Missile track complementing method, device and equipment based on digital main line and medium

Abstract

The invention provides a missile track completion method and device based on a digital main line, equipment and a medium, and the method comprises the steps: obtaining underwater track data of a submarine-launched missile from the digital main line, inputting the underwater track data into a target feature value completion model, and completing the target feature value of the submarine-launched missile according to the target feature value completion model; after processing of the target characteristic value completion model, outputting completion data corresponding to the underwater track data; wherein hyper-parameters of the target feature value completion model are obtained through a particle swarm algorithm, and the hyper-parameters are used for training the initial feature value completion model to obtain the target feature value completion model. Visibly, according to the embodiment of the invention, in the training process of the initial feature value completion model, the particle swarm optimization is introduced to optimize the hyper-parameter, so that the accuracy of the trained target feature value completion model is improved, and the completion data corresponding to the output underwater track data after the processing of the target feature value completion model is more accurate.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular to a missile track completion method, device, equipment and medium based on a digital main line. Background Art

[0002] With the continuous development of modern military technology, submarine-launched missiles play a vital role in national defense strategy. Submarine-launched missiles have the characteristics of strong concealment, high strike accuracy and flexibility, and are widely used in strategic weapons and equipment of various countries.

[0003] At present, submarine-launched missiles mainly rely on inertial navigation systems (INS) and global positioning systems (GPS) for navigation and control. However, due to the complex underwater environment and limited signal propagation, the performance of navigation and control is affected, resulting in missing underwater track data of submarine-launched missiles. The accuracy of existing underwater track data completion methods is low, resulting in inaccurate completion data of underwater track data. Therefore, how to improve the accuracy of underwater track data completion data has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In order to solve the above technical problems, the embodiments of the present disclosure provide a missile track completion method, device, equipment and medium based on digital main line.

[0005] In a first aspect, the present disclosure provides a missile track completion method based on a digital main line, the method comprising:

[0006] Acquire underwater track data of a submarine-launched missile from the digital main line; wherein the underwater track data has missing characteristic values;

[0007] The underwater track data is input into the target eigenvalue completion model, and after being processed by the target eigenvalue completion model, the output is the completion data corresponding to the underwater track data; wherein, the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network.

[0008] In an optional implementation manner, before the underwater track data is input into the target feature value completion model and, after being processed by the target feature value completion model, is output as the completion data corresponding to the underwater track data, the method further includes:

[0009] The particle swarm algorithm is used to optimize the initial hyperparameter values ​​of the initial eigenvalue completion model to obtain the optimized hyperparameter values;

[0010] Based on the optimized hyperparameter values, the initial eigenvalue completion model is trained to obtain a trained eigenvalue completion model;

[0011] The mean square error of the generator in the trained feature value completion model is used as the particle fitness value of the particle swarm algorithm, and based on the particle fitness value, the optimized hyperparameter value is adjusted to obtain a target hyperparameter value;

[0012] Determine whether the particle swarm algorithm satisfies a preset termination condition. If the particle swarm algorithm satisfies the preset termination condition, terminate the particle swarm algorithm, and use the trained eigenvalue completion model obtained by training the initial eigenvalue completion model based on the target hyperparameter value as the target eigenvalue completion model.

[0013] In an optional implementation manner, the method further includes:

[0014] If the particle swarm algorithm does not meet the preset termination condition, the target hyperparameter value is used to update the optimized hyperparameter value, and the step of training the initial eigenvalue completion model based on the optimized hyperparameter value to obtain the trained eigenvalue completion model is continued until the particle swarm algorithm meets the preset termination condition.

[0015] In an optional implementation, the initial eigenvalue completion model is trained based on the optimized hyperparameter value to obtain the trained eigenvalue completion model, and the method further includes:

[0016] Based on the underwater complete track data samples of the submarine-launched missile, the corresponding mask samples and the random noise samples, the data samples to be interpolated are constructed;

[0017] Accordingly, the initial eigenvalue completion model is trained based on the optimized hyperparameter value to obtain a trained eigenvalue completion model, including:

[0018] Fixing the optimized hyperparameter values ​​and the parameter values ​​of the generator of the initial eigenvalue completion model, taking the complete track data sample and the completed data sample output by the generator as input, training the discriminator in the initial eigenvalue completion model, and obtaining a trained discriminator;

[0019] The optimized hyperparameter values ​​and the trained discriminator parameter values ​​are fixed, the data samples to be interpolated and the corresponding mask samples are taken as input, and the generator in the initial eigenvalue completion model is trained to obtain the trained generator; wherein the trained discriminator and the trained generator are used to construct the trained eigenvalue completion model.

[0020] In an optional implementation, the parameter value of the discriminator in the initial eigenvalue completion model is adjusted based on a gradient penalty mechanism.

[0021] In an optional implementation manner, after the underwater track data is input into a target feature value completion model and, after being processed by the target feature value completion model, is output as completion data corresponding to the underwater track data, the method further includes:

[0022] The supplementary data is sent to the digital main line to generate a complete underwater navigation trajectory of the submarine-launched missile.

[0023] In an optional implementation, the underwater complete track data sample is a sample obtained after preprocessing the initial underwater complete track data sample of the submarine-launched missile obtained in the digital main line, and the preprocessing includes performing cluster analysis on the initial underwater complete track data sample, and applying the Euclidean distance including the time factor in the cluster analysis to eliminate outliers in the initial underwater complete track data sample.

[0024] In a second aspect, the present disclosure provides a missile track completion device based on a digital main line, the device comprising:

[0025] An acquisition module is used to acquire underwater track data of a submarine-launched missile from a digital main line; wherein the underwater track data has missing characteristic values;

[0026] A processing module is used to input the underwater track data into a target eigenvalue completion model, and after being processed by the target eigenvalue completion model, output the completion data corresponding to the underwater track data; wherein, the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network.

[0027] In a third aspect, the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device implements the above method.

[0028] In a fourth aspect, the present disclosure provides a missile track completion device based on a digital main line, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above method is implemented when the processor executes the computer program.

[0029] In a fifth aspect, the present disclosure provides a computer program product, wherein the computer program product comprises a computer program / instructions, and the computer program / instructions implement the above method when executed by a processor.

[0030] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has at least the following advantages:

[0031] The disclosed embodiment provides a missile track completion method based on a digital main line, which obtains underwater track data of a submarine-launched missile from the digital main line, wherein there is a missing eigenvalue in the underwater track data, and the underwater track data is input into a target eigenvalue completion model, and after being processed by the target eigenvalue completion model, the output is the completion data corresponding to the underwater track data; wherein the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network. It can be seen that in the disclosed embodiment, in the training process of the initial eigenvalue completion model, a particle swarm algorithm is introduced to optimize the hyperparameters, thereby improving the accuracy of the target eigenvalue completion model after training, so that after being processed by the target eigenvalue completion model, the completion data corresponding to the output underwater track data is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0034] Figure 1 A flowchart of a missile track completion method based on a digital main line provided in an embodiment of the present disclosure;

[0035] Figure 2 A flowchart of another missile track completion method based on a digital main line provided by an embodiment of the present disclosure;

[0036] Figure 3 A schematic diagram of the structure of a missile track completion device based on a digital main line provided in an embodiment of the present disclosure;

[0037] Figure 4 A schematic structural diagram of a missile track completion device based on a digital main line provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0038] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0039] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0040] With the continuous development of modern military technology, submarine-launched missiles play a vital role in national defense strategy. Submarine-launched missiles have the characteristics of strong concealment, high strike accuracy, and maneuverability, and are widely used in the strategic weapons and equipment of various countries. Existing submarine-launched missiles mainly rely on a combination of inertial navigation systems (INS) and global positioning systems (GPS) for navigation and control. The inertial navigation system calculates the trajectory of the missile by detecting its acceleration and angular velocity; the global positioning system provides accurate location information through satellite signals. Although this combined method shows high accuracy and reliability during the missile flight phase, in the initial underwater stage of the missile launch, due to problems such as complex environment and limited signal propagation, as described below:

[0041] 1. Accumulation of inertial navigation system errors: Inertial navigation systems will produce accumulated errors during long-term operation, especially in underwater environments, where the accumulated errors are more obvious, affecting the accuracy of the trajectory. Errors in the underwater stage are difficult to effectively correct using traditional methods, causing the missile to deviate from the predetermined trajectory.

[0042] 2. Environmental interference: The complex ocean environment, such as water currents, temperature changes, and electromagnetic interference, greatly affects the effectiveness of existing navigation and trajectory enhancement methods. These environmental factors often lead to the accumulation of errors in navigation data, reducing the accuracy of missile trajectories.

[0043] Based on the above problems, there is a lack of underwater track data for submarine-launched missiles. The lack of underwater track data directly affects the navigation system's response speed to environmental changes and decision-making accuracy.

[0044] To this end, the embodiment of the present disclosure provides a missile track completion method based on a digital main line, which obtains underwater track data of a submarine-launched missile from the digital main line, wherein there is a missing eigenvalue in the underwater track data, and the underwater track data is input into a target eigenvalue completion model, and after being processed by the target eigenvalue completion model, the output is the completion data corresponding to the underwater track data; wherein the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network. It can be seen that in the embodiment of the present disclosure, in the training process of the initial eigenvalue completion model, a particle swarm algorithm is introduced to optimize the hyperparameters, thereby improving the accuracy of the target eigenvalue completion model after training, so that after being processed by the target eigenvalue completion model, the completion data corresponding to the output underwater track data is more accurate.

[0045] Based on this, the embodiment of the present disclosure provides a missile track completion method based on a digital main line, referring to Figure 1 , is a flow chart of a missile track completion method based on a digital main line provided by an embodiment of the present disclosure, the method comprising:

[0046] S101: Obtain underwater track data of submarine-launched missiles from the digital mainline.

[0047] Among them, there are missing characteristic values ​​in the underwater track data.

[0048] The Digital Thread is a comprehensive information management platform that aims to cover the entire process from research and development and design to production, testing and operation of weapons and equipment. It integrates multiple data sources and processing modules to support real-time transmission, analysis and application of data.

[0049] In the disclosed embodiment, the underwater track data of the submarine-launched missile can be collected through the meta-model in the digital main line. Specifically, the underwater track data of the submarine-launched missile in the digital main line comes from the submarine-launched missile through the inertial navigation system (INS) and multiple sensors, such as accelerometers, gyroscopes, etc., and due to the complex environment and equipment limitations, the underwater track data of the submarine-launched missile has missing feature values.

[0050] In the disclosed embodiment, the underwater track data of the submarine-launched missile is obtained as follows: first, the initial underwater track data of the submarine-launched missile collected by the inertial navigation system (INS) and multiple sensors (such as accelerometers, gyroscopes, etc.) is collected through the digital main line. Then, the initial underwater track data is pre-processed by using the digital main line to remove abnormal values, so as to obtain the underwater track data of the submarine-launched missile.

[0051] In actual application scenarios, the process of acquiring underwater track data of submarine-launched missiles is as follows:

[0052] The submarine-launched missile collects the initial underwater track data of the submarine-launched missile through the inertial navigation system and multiple sensors. Due to the complex environment and equipment limitations, these data have missing feature values. The collected data is transmitted to the ground command center in real time through an encrypted link to ensure the integrity and confidentiality of the data. The transmitted data is integrated into the meta-model of the digital main line, and is first parsed and preliminarily stored. The system decodes the raw data from the transmission format (such as binary stream or compressed package) through the preset parsing rules. Then, the system performs data format conversion to unify the data of different sensors (such as acceleration, angular velocity, etc.) into a standardized structured data format. These standardized data contain accurate timestamps, data source information, and specific values ​​(such as time-data source-latitude and longitude-acceleration-angular velocity) to ensure the consistency and integrity of subsequent processing. Finally, the data preprocessing module automatically identifies and cleans up outliers, removes erroneous data caused by sensor failure or environmental interference, and finally obtains the underwater track data of the submarine-launched missile.

[0053] Specifically, the above preprocessing method is the same as the method for preprocessing the initial underwater complete track data sample in the following embodiment, and specifically refer to the method for preprocessing the initial underwater complete track data sample.

[0054] S102: Input the underwater track data into the target eigenvalue completion model, and after being processed by the target eigenvalue completion model, output is the completion data corresponding to the underwater track data.

[0055] Among them, the hyperparameters of the target eigenvalue completion model are obtained through the particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model. The target eigenvalue completion model is constructed based on the generative adversarial network.

[0056] In the disclosed embodiment, the target feature value completion model includes a generator and a discriminator, wherein the generator is used to generate completion data corresponding to the underwater track data based on the input underwater track data with confirmed feature values, and the discriminator is used to determine whether the input data is real or generated by the generator, and the performance of the generator is enhanced by learning the features that distinguish between real data and generated data.

[0057] Specifically, the target feature value completion model uses a hyperbolic tangent activation function in the output layers of the generator and discriminator in the generative adversarial network.

[0058] In the disclosed embodiment, the underwater track data is input into the target feature value completion model. The target feature value completion model gradually fills in the missing feature values ​​in the underwater track data through the adversarial process of the generator and the discriminator to output the completed data corresponding to the underwater track data.

[0059] In the disclosed embodiment, a particle swarm optimization algorithm is introduced during the training process of the initial eigenvalue completion model to optimize the hyperparameters of the initial eigenvalue completion model. Specifically, the particle swarm optimization algorithm uses a global search strategy to find the optimal hyperparameter configuration to improve the performance of the target eigenvalue completion model when processing missing eigenvalues ​​in the underwater track data of submarine-launched missiles. Furthermore, the initial eigenvalue completion model continues to be trained to further improve its accuracy and robustness.

[0060] The target feature value completion model is obtained through the following steps 1 to 4:

[0061] Step 1: Use the particle swarm algorithm to optimize the initial hyperparameter values ​​of the initial eigenvalue completion model to obtain the optimized hyperparameter values.

[0062] In the disclosed embodiment, the initial eigenvalue completion model is a basic model (ie, a pre-trained model) of the target eigenvalue completion model. The target eigenvalue completion model can be obtained by training the initial eigenvalue completion model.

[0063] In the disclosed embodiment, the initial hyperparameter value is obtained by initializing the hyperparameters of the initial eigenvalue completion model. Specifically, the initial hyperparameter value may be a weight attenuation coefficient and a gradient penalty coefficient that affect the mean square error part of the generator loss function in the initial eigenvalue completion model. Specifically, the initial hyperparameter value may be optimized by a particle swarm algorithm to obtain optimized hyperparameters.

[0064] In the disclosed embodiment, the parameters of the initial eigenvalue completion model and various settings of the particle swarm are first initialized to obtain the initial hyperparameter values ​​of the initial eigenvalue completion model and the parameters of the initial eigenvalue completion model. Then the particle swarm algorithm enters a dynamic iterative update phase. In this phase, the parameters of the initial eigenvalue completion model are first fixed, and then the initial hyperparameter values ​​are optimized according to the fitness function of the particle swarm algorithm to obtain the optimized hyperparameter values.

[0065] Specifically, multiple particles in the particle swarm algorithm are initialized to different hyperparameter combinations, and their performance on the initial eigenvalue completion model is evaluated through the fitness function. Then, based on the best performance of each particle and the entire particle swarm, the particle's velocity and position are updated, and this process is iterated repeatedly until the stop condition is met. Ultimately, the particle swarm algorithm can converge to the optimal or approximately optimal hyperparameter value (i.e., the optimized hyperparameter value), thereby significantly improving the performance of the initial eigenvalue completion model.

[0066] Step 2: Based on the optimized hyperparameter values, the initial eigenvalue completion model is trained to obtain a trained eigenvalue completion model.

[0067] In the disclosed embodiment, the initial eigenvalue completion model is first switched to a fixed hyperparameter state, and the initial eigenvalue completion model is trained using the optimized hyperparameter values. After each round of training, the corresponding eigenvalue completion model generator calculates the mean square error and feeds it back to the particle swarm algorithm to provide a new optimization direction. The particle swarm algorithm searches for the optimal hyperparameters for the next round based on the updated fitness value. This alternating process continues until a preset termination condition is met, such as reaching the maximum number of iterations or the loss function converges to a stable level. Through this cyclic mode, the hyperparameters continue to approach the ideal values, while ensuring that the model training process is effectively fed back, thereby ultimately obtaining the trained eigenvalue completion model.

[0068] Before training the initial eigenvalue completion model based on the optimized hyperparameter values ​​and obtaining the trained eigenvalue completion model, it is also necessary to construct the data samples to be interpolated based on the underwater complete track data samples of the submarine-launched missile, the corresponding mask samples and the random noise samples.

[0069] Among them, the underwater complete track data sample of the submarine-launched missile is a sample of the complete track data record of the submarine-launched missile during its underwater operation, including timestamp data, data source information and specific numerical data (such as acceleration, angular velocity).

[0070] In an optional implementation, the underwater complete track data sample is a sample obtained after preprocessing the initial underwater complete track data sample of the submarine-launched missile obtained in the digital main line, and the preprocessing includes clustering analysis of the initial underwater complete track data sample, and applying the Euclidean distance including the time factor in the cluster analysis to eliminate outliers in the initial underwater complete track data sample.

[0071] The initial underwater complete track data sample refers to the complete track data sample of the submarine-launched missile obtained in the digital main line without outlier processing.

[0072] In the disclosed embodiment, after obtaining the initial underwater complete track data sample of the submarine-launched missile from the digital main line, the initial underwater complete track data sample is preprocessed to obtain the underwater complete track data sample. Specifically, the preprocessing method can be obtained by applying a cluster analysis of the Euclidean distance including the time factor.

[0073] In actual application scenarios, when analyzing the initial underwater complete track data samples of submarine-launched missiles collected through the digital main line, since the data points in the initial underwater complete track data samples are collected in time sequence, the time dimension should be considered in anomaly detection. However, when the traditional Euclidean distance measurement method is applied to the above initial underwater track data for DBSCAN clustering, since DBSCAN clustering does not take into account the order of time series, it is easy to cause errors. For example, when using Euclidean distance for calculation, although the distance between abnormal data points and normal data points is reachable, the time interval is actually very far, which makes it difficult for the DBSCAN algorithm to accurately identify the location of the outlier. In addition, the traditional algorithm has a huge amount of calculation when processing a large amount of data, so it is necessary to adjust the distance calculation method to better adapt to the time series characteristics of underwater track data.

[0074] Therefore, when processing the longitude, latitude and depth data in the initial underwater complete track data sample, the Haversing formula is first used to calculate the surface distance d between two data points in the initial underwater complete track data sample:

[0075]

[0076] Among them, φ1 and φ2 are the latitudes of the two data points, Δφ is the difference between the latitudes of the two data points, Δλ is the difference between the longitudes of the two data points, and r is the radius of the earth.

[0077] Taking the depth difference into account, we can get an approximate value of the three-dimensional distance d 3D :

[0078]

[0079] Where Δh is the depth difference between two data points.

[0080] In the embodiment of the present disclosure, the longitude, latitude and depth data are processed as one-dimensional distance data, and expanded into a two-dimensional data set by adding a serialized time index. In order to eliminate the dimension difference between the distance value and the time index, the Euclidean distance between the data points (i.e., the above d 3D ), and normalize the distance, and normalize the time index. Finally, combine the normalized distance value and the time index, and use the combined Euclidean distance (i.e., the Euclidean distance including the time factor) as a new similarity metric for data classification. The specific steps are as follows:

[0081] Step 2.1: Convert the time series and distance data into a two-dimensional dataset DT.

[0082] Step 2.2: For each distance value d i and time index t i Perform normalization.

[0083]

[0084] in, is the distance d 3D The normalized value of is the normalized value of the time index, max(D) and min(D) are the maximum and minimum values ​​in the distance dataset, and max(T) and min(T) are the maximum and minimum values ​​in the time index dataset. Thus, the two-dimensional dataset DT is obtained:

[0085]

[0086] in, is the distance d 3D The normalized value of The normalized value of the time index.

[0087] Step 2.3: Calculate the Euclidean distance D(t) between the normalized data points including the time factor s ,t d ).

[0088]

[0089] in, is the distance d corresponding to the two data points 3D The normalized value of are the normalized values ​​of the time indexes corresponding to the two data points.

[0090] Step 2.4: Use D(t s ,t d ) is used as a new metric to classify the data, so as to preprocess the initial underwater complete track data samples and then obtain the underwater complete track data samples.

[0091] For example, the Euclidean distance D(t s ,t d ) as an indicator for preprocessing as follows:

[0092] First, all data points that are beyond the underwater navigation range of submarine-launched missiles are removed to eliminate outliers that affect the analysis results, and all data with close time and almost the same longitude, latitude, depth, speed and heading are deleted to reduce redundancy and improve processing efficiency.

[0093] Secondly, the DBSCAN algorithm is used for cluster analysis:

[0094] Specifically, we start with the first data point d1 of the two-dimensional data set DT and assign it to the first cluster C1. Then, for each subsequent data point di, if the distance between di and the previous data point d_(i-1) does not exceed the preset minimum neighborhood radius eps, di is assigned to the same cluster; otherwise, a new cluster is created. After clustering, the number of data points in each cluster is calculated. If the number of data points in a cluster is less than the minimum number of samples MinPts, the cluster is considered as noise or outliers, and the above steps are repeated until all data points are classified.

[0095] The mask sample corresponds to the above-mentioned underwater complete track data sample, and is used to indicate whether the data points in the underwater complete track data sample are missing. If in the complete track data sample, the identifier of data point A is 0, then the corresponding data point in the mask sample is set to the missing state. If the identifier of data point B is 1, then the corresponding data point in the mask sample is set to the observed state, thereby constructing the data sample to be interpolated.

[0096] The data samples to be interpolated include the non-missing values ​​of the underwater complete track data samples and the missing values ​​replaced by random noise samples.

[0097] Specifically, the construction formula of the data sample to be interpolated is as follows:

[0098]

[0099] Among them, ⊙ and Represents multiplication and addition operations between elements, represents the data sample to be interpolated corresponding to the jth underwater complete track data sample, represents the mask sample corresponding to the jth underwater complete track data sample, represents the jth underwater complete track data sample, Represents the random noise sample corresponding to the jth underwater complete track data sample.

[0100] Exemplarily, the j-th underwater complete track data sample is [50, 60, 120, 130], and the corresponding mask sample is [1, 0, 1, 0]. Accordingly, the data sample to be interpolated corresponding to the j-th underwater complete track data sample can be [50, 0.005, 120, 0.01].

[0101] Accordingly, the initial eigenvalue completion model is trained based on the optimized hyperparameter value to obtain a trained eigenvalue completion model, including:

[0102] Firstly, the optimized hyperparameter values ​​and the parameter values ​​of the generator of the initial eigenvalue completion model are fixed, the complete track data samples and the completed data samples output by the generator are taken as input, and the discriminator in the initial eigenvalue completion model is trained to obtain the trained discriminator.

[0103] In order to ensure the smoothness and continuity of the discriminator in the initial eigenvalue completion model during training, a gradient penalty mechanism can be introduced to effectively avoid gradient surges. The weights of the discriminator are no longer clipped to meet the Lipschitz continuity constraint, but a gradient penalty method is used, that is, this constraint is achieved by limiting the size of the weight gradient. In an optional implementation, the parameter value of the discriminator in the initial eigenvalue completion model is adjusted based on the gradient penalty mechanism.

[0104] Specifically, if the gradient norm of the discriminator deviates from 1, it will be penalized. This strategy not only improves the stability of the discriminator training in the initial eigenvalue completion model, but also overcomes the instability factors that may be caused by weight clipping.

[0105] The gradient penalty mechanism is used to suppress the excessive gradient of the discriminator network to avoid instability in the initial eigenvalue completion model during training. By adding a gradient penalty term to the loss function, the gradient of the discriminator network can be effectively controlled to keep it within a reasonable range, thereby improving the training effect of the initial eigenvalue completion model.

[0106] Among them, the loss function corresponding to the discriminator is:

[0107]

[0108] Among them, ⊙ and represents the multiplication and subtraction operations between elements, L1 represents the loss function corresponding to the judgement, mb represents the number of samples, represents the mask sample corresponding to the jth underwater complete track data sample, express is the probability of the true sample, represents the jth underwater complete track data sample, express is the probability of the true sample, For the general and The interpolated value generated by the generator input to the initial eigenvalue completion model, λ is the gradient penalty coefficient of the mean square error part of the generator loss function in the initial eigenvalue completion model.

[0109] Furthermore, the parameters of the discriminator in the initial feature value completion model are adjusted based on the loss value of the loss function corresponding to the discriminator to obtain a trained discriminator.

[0110] Furthermore, after obtaining the trained discriminator, the optimized hyperparameter values ​​and the parameter values ​​of the trained discriminator are fixed, the data samples to be interpolated and the corresponding mask samples are taken as input, and the generator in the initial eigenvalue completion model is trained to obtain the trained generator.

[0111] Among them, the trained discriminator and the trained generator are used to construct a trained feature value completion model.

[0112] Among them, the loss function corresponding to the generator is:

[0113]

[0114] Among them, ⊙ and represents the multiplication and subtraction operations between elements, L2 represents the loss function corresponding to the generator, mb represents the number of samples, represents the mask sample corresponding to the jth underwater complete track data sample, express is the probability of the true sample, represents the jth underwater complete track data sample, For the general and The interpolation value generated by the generator input to the initial eigenvalue completion model, α is the weight attenuation coefficient of the mean square error part in the generator loss function in the initial eigenvalue completion model.

[0115] Furthermore, the parameters of the generator in the initial feature value completion model are adjusted based on the loss value of the loss function corresponding to the generator to obtain the trained generator.

[0116] Step 3: The mean square error of the generator in the trained eigenvalue completion model is used as the particle fitness value of the particle swarm algorithm, and based on the particle fitness value, the optimized hyperparameter value is adjusted to obtain the target hyperparameter value.

[0117] In the disclosed embodiment, since its goal is to interpolate missing values, it is hoped that the generated data is as close to the real original data as possible, and the mean square error of the generator directly measures the square of the average difference between the data generated by the generator and the real data. Therefore, in the selection of the fitness function of the particle swarm algorithm, the mean square error of the generator can be used as an indicator for evaluating the fitness of the particles.

[0118] Specifically, the above-mentioned and The mean square error of the generator is determined and used as the particle fitness value of the particle swarm algorithm.

[0119] If the particle fitness value is less than the particle individual optimal fitness value, then the particle individual optimal fitness value is equal to the particle fitness value, and the particle individual optimal position is equal to the particle current position.

[0120] If the particle fitness value is less than the global optimal fitness value, the global optimal fitness value is equal to the particle fitness value, and the global optimal position is equal to the current position of the particle.

[0121] Update the positions and velocities of all particles to obtain the target hyperparameter values.

[0122] Step 4: Determine whether the particle swarm algorithm satisfies the preset termination condition. If the particle swarm algorithm satisfies the preset termination condition, the particle swarm algorithm is terminated, and the trained eigenvalue completion model obtained by training the initial eigenvalue completion model based on the target hyperparameter value is used as the target eigenvalue completion model.

[0123] Among them, the preset termination conditions include reaching a preset maximum number of iterations or the loss function value corresponding to the generator converges to a certain stable level.

[0124] In the disclosed embodiment, during the operation of the particle swarm algorithm, a preset termination condition is set. For example, when the maximum number of iterations is reached, the preset termination condition needs to be monitored in real time. If the preset termination condition is met, the particle swarm algorithm is terminated and further calculations and iterations are stopped. After the particle swarm algorithm is completed, the trained feature value obtained by training using the final determined hyperparameter value (i.e., the above-mentioned target hyperparameter value) is used to complete the model.

[0125] If the particle swarm algorithm does not meet the preset termination condition, the target hyperparameter value is used to update the optimized hyperparameter value, and the steps of training the initial eigenvalue completion model based on the optimized hyperparameter value are continued to obtain the trained eigenvalue completion model until the particle swarm algorithm meets the preset termination condition.

[0126] If the particle swarm algorithm does not meet the preset termination condition, it means that the particle swarm algorithm needs to continue to optimize. In this case, the optimized hyperparameter value will be updated according to the target hyperparameter value first, and then the initial eigenvalue completion model will be trained using the optimized hyperparameter value to obtain the trained eigenvalue completion model until the particle swarm algorithm finally meets the preset termination condition.

[0127] The completed data is sent to the digital main line to generate the complete underwater navigation trajectory of the submarine-launched missile.

[0128] In the disclosed embodiment, the completed data is integrated back into the meta-model of the digital main line to ensure its accuracy and continuity. The generated complete underwater navigation trajectory of the submarine-launched missile can provide detailed navigation position, speed and attitude information, and also provide reliable data support for subsequent trajectory optimization and combat strategy formulation.

[0129] In the missile track completion method based on the digital main line provided by the embodiment of the present disclosure, the underwater track data of the submarine-launched missile is obtained from the digital main line, wherein there is a missing eigenvalue in the underwater track data, and the underwater track data is input into the target eigenvalue completion model, and after being processed by the target eigenvalue completion model, the output is the completion data corresponding to the underwater track data; wherein, the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network. It can be seen that in the embodiment of the present disclosure, in the training process of the initial eigenvalue completion model, the particle swarm algorithm is introduced to optimize the hyperparameters, thereby improving the accuracy of the target eigenvalue completion model after training, so that after being processed by the target eigenvalue completion model, the completion data corresponding to the output underwater track data is more accurate.

[0130] Based on the description of the embodiment of the missile track completion method based on the digital main line, the embodiment of the present disclosure also provides a missile track completion method based on the digital main line, referring to Figure 2 , is a flow chart of another missile track completion method based on digital main line provided by an embodiment of the present disclosure. The method includes:

[0131] Firstly, the underwater complete track data samples of submarine-launched missiles are obtained from the digital platform, and the underwater complete track data samples of submarine-launched missiles, the corresponding mask samples and the random noise samples are used to construct the data samples to be interpolated, wherein the underwater complete track data samples are the samples obtained after preprocessing the initial underwater complete track data samples of submarine-launched missiles obtained from the digital main line, and the preprocessing includes clustering analysis of the initial underwater complete track data samples, and applying the Euclidean distance including the time factor in the clustering analysis to remove the outliers in the initial underwater complete track data samples, and initializing the parameters of the initial eigenvalue completion model, and initializing the particle swarm, the individual optimal position and the global optimal position to obtain the initial hyperparameter values ​​of the initial eigenvalue completion model and the parameters of the initial eigenvalue completion model.

[0132] Then, the parameters of the initial eigenvalue completion model are fixed, and the initial hyperparameter values ​​are optimized according to the fitness function of the particle swarm algorithm to obtain the optimized hyperparameter values. Based on the optimized hyperparameter values, the initial eigenvalue completion model is trained to obtain the trained eigenvalue completion model. The mean square error of the generator in the trained eigenvalue completion model is used as the particle fitness value of the particle swarm algorithm, and based on the particle fitness value, the optimized hyperparameter values ​​are adjusted to obtain the target hyperparameter values. It is determined whether the particle swarm algorithm meets the preset termination condition. If the particle swarm algorithm meets the preset termination condition, the particle swarm algorithm is terminated, and the trained eigenvalue completion model obtained by training the initial eigenvalue completion model based on the target hyperparameter value is used as the target eigenvalue completion model. If the particle swarm algorithm does not meet the preset termination condition, the optimized hyperparameter value is updated to facilitate the target hyperparameter value, and the steps of training the initial eigenvalue completion model based on the optimized hyperparameter value to obtain the trained eigenvalue completion model are continued until the particle swarm algorithm meets the preset termination condition.

[0133] It can be seen that in the embodiment of the present disclosure, during the training process of the initial eigenvalue completion model, a particle swarm algorithm is introduced to optimize hyperparameters, thereby improving the accuracy of the target eigenvalue completion model after training, so that after being processed by the target eigenvalue completion model, the completion data corresponding to the output underwater track data is more accurate.

[0134] Based on the above method embodiment, the present disclosure also provides a missile track completion device based on digital main line, referring to Figure 3 , is a structural schematic diagram of a missile track completion device based on a digital main line provided by an embodiment of the present disclosure, the device comprising:

[0135] The acquisition module 301 is used to acquire underwater track data of a submarine-launched missile from the digital main line; wherein the underwater track data has missing characteristic values;

[0136] The processing module 302 is used to input the underwater track data into the target eigenvalue completion model, and after being processed by the target eigenvalue completion model, output the completion data corresponding to the underwater track data; wherein, the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network.

[0137] In an optional implementation, the device further includes:

[0138] An optimization module is used to optimize the initial hyperparameter values ​​of the initial eigenvalue completion model using a particle swarm algorithm to obtain optimized hyperparameter values;

[0139] A training module, used for training the initial eigenvalue completion model based on the optimized hyperparameter value to obtain a trained eigenvalue completion model;

[0140] An adjustment module, used to use the mean square error of the generator in the trained feature value completion model as the particle fitness value of the particle swarm algorithm, and adjust the optimized hyperparameter value based on the particle fitness value to obtain a target hyperparameter value;

[0141] A determination module is used to determine whether the particle swarm algorithm satisfies a preset termination condition. If the particle swarm algorithm satisfies the preset termination condition, the particle swarm algorithm is terminated, and the trained eigenvalue completion model obtained by training the initial eigenvalue completion model based on the target hyperparameter value is used as the target eigenvalue completion model.

[0142] In an optional implementation, the device further includes:

[0143] An updating module is used for updating the optimized hyperparameter value in favor of the target hyperparameter value if the particle swarm algorithm does not meet the preset termination condition, and continuing to perform the step of training the initial eigenvalue completion model based on the optimized hyperparameter value to obtain the trained eigenvalue completion model, until the particle swarm algorithm meets the preset termination condition.

[0144] In an optional implementation manner, the device further includes:

[0145] A construction module is used to construct a data sample to be interpolated based on the underwater complete track data sample of the submarine-launched missile, the corresponding mask sample and the random noise sample;

[0146] Accordingly, the training module includes:

[0147] A first training submodule is used to fix the optimized hyperparameter values ​​and the parameter values ​​of the generator of the initial eigenvalue completion model, take the complete track data sample and the completed data sample output by the generator as input, train the discriminator in the initial eigenvalue completion model, and obtain a trained discriminator;

[0148] The second training submodule is used to fix the optimized hyperparameter values ​​and the parameter values ​​of the trained discriminator, take the data samples to be interpolated and the corresponding mask samples as input, train the generator in the initial eigenvalue completion model, and obtain the trained generator; wherein the trained discriminator and the trained generator are used to construct the trained eigenvalue completion model.

[0149] In an optional implementation, the parameter value of the discriminator in the initial eigenvalue completion model is adjusted based on a gradient penalty mechanism.

[0150] In an optional implementation, the device further includes:

[0151] A generation module is used to send the supplementary data to the digital main line to generate a complete underwater navigation trajectory of the submarine-launched missile.

[0152] In an optional implementation, the underwater complete track data sample is a sample obtained after preprocessing the initial underwater complete track data sample of the submarine-launched missile obtained in the digital main line, and the preprocessing includes performing cluster analysis on the initial underwater complete track data sample, and applying the Euclidean distance including the time factor in the cluster analysis to eliminate outliers in the initial underwater complete track data sample.

[0153] In the missile track completion device based on the digital main line provided by the embodiment of the present disclosure, the underwater track data of the submarine-launched missile is obtained from the digital main line, wherein there is a missing eigenvalue in the underwater track data, and the underwater track data is input into the target eigenvalue completion model, and after being processed by the target eigenvalue completion model, the output is the completion data corresponding to the underwater track data; wherein, the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network. It can be seen that in the embodiment of the present disclosure, in the training process of the initial eigenvalue completion model, the particle swarm algorithm is introduced to optimize the hyperparameters, thereby improving the accuracy of the target eigenvalue completion model after training, so that after being processed by the target eigenvalue completion model, the completion data corresponding to the output underwater track data is more accurate.

[0154] In addition to the above-mentioned method and apparatus, the embodiments of the present disclosure also provide a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device implements the missile track completion method based on the digital main line described in the embodiments of the present disclosure.

[0155] The embodiment of the present disclosure also provides a computer program product, which includes a computer program / instruction. When the computer program / instruction is executed by a processor, the method for completing the missile track based on the digital main line described in the embodiment of the present disclosure is implemented.

[0156] In addition, the embodiment of the present disclosure also provides a missile track completion device based on a digital main line, see Figure 4 As shown, it may include:

[0157] Processor 401, memory 402, input device 403 and output device 404. The number of processors 401 in the missile track completion device based on digital main line can be one or more. Figure 4 In some embodiments of the present disclosure, the processor 401, the memory 402, the input device 403 and the output device 404 may be connected via a bus or other means, wherein: Figure 4 The example of connecting through bus is taken in the following.

[0158] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing of the missile track completion device based on the digital main line by running the software programs and modules stored in the memory 402. The memory 402 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc. In addition, the memory 402 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. The input device 403 can be used to receive input digital or character information, and generate signal input related to user settings and function control of the missile track completion device based on the digital main line.

[0159] Specifically in this embodiment, the processor 401 will load the executable files corresponding to the processes of one or more applications into the memory 402 according to the following instructions, and the processor 401 will run the applications stored in the memory 402, thereby realizing the various functions of the above-mentioned missile track completion device based on the digital main line.

[0160] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0161] The above description is only a specific embodiment of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A missile track completion method based on digital main line, characterized in that: The method comprises: Acquire underwater track data of a submarine-launched missile from the digital main line; wherein the underwater track data has missing characteristic values; The underwater track data is input into the target eigenvalue completion model, and after being processed by the target eigenvalue completion model, the output is the completion data corresponding to the underwater track data; wherein, the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network.

2. The method according to claim 1, characterized in that Before the underwater track data is input into the target feature value completion model and is processed by the target feature value completion model and then output as the completion data corresponding to the underwater track data, the method further includes: The particle swarm algorithm is used to optimize the initial hyperparameter values ​​of the initial eigenvalue completion model to obtain the optimized hyperparameter values; Based on the optimized hyperparameter values, the initial eigenvalue completion model is trained to obtain a trained eigenvalue completion model; The mean square error of the generator in the trained feature value completion model is used as the particle fitness value of the particle swarm algorithm, and based on the particle fitness value, the optimized hyperparameter value is adjusted to obtain a target hyperparameter value; Determine whether the particle swarm algorithm satisfies a preset termination condition. If the particle swarm algorithm satisfies the preset termination condition, terminate the particle swarm algorithm, and use the trained eigenvalue completion model obtained by training the initial eigenvalue completion model based on the target hyperparameter value as the target eigenvalue completion model.

3. The method according to claim 2, characterized in that The method further comprises: If the particle swarm algorithm does not meet the preset termination condition, the target hyperparameter value is used to update the optimized hyperparameter value, and the step of training the initial eigenvalue completion model based on the optimized hyperparameter value to obtain the trained eigenvalue completion model is continued until the particle swarm algorithm meets the preset termination condition.

4. The method according to claim 2, characterized in that Before the initial eigenvalue completion model is trained based on the optimized hyperparameter value to obtain the trained eigenvalue completion model, the method further includes: Based on the underwater complete track data samples of submarine-launched missiles, the corresponding mask samples and random noise samples, the data samples to be interpolated are constructed; Accordingly, the initial eigenvalue completion model is trained based on the optimized hyperparameter value to obtain a trained eigenvalue completion model, including: Fixing the optimized hyperparameter values ​​and the parameter values ​​of the generator of the initial eigenvalue completion model, taking the complete track data sample and the completed data sample output by the generator as input, training the discriminator in the initial eigenvalue completion model, and obtaining a trained discriminator; The optimized hyperparameter values ​​and the trained discriminator parameter values ​​are fixed, the data samples to be interpolated and the corresponding mask samples are taken as input, and the generator in the initial eigenvalue completion model is trained to obtain the trained generator; wherein the trained discriminator and the trained generator are used to construct the trained eigenvalue completion model.

5. The method according to claim 2, characterized in that: The parameter value of the discriminator in the initial eigenvalue completion model is adjusted based on a gradient penalty mechanism.

6. The method according to claim 1, characterized in that After the underwater track data is input into the target feature value completion model and is processed by the target feature value completion model and output as the completion data corresponding to the underwater track data, the method further includes: The supplementary data is sent to the digital main line to generate a complete underwater navigation trajectory of the submarine-launched missile.

7. The method according to claim 4, characterized in that The underwater complete track data samples are samples obtained after preprocessing the initial underwater complete track data samples of the submarine-launched missile obtained in the digital main line. The preprocessing includes clustering analysis of the initial underwater complete track data samples, and applying the Euclidean distance including the time factor in the cluster analysis to eliminate outliers in the initial underwater complete track data samples.

8. A missile track completion device based on digital main line, characterized in that: The device comprises: An acquisition module is used to acquire underwater track data of a submarine-launched missile from a digital main line; wherein the underwater track data has missing characteristic values; A processing module is used to input the underwater track data into a target eigenvalue completion model, and after being processed by the target eigenvalue completion model, output the completion data corresponding to the underwater track data; wherein, the hyperparameters of the target eigenvalue completion model are obtained by a particle swarm algorithm, and the hyperparameters are used to train the initial eigenvalue completion model to obtain the target eigenvalue completion model, and the target eigenvalue completion model is constructed based on a generative adversarial network.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device implements the method according to any one of claims 1 to 7.

10. A missile track completion device based on digital main line, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images