Interference strategy prediction method and device and storage medium

By pre-training the radar system's strategy prediction network and online fine-tuning of real-time interactive data, the problem of insufficient anti-interference performance of the radar system in complex interference environments is solved, the accuracy of the model and training efficiency are improved, and the radar system is enhanced. Adaptability.

CN118917361APending Publication Date: 2024-11-08SHENZHEN RES INST OF BIG DATA
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411081325.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing radar systems have insufficient anti-interference performance in complex interference environments, especially online learning methods require a large amount of adversarial sample data for model training, which has low training efficiency and serious overfitting problems, which affects the accuracy of interference strategy prediction.

Method used

By collecting historical interaction data between radar and jammer for pre-training, a policy prediction network is obtained, and the adapter module is fine-tuned online in real-time interactive data, the predicted interference strategy is output, and the strategy prediction is predicted during the online fine-tuning training process. The parameters of the network remain unchanged, and only a small number of parameters of the training adapter module is iterated.

Benefits of technology

It improves the accuracy and training efficiency of model prediction, increases the data volume of training samples, and makes the radar system have stronger adaptability and learning ability in complex interference environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917361A_ABST
    Figure CN118917361A_ABST
Patent Text Reader

Abstract

The invention provides an interference strategy prediction method and apparatus, and a storage medium. The method comprises the steps of collecting historical interaction data between a radar and a jammer; pre-training the to-be-trained network based on the historical interaction data to obtain a strategy prediction network; and on the basis of real-time interaction data between the radar and the jammer, carrying out on-line fine tuning training on the adapter module in the strategy prediction network, and outputting a predicted interference strategy. The strategy prediction network of the main framework is pre-trained by collecting the existing historical interaction data, the strategy prediction network can preliminarily learn various interference strategies, the network is subjected to online fine tuning training in an actual combat environment, and the real-time interference strategies can be predicted through pre-training of the historical interaction data and online training of real-time interaction data. The data volume of training samples is increased; and the accuracy of model prediction is improved. And in the online fine tuning training process, the parameters of the strategy prediction network are kept unchanged, and only a small number of parameters of the adapter module are iteratively trained, so that the model training efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of radar technology, and in particular to an interference strategy prediction method, device and storage medium. Background Art

[0002] In recent years, with the continuous improvement of the software, hardware and intelligence level of jammers, the electromagnetic environment faced by radars has become increasingly complex. If radars only use fixed anti-interference strategies and can only deal with certain types of interference, the anti-interference performance of radars will be seriously reduced.

[0003] In order to improve the adaptability and learning ability of radar in complex interference environments, frequency agility (FA) technology is introduced into radar systems, which adopts a method of dynamic frequency hopping on a wide spectrum. For example, sparse decomposition technology usually focuses on a limited number of jammer types and is limited in applicability in various interference environments. In addition, various online learning techniques have also been integrated into the design of frequency hopping strategies, treating anti-interference work as a sequential decision process. For example, model-free reinforcement learning (RL) methods, dual network DQN (D3QN), proximal policy optimization (PPO), and bandit algorithms of situational Thompson sampling (TS) are applied to optimize anti-interference strategies. However, a large amount of adversarial sample data is required for model training in online learning methods, and the training efficiency of model parameters is low. When the amount of adversarial sample data collected is small, it will lead to model overfitting, affecting the accuracy of interference strategy prediction. Summary of the invention

[0004] The present invention provides an interference strategy prediction method, device and storage medium, aiming to solve at least one of the above technical problems.

[0005] The present invention provides an interference strategy prediction method, comprising:

[0006] Collect historical interaction data between radar and jammer;

[0007] Based on the historical interaction data, pre-training the network to be trained to obtain a strategy prediction network;

[0008] Based on the real-time interaction data between the radar and the jammer, the adapter module in the strategy prediction network is fine-tuned online and the predicted jamming strategy is output;

[0009] The adapter module is associated and configured in the strategy prediction network after the pre-training is completed; and the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process.

[0010] According to an interference strategy prediction method provided by the present invention, the strategy prediction network includes a first linear layer and a plurality of cascaded transformer layers; wherein the output of any transformer layer is used as the input of the next transformer layer; and the last transformer layer is connected to the first linear layer;

[0011] The pre-training of the network to be trained based on the historical interaction data to obtain a strategy prediction network includes:

[0012] Input the historical interaction data into the first transformer layer for feature conversion processing, and input the feature information output by each transformer layer into the next transformer layer until the feature information output by the last transformer layer is obtained;

[0013] Input the feature information output by the last transformer layer into the first linear layer to predict the jammer action and obtain the strategy prediction result;

[0014] Determining a first loss value based on the strategy prediction result;

[0015] Based on the first loss value, the parameters of the network to be trained are optimized and updated to obtain the strategy prediction network.

[0016] According to an interference strategy prediction method provided by the present invention, any of the transformer layers includes a first normalization layer, an attention layer, a first connection layer, a second normalization layer, a multilayer perceptron and a second connection layer:

[0017] The first normalization layer is connected to the attention layer;

[0018] The attention layer is connected to the first connection layer, wherein the first connection layer is used to connect the features output by the attention layer and the features input to the first normalization layer;

[0019] The second normalization layer is connected to the first connection layer and the multi-layer perceptron respectively;

[0020] The multilayer perceptron is connected to the second connection layer, wherein the second connection layer is used to connect the features output by the multilayer perceptron and the features input to the output of the first connection layer.

[0021] According to an interference strategy prediction method provided by the present invention, the historical interaction data is input into the first transformer layer for feature conversion processing, including:

[0022] Inputting the historical interaction data into the first normalization layer for normalization processing to obtain first normalized feature information;

[0023] Inputting the first normalized feature information into the attention layer for attention learning to obtain attention feature information;

[0024] Inputting the historical interaction data and the attention feature information into the first connection layer for feature splicing to obtain first splicing feature information;

[0025] Inputting the first splicing feature information into the second normalization layer for normalization processing to obtain second normalized feature information;

[0026] Inputting the second normalized feature information into a multi-layer perceptron for feature conversion processing to obtain target feature information;

[0027] The target feature information and the first splicing feature information are input into the second connection layer for feature splicing to obtain second splicing feature information, and the second splicing feature information is used as the feature information output by the first transformer layer.

[0028] According to an interference strategy prediction method provided by the present invention, the multi-layer perceptron includes a second linear layer, a first activation layer and a third linear layer;

[0029] wherein the second linear layer is connected to the second normalization layer and the first activation layer respectively;

[0030] The first activation layer is connected to the third linear layer;

[0031] The third linear layer is connected to the second connection layer, wherein a feature dimension corresponding to a feature input to the second linear layer is the same as a feature dimension of a feature output by the third linear layer.

[0032] According to an interference strategy prediction method provided by the present invention, the multilayer perceptron in each transformer layer is connected to the adapter module;

[0033] The adapter module includes an encoder, a second activation layer, a decoder and a third connection layer:

[0034] The encoder is connected to the second activation layer;

[0035] The second activation layer is connected to the decoder;

[0036] The decoder is connected to the third connection layer.

[0037] According to a jamming strategy prediction method provided by the present invention, the adapter module in the strategy prediction network is subjected to online fine-tuning training based on the real-time interaction data between the radar and the jammer, comprising:

[0038] Inputting the real-time interaction data into a strategy prediction network including an adapter module to obtain an interference strategy;

[0039] Based on the interference strategy, the parameters of the adapter module are optimized online, wherein the parameters of the strategy prediction network remain unchanged.

[0040] According to an interference strategy prediction method provided by the present invention, after the interference strategy is obtained by output prediction, the method further includes:

[0041] Based on the interference strategy, an optimal strategy for the radar is determined in an action space preset by the radar.

[0042] The present invention also provides an interference strategy prediction device, comprising:

[0043] A collection module, used to collect historical interaction data between radar and jammer;

[0044] A pre-training module, used for pre-training the network to be trained based on the historical interaction data to obtain a strategy prediction network;

[0045] An online fine-tuning module, used for performing online fine-tuning training on the adapter module in the strategy prediction network based on the real-time interaction data between the radar and the jammer, and outputting the predicted jamming strategy;

[0046] The adapter module is associated and configured in the strategy prediction network after the pre-training is completed; and the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process.

[0047] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the interference strategy prediction methods described above is implemented.

[0048] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the interference strategy prediction method described in any one of the above is implemented.

[0049] The interference strategy prediction method, device and storage medium provided by the present invention include: collecting historical interaction data between radar and jammer; pre-training the network to be trained based on the historical interaction data to obtain a strategy prediction network; based on the real-time interaction data between the radar and the jammer, online fine-tuning training is performed on the adapter module in the strategy prediction network, and the predicted interference strategy is output; wherein the adapter module is associated and configured in the strategy prediction network after the pre-training is completed; the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process. The present invention collects existing historical interaction data and pre-trains the strategy prediction network of the main framework, so that the strategy prediction network can initially learn various interference strategies, and then in the actual combat environment, the network is fine-tuned online, and the data volume of the training sample is increased through pre-training of historical interaction data and online training of real-time interaction data, thereby improving the accuracy of model prediction. In addition, the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process, and only a small number of parameters of the adapter module are iteratively trained to improve the efficiency of model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 It is a flow chart of the interference strategy prediction method provided by the present invention;

[0052] Figure 2 is a model structure diagram of a strategy prediction network provided by an embodiment of the present invention;

[0053] Figure 3 is a model structure diagram of a strategy prediction network including an adapter module provided by an embodiment of the present invention;

[0054] Figure 4 It is a diagram showing the effects of different online learning methods provided by an embodiment of the present invention in a static environment;

[0055] Figure 5 It is a diagram showing the effects of different online learning methods provided by an embodiment of the present invention in a non-static environment;

[0056] Figure 6 is a performance effect diagram of three algorithms provided by an embodiment of the present invention;

[0057] Figure 7is a comparison diagram of the number of parameters required to be updated for OM with adaptor and OM without adaptor provided in an embodiment of the present invention;

[0058] Figure 8 It is a structural schematic diagram of the interference strategy prediction device provided by the present invention;

[0059] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] It should be understood that, although the terms first, second, etc. may be used to describe various information in one or more embodiments of the present invention, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of the present invention, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "when..." or "when...".

[0062] Figure 1 FIG. 1 is a flow chart of the interference strategy prediction method provided by the present invention. Figure 1 As shown, the interference strategy prediction method includes:

[0063] Step S11, collecting historical interaction data between the radar and the jammer;

[0064] It should be noted that during the confrontation between radar and jammer, the radar action space is defined in advance as The action space of the jammer is defined as Among them, the radar action of the nth pulse Each pulse contains M sub-pulses. Jammer Action In the interaction of the nth pulse, the radar moves from the action space Select action a n , while the jammer changes from its action space Select action b n .

[0065] Action strategies of both parties: At pulse n, the radar follows the strategy x of the current roundn Select Action Interference opportunities according to strategy y n Selecting a Disruption Strategy It should be noted that the system has certain history-based characteristics, that is, the system stores the historical interaction data of radar and jammer with a length of k.

[0066] Step S12, pre-training the network to be trained based on historical interaction data to obtain a strategy prediction network;

[0067] It should be noted that the strategy prediction network includes a first linear layer and several cascaded transformer layers; wherein the output of any transformer layer is used as the input of the next transformer layer; the last transformer layer is connected to the first linear layer. The specific network structure is specifically described in the following embodiments and will not be repeated here.

[0068] Specifically, the historical interaction data is input into the first transformer layer for feature conversion processing, and the feature information output by each transformer layer is input into the next transformer layer until the feature information output by the last transformer layer is obtained; and then the feature information output by the last transformer layer is input into the first linear layer to predict the jammer action, and the strategy prediction result is obtained. Furthermore, based on the strategy prediction result, the first loss value is determined, where the loss value is calculated as follows:

[0069]

[0070] in, represents an interfering action b n The associated one-hot encoded vector, represents the distribution of interference strategies predicted by the network, ω represents the parameters of the network, and n is the number of rounds of back propagation parameter updates of the neural network. Then, based on the first loss value, the parameters of the training network are optimized and updated to obtain a strategy prediction network, so that the strategy prediction network has a preliminary understanding of all interference strategies.

[0071] Step S13, based on the real-time interaction data between the radar and the jammer, the adapter module in the strategy prediction network is fine-tuned online and the predicted jamming strategy is output.

[0072] It should be noted that compared with the pre-training stage, fine-tuning training is usually only performed on a very small batch of real-time interaction data. The fine-tuning process usually updates all parameters of the entire model after each interaction between the radar and the jammer, which leads to low efficiency of the fine-tuning process, and a large number of neural network parameters are updated on small batches of data, which can easily cause parameter overfitting.

[0073] In this embodiment, the policy prediction network pre-training is completed, and the adapter module is associated with the policy prediction network pre-training. During the online fine-tuning training, the parameters of the original network remain unchanged, that is, the original parameters of the policy prediction network remain unchanged during the online fine-tuning training. Only a small number of new parameters introduced in the adapter module participate in the training at each iteration, thereby improving the efficiency of the model online training.

[0074] Specifically, the real-time interaction data is input into the policy prediction network including the adapter module to obtain the interference strategy; then based on the interference strategy, the parameters of the adapter module are optimized online, wherein the parameters of the policy prediction network that has been pre-trained do not participate in the update of the gradient descent algorithm, and the original values ​​of the parameters are maintained. In each round of interaction, the adapter module needs to calculate the gradient of the parameters in the adapter module, and use the gradient back propagation algorithm to optimize and update the parameter values.

[0075] In one embodiment, after the jammer's jamming strategy is predicted, the optimal strategy of the radar is determined in the action space preset by the radar. For example, the optimal strategy of the radar is determined in combination with a benefit matrix, where the benefit matrix refers to a matrix including all action benefit values ​​of both parties.

[0076] The embodiment of the present invention, through the above scheme, includes: collecting historical interaction data between radar and jammer; pre-training the network to be trained based on the historical interaction data to obtain a strategy prediction network; based on the real-time interaction data between the radar and the jammer, online fine-tuning training of the adapter module in the strategy prediction network, and outputting the predicted interference strategy; wherein the adapter module is associated and configured in the strategy prediction network after the pre-training is completed; the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process. The present invention pre-trains the strategy prediction network of the main framework by collecting existing historical interaction data, so that the strategy prediction network can initially learn various interference strategies, and then in the actual combat environment, the network is fine-tuned online, and the data volume of the training sample is increased through the pre-training of historical interaction data and the online training of real-time interaction data, thereby improving the accuracy of model prediction. And the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process, and only a small number of parameters of the adapter module are iteratively trained to improve the efficiency of model training.

[0077] Further, refer to Figure 2 , Figure 2It is a model structure diagram of a strategy prediction network provided by an embodiment of the present invention. Any transformer layer includes a first normalization layer, an attention layer, a first connection layer, a second normalization layer, a multi-layer perceptron, and a second connection layer: the first normalization layer is connected to the attention layer; the attention layer is connected to the first connection layer, wherein the first connection layer is used to connect the features output by the attention layer and the features input to the first normalization layer; the second normalization layer is respectively connected to the first connection layer and the multi-layer perceptron; the multi-layer perceptron is connected to the second connection layer, wherein the second connection layer is used to connect the features output by the multi-layer perceptron and the features input to the output of the first connection layer.

[0078] Optionally, the multilayer perceptron includes a second linear layer, a first activation layer, and a third linear layer; wherein the second linear layer is connected to the second normalization layer and the first activation layer respectively; the first activation layer uses a nonlinear activation function to implement nonlinear mapping, such as ReLU (Rectified Linear Unit), Sigmoid or Tanh functions, so that the model can capture more complex feature relationships and enhance the model's expressiveness. The first activation layer is connected to the third linear layer; the third linear layer is connected to the second connection layer, wherein the feature dimension corresponding to the feature input to the second linear layer is the same as the feature dimension of the feature output by the third linear layer, that is, the third linear layer restores the feature dimension to a tensor that matches the feature dimension input to the second linear layer.

[0079] Further, refer to Figure 3 , Figure 3 It is a model structure diagram of a strategy prediction network including an adapter module provided by an embodiment of the present invention. The multi-layer perceptron in each transformer layer is connected to the adapter module; the adapter module includes an encoder, a second activation layer, a decoder and a third connection layer: the encoder is connected to the second activation layer; the second activation layer is connected to the decoder; the decoder is connected to the third connection layer.

[0080] In one embodiment of the present invention, the historical interaction data is input into the first transformer layer for feature conversion processing, including:

[0081] The historical interaction data is input into the first normalization layer for normalization processing to obtain the first normalized feature information; the first normalized feature information is input into the attention layer for attention learning to obtain the attention feature information; the historical interaction data and the attention feature information are input into the first connection layer for feature splicing to obtain the first splicing feature information; the first splicing feature information is input into the second normalization layer for normalization processing to obtain the second normalized feature information; the second normalized feature information is input into the multi-layer perceptron for feature conversion processing to obtain the target feature information; the target feature information and the first splicing feature information are input into the second connection layer for feature splicing to obtain the second splicing feature information, and the second splicing feature information is used as the feature information output by the first transformer layer.

[0082] Specifically, the historical interaction data is input into the first normalization layer for normalization processing to obtain the first normalized feature information. Layer normalization helps to adjust the scale of the activation value so that it has better numerical stability; the first normalized feature information is then input into the attention layer for attention learning to obtain the attention feature information, wherein the attention layer is a multi-head attention layer, in which the self-attention mechanism is further expanded. By dividing the input data into multiple "heads", each "head" learns different information, and the model can capture sequence information from different angles and contexts, so that the model can better understand the contribution of each fragment to the accuracy of the output interference strategy, thereby improving the model's ability to process complex sequence data. Further, the historical interaction data and attention feature information are input into the first connection layer for feature splicing to obtain the first splicing feature information. The use of residual connection helps to alleviate the gradient vanishing problem in deep networks. Further, the first splicing feature information is input into the second normalization layer for normalization processing to obtain the second normalized feature information. In order to extract and convert important information in the input data and obtain target feature information that can be used for subsequent tasks (such as classification, regression, etc.), the second normalized feature information is input into the multi-layer perceptron for feature conversion processing to obtain target feature information, and then the target feature information and the first concatenated feature information are input into the second connection layer for feature concatenation to obtain second concatenated feature information, and the second concatenated feature information is used as the feature information output by the first transformer layer. The feature information output by the first transformer layer will be subsequently input into the second transformer layer for feature processing.

[0083] The embodiment of the present invention collects existing historical interaction data through the above scheme, and pre-trains the strategy prediction network of the main framework, so that the strategy prediction network can preliminarily learn various interference strategies.

[0084] In one embodiment, as a specific example, the experimental parameters are set as shown in Table 1:

[0085] Table 1 Parameter settings of radar and jammer

[0086] parameter Numeric Radar antenna gain 30dB Radar transmission power 10KW Jammer transmission power 1KW Sub-pulse number 4 Sub-pulse bandwidth 2MHz Sub pulse width 3μs Pulse repetition frequency 1000HZ Carrier frequency 10GHz distance 100KM

[0087] Interference Environment: Considering the inherent complexity in intelligent point perturbation, two different interference environments are used to demonstrate the advantages of the algorithm in the proposed embodiment of the present invention. In the static environment, a fixed interference strategy is consistently applied to all iterations. In the non-steady-state environment, the interference strategy is dynamically adjusted, reflecting the frequency occurrence within a moving window of the last 5 interactions. Specifically, the highest and second highest frequencies are assigned interference probabilities of 0.7 and 0.3, respectively.

[0088] The radar uses a total of 8 algorithms, including four algorithms based on reinforcement learning: Deep Q Network (DQN), Deep Q Network with Intrinsic Curiosity Module (ICM-DQN), Proximal Policy Optimization (PPO), Trust Region Policy Optimization (TRPO). Two algorithms based on Bandit: Thompson Sampling (TS) and Upper Confidence Bound algorithm (UCB); a random algorithm Random, and the interference strategy prediction method OM-adapter proposed in this invention.

[0089] There are two types of radar action spaces: |A R |=3 4 and |A R |=50 4 .

[0090] The reward of each round is used as the judgment standard. The higher the reward of each round, the better the algorithm effect. The reward formula is as follows.

[0091]

[0092] Where c represents the preset threshold of SINR, SINR(a n ,b n ) is related to action a n and b nRelated SINR, SINR (Signal-to-Interference plus Noise Ratio) is a metric used to describe signal quality in radar communication systems.

[0093]

[0094] Where P R , P J and Represent the received power of radar, jammer and noise respectively. Function Determine whether the mth sub-pulse radar is interfered by the jammer.

[0095] Targeting |A R |=3 4 :Reference Figure 4 and Figure 5 , Figure 4 is a diagram showing the effects of different online learning methods provided by an embodiment of the present invention in a static environment. Figure 5 is a diagram showing the effects of different online learning methods provided by an embodiment of the present invention in a non-static environment. In an experiment conducted in a static environment, Figure 4 As shown in , all algorithms eventually converge to the optimal reward after giving enough iterations. Figure 5 In the non-static environment, only the algorithm proposed in the embodiment of the present invention achieves the optimal reward. The algorithm based on Bandit performs poorly in the non-static environment, and the algorithm based on reinforcement learning cannot obtain rewards comparable to those of the method in the embodiment of the present invention. In addition, the number of interaction rounds required for different types of methods to achieve stable high returns is shown in Table 2. By comparison, it can be seen that the method proposed in the embodiment of the present invention is significantly superior to all other comparison algorithms in terms of sample efficiency.

[0096] Table 2 The number of interaction rounds required for different types of methods to achieve stable high returns

[0097]

[0098] In addition, for This huge action space is usually beyond the reach of most traditional online algorithms due to their high computational complexity. The OM-adaptor interference strategy prediction algorithm proposed in the embodiment of the present invention still identifies the optimal strategy in this complex environment. Figure 6 , Figure 6is a performance effect diagram of three algorithms provided in an embodiment of the present invention, where OM with adaptor refers to the algorithm network with an adapter module in the embodiment of the present invention, OM without adaptor refers to the transformer network without an adapter module, and Random refers to a random algorithm. The results show that both OM with adaptor and OM without adaptor can effectively resist interference strategies. However, referring to Figure 7 , Figure 7 3 is a comparison chart of the number of parameters required to be updated for OM with adaptor and OM without adaptor provided in an embodiment of the present invention. The results show that the OM with adaptor algorithm of the embodiment of the present invention can achieve excellent anti-interference performance while significantly reducing the computing requirements.

[0099] The algorithm of the embodiment of the present invention integrates an adapter module, which can further improve parameter efficiency. Compared with the traditional online algorithm in this field, the algorithm network of the embodiment of the present invention shows significant improvements in both sample and computational efficiency.

[0100] The interference strategy prediction device provided by the present invention is described below. The interference strategy prediction device described below and the interference strategy prediction method described above can be referenced to each other.

[0101] Figure 8 is a schematic diagram of the structure of the interference strategy prediction device provided by the present invention, such as Figure 8 As shown, an interference strategy prediction device according to an embodiment of the present invention includes:

[0102] A collection module 21, used to collect historical interaction data between the radar and the jammer;

[0103] A pre-training module 22 is used to pre-train the network to be trained based on historical interaction data to obtain a strategy prediction network;

[0104] An online fine-tuning module 23 is used to perform online fine-tuning training on the adapter module in the strategy prediction network based on the real-time interaction data between the radar and the jammer, and output the predicted jamming strategy;

[0105] Among them, the adapter module is associated and configured in the policy prediction network after the pre-training is completed; the parameters of the policy prediction network remain unchanged during the online fine-tuning training process.

[0106] The interference strategy prediction device also includes:

[0107] The policy prediction network includes a first linear layer and several cascaded transformer layers; the output of any transformer layer serves as the input of the next transformer layer; and the last transformer layer is connected to the first linear layer.

[0108] The interference strategy prediction device also includes:

[0109] Any transformer layer includes the first normalization layer, the attention layer, the first connection layer, the second normalization layer, the multi-layer perceptron and the second connection layer:

[0110] The first normalization layer is connected to the attention layer;

[0111] The attention layer is connected to the first connection layer, wherein the first connection layer is used to connect the features output by the attention layer and the features input to the first normalization layer;

[0112] The second normalization layer is connected to the first connection layer and the multi-layer perceptron respectively;

[0113] The multilayer perceptron is connected to the second connection layer, wherein the second connection layer is used to connect the features output by the multilayer perceptron and the features input to the output of the first connection layer.

[0114] The interference strategy prediction device also includes:

[0115] The multilayer perceptron includes a second linear layer, a first activation layer, and a third linear layer;

[0116] Among them, the second linear layer is connected to the second normalization layer and the first activation layer respectively;

[0117] The first activation layer is connected to the third linear layer;

[0118] The third linear layer is connected to the second connection layer, wherein the feature dimension corresponding to the feature input to the second linear layer is the same as the feature dimension of the feature output by the third linear layer.

[0119] The interference strategy prediction device also includes:

[0120] The multilayer perceptron in each transformer layer is connected to the adapter module;

[0121] The adapter module consists of an encoder, a second activation layer, a decoder, and a third connection layer:

[0122] The encoder is connected to the second activation layer;

[0123] The second activation layer is connected to the decoder;

[0124] The decoder is connected to the third connection layer.

[0125] The online fine-tuning module 23 is also used for:

[0126] The historical interaction data is input into the first transformer layer for feature conversion processing, and the feature information output by each transformer layer is input into the next transformer layer until the feature information output by the last transformer layer is obtained;

[0127] The feature information output by the last transformer layer is input into the first linear layer to predict the jammer action and obtain the strategy prediction result;

[0128] Based on the strategy prediction result, determining a first loss value;

[0129] Based on the first loss value, the parameters of the network to be trained are optimized and updated to obtain a strategy prediction network.

[0130] The online fine-tuning module 23 is also used for:

[0131] Input the historical interaction data into the first normalization layer for normalization processing to obtain the first normalized feature information;

[0132] Input the first normalized feature information into the attention layer for attention learning to obtain attention feature information;

[0133] Inputting historical interaction data and attention feature information into the first connection layer for feature splicing to obtain first splicing feature information;

[0134] Inputting the first splicing feature information into the second normalization layer for normalization processing to obtain second normalized feature information;

[0135] Inputting the second normalized feature information into a multi-layer perceptron for feature conversion processing to obtain target feature information;

[0136] The target feature information and the first spliced ​​feature information are input into the second connection layer for feature splicing to obtain second spliced ​​feature information, and the second spliced ​​feature information is used as the feature information output by the first transformer layer.

[0137] The online fine-tuning module 23 is also used for:

[0138] Inputting the real-time interaction data into a strategy prediction network including an adapter module to obtain an interference strategy;

[0139] Based on the interference strategy, the parameters of the adapter module are optimized online, where the parameters of the policy prediction network remain unchanged.

[0140] The interference strategy prediction device also includes:

[0141] The determination module is used to determine the optimal strategy of the radar in the action space preset by the radar based on the interference strategy.

[0142] It should be noted here that the above-mentioned device provided in the embodiment of the present invention can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.

[0143] Fig. 9 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig. 9 As shown, the electronic device may include: a processor 310, a memory 320, a communication interface 330 and a communication bus 340, wherein the processor 310, the memory 320 and the communication interface 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 320 to execute the interference strategy prediction method.

[0144] In addition, the logic instructions in the above-mentioned memory 320 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0145] On the other hand, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the interference strategy prediction method provided by the above methods.

[0146] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0147] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting interference strategies, It is characterized in that include: Collect historical interaction data between radar and jammer; Based on the historical interaction data, pre-training the network to be trained to obtain a strategy prediction network; Based on the real-time interaction data between the radar and the jammer, the adapter module in the strategy prediction network is fine-tuned online and the predicted jamming strategy is output; The adapter module is associated and configured in the strategy prediction network after the pre-training is completed; and the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process.

2. The interference strategy prediction method according to claim 1, It is characterized in that The strategy prediction network includes a first linear layer and a plurality of cascaded transformer layers; wherein the output of any transformer layer is used as the input of the next transformer layer; and the last transformer layer is connected to the first linear layer; The pre-training of the network to be trained based on the historical interaction data to obtain a strategy prediction network includes: Input the historical interaction data into the first transformer layer for feature conversion processing, and input the feature information output by each transformer layer into the next transformer layer until the feature information output by the last transformer layer is obtained; Input the feature information output by the last transformer layer into the first linear layer to predict the jammer action and obtain the strategy prediction result; Determining a first loss value based on the strategy prediction result; Based on the first loss value, the parameters of the network to be trained are optimized and updated to obtain the strategy prediction network.

3. The interference strategy prediction method according to claim 2, It is characterized in that Any of the transformer layers includes a first normalization layer, an attention layer, a first connection layer, a second normalization layer, a multilayer perceptron, and a second connection layer: The first normalization layer is connected to the attention layer; The attention layer is connected to the first connection layer, wherein the first connection layer is used to connect the features output by the attention layer and the features input to the first normalization layer; The second normalization layer is connected to the first connection layer and the multi-layer perceptron respectively; The multilayer perceptron is connected to the second connection layer, wherein the second connection layer is used to connect the features output by the multilayer perceptron and the features input to the output of the first connection layer.

4. The interference strategy prediction method according to claim 3, It is characterized in that The inputting the historical interaction data into the first transformer layer for feature conversion processing includes: Inputting the historical interaction data into the first normalization layer for normalization processing to obtain first normalized feature information; Inputting the first normalized feature information into the attention layer for attention learning to obtain attention feature information; Inputting the historical interaction data and the attention feature information into the first connection layer for feature splicing to obtain first splicing feature information; Inputting the first splicing feature information into the second normalization layer for normalization processing to obtain second normalized feature information; Inputting the second normalized feature information into a multi-layer perceptron for feature conversion processing to obtain target feature information; The target feature information and the first splicing feature information are input into the second connection layer for feature splicing to obtain second splicing feature information, and the second splicing feature information is used as the feature information output by the first transformer layer.

5. The interference strategy prediction method according to claim 3, It is characterized in that The multilayer perceptron includes a second linear layer, a first activation layer and a third linear layer; wherein the second linear layer is connected to the second normalization layer and the first activation layer respectively; The first activation layer is connected to the third linear layer; The third linear layer is connected to the second connection layer, wherein a feature dimension corresponding to a feature input to the second linear layer is the same as a feature dimension of a feature output by the third linear layer.

6. The interference strategy prediction method according to claim 3, It is characterized in that The multilayer perceptron in each transformer layer is connected to the adapter module; The adapter module includes an encoder, a second activation layer, a decoder and a third connection layer: The encoder is connected to the second activation layer; The second activation layer is connected to the decoder; The decoder is connected to the third connection layer.

7. The interference strategy prediction method according to claim 6, It is characterized in that The online fine-tuning training of the adapter module in the strategy prediction network based on the real-time interaction data between the radar and the jammer includes: Inputting the real-time interaction data into a strategy prediction network including an adapter module to obtain an interference strategy; Based on the interference strategy, the parameters of the adapter module are optimized online, wherein the parameters of the strategy prediction network remain unchanged.

8. The interference strategy prediction method according to claim 1, It is characterized in that After the output prediction obtains the interference strategy, the method further includes: Based on the interference strategy, an optimal strategy for the radar is determined in an action space preset by the radar.

9. An interference strategy prediction device, It is characterized in that include: A collection module, used to collect historical interaction data between radar and jammer; A pre-training module, used for pre-training the network to be trained based on the historical interaction data to obtain a strategy prediction network; An online fine-tuning module, used for performing online fine-tuning training on the adapter module in the strategy prediction network based on the real-time interaction data between the radar and the jammer, and outputting the predicted jamming strategy; The adapter module is associated and configured in the strategy prediction network after the pre-training is completed; and the parameters of the strategy prediction network remain unchanged during the online fine-tuning training process.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the interference strategy prediction method according to any one of claims 1 to 8 is implemented.

Citation Information

Cited By

  • Interference resource scheduling method, device and equipment for air-ground integrated network

    CN119767415A