing the preprocessing results into a training set and a validation set; S2: Building a plurality of long short-term memory networks, and then connecting the plurality of the long short-term memory networks in series to form a first network module, and setting a value range of a hyperparameter of the first network module; S3: assigning the activation function of the Kolmogorov-Arnold network to the residual activation function to obtain the second network module; S4: establishing a communication connection between the input end of the second network module and the output end of the first network module to obtain a geomagnetic modeling system to be optimized; S5: Optimizing the model parameters of the geomagnetic modeling system to be optimized by using the training set and the verification set within the hyperparameter value range through a sparrow search algorithm to obtain a geomagnetic modeling system.

[0008] The geomagnetic modeling method based on long short-term memory network optimization disclosed in the present invention aims at the technical defects existing in the prior art. A plurality of long short-term memory networks are built through step S2, and then the plurality of long short-term memory networks are connected in series to form a first network module. Then, the Kolmogorov-Arnold network (second network module) after the activation function is modified through step S3 is established with the first network module for communication connection, and finally an optimization mechanism for jointly optimizing the long short-term memory network with the sparrow search algorithm and the Kolmogorov-Arnold network is formed. A high-precision geomagnetic modeling system can be constructed, which solves the problems of low precision of traditional geomagnetic models and gradient disappearance or explosion and weak nonlinear fitting ability in geomagnetic modeling based on traditional long short-term memory networks.

[0009] In one possible implementation, the satellite magnetic survey data includes three components of the magnetic field in the Universal Coordinated Time, longitude, latitude, spheroidal distance, and north-east coordinate system; the data set constructed based on this solution has a simple structure and is easy to screen data, which can reduce the computational complexity brought about by parameter optimization, thereby improving the efficiency of parameter optimization.

[0010] In a possible implementation, step S1 includes the following steps: S11: acquiring the satellite magnetic survey data, and selecting satellite magnetic survey data having a stable magnetic field from the satellite magnetic survey data according to the influence of light on the magnetic field; S12: using the world coordinated time, longitude, latitude, and spherical center distance in the satellite magnetic survey data with a stable magnetic field as input parameters of the CHAOS model to obtain three components of the model magnetic field through the CHAOS model, and obtaining the three-dimensional coordinates of the position point according to the spherical center distance, longitude, and latitude; S13: Using longitude, latitude, spherical center distance, three-dimensional coordinates of the location point, and three components of the model magnetic field as input data, and the three components of the magnetic field in the north-east coordinate system as output data, a synthetic data set is created; S14: normalizing the data set, and then setting the observation window size to produce a time series data set; S15: taking the time series data set produced by the step S14 as the preprocessing result, and dividing it into a training set and a validation set in proportion; This scheme filters data based on the impact of light on the magnetic field when screening data, which can reduce the impact of light on the magnetic field, thereby screening out stable fields in the magnetic field, and then obtaining the three components of the model magnetic field and the three-dimensional coordinates of the location points, and finally forming a data set. The data set has nine input quantities (longitude, latitude, distance from the center of the sphere, three-dimensional coordinates of the location points, and three components of the model magnetic field) and three output quantities (three components of the magnetic field in the north-east coordinate system), thereby ensuring that the input data is compatible with the long short-term memory network, further reducing the computational complexity of parameter optimization.

[0011] In a possible implementation, the residual activation function is calculated as follows: , In the formula, represents the basis functions of the B-spline function; Represents the linear combination coefficient, which is a parameter that needs to be optimized for the model parameters; The residual activation function constructed by this scheme can further enhance the nonlinear capability of the system, thereby further improving the accuracy of the geomagnetic model of the geomagnetic modeling system.

[0012] In a possible implementation manner, step S5 includes the following steps: S51: setting the warning value, the ratio of discoverers, the ratio of anti-predators and the maximum iteration round of the sparrow search algorithm, and determining the population size of the sparrow search algorithm, and then determining the number of discoverers and the number of anti-predators according to the ratio of discoverers and the ratio of anti-predators, and then executing the next step; S52: taking out position coordinates of a number equal to the population size from the hyperparameter value range as the current position coordinates of all individuals in the current population, and then executing the next step; S53: respectively taking the current position coordinates of all individuals in the current population as the hyperparameters of the first network module, optimizing the model parameters of the geomagnetic modeling system to be optimized using the training set to obtain an optimized system of each current position coordinate, and then executing the next step; S54: using the verification set, respectively calculating the root mean square error of the optimization system of each current position coordinate to obtain the corresponding fitness, and then executing the next step; S55: sorting all fitness obtained in step S54 to determine the order of the current positions of all individuals in the current population, and then executing the next step; S56: Determine whether the current iteration round is equal to the maximum iteration round; If so, the optimization system of the position coordinates of the individual currently having the best fitness is used as the geomagnetic modeling system; If not, proceed to the next step; S57: using the finder update formula of the sparrow search algorithm to update the positions of all finders determined by the step S51, using the anti-predator update formula of the sparrow search algorithm to update the positions of all anti-predators determined by the step S51, using the joiner update formula of the sparrow search algorithm to determine the joiners of the finder in the current population and update the position of each joiner, and then executing the next step; S58: removing all individuals whose current position coordinates are not within the range of the hyperparameter value to update the population, and then returning to execute step S53; This scheme optimizes the model parameters of the geomagnetic modeling system to be optimized using the training set and the validation set through the sparrow search algorithm within the range of hyperparameter values, thereby achieving the completion and optimization of the long short-term memory network, so that the obtained geomagnetic modeling system has nonlinear capabilities and the modeling accuracy is further improved.

[0013] In a possible implementation, the finder update formula is as follows: , In the formula, represents the standard uniform distribution; represents the standard normal distribution; represents the number of said discoverers; The spatial dimension representing the range of values ​​of the hyperparameter; Representative of the population Individuals are conducting After the iteration Dimension location information; represents the maximum iteration round; represents the warning value; This solution can quickly and regularly update the location information of the discoverer during the execution of the sparrow search algorithm, so that the efficiency of the algorithm solving or optimizing parameters is further improved.

[0014] In a possible implementation, the anti-predator update formula is as follows: , In the formula, Representative interval Uniform distribution on Representative of the population The anti-predator In the iteration Dimension location information; Representatives conduct After the iteration, the individual position with the global optimal fitness is obtained. Dimension location information; Representative of the population Anti-predator The fitness obtained after iterations; Represents the global optimal fitness of the population; Representatives conduct After the iteration, the individual position with the worst global fitness is obtained. Dimension location information; Represents the global worst fitness of the population; represents the scale factor; Represents any decimal number; This solution can achieve rapid and regular updating of the anti-predator position information during the execution of the sparrow search algorithm, which can ultimately further improve the efficiency of the algorithm in solving or optimizing parameters.

[0015] In a possible implementation, the joiner update formula is as follows: , In the formula, It represents randomly selecting a number between -1 and 1; This scheme can determine the joiners among the discoverers and update the position coordinates of each joiner during the execution of the sparrow search algorithm. It can increase the activity of the individuals in the population to more accurately obtain the optimal fitness within the maximum iteration round range, and complete the model parameter optimization of the geomagnetic modeling system to be optimized.

[0016] Another technical solution of the present invention is to provide a geomagnetic modeling system based on long short-term memory network optimization, which is a geomagnetic modeling system obtained by executing the geomagnetic modeling method based on long short-term memory network optimization disclosed in the present invention.

[0017] The geomagnetic modeling system disclosed in the present invention addresses the technical defects of low geomagnetic model accuracy and long spherical harmonic parameter update time in the prior art, and solves the problems of low accuracy of traditional geomagnetic models and gradient disappearance or explosion and weak nonlinear fitting ability in geomagnetic modeling based on traditional long short-term memory networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of the structure of a geomagnetic modeling system based on long short-term memory network optimization disclosed in an embodiment of the present application; Figure 2 Schematic diagram of the Kolmogorov-Arnold network structure disclosed in this embodiment;

[0019] Figure 3 To construct a flow chart of the geomagnetic modeling method disclosed in the embodiments of the present application; Figure 4 A flowchart corresponding to step S1 in the method for constructing the geomagnetic modeling system disclosed in the embodiment of the present application; Figure 5 A flowchart corresponding to step S5 in the method for constructing the geomagnetic modeling system disclosed in the embodiment of the present application; Figure 6 It is a curve diagram showing the change of mean square error loss of the long short-term memory network model in the prior art with the iteration rounds; Figure 7 A curve diagram showing changes in mean square error loss of the geomagnetic modeling system disclosed in an embodiment of the present application with respect to iteration rounds; Figure 8 for Figure 7 The corresponding curve image is an enlarged view of the segment from 50 to 150 iteration rounds; Fig. 9 It is a curve diagram showing the change of the prediction mean square error of the long short-term memory network model in the prior art with the iteration rounds; Fig.10 This is a graph showing how the predicted mean square error of the geomagnetic modeling system disclosed in an embodiment of the present application changes with the number of iterations. DETAILED DESCRIPTION

[0020] First, those skilled in the art should understand that these implementations are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0021] In the embodiments of the present application, unless otherwise clearly specified and limited, the communication or communication connection between the first feature and the second feature refers to the transmission of information between the first feature and the second feature. This information transmission can be either unidirectional or bidirectional, and the communication connection can be achieved by electrical connection of wires, radio connection, electrical connection of electromagnetic media (such as semiconductors), communication achieved by channels, etc.

[0022] In the embodiments of the present application, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0023] The present application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] See also Figure 1 to Figure 5 As shown, the embodiment of the present application discloses a geomagnetic modeling system based on long short-term memory network optimization, Figure 1 : is a structural diagram of the geomagnetic modeling system, which includes a first network module and a second network module. The input end of the second network module is communicatively connected to the output end of the first network module. The first network module includes a plurality of long short-term memory networks. In this embodiment, three long short-term memory networks are used, and the second network module includes a Kolmogorov-Arnold network (KAN), and the activation function of the Kolmogorov-Arnold network is designated as a residual activation function. The structural diagram of the Kolmogorov-Arnold network is shown in FIG. Figure 2 shown.

[0025] The geomagnetic modeling method will be further disclosed below, see Figure 3 As shown, the method comprises the following steps: S1: Obtain satellite magnetic survey data, perform data preprocessing on the satellite magnetic survey data to obtain preprocessing results, and divide the preprocessing results into a training set and a validation set.

[0026] The satellite magnetic data retrieved in this embodiment include three components of the magnetic field in the universal time coordinate system, longitude, latitude, geocentric distance, and north-east coordinate system. Figure 4 As shown, step S1 in this embodiment includes the following steps: S11: Acquire satellite magnetic survey data, and select satellite magnetic survey data with a stable magnetic field from the satellite magnetic survey data according to the influence of light on the magnetic field.

[0027] Specifically, we can first use the cdfread function in the mathematical experimental model (MATLAB) to read the three components of the magnetic field under the Universal Time Coordinated (UTC), longitude, latitude, geocentric distance and North East Coordinate (NEC coordinate system) in the Common Data Format (CDF) file of the satellite magnetic data in the satellite raw data, convert these data into a .mat format file, and then enter the data screening stage.

[0028] In this embodiment, the process of screening data is to screen out the stable field in the magnetic field by reducing the influence of light on the magnetic field. The specific screening requirements are as follows: Latitude range requirement: [-60,60]; Local time range requirements: 18:00 of the current day to 6:00 of the next day; Latitude range requirement: above 60 degrees; Time range requirement: local time is 20:00 on the current day to 4:00 on the next day.

[0029] Local time can be converted using UTC and longitude. The conversion formula is as follows: , , In the formula, It represents rounding down. A positive time zone represents the eastern time zone, and a negative time zone represents the western time zone. -12 and 12 represent the western 12th zone and the eastern 12th zone respectively.

[0030] S12: The world coordinated time, longitude, latitude, and spherical center distance in the satellite magnetic data with a stable magnetic field are used as input parameters of the CHAOS model (full name: CHAMP / Ørsted / SAC-C model) to obtain the three components of the model magnetic field (i.e., CHAOS_X, CHAOS_Y, and CHAOS_Z) through the CHAOS model, and the three-dimensional coordinates of the position point are obtained according to the spherical center distance, longitude, and latitude. The formula for obtaining the three-dimensional coordinates of the position point according to the spherical center distance, longitude, and latitude is as follows: , In the formula, represents the distance from the center of the ball; represents latitude; Represents longitude.

[0031] S13: The longitude, latitude, spherical center distance, three-dimensional coordinates of the location point, and the three components of the model magnetic field are used as input data, and the three components of the magnetic field in the north-east coordinate system are used as output data to synthesize the data set.

[0032] S14: normalize the data set, and then set the observation window size to produce a time series data set. The normalization process uses Z-Score normalization. The observation window size set in this embodiment is 16, and the time series data set is produced accordingly to more easily capture time correlation.

[0033] S15: The time series data set produced by step S14 is used as a preprocessing result, and is divided into a training set and a validation set in proportion. The division ratio may be 4:1.

[0034] S2: Build several long short-term memory networks, then connect several long short-term memory networks in series to form a first network module, and set the value range of hyperparameters of the first network module.

[0035] In this embodiment, three long short-term memory networks are used to form the first network module, and the input port of the first network module is set to 9 ports for inputting data (i.e., longitude, latitude, spherical center distance, three-dimensional coordinates of the location point, and three components of the model magnetic field) into the first network module. In addition, as an implementation method, in this embodiment, the hyperparameter value range is set as follows: , , , In the formula, Represents the number of network nodes of the LSTM network at the head end; Represents the number of network nodes in the middle long short-term memory network; Represents the number of network nodes of the LSTM network at the end.

[0036] S3: Assign the activation function of the Kolmogorov-Arnold network to the residual activation function to obtain the second network module.

[0037] In this embodiment, the residual activation function is calculated as follows: , In the formula, represents the basis functions of the B-spline function; Represents the linear combination coefficient, which is a parameter that needs to be optimized for the model.

[0038] S4: Establish a communication connection between the input end of the second network module and the output end of the first network module to obtain the geomagnetic modeling system to be optimized.

[0039] S5: The model parameters of the geomagnetic modeling system to be optimized are optimized by using the training set and the validation set within the range of hyperparameter values ​​through the sparrow search algorithm to obtain the geomagnetic modeling system.

[0040] See also Figure 5 As shown, in this embodiment, step S5 includes the following steps: S51: Set the warning value, discoverer ratio, anti-predator ratio and maximum iteration round of the sparrow search algorithm, and determine the population size of the sparrow search algorithm, and then determine the number of discoverers and the number of anti-predators according to the discoverer ratio and the anti-predator ratio, and then execute the next step.

[0041] Specifically in this embodiment, the discoverer ratio is set to 0.7, the anti-predator ratio is set to 0.2, and the maximum iteration round is set to 10.

[0042] S52: Take out position coordinates equal to the population size in the range of hyperparameter values ​​as the current position coordinates of all individuals in the current population, and then execute the next step.

[0043] S53: The current position coordinates of all individuals in the current population are respectively used as the hyperparameters of the first network module, and the model parameters of the geomagnetic modeling system to be optimized are optimized using the training set to obtain the optimized system of each current position coordinate, and then the next step is executed.

[0044] S54: Calculate the root mean square error of the optimization system of each current position coordinate using the verification set to obtain the corresponding fitness, and then execute the next step.

[0045] S55: Sort all fitness obtained in step S54 to determine the order of the current positions of all individuals in the current population, and then execute the next step.

[0046] S56: After going through one iteration round from step S53 to step S58, it is determined whether the current iteration round is equal to the maximum iteration round; If so, the optimization system of the position coordinates of the individual with the best fitness at present is used as the geomagnetic modeling system; If not, proceed to the next step.

[0047] S57: Use the discoverer update formula of the sparrow search algorithm to update the position of all discoverers determined by step S51, use the anti-predator update formula of the sparrow search algorithm to update the position of all anti-predators determined by step S51, use the joiner update formula of the sparrow search algorithm to determine the joiners of the discoverer in the current population and update the position of each joiner, and then execute the next step.

[0048] In this embodiment, the calculation formula of the discoverer update formula is as follows: , In the formula, represents the standard uniform distribution; represents the standard normal distribution; Represents the number of discoverers; The spatial dimension representing the range of hyperparameter values; since three long short-term memory networks are used in this embodiment, the spatial dimension is 3; Representative of the population Individuals are conducting After the iteration Dimension location information; represents the maximum iteration round; Represents the warning value.

[0049] At the same time, the anti-predator update formula is calculated as follows: , In the formula, Representative interval Uniform distribution on Representative of the population The anti-predator In the iteration Dimension location information; Representatives conduct After the iteration, the individual position with the global optimal fitness is obtained. Dimension location information; Representative of the population Anti-predator The fitness obtained after iterations; Represents the global optimal fitness of the population; Representatives conduct After the iteration, the individual position with the worst global fitness is obtained. Dimension location information; Represents the global worst fitness of the population; represents the scale factor; Represents any decimal, that is, a decimal that can take any value. In this embodiment, the value is .

[0050] In addition, in this embodiment, the calculation formula of the joiner update formula is as follows: , In the formula, It represents randomly selecting a number between -1 and 1.

[0051] S58: Remove all individuals whose current position coordinates are not within the range of the hyperparameter value to update the individuals in the population, and then return to execute step S53.

[0052] Figure 6 , Figure 7 and Figure 8 A comparison diagram of the mean square error loss between the geomagnetic modeling system disclosed in this embodiment and a conventional geomagnetic modeling system is shown. Figure 6 It is a curve diagram showing the change of mean square error loss of the long short-term memory network model in the prior art with the iteration rounds. Figure 7 is a curve diagram showing the variation of the mean square error loss of the geomagnetic modeling system disclosed in this embodiment with the iteration rounds, Figure 8 for Figure 7 The corresponding curve image is an enlarged view of the segment with iteration rounds of 50 to 150. Figure 6and Figure 8 It can be seen that when it iterates to 150 rounds, the mean square error loss of the long short-term memory network model of the prior art drops from the initial 3.07 to 0.92, while the mean square error loss of the geomagnetic modeling system disclosed in this embodiment drops from the initial 2.79 to 0.07. It can be seen that the fitting ability of the geomagnetic modeling system disclosed in this embodiment is better than that of the long short-term memory network model of the prior art.

[0053] Fig. 9 and Fig.10 A comparison diagram of the predicted mean square error between the geomagnetic modeling system disclosed in this embodiment and a conventional geomagnetic modeling system is shown. Fig. 9 It is a curve diagram showing the change of the prediction mean square error of the long short-term memory network model in the prior art with the iteration rounds. Fig.10 This is a graph showing the change of the predicted mean square error of the geomagnetic modeling system disclosed in this embodiment with the iteration rounds. Fig. 9 and Fig.10 It can be seen that the prediction mean square error loss of the long short-term memory network model of the prior art drops from the initial 6.12 to 1.82, and the final loss value is still large. Therefore, the model has poor fitting effect in the training set and the validation set. The prediction mean square error loss of the geomagnetic modeling system disclosed in this embodiment drops from the initial 2.32 to 0.10. The final mean square loss of the training set and the validation set of this model is significantly reduced compared with the long short-term memory network model of the prior art. Therefore, the fitting effect of the training set and the validation set is better than that of the long short-term memory network model of the prior art.

[0054] The following table shows the comparison of prediction accuracy between the geomagnetic modeling system disclosed in this embodiment and the long short-term memory network model (hereinafter referred to as LSTM) of the prior art:

[0055] It can be seen from the above table that the geomagnetic modeling system of this embodiment is better than the LSTM modeling.

[0056] It can be seen that the geomagnetic modeling system disclosed in this embodiment has the advantage of high accuracy, and can solve the problems of low accuracy of traditional geomagnetic models and gradient disappearance or explosion and weak nonlinear fitting ability in geomagnetic modeling based on traditional long short-term memory networks.

[0057] In the description of the embodiments of the present application, it should be noted that in the description of the present application, terms such as "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is only for the convenience of description, and does not indicate or imply that the device or component must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present application.

[0058] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" etc. means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0059] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A geomagnetic modeling method based on long short-term memory network optimization, characterized in that: The steps include: S1: acquiring satellite magnetic survey data, performing data preprocessing on the satellite magnetic survey data to obtain preprocessing results, and dividing the preprocessing results into a training set and a validation set; S2: Building a plurality of long short-term memory networks, and then connecting the plurality of the long short-term memory networks in series to form a first network module, and setting a value range of a hyperparameter of the first network module; S3: assigning the activation function of the Kolmogorov-Arnold network to the residual activation function to obtain the second network module; S4: establishing a communication connection between the input end of the second network module and the output end of the first network module to obtain a geomagnetic modeling system to be optimized; S5: Optimizing the model parameters of the geomagnetic modeling system to be optimized by using the training set and the verification set within the hyperparameter value range through a sparrow search algorithm to obtain a geomagnetic modeling system.

2. The geomagnetic modeling method based on long short-term memory network optimization according to claim 1 is characterized in that: The satellite magnetic survey data includes three components of the magnetic field in the Universal Coordinated Time, longitude, latitude, geocentric distance, and north-east coordinate system.

3. The geomagnetic modeling method based on long short-term memory network optimization according to claim 2 is characterized in that: The step S1 comprises the following steps: S11: acquiring the satellite magnetic survey data, and selecting satellite magnetic survey data having a stable magnetic field from the satellite magnetic survey data according to the influence of light on the magnetic field; S12: using the world coordinated time, longitude, latitude, and spherical center distance in the satellite magnetic survey data with a stable magnetic field as input parameters of the CHAOS model to obtain three components of the model magnetic field through the CHAOS model, and obtaining the three-dimensional coordinates of the position point according to the spherical center distance, longitude, and latitude; S13: Using longitude, latitude, spherical center distance, three-dimensional coordinates of the location point, and three components of the model magnetic field as input data, and the three components of the magnetic field in the north-east coordinate system as output data, a synthetic data set is created; S14: normalizing the data set, and then setting the observation window size to produce a time series data set; S15: taking the time series data set produced by the step S14 as the preprocessing result, and dividing it into a training set and a validation set in proportion.

4. The geomagnetic modeling method based on long short-term memory network optimization according to claim 3 is characterized in that: The formula of the residual activation function is as follows: , In the formula, represents the basis functions of the B-spline function; Represents the linear combination coefficient, which is a parameter that needs to be optimized for the model.

5. The geomagnetic modeling method based on long short-term memory network optimization according to any one of claims 2 to 4, characterized in that: The step S5 comprises the following steps: S51: setting the warning value, the ratio of discoverers, the ratio of anti-predators and the maximum iteration round of the sparrow search algorithm, and determining the population size of the sparrow search algorithm, and then determining the number of discoverers and the number of anti-predators according to the ratio of discoverers and the ratio of anti-predators, and then executing the next step; S52: taking out position coordinates of a number equal to the population size from the hyperparameter value range as the current position coordinates of all individuals in the current population, and then executing the next step; S53: respectively taking the current position coordinates of all individuals in the current population as the hyperparameters of the first network module, optimizing the model parameters of the geomagnetic modeling system to be optimized using the training set to obtain an optimized system of each current position coordinate, and then executing the next step; S54: using the verification set, respectively calculating the root mean square error of the optimization system of each current position coordinate to obtain the corresponding fitness, and then executing the next step; S55: sorting all fitness obtained in step S54 to determine the order of the current positions of all individuals in the current population, and then executing the next step; S56: Determine whether the current iteration round is equal to the maximum iteration round; If so, the optimization system of the position coordinates of the individual currently having the best fitness is used as the geomagnetic modeling system; If not, proceed to the next step; S57: using the finder update formula of the sparrow search algorithm to update the positions of all finders determined by the step S51, using the anti-predator update formula of the sparrow search algorithm to update the positions of all anti-predators determined by the step S51, using the joiner update formula of the sparrow search algorithm to determine the joiners of the finder in the current population and update the position of each joiner, and then executing the next step; S58: Remove all individuals whose current position coordinates are not within the range of the hyperparameter value to update the population, and then return to execute step S53.

6. The geomagnetic modeling method based on long short-term memory network optimization according to claim 5 is characterized in that: The formula of the discoverer update formula is as follows: ， In the formula, represents the standard uniform distribution; represents the standard normal distribution; represents the number of said discoverers; The spatial dimension representing the range of values ​​of the hyperparameter; Representative of the population Individuals are conducting After the iteration Dimension location information; represents the maximum iteration round; Represents the warning value.

7. The geomagnetic modeling method based on long short-term memory network optimization according to claim 6 is characterized in that: The anti-predator update formula is as follows: ， In the formula, Representative interval Uniform distribution on Representative of the population The anti-predator In the iteration Dimension location information; Representative of the population Anti-predator The fitness obtained after iterations; Represents the global optimal fitness of the population; Representatives conduct After the iteration, the individual position with the worst global fitness is obtained. Dimension location information; Represents the global worst fitness of the population; represents the scale factor; Represents any decimal number.

8. The geomagnetic modeling method based on long short-term memory network optimization according to claim 6 or 7, characterized in that: The formula for the joiner update formula is as follows: ， In the formula, It represents randomly selecting a number between -1 and 1.

9. A geomagnetic modeling system based on long short-term memory network optimization, characterized in that: The geomagnetic modeling system is a geomagnetic modeling system obtained by executing the geomagnetic modeling method based on long short-term memory network optimization as described in any one of claims 1 to 8.

Images