Communication anti-interference strategy recommendation method based on time sequence knowledge graph

Abstract

The invention relates to a communication anti-interference strategy recommendation method based on a time sequence knowledge graph, and the method specifically comprises the steps: converting sensed environment interference data into a structured knowledge graph, and achieving the precise classification and historical recording of interference types. On the basis of channel environment data monitored in real time, in combination with communication network configuration, historical anti-interference data and environment information of equipment, an optimal anti-interference strategy is intelligently recommended to adapt to a current complex and changeable interference environment, and the robustness of an anti-interference system is improved; historical interference data and a machine learning technology are utilized to predict interference types possibly faced by equipment, and corresponding anti-interference measures are deployed in advance, so that the reliability of a communication system is improved. According to the method, the problems of insufficient accuracy, robustness, safety and service quality when a current interference strategy recommendation algorithm is used for radio communication are solved, the anti-interference efficiency is improved, manual intervention is reduced, and the robustness of an anti-interference system is enhanced.

Description

Technical Field

[0001] The present invention relates to a communication anti - interference technology, and particularly to an intelligent anti - interference strategy recommendation method based on a temporal knowledge graph. Background Art

[0002] Radio communication, as an important information transmission method, plays an irreplaceable role in military, civilian and other fields. However, with the increasingly complex radio environment, radio communication faces various interferences, which may come from multiple aspects such as the natural environment, human interference, electronic devices, etc., seriously affecting the communication quality and even causing communication interruption. Therefore, it is particularly important to study and implement effective radio communication anti - interference strategies.

[0003] As a structured knowledge base, a knowledge graph can represent the complex semantic relationships between entities, concepts, attributes and events in the form of a graph structure, becoming an efficient data management and organization method in the Internet era. On this basis, a temporal knowledge graph adds a time dimension to the knowledge graph, which can capture the changes of entities and relationships in the knowledge graph in a timely manner. At the same time, the structured features of the temporal knowledge graph can well describe the relationships between various interference information in the electromagnetic environment, understand the complexity of the electromagnetic interference environment as a whole, and provide more comprehensive and accurate data characteristics for electromagnetic anti - interference strategy recommendation by combining the captured historical information.

[0004] Researchers have proposed several interference strategy recommendation algorithms, including the DQN agent decision algorithm and the Q - Learning agent decision algorithm, etc. However, these methods often can only make decisions based on the current channel state, and cannot fully consider multi - dimensional information such as environmental factors, historical data and communication networks. This limitation leads to insufficient accuracy and robustness of the decision, and it is difficult to cope with complex and changeable interference environments. In addition, these methods cannot predict future possible interferences and formulate effective anti - interference plans in advance, which may affect the security and service quality of the communication system. Summary of the Invention

[0005] Aiming at the problems of insufficient accuracy, robustness, security, and quality of service in the current interference strategy recommendation algorithm for radio communication, a communication anti-interference strategy recommendation method based on a temporal knowledge graph is proposed. Specifically, it includes: 1. Constructing an intelligent interference knowledge base: By converting the perceived environmental interference data into a structured knowledge graph, accurate classification of interference types and historical records are achieved. 2. Real-time adaptive strategy recommendation: Based on the real-time monitored channel environment data, combined with the communication network configuration of the device, historical anti-interference data, and environmental information, the optimal anti-interference strategy is intelligently recommended to adapt to the current complex and changeable interference environment and improve the robustness of the anti-interference system. 3. Predictive interference response: Using historical interference data and machine learning techniques, the interference types that the device may face within the next 24 hours are predicted, and corresponding anti-interference measures are deployed in advance, thereby improving the reliability of the communication system. This method realizes the accurate identification of communication interference types by constructing a knowledge graph containing rich time information, and can intelligently recommend corresponding anti-interference strategies, improve anti-interference efficiency, reduce manual intervention, and enhance the robustness of the anti-interference system.

[0006] The technical solution of the present invention is as follows:

[0007] A communication anti-interference strategy recommendation method based on a temporal knowledge graph, comprising the following steps:

[0008] Step 1, interference signal processing stage: Perform interference perception and classification. Obtain the original time series of environmental interference, i.e., one-dimensional time series, through an interference perceptron. Decompose the obtained one-dimensional time series by variational mode decomposition operation into several intrinsic mode functions, and extract the local features and time-frequency characteristics of the interference signal; splice the obtained intrinsic mode functions to obtain an intrinsic mode matrix; for the obtained intrinsic mode matrix, use a two-dimensional convolutional neural network to extract its features; input the extracted information into a residual network, and through the deep learning of the residual network, obtain the classification label of the interference signal, and use the obtained label together with the center frequency of each mode obtained by variational mode decomposition as key features;

[0009] Step 2, temporal knowledge graph construction stage: Construct a temporal knowledge graph, which includes the network structure information of the communication network where the current device is located, the record of relevant interference information suffered, and the anti-interference strategy knowledge graph; the knowledge graphs of the anti-interference strategy knowledge graph and the network structure information of the communication network where the current device is located are stored separately, and a timestamp label is added when training the link prediction model subsequently;

[0010] Step 3, Link Prediction Phase: Construct a link prediction model, which includes two modules. The first module is the historical vocabulary generation module. This module automatically identifies timestamp information, counts the historical vocabulary that appears in the temporal knowledge graph before the timestamp, and generates a historical vocabulary according to the incremental maintenance method after obtaining the historical vocabulary set. The historical vocabulary contains the past states and behaviors of the system. The second module is the neural network model for link prediction, which is constructed based on a recurrent neural network and a fully connected layer. First, encode the entities and relationships in the temporal knowledge graph into vector representations. Then, input these vectors into the recurrent neural network to learn the interaction relationships between entities. Finally, fuse the output of the recurrent neural network with the historical vocabulary, calculate the prediction scores of each candidate entity through the Softmax function, and the candidate entity with the highest prediction score is the final prediction result.

[0011] Further, it specifically includes the following steps:

[0012] Step 1, Interference Signal Processing Phase:

[0013] Perform interference perception classification. Obtain the original temporal sequence f of environmental interference, i.e., a one-dimensional temporal sequence, through an interference perceptron. Decompose the obtained one-dimensional temporal sequence through variational mode decomposition operation into several intrinsic mode functions, and extract the local features and time-frequency characteristics of the interference signal:

[0014]

[0015] where, u k (t) is the k-th intrinsic mode function, ω k is the central frequency of the k-th intrinsic mode, δ(t) is the Dirac function, is the first derivative with respect to time t, and j is the imaginary unit;

[0016] Concatenate the obtained intrinsic mode functions u1, u2,..., u k to obtain an intrinsic mode matrix which is a two-dimensional matrix. Each row of this matrix represents the numerical values of the intrinsic mode function sequence at the same sampling point, and each column represents all the sampling point numerical values of the same intrinsic mode function sequence within the unit time point i;

[0017] For the obtained intrinsic mode matrix Use a two-dimensional convolutional neural network to extract features from it; input the extracted information into a residual network, and through the deep learning of the residual network, obtain the classification label of the interference signal, and use the obtained label together with the modal center frequencies ω k ={ω1, ω2,..., ω k} as key features;

[0018] Step 2: Temporal Knowledge Graph Construction Phase

[0019] Represent the head entity, tail entity, relationship, and timestamp information in the fact as a quadruple and construct a temporal knowledge graph; this knowledge graph consists of three core parts: the network structure information of the communication network where the current device is located, the record of interference-related information suffered, and the anti-interference strategy knowledge graph; the network structure information of the communication network where the current device is located refers to the communication Internet where the current device is located, and the network structure information records the connection status of all communication device nodes in the entire communication network; the anti-interference strategy knowledge graph is obtained through entity extraction and relationship extraction from the anti-interference knowledge in the communication field, which records various anti-interference means in the communication anti-interference field and their usage conditions; the interference-related information record refers to the type and frequency band characteristics of the interference signal in the past period of time recorded by the temporal knowledge graph, and the anti-interference means adopted by the system; the knowledge graphs of the anti-interference strategy knowledge graph and the network structure information of the communication network where the current device is located are stored separately, and when training the link prediction model later, a timestamp label is added.

[0020] Step 3: Link Prediction Phase

[0021] Construct a link prediction model. The link prediction model includes two modules. The first module is the historical vocabulary generation module. This module automatically identifies the timestamp information, counts the historical vocabulary that appears in the temporal knowledge graph before the timestamp, and determines the assignment of the historical vocabulary by checking whether there is a tail entity that can form a quadruple with the head entity and the relationship at a certain moment, that is, if the fact exists, the value at the corresponding position in the historical vocabulary is set to 1, otherwise it is set to 0, thus forming a one-hot vector; operate over the entire time period to form a set of historical vocabulary for all moments within this time period. where is an N-dimensional one-hot vector that contains the set of all tail entities that can form facts with the head entity and the relationship in the interference record subgraph at time t i After obtaining the set of historical vocabulary, generate the historical vocabulary according to the incremental maintenance method; the historical vocabulary contains the past state and behavior of the system, providing rich context information for the link prediction model; the second module is the neural network model for link prediction, which is constructed based on a recurrent neural network and a fully connected layer; first, encode the entities and relationships in the temporal knowledge graph into vector representations; then, input these vectors into the recurrent neural network to learn the interaction relationships between entities; finally, fuse the output of the recurrent neural network with the historical vocabulary, and calculate the prediction score of each candidate entity through the Softmax function. The candidate entity with the highest prediction score is the final prediction result.

[0022] The beneficial effects of the present invention are as follows:

[0023] 1. Real-time intelligent decision-making: By using the temporal knowledge graph, the current channel state and historical interference data are analyzed in real time, the interference type is accurately identified, and the optimal anti-interference strategy is intelligently recommended, significantly improving the system's adaptability to the dynamically changing interference environment.

[0024] 2. Predictive maintenance: Based on historical data and machine learning models, future interference is predicted, and preventive measures are taken in advance, effectively reducing the system failure rate and improving the system reliability.

[0025] 3. Self-learning ability: The system can continuously optimize the model by learning new interference data and the effects of strategies, improving the system's performance and robustness. Brief Description of the Drawings

[0026] Figure 1 It is a modeling diagram of the interference perception system of the present invention;

[0027] Figure 2 It is a flow chart of the improved anti-interference strategy recommendation of the present invention. Detailed Embodiment

[0028] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and the detailed implementation manner and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0029] The present invention proposes a communication anti-interference strategy recommendation method based on a temporal knowledge graph. This method mainly involves interference perception classification and the problem of recommending anti-interference strategies for corresponding interference types using knowledge graph data containing time information. As Figure 2 shown, this method consists of three stages, namely the interference signal processing stage, the temporal knowledge graph construction stage, and the link prediction stage.

[0030] As Figure 1 shown, the interference perception classification obtains the original temporal sequence f of the environmental interference through an interference perception machine, that is, a one-dimensional temporal sequence, and performs variational mode decomposition operation on the obtained one-dimensional temporal sequence to decompose it into several intrinsic mode functions, effectively extracting the local features and time-frequency characteristics of the interference signal:

[0031]

[0032] Among them, u k (t) is the k-th intrinsic mode function, ω k is the central frequency of the k-th intrinsic mode, δ(t) is the Dirac function, is the first derivative with respect to time t, and j is the imaginary unit.

[0033] The obtained intrinsic mode functions u1, u2,..., u k are concatenated to obtain the intrinsic mode matrix is a two-dimensional matrix. Each row of this matrix represents the values of the intrinsic mode function sequence at the same sampling point, and each column represents the values of all sampling points of the same intrinsic mode function sequence within the unit time point i.

[0034] For the obtained intrinsic mode matrix use a two-dimensional convolutional neural network (Convolutional Neural Network, CNN) to extract features from it. CNN can efficiently capture local features in images and is suitable for processing signals in the form of two-dimensional matrices. The extracted information is input into the residual network. Through the deep learning of the residual network, the feature extraction ability and classification accuracy of the model are further improved, and the classification label of the interference signal is obtained. The obtained label, together with the center frequencies ω k ={ω1, ω2,..., ω k} are used as key features for subsequent embedding into the constructed temporal knowledge graph, providing rich data support for subsequent interference prediction and strategy recommendation.

[0035] In order to achieve intelligent prediction and strategy recommendation of interference, this method represents the head entity, tail entity, relationship, and timestamp information in the fact in the form of a quadruple and constructs a temporal knowledge graph. This knowledge graph contains three core parts: the network structure information of the communication network where the current device is located, the record of interference-related information suffered, and the anti-interference strategy knowledge graph. The network structure information of the communication network where the current device is located refers to the communication Internet where the current device is located. The network structure information records the connection status of all communication device nodes in the entire communication network; the anti-interference strategy knowledge graph is obtained through entity extraction and relationship extraction from the anti-interference knowledge in the communication field, which records various anti-interference means in the communication anti-interference field and their usage; the interference-related information record refers to the type and frequency band characteristics of the interference signal in the past period of time recorded by the temporal knowledge graph, as well as the anti-interference means adopted by the system. Since the anti-interference strategy knowledge graph and the network structure information of the communication network where the current device is located do not change drastically in a short period of time, these two parts of the knowledge graph are stored separately and timestamp labels are added during subsequent link prediction model training. By embedding the above three knowledge graphs into the temporal knowledge graph, the semantic information of the temporal knowledge graph is enriched, providing a solid foundation for subsequent link prediction.

[0036] The link prediction model of this method includes two modules. The first module is the historical vocabulary generation module, which automatically identifies timestamp information, counts the historical vocabulary that appears in the temporal knowledge graph before this timestamp, and determines the assignment of the historical vocabulary by checking whether there is a tail entity that can form a quadruple with the head entity and the relationship at a certain moment. That is, if this fact exists, the value at the corresponding position in the historical vocabulary is set to 1, otherwise it is set to 0, thus forming a one-hot vector. Operations are performed over the entire time period to form a set of historical vocabulary for all moments within this time period. Among them is an N-dimensional one-hot vector that contains the set of all tail entities that can form facts with the head entity and the relationship in the interference record subgraph at time t i . After obtaining the set of historical vocabulary, the historical vocabulary is generated according to the method of incremental maintenance. The historical vocabulary contains the past states and behaviors of the system, provides rich context information for the link prediction model, and helps the model better understand the current prediction task. By learning the patterns in historical data, the model can better adapt to new and unseen data, improving the generalization ability of the model. The second module is the neural network model for link prediction, which is constructed based on a recurrent neural network and a fully connected layer. First, the entities and relationships in the temporal knowledge graph are encoded into vector representations. Then, these vectors are input into the recurrent neural network to learn the interaction relationships between entities. Finally, the output of the recurrent neural network is fused with the historical vocabulary, and the prediction scores of each candidate entity are calculated through the Softmax function. The candidate entity with the highest prediction score is the final prediction result. Utilizing the powerful learning ability of the neural network, the prediction accuracy of link prediction is improved, providing intelligent decision support for the system.

[0037] The above-described embodiments only represent one implementation manner of the present invention, and its description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A communication anti-interference strategy recommendation method based on time series knowledge graph, characterized in that: The following steps are involved: Step 1, interference signal processing stage: perform interference perception classification, obtain the original time series of environmental interference, that is, one-dimensional time series, through the interference perception machine, decompose the obtained one-dimensional time series into several intrinsic mode functions through variational mode decomposition, and extract the local features and time-frequency characteristics of the interference signal; splice the obtained intrinsic mode functions to obtain the intrinsic mode matrix; for the obtained intrinsic mode matrix, use a two-dimensional convolutional neural network to extract features; input the extracted information into the residual network, and obtain the classification label of the interference signal through deep learning of the residual network, and use the obtained label together with the center frequency of each mode obtained by variational mode decomposition as the key feature; Step 2, time series knowledge graph construction phase: construct a time series knowledge graph, which includes the network structure information of the communication network where the current device is located, the records of related information about the interference suffered, and the anti-interference strategy knowledge graph; the knowledge graphs of the anti-interference strategy knowledge graph and the network structure information of the communication network where the current device is located are stored separately, and the timestamp label is added during the subsequent link prediction model training; Step 3, link prediction phase: build a link prediction model. The link prediction model includes two modules. The first module is a historical vocabulary generation module. This module automatically identifies timestamp information, counts the historical words that appear in the time series knowledge graph before the timestamp, and generates a historical vocabulary according to the incremental maintenance method after obtaining the historical vocabulary set. The historical vocabulary contains the past state and behavior of the system. The second module is a neural network model for link prediction. This module is built based on a recursive neural network and a fully connected layer. First, the entities and relationships in the time series knowledge graph are encoded as vector representations. Then, these vectors are input into the recursive neural network to learn the interaction between entities. Finally, the output of the recursive neural network is fused with the historical vocabulary, and the prediction score of each candidate entity is calculated through the Softmax function. The candidate entity with the highest prediction score is the final prediction result.

2. According to claim 1, the communication anti-interference strategy recommendation method based on time series knowledge graph is characterized in that: The specific steps include: Step 1: Interference signal processing stage: Interference perception classification is performed. The original time series f of environmental interference, i.e., the one-dimensional time series, is obtained through the interference perception machine. The obtained one-dimensional time series is decomposed into several intrinsic mode functions by variational mode decomposition, and the local features and time-frequency characteristics of the interference signal are extracted: Among them, u k (t) is the kth eigenmode function, ω k is the center frequency of the kth eigenmode, δ(t) is the Dirac function, is the first-order derivative with respect to time t, j is an imaginary unit; The obtained intrinsic mode functions u1,u2,...,u k Concatenate and get the eigenmode matrix is a two-dimensional matrix, each row of which represents the value of the intrinsic mode function sequence at the same sampling point, and each column represents the values ​​of all sampling points of the same intrinsic mode function sequence at the unit time point i; For the obtained eigenmode matrix A two-dimensional convolutional neural network is used to extract features; the extracted information is input into the residual network, and the classification label of the interference signal is obtained through deep learning of the residual network, and the obtained label is combined with the center frequency ω of each mode obtained by variational mode decomposition. k ={ω1,ω2,...,ω k } as key features; Step 2: Time series knowledge graph construction phase The head entity, tail entity, relationship, and timestamp information in the fact are represented in the form of quadruples and a time series knowledge graph is constructed; the knowledge graph contains three core parts: the network structure information of the communication network where the current device is located, the interference-related information records suffered, and the anti-interference strategy knowledge graph; the network structure information of the communication network where the current device is located refers to the communication Internet where the current device is located, and the network structure information records the connectivity of all communication device nodes in the entire communication network; the anti-interference strategy knowledge graph is obtained from the anti-interference knowledge in the communication field through entity extraction and relationship extraction, which records various anti-interference means in the communication anti-interference field and their use; the interference-related information records refer to the types and frequency band characteristics of interference signals in the past period of time recorded in the time series knowledge graph, as well as the anti-interference means adopted by the system; the knowledge graphs of the two parts, the anti-interference strategy knowledge graph and the network structure information of the communication network where the current device is located, are stored separately, and the timestamp label is added when the link prediction model is trained later; Step 3: Link prediction phase Construct a link prediction model. The link prediction model includes two modules. The first module is the historical vocabulary generation module. This module automatically identifies timestamp information, counts the historical vocabulary that appears in the time series knowledge graph before the timestamp, and determines the assignment of the historical vocabulary by checking whether there is a tail entity that can form a quadruple with the head entity and relationship at a certain moment. That is, if the fact exists, the value of the corresponding position in the historical vocabulary is set to 1, otherwise it is set to 0, thus forming a unique hot vector; operate on the entire time period to form a historical vocabulary set for all moments in the time period. in is an N-dimensional one-hot vector that contains the i The interference record subgraph at each moment records the set of tail entities that can constitute facts with the head entity and relationship; after obtaining the historical vocabulary set, the historical vocabulary is generated according to the incremental maintenance method; the historical vocabulary contains the past state and behavior of the system, providing rich contextual information for the link prediction model; the second module is a neural network model for link prediction, which is built based on a recursive neural network and a fully connected layer; first, the entities and relationships in the temporal knowledge graph are encoded as vector representations; then, these vectors are input into the recursive neural network to learn the interaction between entities; finally, the output of the recursive neural network is fused with the historical vocabulary, and the prediction score of each candidate entity is calculated through the Softmax function, and the candidate entity with the highest prediction score is the final prediction result.

Images