Interference resource planning method based on knowledge graph and space-time diagram convolutional neural network, medium and equipment

Abstract

The invention discloses an interference resource planning method based on a knowledge graph and space-time diagram convolutional neural network. A head entity, a relation and a tail entity gt; constructing an interference pattern decision knowledge graph in a triple form; constructing an interference pattern decision knowledge graph embedding representation model, and generating embedding features representing semantic association; constructing interference pattern decision knowledge fusion based on a multi-layer perceptron, and combining attribute characteristics of interference pattern decision with a historical space-time network interference efficiency evaluation sequence; constructing a space-time diagram convolutional neural network to model embedded features, extracting space correlation characteristics between nodes, capturing a time evolution rule of node attributes, and dynamically predicting interference efficiency; and reasoning an optimal interference pattern in combination with the interference efficiency prediction result and the real-time situation information of the target equipment. The method can effectively predict the interference efficiency evaluation of different interference devices, significantly reduces the prediction error of the interference efficiency evaluation, and improves the prediction precision and model stability.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to an interference resource planning method, medium and device based on a knowledge graph and a spatiotemporal graph convolutional neural network. Background Art

[0002] Against the backdrop of the rapid development of the low-altitude economy, the deep integration of low-altitude platforms, drones, satellites and ground communication systems has made the spectrum environment increasingly complex. Malicious sensing devices (such as malicious sensing drones) have become a major potential threat to public security by locating, tracking and illegally obtaining target activity information, especially in protecting critical infrastructure (such as surface operating platforms) from interference from malicious sensing drones. Therefore, in order to reduce the threat of malicious sensing drones to the safety of surface operating platforms, the monitoring and countermeasure platforms for malicious sensing drones will use different active and passive interference styles to interfere with malicious sensing drone devices. The traditional interference style decision method relies on the combination of prior expert knowledge base and template matching. However, this method makes interference decisions based on state data to fill in query templates. When the data deviation is large or the filled template is undefined, it is not possible to make accurate interference decisions. Summary of the invention

[0003] In response to the interference style decision-making problem in malicious perception defense scenarios, the present invention provides an interference resource planning method, medium and equipment based on knowledge graph and spatiotemporal graph convolutional neural network, which can effectively predict the interference effectiveness evaluation of different interference devices, significantly reduce the interference effectiveness evaluation prediction error, and improve the prediction accuracy and model stability.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0005] A method for interference resource planning based on knowledge graph and spatiotemporal graph convolutional neural network, the method comprising the following steps:

[0006] Step 1: Aiming at the game confrontation scenario of malicious perception defense, the interference pattern decision knowledge graph is constructed in the form of <head entity, relationship, tail entity> triples to systematically represent the multi-dimensional and complex relationship between the monitoring and countermeasure platform, malicious perception equipment and interference patterns;

[0007] Step 2: Construct an interference style decision knowledge graph embedding representation model, initialize the embedding of entities and relations, and map the entities and relations to a high-dimensional vector space through a convolutional neural network to generate embedding features that represent semantic associations;

[0008] Step 3: Construct the interference style decision knowledge fusion based on multi-layer perceptron, combine the attribute characteristics of interference style decision with the historical spatiotemporal network interference effectiveness evaluation sequence, and fuse multi-source heterogeneous data;

[0009] Step 4: Construct a spatiotemporal graph convolutional neural network to model the embedded features, extract the spatial correlation characteristics between nodes through the graph convolution module, and combine the gated recurrent unit to capture the time evolution of node attributes and dynamically predict the interference effectiveness;

[0010] Step 5: Combine the jamming effectiveness prediction results and the real-time situation information of the target device to infer the optimal jamming pattern.

[0011] Furthermore, step 1 includes the following sub-steps:

[0012] Step 1.1: Construct a game confrontation model for malicious perception defense, and set the malicious perception drone equipment set U = {u 1 ,u 2 ,…,u k} and the monitoring and countermeasure platform set P = {p 1 ,p 2 ,...,p m}, k = 1, 2, ..., K, m = 1, 2, ..., M, K is the number of malicious sensing devices onboard drones; M is the number of monitoring and countermeasure platforms; each monitoring and countermeasure platform can select a set of interference patterns J = {n 1 ,n 2 ,...,n m}Including active jammers and various passive jammers, n=1,2,...,N J , N J The number of jamming pattern sets selectable for each jamming device;

[0013] Step 1.2: For the game confrontation scenario of malicious sensing defense, construct a triple describing the relationship between malicious sensing devices, interference devices and interference patterns; set the knowledge triple <e h ,r,e t > represents the head entity, relationship and tail entity in the triple relationship; the entity set E includes various entities related to the interference spectrum; the relationship set R is used to describe various relationships between entities;

[0014] Step 1.3: Based on the interference pattern decision knowledge graph generated by the triples, multiple triples are formed including <monitoring and countermeasure platform_malicious interference drone_interference pattern, interference pattern, noise suppression interference>, <monitoring and countermeasure platform_malicious interference drone_interference pattern, interference quantity, interference quantity value>, <monitoring and countermeasure platform_malicious interference drone_interference pattern, malicious interference drone, malicious interference drone number>; set the positive sample triple set to T KG , which contains triples with correct correspondences; by removing the head entity e in the positive sample triple h , tail entity e t or relation r, and use other head entities e′ h , tail entity e′ t Or relation r′ is replaced to obtain the negative sample triple set T′ of the wrong corresponding association relationship KG , denoted as T′ KG ={ <e′ h ,r,e t >|e h ∈E}∪{ <e h ,r′,e t >|r∈R}∪{{ <e h ,r,e′ t >|e t ∈E}}.

[0015] Furthermore, the entity set E includes various entities related to the interference spectrum, specifically including the monitoring and countermeasure platform number, the malicious perception equipment number, and the interference pattern number; the interference patterns include internal active suppression and deception interference equipment and foil centroid interference, foil dilution interference, infrared centroid interference, and infrared dilution interference; the interference parameter values ​​include interference distance values ​​and interference quantity values.

[0016] Furthermore, step 2 includes the following sub-steps:

[0017] Step 2.1: Set each entity as a c-dimensional vector and set up an independent semantic space for each relation r; for any given triple <e h ,r,e t >, set the projection matrix M r , the projection matrix M r The value of is randomly initialized and continuously updated during the model training process. The projection matrix M r The dimension of is set to the product of the dimension of the entity embedding encoding column vector and the dimension of the relation embedding encoding column vector; by multiplying the head entity embedding encoding column vector h by the shadow matrix M r , the tail entity embedding encoding column vector t multiplied by M r , so that the head entity h r and tail entity tr Projecting from entity space to relational space, we get h r and t r ,h r =M r h and t r =M r t;

[0018] Step 2.2: The triple association relationship is obtained by converting the head entity vector h r Add the mapping result of the relationship vector r to the tail entity vector t r Align so that the triplet representation satisfies h r +r≈t r ; For positive sample triples <e h ,r,e t >∈T, set the interference style decision knowledge graph embedding representation model embedding objective function d <e h ,r,e t >=||h r +rt r ||≈0; for negative sample triples <e′ h ,r′,e′ t >∈T′, set the embedding objective function d <e′ h ,r′,e′ t >=||h′ r +rt′ r ||>>0; Set the score function f KG <e h ,r,e t >, used to calculate the head entity h r Through the relationship r to the end entity t r The distance between and is used to evaluate the triples:

[0019] Step 2.3: Set the loss function L KG , calculate the positive sample score function value d <e h ,r,e t > and the negative sample score function value d <e′ h ,r′,e′ t >The difference between the two; set the interval hyperparameter γ to reduce the distance between the positive and negative sample scores; set the function [x] + =max(0,x), which is used to calculate the maximum value between 0 and x, and limit the loss function value to a non-negative interval; the gradient descent method is used to train the interference style decision knowledge graph embedding representation model to optimize the loss function L KG , so that the positive sample triplet gets a lower score value, while the negative sample triplet gets a higher score value:

[0020] Step 2.4: Facing the interference style decision knowledge graph, embed the triples in the interference style decision knowledge graph through the knowledge graph-based embedding model <e h ,r,e t > Initialize to a 3D vector v h ,v r ,v t ; Assume Φ KG represents the set of filters in the convolutional layer of the knowledge graph for interference style decision, τ KG Represents the number of filters. Multiple filters perform convolution operations on three-dimensional vectors and are concatenated into a vector with dimension A single vector of Φ to retain the diverse information extracted by different filters; Φ KG τ KG Represents the vector dimension obtained by sequentially concatenating the features extracted by each filter; the single vector obtained by concatenation is concatenated with the weight parameter matrix Perform dot multiplication and return the triple score f CNN-KG <e h ,r,e t >, used to distinguish the triples of correct correspondence from the triples of incorrect correspondence, so that the score of the positive sample triple is lower than that of the negative sample triple: f CNN-KG <e h ,r,e t > = concat(σ KG ([v h ,v r ,v t ]*Φ KG ))·u KG , σ KG represents the convolution function;

[0021] Step 2.5: Set the loss function of the knowledge graph feature embedding based on the convolutional neural network to It is used to calculate the loss of positive and negative triplets, where the positive sample triplet sets the label l to +1, the negative sample triplet sets the label l to -1, and λ KG is the regularization coefficient; the training loss is minimized by adaptive moment estimation to obtain the embedding vectors of the corresponding head entity, relation, and tail entity, where the embedding vector of the head entity is used as the feature vector.

[0022] Step 3 further comprises the following steps:

[0023] The attribute features of the interference style decision knowledge e KG Interference effectiveness evaluation sequence x with historical spatiotemporal network t , through the weight matrix w s Perform weighted summation and add the corresponding bias vector bs ; Set the activation function ReLu(·) to perform nonlinear transformation on the weighted input to generate the output of the current layer; recursively map layer by layer to establish the mapping relationship between input features and output values: X t =ReLu(e KG x t w s +b s ).

[0024] Step 4 further comprises the following steps:

[0025] Step 4.1: Construct the spatiotemporal graph network model G KG =(V KG ,E KG ,A KG +, node set V KG represents the set of network nodes of the interference pattern decision model for malicious sensing devices, The number of node sets is Edge set E KG It represents the association relationship set of the same type of interference patterns, UAV-borne malicious perception devices, and monitoring and countermeasure platforms based on the semantic consistency of interference patterns, including the UAV-borne malicious perception device U k Monitoring and Countermeasure Platform m , UAV-mounted malicious sensing equipment k With interference pattern G n and Monitoring and Countermeasure Platform P m With interference pattern G n The relationship between Constructing the spatiotemporal graph network G KG The adjacency matrix of Used to describe whether there is a relationship between graph nodes;

[0026]

[0027] In the formula, A ij =1 means that the mth monitoring and countermeasure platform implements the nth interference pattern on the ith UAV-borne malicious sensing device;

[0028] Step 4.2: Set the spatiotemporal graph network data as the graph model G KG The interference effectiveness evaluation value calculated after the mth monitoring and countermeasure platform implements the nth interference style on the i-th UAV-mounted malicious sensing device at time t; the feature vector dimension of the graph node is set to c; in the time step t, the spatiotemporal graph network model graph model G KG The characteristic is expressed as Set the mapping function f KG The historical spatiotemporal network interference effectiveness evaluation sequence Future interference effectiveness evaluation prediction mapped to spatiotemporal network According to the observed T KG The historical interference effectiveness evaluation value predicts the next T K ' G The interference effectiveness evaluation value of the step is:

[0029]

[0030] Where W KG is the trainable parameter matrix of the mapping function, T KG is the length of the historical spatiotemporal network sequence, T K ' G is the length of the target spatiotemporal network sequence to be predicted;

[0031] Step 4.3: Based on the topological structure of the graph model, a neighborhood aggregation strategy is used to update the features of graph nodes and their neighbors; the spatiotemporal graph convolutional neural network architecture is expressed as:

[0032] In the formula, represents the adjacency matrix after adding self-loops, for The degree matrix, W l,KG is the weight matrix of the lth layer, y l ′ is the output feature matrix of the lth layer, and y′ 0 =X t ;

[0033] Step 4.4: Extract node attribute features in the interference style decision knowledge graph through the graph convolutional neural network and pass it as input to the GRU to learn the time evolution law of the node label and predict the interference effectiveness evaluation value; use the graph convolutional neural network to extract the node relationship information in the graph structure and model the spatial dependency of node features; use GRU to capture the dynamic changes of node labels over time and generate the predicted interference effectiveness evaluation value;

[0034] u t =σ(W u gc([X t ,h t-1 ],A)+b u )

[0035] r t =σ(W r gc([X t ,h t-1 ],A)+b r )

[0036] c t =tanh(W c gc([Xt ,(r t ⊙h t-1 )],A)+b c )

[0037] h t =u t ⊙h t-1 +(1-u t )⊙c t

[0038] In the formula, r t represents the reset gate, u t represents the update gate, where the reset gate combines the information in the memory with the current time step information, and the update gate is used to select memory or forget information; the activation function σ(·) represents the gating signal; c t Indicates the instantaneous state of the current time step; in the memory update phase, when u t When u is used as a forget gate, t ⊙h t-1 Forget unimportant previous information to update memory; when u t As a memory gate, (1-u t )⊙c t Remember current information; W u , W r , W c represents the weight matrix, b u 、b r 、b c represents the parameter matrix; h t represents the output at time t; h t Input to the fully connected layer to generate the predicted interference effectiveness evaluation value

[0039] Step 4.5: Calculate the actual interference effectiveness evaluation value y of the known node t Interference effectiveness prediction value The mean square error between them is used to construct the loss function of the graph convolutional neural network interference style decision model based on knowledge graph embedding: In the formula, and Represent the true value and predicted value of the j-th time sample on node m, respectively; y and They are and The set of KG is the regularization coefficient; L reg is the regularization loss term.

[0040] Furthermore, in step 4.5, for the optimized model, the root mean square error between the true value of the known graph node and its interference effectiveness prediction value is used as the evaluation index of the transmission performance prediction.

[0041] In a second aspect, the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the interference resource planning method based on the knowledge graph and spatiotemporal graph convolutional neural network as described above.

[0042] In a third aspect, the present invention discloses an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the interference resource planning method based on the knowledge graph and spatiotemporal graph convolutional neural network as described above is implemented.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The interference resource planning method, medium and device based on knowledge graph and spatiotemporal graph convolutional neural network of the present invention, aiming at the interference style decision problem of the monitoring and countermeasure platform on the malicious sensing device, constructs the game confrontation scenario of the monitoring and countermeasure platform and the malicious sensing device, combines the space, position relationship and performance parameters of the malicious sensing device and the monitoring and countermeasure platform, simulates the interference effectiveness evaluation value corresponding to the interference style, and generates the interference confrontation situation. Secondly, the present invention constructs the interference style decision knowledge graph to describe the complex relationship between the monitoring and countermeasure platform, the malicious sensing device, and the interference style, and generates the embedded feature vector describing the complex relationship through the knowledge graph embedding algorithm. Then, the present invention extracts the node attribute features in the interference style decision knowledge graph through the graph convolutional neural network, and uses it as the input of the gated recurrent unit module, learns the time evolution law of the node attribute, that is, the interference effect, realizes the prediction of the interference effectiveness evaluation value, and thus infers the optimal interference style. The proposed algorithm can accurately predict the interference effectiveness evaluation, thereby realizing the effective planning of interference resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of the main steps of an interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network of the present invention.

[0046] Figure 2 This is a specific operation flow chart of an interference resource planning method based on a knowledge graph and a spatiotemporal graph convolutional neural network of the present invention.

[0047] Figure 3 This is an example diagram of the architecture of an interference resource planning method based on a knowledge graph and a spatiotemporal graph convolutional neural network of the present invention.

[0048] Figure 4 This is an example diagram of a game confrontation scenario for malicious perception defense according to the present invention.

[0049] Figure 5 This is an example diagram of the interference style decision knowledge graph generated based on triples in the present invention.

[0050] Figure 6 This is a structural diagram of the interference style decision knowledge graph embedding model of the present invention. DETAILED DESCRIPTION

[0051] The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0052] Embodiment 1

[0053] This embodiment proposes an interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network. By constructing an interference style decision knowledge graph, the system characterizes the multi-dimensional complex relationship between the monitoring and countermeasure platform, malicious perception equipment and interference style, and uses the knowledge graph embedding algorithm to map entities and relationships to a high-dimensional vector space to generate embedded features that can represent semantic associations. On this basis, the spatiotemporal graph convolutional neural network is used to model the embedded features, and the spatial correlation characteristics between nodes are extracted through the graph convolution module. The gated recurrent unit is combined to capture the time evolution of node attributes, thereby realizing the dynamic prediction of interference effectiveness. The algorithm further combines the interference effectiveness prediction results and the real-time situation information of the target device to infer the optimal interference style. Its main steps, operation process and architecture are as follows: Figure 1-3 As shown, the specific steps include the following steps.

[0054] Step 1: In the context of low-altitude intelligent networking, the present invention aims at the safety protection of surface operating platforms and solves the potential threat posed by malicious sensing devices on drones to the security and privacy of the platforms; a variety of effective interference patterns are adopted through the monitoring and countermeasure platform to prevent these devices from detecting and identifying the platform, thereby ensuring the safety and confidentiality of the platform; in order to effectively prevent malicious sensing drones from successfully detecting and identifying targets, we implement active interference suppression, deceptive interference and passive interference on malicious sensing drones through the monitoring and countermeasure platform, making it impossible for them to successfully identify the target, thereby achieving effective protection of the surface operating platform.

[0055] like Figure 4As shown in the figure, in malicious sensing defense, the game confrontation process simulates the interaction between malicious sensing devices and surface monitoring and countermeasure platforms to study the dynamic strategy adjustment of both parties under different mission objectives and constraints; through information confrontation and decision optimization, both parties continuously adjust their action strategies to achieve the optimal defense effect of the defense platform. In the search phase of malicious sensing devices, the monitoring and countermeasure platform forms a large number of false targets by releasing interference equipment such as diluting chaff and active deception, while trying to hide its own characteristics. In order to further interfere with the detection capability of malicious sensing devices, the monitoring and countermeasure platform can release centroid chaff to force malicious sensing drones to misidentify targets; when the platform detects that the malicious sensing device is tracking a target, it will destroy its detection accuracy by releasing centroid chaff, and further interfere with the target recognition capability of the device by using active interference patterns (such as active suppression and deception); in addition, the chaff cloud released by the monitoring and countermeasure platform can have a scattering cross-section larger than the ship itself, thereby inducing malicious sensing drones to change the target tracking trajectory; the effective application of this interference pattern can significantly inhibit the detection and identification capabilities of malicious sensing drones to the platform, thereby achieving effective protection of the safety and privacy of the surface operation platform.

[0056] The interference pattern decision system of malicious sensing devices includes the malicious sensing drone device set U = {u 1 ,u 2 ,...,u k}, k = 1, 2, ..., K, the set includes K unmanned aerial vehicle malicious sensing devices; the monitoring and countermeasure platform set P = {p 1 ,p 2 ,...,p m}, m = 1, 2, ..., M, the set includes M monitoring and countermeasure platforms; each monitoring and countermeasure platform can select a set of interference patterns J = {n 1 ,n 2 ,...,n m},n=1,2,...,N J , including active jammers (such as suppression and deception jammers) and a variety of passive jammers; each jammer can select a jammer pattern set to N J kind.

[0057] like Figure 5 As shown, in order to characterize the relationship between malicious perception devices, monitoring and countermeasure platforms, and interference patterns, the present invention constructs an interference pattern decision knowledge graph. Figure 5 It is only used to represent the knowledge graph, and the text in the graph has no actual impact on the technical solution of the present invention. By constructing a knowledge triple containing information such as malicious sensing devices, interference devices, and interference patterns <e h ,r,e t>, systematically describe the complex relationship between interference devices and malicious sensing devices; set { <e h ,r,e t >|e h ,e t ∈E; r∈R} represents the head entity, relationship and tail entity in the triple relationship; the entity set E includes various entities related to the interference spectrum, such as the monitoring and countermeasure platform number, the malicious perception device number, the interference style number including the interference equipment and chaff centroid interference, chaff dilution interference, infrared centroid interference, infrared dilution interference for internal active suppression and deception, and the interference parameter value including the interference distance value and the interference quantity value. The relationship set R describes the various relationships between entities, such as the monitoring and countermeasure platform, the interference parameters including the interference distance, the interference quantity, etc.; the above interference style decision knowledge graph modeling method generates multiple triples such as <monitoring and countermeasure platform_malicious interference drone_interference style, interference style, noise suppression interference>, <monitoring and countermeasure platform_malicious interference drone_interference style, interference quantity, interference quantity value>, <monitoring and countermeasure platform_malicious interference drone_interference style, malicious interference drone, malicious interference drone number>. Each knowledge triple <e h ,r,e t > represents a slave entity e h To e t The fact of the relationship r is that when the monitoring and countermeasure platform selects noise suppression interference on the malicious sensing device.

[0058] Step 2: Based on the framework of step 1, construct an interference style decision knowledge graph embedding representation model, initialize the embedding of entities and relationships, and map the entities and relationships to a high-dimensional vector space through a convolutional neural network to generate embedded features that can represent semantic associations.

[0059] Before using the triple information in the knowledge graph for calculation, we first need to abstract the entities and relations in the triple into a real-valued vector. The specific operation is to set each entity as a c-dimensional vector and set an independent semantic space for each relationship r. For any given triple <e h ,r,e t >, set the mapping matrix M r , the head entity e h and tail entity e t Mapping from the entity space to the relation space where the relation r is located, we get h r and t r ; Projection matrix M r The value of is randomly initialized and continuously updated during the model training process; the projection matrix M rThe dimension of is set to the product of the dimension of the entity embedding encoding column vector and the dimension of the relation embedding encoding column vector; by multiplying the head entity embedding encoding column vector h by the shadow matrix M r , the tail entity embedding encoding column vector t multiplied by M r , so that the head entity h r and tail entity t r Projecting from entity space to relational space, the above process can also be expressed as: r =M r h and t r =M r t; This mapping mechanism enables the same entity to have different vector representations in different relationships, thereby effectively capturing the complex semantics of entities in multi-dimensional relationships;

[0060] Next, we will pass the head entity vector h r Add the mapping result with the relationship vector r, and try to match the tail entity vector t r Align so that the triplet representation satisfies h r +r≈t r ; For positive sample triples <e h ,r,e t >∈T, set the embedding objective function d <e h ,r,e t >=||h r +rt r ||≈0; for negative sample triples <e′ h ,r′,e′ t >∈T′, set the embedding objective function d <e′ h ,r′,e′ t >=||h′ r +rt′ r ||>>0.; Set the score function f KG <e h ,r,e t >, used to calculate the head entity h r Through the relationship r to the end entity t r The distance is evaluated by calculating the square of the two norms of the triples; the above process can also be expressed as:

[0061] Set the loss function L KG , calculate the positive sample score function value d <e h ,r,e t > and the negative sample score function value d <e′ h ,r′,e′ t >The difference between the two; set the interval hyperparameter γ to adjust the gap between the positive and negative sample scores; set the function [x] +=max(0,x), which is used to calculate the maximum value between 0 and x, limiting the loss function value to a non-negative interval; the model is trained using the gradient descent method to optimize the loss function L KG , which aims to make the positive sample triplet obtain a lower score value and the negative sample triplet obtain a higher score value;

[0062] like Figure 6 As shown in the figure, facing the interference style decision knowledge graph, the triples in the interference style decision knowledge graph are embedded in the knowledge graph based on the knowledge graph. h ,r,e t > Initialize to a 3D vector v h ,v r ,v t ;v h ,v r ,v t , and form a matrix The i-th row H(i,:) in the matrix H corresponds to the embedding vector of the i-th triple. In order to capture the global relationship between the entity embedding vectors, the interference spectrum knowledge graph embedding model based on convolutional neural network embeds each triple network into three-segment network and feeds it to the convolution layer to achieve global information integration between different entities and their relationship dimensions, thereby effectively capturing and modeling the correlation and interaction between entities. Specifically, the convolution filter is applied on the rows of the matrix H. Then, through the dot product operation with the triplet embedding vector [v h ,v r ,v t ]·w, capturing the global relationship between embedded triplets. This convolution process can effectively integrate the attribute information of the head entity, relation, and tail entity in the triplet to generate a new feature vector Specifically, each element in the vector is mapped through a convolution operation and a nonlinear activation function to generate the final embedded representation. This process can be expressed as: i =σ(ω·H(i,:)+b); set Φ KG represents the set of filters in the convolutional layer of the knowledge graph for interference style decision, τ KG Represents the number of filters. Multiple filters are convolved and connected in series through the concat(·) function to form a dimension of A single vector retains the diverse information extracted by different filters; this feature vector and the weight parameter matrix Perform dot multiplication and return the triple score; this score is used to distinguish the triples with correct correspondence from the triples with incorrect correspondence, ensuring that the score of the positive sample triple is lower than that of the negative sample triple. The above process can also be expressed as: CNN-KG <e h ,r,et > = concat(σ KG ([v h ,v r ,v t ]*Φ KG ))·u KG ;

[0063] The loss function of the knowledge graph feature embedding based on the convolutional neural network for interference style decision is set as It is used to calculate the loss of positive and negative triplets, where the positive sample triplet sets the label l to +1, the negative sample triplet sets the label l to -1, and λ KG is the regularization coefficient; the positive and negative sample triplet labels are expressed as: By minimizing the training loss through adaptive moment estimation, the embedding vectors of the corresponding head entity, relation, and tail entity are obtained, where the embedding vector of the head entity is used as the feature vector.

[0064] Step 3: Disturb the attribute features of style decision knowledge KG Interference effectiveness evaluation sequence x with historical spatiotemporal network t , through the weight matrix w s Perform weighted summation and add the corresponding bias vector b s ; Set the activation function ReLu(·) to perform nonlinear transformation on the weighted input to generate the output of the current layer; recursively map layer by layer to establish the mapping relationship between input features and output values. The above process can also be expressed as: X t =ReLu(e KG x t w s +b s ).

[0065] Step 4: To describe the multi-level relationship network between the malicious sensing equipment onboard drones, monitoring and countermeasure platforms, and their interference patterns, a spatiotemporal graph network model G is constructed. KG =(V KG ,E KG ,A KG ); Set the node set V KG and edge set E KG ; Set A to be the spatiotemporal graph network G KG The adjacency matrix of ; V is set to represent the set of network nodes of the interference pattern decision model for malicious sensing devices, expressed as Set the number of node sets to Setting E represents the association relationship set of the same type of interference patterns, UAV-borne malicious perception devices, and monitoring and countermeasure platforms based on the semantic consistency of the interference patterns, including the UAV-borne malicious perception device U k Monitoring and Countermeasure Platform m, UAV-mounted malicious sensing equipment k With interference pattern G n and Monitoring and Countermeasure Platform P m With interference pattern G n The relationship between Constructing the adjacency matrix of a spatiotemporal graph network Used to describe whether there is a relationship between graph nodes; set A ij =1 indicates that the mth monitoring and countermeasure platform implements the nth interference pattern on the ith UAV-borne malicious sensing device; the above process can also be expressed as:

[0066]

[0067] Set the spatiotemporal graph network data as the graph model G KG The interference effectiveness evaluation value calculated after the mth monitoring and countermeasure platform implements the nth interference style on the i-th UAV-mounted malicious sensing device at time t; set in time step t, the spatiotemporal graph network model graph model G KG The characteristic is expressed as In order to predict the interference effectiveness evaluation value at the future time step t, the mapping function f is set KG The historical spatiotemporal network interference effectiveness evaluation sequence Future interference effectiveness evaluation prediction mapped to this spatiotemporal network According to the T observed by using experience or virtual-real game confrontation training KG The historical interference effectiveness evaluation value predicts the next T K ' G The interference effectiveness evaluation value of the step; Set the mapping function to train the parameter matrix W KG ; Set the length T of the historical space-time network sequence KG ; Set the length T of the target spatiotemporal network sequence to be predicted K ' G ; The above process can also be expressed as:

[0068]

[0069] Based on the topological structure of the graph model, the neighborhood aggregation strategy is used to update the features of the graph nodes and their neighbors to achieve effective integration of feature information; the adjacency matrix after adding the self-loop is set to set up for The degree matrix of l,KG is the weight matrix of layer I; set y l ′ is the output feature matrix of layer I, and y′ 0 =X t ; The spatiotemporal graph convolutional neural network architecture can be expressed as:

[0070] Step 4.4: Extract node attribute features in the interference style decision knowledge graph through the graph convolutional neural network, and pass it as input to the GRU to learn the time evolution law of the node label, that is, the interference effectiveness evaluation, so as to predict the interference effectiveness evaluation value; set the graph convolutional neural network to extract the node relationship information in the graph structure and model the spatial dependency of the node features; set the GRU to capture the dynamic changes of the node label over time; set the key components of the GRU including resetting the gate r t and update gate u t , where the reset gate combines the information in the memory with the current time step information, and the update gate is used to select memory or forget information; the activation function σ(·) is set to represent the gating signal; c is set t Indicates the instantaneous state of the current time step; in the memory update phase, when u t When u is used as a forget gate, t ⊙h t-1 Forget unimportant previous information to update memory; when u t As a memory gate, (1-u t )⊙c t Remember the current information; set the weight and parameter matrix W u , W r , W c and b u , b r , b c ; Set h t represents the output at time t; h t Input to the fully connected layer to generate the predicted interference effectiveness evaluation value The above process can also be expressed as:

[0071] u t =σ(W u gc([X t ,h t-1 ],A)+b u ),

[0072] r t =σ(W r gc([X t ,h t-1 ],A)+b r ),

[0073] c t =tanh(W c gc([X t ,(r t ⊙h t-1 )],A)+b c ),

[0074] h t =u t ⊙h t-1 +(1-u t )⊙c t .

[0075] set up and Represent the true value and predicted value of the j-th time sample on node m respectively; set y and They are and A collection of; set λ KG is the regularization coefficient; set L reg is the regularization loss term; calculate the true interference effectiveness evaluation value y of the known node t Interference effectiveness prediction value The mean square error between them sets the loss function of the graph convolutional neural network interference style decision model based on knowledge graph embedding, which can be expressed as:

[0076] The present invention focuses on the malicious perception drone scene and conducts research on the interference style decision of the surface monitoring and countermeasure platform. The decision-making process involves multiple interrelated and highly complex multidimensional problems, covering multidimensional features such as time, space, and spectrum, and the interaction between devices. Specifically, the game confrontation process includes malicious perception drone equipment, surface monitoring and countermeasure platforms, and various interference devices, and each device not only has specific spatial location information, but also is related to multiple attributes such as interference style and spectrum resources. Therefore, in order to accurately characterize the complex correlation between multidimensional features, it is necessary to characterize the characteristics of different correlations through the knowledge graph embedding method of interference style decision, and then realize the unified modeling of equipment, interference style and various attributes. The present invention uses the interference effectiveness evaluation time series data calculated by the surface monitoring and countermeasure platform for the release of different interference styles of malicious perception drone equipment as the model input, combines the historical interference confrontation situation data, and adopts the interference effectiveness prediction method based on knowledge graph and spatiotemporal graph convolutional neural network to plan the future interference game situation.

[0077] Embodiment 2

[0078] This embodiment proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the interference resource planning method based on the knowledge graph and the spatiotemporal graph convolutional neural network as described in the first embodiment.

[0079] Embodiment 3

[0080] This embodiment proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the interference resource planning method based on the knowledge graph and spatiotemporal graph convolutional neural network as described in Embodiment 1 is implemented.

[0081] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0082] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0083] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0085] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0086] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for interference resource planning based on knowledge graph and spatiotemporal graph convolutional neural network, characterized in that: The method comprises the following steps: Step 1: Aiming at the game confrontation scenario of malicious perception defense, a knowledge graph of interference pattern decision is constructed in the form of <head entity, relationship, tail entity> triples to systematically represent the multi-dimensional and complex relationship between the monitoring and countermeasure platform, malicious perception equipment and interference patterns; Step 2: Construct an interference style decision knowledge graph embedding representation model, initialize the embedding of entities and relations, and map the entities and relations to a high-dimensional vector space through a convolutional neural network to generate embedding features that represent semantic associations; Step 3: Construct the interference style decision knowledge fusion based on multi-layer perceptron, combine the attribute characteristics of interference style decision with the historical spatiotemporal network interference effectiveness evaluation sequence, and fuse multi-source heterogeneous data; Step 4: Construct a spatiotemporal graph convolutional neural network to model the embedded features, extract the spatial correlation characteristics between nodes through the graph convolution module, and combine the gated recurrent unit to capture the time evolution of node attributes and dynamically predict the interference effectiveness; Step 5: Combine the jamming effectiveness prediction results and the real-time situation information of the target device to infer the optimal jamming pattern.

2. The interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network according to claim 1 is characterized in that: Step 1 includes the following sub-steps: Step 1.1: Construct a game confrontation model for malicious perception defense, and set the malicious perception drone equipment set U = {u1,u2,…,u k } and the set of monitoring and countermeasure platforms P = {p1,p2,…,p m }, k = 1, 2, ..., K, m = 1, 2, ..., M, K is the number of malicious sensing devices onboard drones; M is the number of monitoring and countermeasure platforms; each monitoring and countermeasure platform can select a set of interference patterns J = {n1, n2, ..., n m }Including active jammers and various passive jammers, n=1,2,…,N J , N J The number of jamming pattern sets selectable for each jamming device; Step 1.2: For the game confrontation scenario of malicious sensing defense, construct a triple that describes the relationship between malicious sensing devices, interference devices, and interference patterns; Setting knowledge triples <e h ,r,e t > represents the head entity, relationship and tail entity in the triple relationship; the entity set E includes various entities related to the interference spectrum; the relationship set R is used to describe various relationships between entities; Step 1.3: Based on the interference pattern decision knowledge graph generated by the triples, multiple triples are formed including <monitoring and countermeasure platform_malicious interference drone_interference pattern, interference pattern, noise suppression interference>, <monitoring and countermeasure platform_malicious interference drone_interference pattern, interference quantity, interference quantity value>, <monitoring and countermeasure platform_malicious interference drone_interference pattern, malicious interference drone, malicious interference drone number>; set the positive sample triple set to T KG , which contains triples with correct correspondences; by removing the head entity e in the positive sample triple h , tail entity e t or relation r, and use other head entities e′ h , tail entity e′ t Or relation r′ is replaced to obtain the negative sample triple set T′ of the wrong corresponding association relationship KG , denoted as T′ KG ={ <e′ h ,r,e t >|e h ∈E}∪{ <e h ,r′,e t >|r∈R}∪{{ <e h ,r,e′ t >|e t ∈E}}.

3. The interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network according to claim 2 is characterized in that: The entity set E includes various entities related to the interference spectrum, specifically including the monitoring and countermeasure platform number, the malicious perception equipment number, and the interference style number; the interference style includes the interference equipment for internal active suppression and deception and the chaff centroid interference, chaff dilution interference, infrared centroid interference, and infrared dilution interference; the interference parameter value includes the interference distance value and the interference quantity value.

4. The interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network according to claim 1 is characterized in that: Step 2 includes the following sub-steps: Step 2.1: Set each entity as a c-dimensional vector and set up an independent semantic space for each relation r; for any given triple <e h ,r,e t >, set the projection matrix M r , the projection matrix M r The value of is randomly initialized and continuously updated during the model training process. The projection matrix M r The dimension of is set to the product of the dimension of the entity embedding encoding column vector and the dimension of the relation embedding encoding column vector; by multiplying the head entity embedding encoding column vector h by the shadow matrix M r , the tail entity embedding encoding column vector t multiplied by M r , so that the head entity h r and tail entity t r Projecting from entity space to relational space, we get h r and t r ,h r =M r h and t r =M r t; Step 2.2: The triple association relationship is obtained by converting the head entity vector h r Add the mapping result of the relationship vector r to the tail entity vector t r Align so that the triplet representation satisfies h r +r≈t r ; For positive sample triples <e h ,r,e t >∈T, set the interference style decision knowledge graph embedding representation model embedding objective function d <e h ,r,e t >=||h r +rt r ||≈0; for negative sample triples <e h ′,r′,e′ t >∈T′, set the embedding objective function d <e′ h ,r′,e′ t >=||h′ r +rt′ r ||>>0; Set the score function f KG <e h ,r,e t >, used to calculate the head entity h r Through the relationship r to the end entity t r The distance between and is used to evaluate the triples: Step 2.3: Set the loss function L KG , calculate the positive sample score function value d <e h ,r,e t > and the negative sample score function value d <e h ′,r′,e t >The difference between the two; set the interval hyperparameter γ to reduce the distance between the positive and negative sample scores; set the function [x] + =max(0,x), used to calculate the maximum value between 0 and x, limiting the loss function value to a non-negative interval; The gradient descent method is used to train the interference style decision knowledge graph embedding representation model to optimize the loss function L KG , so that the positive sample triplet gets a lower score value, while the negative sample triplet gets a higher score value: Step 2.4: Facing the interference style decision knowledge graph, embed the triples in the interference style decision knowledge graph through the knowledge graph-based embedding model <e h ,r,e t > Initialize to a 3D vector v h ,v r ,v t ; Assume Φ KG represents the set of filters in the convolutional layer of the knowledge graph for interference style decision, τ KG Represents the number of filters. Multiple filters perform convolution operations on three-dimensional vectors and are concatenated into a vector with dimension A single vector of Φ to retain the diverse information extracted by different filters; Φ KG τ KG Represents the vector dimension obtained by sequentially concatenating the features extracted by each filter; the single vector obtained by concatenation is concatenated with the weight parameter matrix Perform dot multiplication and return the triple score f CNN-KG <e h ,r,e t >, used to distinguish the triples of correct correspondence from the triples of incorrect correspondence, so that the score of the positive sample triple is lower than that of the negative sample triple: f CNN-KG <e h ,r,e t > = concat(σ KG ([v h ,v r ,v t ]*Φ KG ))·u KG , σ KG represents the convolution function; Step 2.5: Set the loss function of the knowledge graph feature embedding based on the convolutional neural network to It is used to calculate the loss of positive and negative triplets, where the positive sample triplet sets the label l to +1, the negative sample triplet sets the label l to -1, and λ KG is the regularization coefficient; the training loss is minimized by adaptive moment estimation to obtain the embedding vectors of the corresponding head entity, relation, and tail entity, where the embedding vector of the head entity is used as the feature vector.

5. The interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network according to claim 1 is characterized in that: Step 3 further comprises the following steps: The attribute features of the interference style decision knowledge e KG Interference effectiveness evaluation sequence x with historical spatiotemporal network t , through the weight matrix w s Perform weighted summation and add the corresponding bias vector b s ; Set the activation function ReLu(·) to perform nonlinear transformation on the weighted input to generate the output of the current layer; recursively map layer by layer to establish the mapping relationship between input features and output values: X t =ReLu(e KG x t w s +b s ).

6. The interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network according to claim 1 is characterized in that: Step 4 further includes the following steps: Step 4.1: Construct the spatiotemporal graph network model G KG =(V KG ,E KG ,A KG ), node set V KG represents the set of network nodes of the interference pattern decision model for malicious sensing devices, The number of node sets is Edge set E KG It represents the association relationship set of the same type of interference patterns, UAV-borne malicious perception devices, and monitoring and countermeasure platforms based on the semantic consistency of interference patterns, including the UAV-borne malicious perception device U k Monitoring and Countermeasure Platform P m , UAV-mounted malicious sensing equipment k With interference pattern G n and monitoring and countermeasure platform P m With interference pattern G n The relationship between Constructing the spatiotemporal graph network G KG The adjacency matrix of Used to describe whether there is a relationship between graph nodes; In the formula, A ij =1 means that the mth monitoring and countermeasure platform implements the nth interference pattern on the ith UAV-borne malicious sensing device; Step 4.2: Set the spatiotemporal graph network data as the graph model G KG The interference effectiveness evaluation value calculated after the mth monitoring and countermeasure platform implements the nth interference style on the i-th UAV-mounted malicious sensing device at time t; the feature vector dimension of the graph node is set to c; in the time step t, the spatiotemporal graph network model graph model G KG The characteristic is expressed as Set the mapping function f KG The historical spatiotemporal network interference effectiveness evaluation sequence Future interference effectiveness evaluation prediction mapped to spatiotemporal network According to the observed T KG The historical interference effectiveness evaluation value predicts the next T K ' G The interference effectiveness evaluation value of the step is: Where W KG is the trainable parameter matrix of the mapping function, T KG is the length of the historical spatiotemporal network sequence, T K ' G is the length of the target spatiotemporal network sequence to be predicted; Step 4.3: Based on the topological structure of the graph model, a neighborhood aggregation strategy is used to update the features of graph nodes and their neighbors; the spatiotemporal graph convolutional neural network architecture is expressed as: In the formula, represents the adjacency matrix after adding self-loops, The degree matrix, W l,KG is the weight matrix of the lth layer, y l ′ is the output feature matrix of the lth layer, and y′0=X t ; Step 4.4: Extract node attribute features in the interference style decision knowledge graph through the graph convolutional neural network and pass it as input to the GRU to learn the time evolution law of the node label and predict the interference effectiveness evaluation value; use the graph convolutional neural network to extract the node relationship information in the graph structure and model the spatial dependency of node features; use GRU to capture the dynamic changes of node labels over time and generate the predicted interference effectiveness evaluation value; u t =σW u gc([X t ,h t-1 ],A)+b u ) r t =σ(W r gc([X t ,h t-1 ],A)+b r ) c t =tanh(W c gc([X t ,(r t ⊙h t-1 )],A)+b c ) h t =u t ⊙h t-1 +(1-u t )⊙c t In the formula, r t represents the reset gate, u t represents the update gate, where the reset gate combines the information in the memory with the current time step information, and the update gate is used to select memory or forget information; the activation function σ(·) represents the gating signal; c t Indicates the instantaneous state of the current time step; in the memory update phase, when u t When u is used as a forget gate, t ⊙h t-1 Forget unimportant previous information to update memory; when u t As a memory gate, (1-u t )⊙c t Remember current information; W u , W r , W c represents the weight matrix, b u 、b r 、b c represents the parameter matrix; h t represents the output at time t; h t Input to the fully connected layer to generate the predicted interference effectiveness evaluation value Step 4.5: Calculate the actual interference effectiveness evaluation value y of the known node t Interference effectiveness prediction value The mean square error between them is used to construct the loss function of the graph convolutional neural network interference style decision model based on knowledge graph embedding: In the formula, and Represent the true value and predicted value of the j-th time sample on node m, respectively; y and They are and The set of KG is the regularization coefficient; L reg is the regularization loss term.

7. The interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network according to claim 6 is characterized in that: In step 4.5, for the optimized model, the root mean square error between the true value of the known graph node and its interference effectiveness prediction value is used as an evaluation indicator for transmission performance prediction.

8. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the interference resource planning method based on knowledge graph and spatiotemporal graph convolutional neural network as described in any one of claims 1 to 7.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the interference resource planning method based on the knowledge graph and spatiotemporal graph convolutional neural network as described in any one of claims 1 to 7 is implemented.

Images