An interference relationship identification method based on a knowledge graph

Abstract

This invention discloses a knowledge graph-based interference relationship identification method, comprising: real-time acquisition of at least one first parameter of a first device to be identified, and at least one second parameter of a second device to be identified; constructing a first triplet for the first device and a second triplet for the second device based on the first parameter of the first device and the second parameter of the second device, respectively; inputting the first triplet and the second triplet together into a trained interference identification network, wherein the trained interference identification network preprocesses each input first triplet and second triplet to obtain preprocessed first triplets and preprocessed second triplets; and outputting an identification result indicating whether there is interference between the first device and the second device based on the preprocessed first triplet and preprocessed second triplet. This invention can efficiently analyze interference relationships within a large-scale system with low computational cost.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wireless communication, and particularly relates to a knowledge graph-based interference relationship identification method. BACKGROUND

[0002] In recent years, wireless communication systems are developing towards the direction of integration of air, space, land and sea. In the current situation of tight spectrum availability, different types of wireless communication protocols are constantly exploring the possibility of fusion coexistence in pursuit of large bandwidth and low delay. It can be predicted that the spectrum in the future will be even tighter, and wireless systems will encounter not only interference from legal communication devices but also interference from malicious interference devices that aim to disrupt communication systems. Therefore, it is particularly important to find such interference relationships.

[0003] In the past, in order to cope with complex and variable interference situations, researchers have made many efforts in frequency hopping spread spectrum technology, waveform anti-jamming, and anti-jamming decision-making. Frequency hopping spread spectrum technology focuses more on resisting interference. Frequency hopping mainly changes the center frequency of communication quickly and jumps within a larger bandwidth to avoid interference. Spread spectrum technology mainly converts information into spread spectrum codes to communicate at a fixed frequency at the cost of speed or occupying more bandwidth. The receiving party can analyze the transmitted information through the sequence correlation to achieve anti-jamming.

[0004] The above-mentioned technologies have now become part of many communication systems, but the technology itself has been relatively mature, and new anti-jamming ideas need to be added to keep up with the development of today's interference technology. Therefore, waveform-based anti-jamming technology has emerged, including adaptive theory and processing in space-time-frequency code domain. Through the above processing, communication systems have achieved good results in dealing with regular interference such as single tone, multi-tone, narrowband, and wideband.

[0005] With the advent of cognitive radio theory, interference technology has also developed, developing into smart jamming, deception jamming, and perception-based modulation jamming. This poses new challenges to existing communication systems, which need to add an "eye" for spectrum perception and a "brain" for decision-making to deal with interference. Therefore, modern anti-jamming communication introduces neural networks with strong reasoning ability, and many scholars have studied it and achieved many results.

[0006] Returning to the background of interference relationships, it is proposed in the context of current communication systems continuously enhancing their perception and decision-making capabilities to solve the problem of how to discover the interference relationship between devices to facilitate subsequent minimization of the malicious impact of interference.

[0007] Traditional interference recognition technology mainly focuses on judging whether the signal is interference, what type of interference, and the parameters of the interference. The energy detection algorithm is mainly used to detect noise signals. Energy detection has become the most commonly used method in this field because it does not require any prior feature information about user signals. After research, researchers have summarized several signal indicators that can effectively distinguish different types of interference: single frequency energy concentration C, average spectral flatness coefficient Fc, fractional Fourier energy concentration, spectral moment kurtosis coefficient, spectral moment skewness coefficient, spectral R parameter, etc. Determining the type also facilitates the extraction of various interference parameters based on time-frequency characteristics to facilitate accurate interference avoidance by the communication system.

[0008] With the evolution of interference technology, interference signals are no longer random in several fixed interference patterns, or inefficient interference strategies such as full-band interference. New interference opportunities perceive the surrounding spectrum environment and make decisions on interference patterns based on swarm intelligence algorithms and neural networks. The ultimate goal is to maximize interference efficiency. Therefore, traditional interference recognition based on time-frequency characteristics and high-order statistics cannot meet the challenge of intelligent jammers.

[0009] In the face of the above challenges, many scholars have proposed interference recognition based on decision trees, neural networks, and other technologies. The above research has achieved good results. However, most of the research is aimed at identifying a jammer. When there are a large number of legal devices and malicious jammers in the environment, existing research is not applicable. In addition, research based on decision trees and neural networks will be limited by the technology itself, making it difficult to fully utilize its capabilities in complex real-world environments. That is, existing technologies do not have an efficient method for analyzing interference relationships in a large-scale system, and traditional interference analysis focuses on analyzing the type and index of interference, which also leads to a large amount of calculation. SUMMARY

[0010] To solve the above problems in the prior art, the present application provides a knowledge graph-based interference relationship recognition method.

[0011] The technical problem to be solved by the present application is solved by the following technical solutions:

[0012] The present application provides a knowledge graph-based interference relationship recognition method, comprising:

[0013] Real-time acquisition of at least one first parameter of a first device to be identified and at least one second parameter of a second device to be identified;

[0014] constructing a first triad of the first device and a second triad of the second device according to the first parameter of the first device and the second parameter of the second device respectively;

[0015] inputting the first triad and the second triad into a trained interference recognition network, the trained interference recognition network respectively pre-processing each first triad and second triad inputted, and obtaining a pre-processed first triad and a pre-processed second triad correspondingly; wherein the trained interference recognition network is obtained by training an initialized interference recognition network according to a training set containing positive triads and negative triads and a loss function

[0016] the trained interference recognition network outputs an identification result of whether the first device and the second device have interference according to the pre-processed first triad and the pre-processed second triad.

[0017] Compared with the prior art, the beneficial effects of the present application are:

[0018] The interference relationship recognition method based on a knowledge graph provided by the present application focuses on the interference relationship between devices in the system, and does not care about specific interference types, so when processing the interference relationship between devices, it is more suitable for parallel computing, and when calculating a single interference relationship, the amount of calculation is greatly reduced compared with traditional algorithms for calculating interference types, so that more reliable interference relationship recognition can be achieved faster and better, and when the interference relationship recognition method based on a knowledge graph provided by the present application is deployed in a system, all devices in the system can be efficiently monitored, so that interference events such as malicious interference and interference between legal devices can be discovered in time.

[0019] The present application will be further described in detail below with reference to the accompanying drawings and the present application. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a flowchart of an interference relationship recognition method based on a knowledge graph provided by an embodiment of the present application;

[0021] Figure 2 is a schematic diagram of an example of mapping entities and relationships into embedding vectors provided by an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of an example of complexifying embedding vectors of relationships provided by an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of an interference recognition flow of a system provided by an embodiment of the present application;

[0024] Figure 5 is a performance comparison result of different methods provided by an embodiment of the present application on the same interference data set. DETAILED DESCRIPTION

[0025] The present application will be further described in detail below in conjunction with specific embodiments, but the embodiments of the present application are not limited thereto.

[0026] Figure 1 is a flowchart of a knowledge graph-based interference relationship identification method provided by an embodiment of the present application, and the method comprises the following steps:

[0027] S101, acquiring at least one first parameter of a first device to be identified and at least one second parameter of a second device to be identified in real time.

[0028] For example, the number of devices and the parameters of each device can be acquired in real time from the system to obtain a formatted text, which records the known devices in the system and the unknown devices accessing the system, and the parameters of each device. By analyzing these texts, the devices to be identified and the parameters of each device to be identified can be obtained. The parameters of each device may, for example, include position, motion condition, communication period, communication frequency band, communication state, etc. For example, when there is a known device in the system and two devices are newly accessed to the system in sequence, the known device can be taken as a device to be identified, and the device first accessing the system among the two newly accessed devices can be taken as another device to be identified, so as to identify whether there is interference between the two devices through subsequent steps. Then, the known device can be taken as a device to be identified again, and the device last accessing the system among the two newly accessed devices can be taken as another device to be identified, so as to identify whether there is interference between the two devices through subsequent steps. Then, the two newly accessed devices can be taken as two devices to be identified respectively, so as to identify whether there is interference between the two devices through subsequent steps.

[0029] Here, the first device to be identified is one of the two devices to be identified, and the second device to be identified is the other of the two devices to be identified. The first parameter is the parameter of one of the two devices to be identified, and the second parameter is the parameter of the other of the two devices to be identified.

[0030] S102, constructing a first triple of the first device and a second triple of the second device according to the first parameter of the first device and the second parameter of the second device.

[0031] Specifically, the above S102 is implemented through the following steps:

[0032] S1021, for each first parameter, taking the identity of the first device as one of the head entity and the tail entity, taking the parameter name of the first parameter as the relationship, and taking the parameter value of the first parameter as the other of the tail entity and the head entity, to obtain a first triple of the first device.

[0033] Here, the identity of the device can be the name of the device, or an identity generated for the device.

[0034] S1022, for each second parameter, taking the identity of the second device as one of the head entity and the tail entity, taking the parameter name of the second parameter as the relationship, and taking the parameter value of the second parameter as the other of the tail entity and the head entity, to obtain a second triple of the second device.

[0035] Here, for each device, the device has several parameters, and several triples of the device can be constructed accordingly. For example, when the identity of a device is "device 1", the device has two parameters, and the names of the two parameters are "position" and "motion state" respectively, the parameter value of "position" is "(x, y, z)", and the parameter value of "motion state" is "static", then two triples of the device can be constructed, i.e. <device 1, position, (x, y, z)> and <device 1, motion state, static>.

[0036] S103, input the first triple and the second triple into the trained interference recognition network, and the trained interference recognition network respectively preprocesses each first triple and second triple input, to obtain a preprocessed first triple and a preprocessed second triple; wherein the trained interference recognition network is obtained by training an initialized interference recognition network according to a training set containing positive triples and negative triples, and a loss function.

[0037] Here, the initialized interference recognition network is constructed according to the RotatE algorithm, and the RotatE algorithm is used to find the interference relationship between devices. The initialized interference recognition network has a first initial mapping vector and a second initial mapping vector, for example, the mapping dimension of the first initial mapping vector is 2000, and the mapping dimension of the second initial mapping vector is 1000. The trained interference recognition network has a trained first mapping vector and a trained second mapping vector.

[0038] Here, a positive triple refers to a correct triple constructed, and a negative triple refers to an incorrect triple formed by replacing the tail entity in a positive triple with an incorrect tail entity.

[0039] Specifically, the above S103 can be realized by the following steps:

[0040] S1031, the trained interference identification network maps the head entity and the tail entity in the triple to the first dimension by using the trained first mapping vector, to obtain the real part embedding vector and the imaginary part embedding vector of the head entity, and the real part embedding vector and the imaginary part embedding vector of the tail entity.

[0041] S1032, the trained interference identification network maps the relationship in the triple to the second dimension by using the trained second mapping vector, to obtain the embedding vector of the relationship; wherein the second dimension is less than the first dimension.

[0042] For example, as shown in the figure, Figure 2 for a triple, assuming that the mapping dimension of the first mapping vector is 2000, and the mapping dimension of the second mapping vector is 1000, through the first mapping vector, the head entity and the tail entity in the triple can be mapped into embedding vectors with a dimension of 2000, to obtain the real part embedding vector and the imaginary part embedding vector of the head entity, and the real part embedding vector and the imaginary part embedding vector of the tail entity, and through the second mapping vector, the relationship in the triple can be mapped into an embedding vector with a dimension of 1000, and then the initialized embedding vector after mapping can also be initialized by uniform distribution.

[0043] S104, the trained interference identification network outputs the recognition result of whether the first device and the second device have interference according to the preprocessed first triple and the preprocessed second triple.

[0044] Here, each preprocessed triple includes: the real part embedding vector and the imaginary part embedding vector of the head entity, the real part embedding vector and the imaginary part embedding vector of the tail entity, and the embedding vector of the relationship; based on this, the above S104 is realized by the following steps:

[0045] S1041, for each preprocessed triple, the embedding vector of the relationship in the preprocessed triple is rotated element by element to obtain the real part embedding vector and the imaginary part embedding vector of the relationship in the preprocessed triple.

[0046] Here, through step S1041, the embedding vector of the relationship in a preprocessed triple can be complex. For example, Figure 3 the schematic diagram for complexing the embedding vector of the relationship, wherein the dimension of the embedding vector of the relationship is 1000, denotes a multiplier, Sin(*) is a sine function, and Cos(*) is a cosine function.

[0047] S1042, according to all real part embedding vectors and all imaginary part embedding vectors in the preprocessed triple, the distance score of the preprocessed triple is calculated.

[0048] Specifically, S1042 is implemented through the following steps:

[0049] S1. Calculate the real part score of the preprocessed triple based on the real part embedding vectors of the head entity, the tail entity, and the relation in the preprocessed triple.

[0050] Specifically, the real part score of a preprocessed triplet is calculated as follows: The first product is obtained by multiplying the real part embedding vector of the head entity and the real part embedding vector of the tail entity in the preprocessed triplet; the second product is obtained by multiplying the imaginary part embedding vector of the head entity and the imaginary part embedding vector of the tail entity in the preprocessed triplet; the real part score is obtained by subtracting the first product, the second product, and the real part embedding vector of the relation in the preprocessed triplet. For example, the real part score s of a preprocessed triplet... real The expression is: s real =h real *r real -h imag *r imag -t real .

[0051] S2. Calculate the imaginary part score of the preprocessed triple based on the imaginary part embedding vectors of the head entity, the tail entity, and the relation in the preprocessed triple.

[0052] Specifically, the imaginary part score of a preprocessed triplet is calculated as follows: The first product is obtained by multiplying the real embedding vector of the head entity and the real embedding vector of the tail entity in the preprocessed triplet; the second product is obtained by multiplying the imaginary embedding vector of the head entity and the imaginary embedding vector of the tail entity in the preprocessed triplet; the imaginary part score is obtained by subtracting the first product, the second product, and the imaginary embedding vector of the relation in the preprocessed triplet. For example, the imaginary part score s of a preprocessed triplet... imag The expression is: s imag =h real *r real +h imag *r imag -t imag .

[0053] S3. Apply the L1 norm to the real part score and the imaginary part score respectively, and then sum them. Based on the sum, obtain the distance score of the preprocessed triple.

[0054] For example, the expression for the distance score s of a preprocessed triple is: s 1 =|s real |+|s imap |, where s 1 Let |s be the sum of the preprocessed triples. real | is the L1 norm of the real part score of the preprocessed triplet, |s imap | is the L1 norm of the imaginary part score of the preprocessed triplet, assuming s 1 If the dimension is 1000, then s 1 i s 1 The value of the i-th dimension, where γ is a preset value.

[0055] S1043. Sum the distance scores of the preprocessed first triplet and the preprocessed second triplet, and then average them to obtain the mean score.

[0056] For example, when there are two preprocessed first triplets and two preprocessed second triplets, the distance scores of these five triplets can be summed and averaged to obtain a mean score.

[0057] S1044. Based on the average score and a preset threshold, output the identification result of whether there is interference between the first device and the second device.

[0058] Specifically, when the average score is greater than a preset threshold, the system outputs a recognition result indicating no interference between the first and second devices; when the average score is less than or equal to the preset threshold, the system outputs a recognition result indicating interference between the first and second devices.

[0059] Here, the preset threshold can be set arbitrarily according to actual needs, and there are no restrictions on it.

[0060] In some embodiments, after S104 described above, the method further includes:

[0061] S105. When the identification result indicates that there is interference between the first device and the second device, the identifier of the first device is used as one of the head entity or tail entity, the identifier of the second device is used as the other of the head entity or tail entity, and "interference" is used as a relation to obtain a third triplet, and the third triplet is updated to the knowledge graph.

[0062] For example, when the first device is identified as "Device 1" and the second device is identified as "Device 2", then a third triplet is constructed as: <Device 1, Interference, Device 2>.

[0063] Here, the knowledge graph mentioned above is the knowledge graph composed of the first triplet and the second triplet.

[0064] Here, after executing S105 above, the parameters of the device to be identified can be obtained again to continue the interference identification between devices. That is, it can return to S101 above to continue the next interference identification between devices.

[0065] For example, Figure 4 This is a schematic diagram of the interference identification process of a system. For example... Figure 5 As shown, firstly, by acquiring scene data, we can obtain perception data A, perception data B, and perception data C. By performing entity mining on perception data A, perception data B, and perception data C, we can obtain the devices in the system. By performing attribute mining on the devices (i.e., attribute detection / parameter detection), we can obtain the location, frequency band, power, and other parameters of each device. Based on these device parameters, we construct triples, which constitute a knowledge graph. By using these triples for knowledge reasoning, we can achieve the identification of interference states between devices.

[0066] In some embodiments, prior to S103 above, the following steps may be included:

[0067] S11. Obtain a training set containing multiple positive triples and multiple negative triples.

[0068] S12. During the c-th training, select at least one positive triplet from the training set as the positive triplet for the c-th training, select at least one negative triplet from the training set as the negative triplet for the c-th training, and obtain the interference recognition network obtained during the (c-1)-th training; c is an integer greater than or equal to 1. When c is 1, the interference recognition network obtained during the (c-1)-th training is the initialized interference recognition network.

[0069] Here, the number of positive triples used in each training iteration can be the same as or different from the number of negative triples. For example, in each training iteration, p positive triples and p negative triples can be used, or 1 positive triplet and q negative triples can be used, and so on, where p and q are both positive integers greater than 1. This embodiment of the invention does not limit this. The positive and negative triples used in each training iteration are randomly selected from the training set.

[0070] S13. Input the positive triplet from the cth training session and the negative triplet from the cth training session into the interference recognition network obtained from the (c-1)th training session to obtain the sum of the distance scores of the positive triplet from the cth training session and the sum of the distance scores of the negative triplet from the cth training session.

[0071] The principle of calculating the distance score for each triplet has been explained in detail above and will not be repeated here.

[0072] S14. Determine the loss value for the c-th training iteration based on the sum of the loss function and the distance score.

[0073] For example, the expression for the loss value of the c-th training iteration is:

[0074]

[0075] Among them, L c Let be the loss value for the c-th training iteration, γ and k are preset values, σ is the sigmoid function, m is the number of positive triples in the positive triples of the c-th training iteration, n is the number of negative triples in the negative triples of the c-th training iteration, and dis... r (h j ,t j ) represents the distance score for the j-th positive triple, and dis is the distance score for the j-th positive triple. r (h′ i ,t′ i ) represents the distance score of the i-th negative triplet, where m and n are both positive integers greater than 0.

[0076] S15. Based on the loss value of the c-th training, adjust the first and second mapping vectors of the interference recognition network obtained from the (c-1)-th training to obtain the interference recognition network obtained from the c-th training with the first and second mapping vectors adjusted in the c-th training. Iterate in this way until a trained interference recognition network with the trained first and second mapping vectors is obtained.

[0077] Here, training can be stopped when the preset maximum number of training iterations or preset training accuracy is reached, and the interference recognition network obtained from the last training iteration can be used as the trained interference recognition network.

[0078] Here, a test set can also be set up to test the currently trained interference recognition network (model) after a period of training, so as to know the current training level of the interference recognition network.

[0079] The interference relationship identification method based on knowledge graphs provided by this invention focuses on the interference relationships between devices within a system, without concern for specific interference types. Therefore, it is more suitable for parallel computing when processing interference relationships between devices. Furthermore, when calculating a single interference relationship, the computational load is significantly reduced compared to traditional algorithms for calculating interference types, enabling faster and more reliable interference relationship identification. Moreover, when the interference relationship identification method based on knowledge graphs provided by this invention is deployed in a system, all devices within the system can be efficiently monitored, thereby enabling timely detection of interference events such as malicious interference and interference between legitimate devices.

[0080] To further illustrate the technical effectiveness of the knowledge graph-based interference relationship identification method provided by this invention, some performance comparison test results are presented below. Tables 1 and 2 show the training computer configuration and training parameters, respectively. Figure 5 This is a performance comparison of different methods on the same interference dataset.

[0081] Table 1

[0082] Operating system Win10 22H2 CPU Intel i7-6950X GPU NVIDIA 3090 Memory 64G torch version 1.13.1 cuda version 12.2

[0083] Table 2

[0084]

[0085]

[0086] pass Figure 5 It can be observed that due to the RoteE algorithm's support for three relation patterns, it can achieve more reliable and faster identification of interfering relations. Compared to the TransE algorithm, RoteE achieves better performance with shorter training time. Conversely, compared to the ComplEx and DistMult algorithms, these algorithms, being older, have poorer ability to identify interfering relations, resulting in significantly inferior training results compared to RoteE. This verifies the superiority of the knowledge graph-based interfering relation identification method of this invention.

[0087] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0088] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0089] In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. While different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce a good effect.

[0090] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for identifying interference relationships based on knowledge graphs, characterized in that, include: In real time, at least one first parameter of the first device to be identified and at least one second parameter of the second device to be identified are acquired. Based on the first parameter of the first device and the second parameter of the second device, construct the first triplet of the first device and the second triplet of the second device respectively; The first triplet and the second triplet are input together into the trained interference recognition network. The trained interference recognition network preprocesses each input first triplet and second triplet to obtain preprocessed first triplets and preprocessed second triplets. The trained interference recognition network is obtained by training an initialized interference recognition network based on a training set containing positive and negative triplets and a loss function. The trained interference identification network outputs an identification result as to whether there is interference between the first device and the second device, based on the preprocessed first triplet and the preprocessed second triplet. Each preprocessed triple includes: the real and imaginary embedding vectors of the head entity, the real and imaginary embedding vectors of the tail entity, and the embedding vector of the relation; the trained interference recognition network preprocesses each input triple to obtain the preprocessed triples using the following method: The trained interference recognition network uses a trained first mapping vector to map the head entity and tail entity in the triple to the first dimension, respectively, to obtain the real part embedding vector and imaginary part embedding vector of the head entity, and the real part embedding vector and imaginary part embedding vector of the tail entity. The trained interference recognition network uses a trained second mapping vector to map the relations in the triples to a second dimension, obtaining the embedding vector of the relations; wherein the second dimension is smaller than the first dimension; The trained interference detection network, based on the preprocessed first triplet and the preprocessed second triplet, outputs a result indicating whether there is interference between the first device and the second device, including: For each preprocessed triplet, the embedding vector of the relation in the preprocessed triplet is rotated element by element to obtain the real part embedding vector and the imaginary part embedding vector of the relation in the preprocessed triplet. Calculate the distance score of the preprocessed triples based on all real part embedding vectors and all imaginary part embedding vectors in the preprocessed triples; The average score is obtained by summing the distance scores of the preprocessed first triplet and the preprocessed second triplet. Based on the average score and a preset threshold, the identification result indicating whether there is interference between the first device and the second device is output.

2. The knowledge graph-based interference relationship identification method according to claim 1, characterized in that, The step of constructing a first triplet for the first device and a second triplet for the second device based on the first parameter of the first device and the second parameter of the second device, respectively, includes: For each first parameter, the identifier of the first device is used as one of the head entity and the tail entity, the parameter name of the first parameter is used as the relation, and the parameter value of the first parameter is used as the other of the tail entity and the head entity, thus obtaining a first triplet of the first device. For each second parameter, the identifier of the second device is used as one of the head entity and the tail entity, the parameter name of the second parameter is used as the relation, and the parameter value of the second parameter is used as the other of the tail entity and the head entity, thus obtaining a second triplet of the second device.

3. The knowledge graph-based interference relationship identification method according to claim 1, characterized in that, The step of calculating the distance score of the preprocessed triples based on all real part embedding vectors and all imaginary part embedding vectors in the preprocessed triples includes: The real part score of the preprocessed triple is calculated based on the real part embedding vector of the head entity, the real part embedding vector of the tail entity, and the real part embedding vector of the relation in the preprocessed triple. The imaginary part score of the preprocessed triple is calculated based on the imaginary part embedding vector of the head entity, the imaginary part embedding vector of the tail entity, and the imaginary part embedding vector of the relation in the preprocessed triple. The real part score and the imaginary part score are respectively subjected to L1 norm and then summed. The distance score of the preprocessed triple is obtained based on the sum.

4. The knowledge graph-based interference relationship identification method according to claim 3, characterized in that, The real part score and the imaginary part score of the preprocessed triplet are calculated as follows: The first product is obtained by multiplying the real part embedding vector of the head entity in the preprocessed triplet with the real part embedding vector of the tail entity in the preprocessed triplet. The second product is obtained by multiplying the imaginary embedding vector of the head entity in the preprocessed triplet with the imaginary embedding vector of the tail entity in the preprocessed triplet. The real part score is obtained by subtracting the real part embedding vectors of the relations in the first product, the second product, and the preprocessed triplet. The imaginary part score is obtained by subtracting the imaginary part embedding vectors of the relations in the first product, the second product, and the preprocessed triplet.

5. The knowledge graph-based interference relationship identification method according to claim 1, characterized in that, The step of outputting the identification result indicating whether there is interference between the first device and the second device based on the average score and a preset threshold includes: When the average score is greater than the preset threshold, the recognition result without interference between the first device and the second device is output. When the average score is less than or equal to the preset threshold, an identification result indicating that there is interference between the first device and the second device is output.

6. The knowledge graph-based interference relationship identification method according to claim 1, characterized in that, Before inputting the first triplet and the second triplet together into the trained interference detection network, and before the trained interference detection network preprocesses each input first triplet and second triplet to obtain the preprocessed first triplet and preprocessed second triplet, the method further includes: Obtain the training set containing multiple positive triples and multiple negative triples; During the c-th training iteration, at least one positive triplet is selected from the training set as the positive triplet for the c-th training iteration, and at least one negative triplet is selected from the training set as the negative triplet for the c-th training iteration. The interference recognition network obtained during the (c-1)-th training iteration is then obtained. c is an integer greater than or equal to 1. When c is 1, the interference recognition network obtained during the (c-1)-th training iteration is the initialized interference recognition network. The positive triples from the c-th training and the negative triples from the c-th training are input together into the interference recognition network obtained from the (c-1)-th training to obtain the sum of the distance scores of the positive triples from the c-th training and the sum of the distance scores of the negative triples from the c-th training. The loss value for the c-th training iteration is determined based on the sum of the loss function and the distance score. Based on the loss value of the c-th training, the first and second mapping vectors of the interference recognition network obtained from the (c-1)-th training are adjusted to obtain the interference recognition network obtained from the c-th training with the first and second mapping vectors adjusted in the c-th training. This process is repeated until the trained interference recognition network with the trained first and second mapping vectors is obtained.

7. The knowledge graph-based interference relationship identification method according to claim 6, characterized in that, The expression for the loss value of the c-th training iteration is: ； in, Let c be the loss value of the training iteration. and All are preset values. For the sigmoid function, Let be the number of positive triples in the c-th training iteration. Let be the number of negative triples in the negative triplet set of the c-th training iteration. For the first Distance score for each positive triplet For the first Distance score for each negative triplet and All are positive integers greater than 0.

8. The knowledge graph-based interference relationship identification method according to claim 1, characterized in that, After the trained interference detection network outputs a result indicating whether there is interference between the first device and the second device based on the preprocessed first triplet and the preprocessed second triplet, the method further includes: When the identification result indicates that there is interference between the first device and the second device, the identifier of the first device is used as one of the head entity or the tail entity, the identifier of the second device is used as the other of the head entity or the tail entity, and "interference" is used as a relation to obtain a third triplet, and the third triplet is updated to the knowledge graph. Reacquire the parameters of the device to be identified and continue the interference identification between devices.

Images