A mode selection method and device for a d2d assisted power internet of things

Abstract

The application discloses a mode selection method and device for D2D assisted power Internet of Things, comprising: constructing a graph neural network model, modeling the interference relationship between D2D user pairs and cellular users and their channels in the power network into a graph structure in a graph construction module, and encoding the node features and edge features in the graph structure into feature vectors, using a graph convolution neural network to extract features of the feature vectors in a graph convolution module to obtain node hidden states including D2D user pair nodes and cellular user nodes, and inputting the node hidden states into a classifier in a classification decision module to calculate a channel selection variable to determine the selected channel mode; and deploying and applying the graph neural network model after training. The application optimizes the mode selection of D2D communication by using a graph neural network and maximizes the system throughput, further reduces the calculation complexity and improves the calculation speed, and can greatly improve the overall performance of the communication system.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, and specifically relates to a mode selection method and apparatus for D2D-assisted power Internet of Things. Background Technology

[0002] Device-to-device (D2D) communication technology is the fifth generation (5G). th One of the key technologies in Generation 5G mobile communication is Direct-to-Demand (D2D). D2D refers to a communication mode where end users communicate directly with each other, bypassing the base station without needing to be relayed by the base station. D2D communication can use both licensed and unlicensed frequency band resources, thus effectively solving the problem of scarce wireless resources. It can improve system bandwidth utilization and increase information transmission rate. User terminals performing D2D communication can transmit information with low power, effectively saving energy consumption. Furthermore, the transmission latency between users is very low, which contributes to the increase in transmission rate. In addition, direct communication between terminals is not controlled by the base station, thus reducing the base station load.

[0003] D2D communication offers many advantages, but its introduction into cellular networks also brings numerous challenges. When D2D users utilize unlicensed frequency bands, it becomes difficult to detect and control user terminals, leading to degraded network performance. Conversely, while licensed frequency bands can improve resource utilization, they introduce new interference into the existing network. This interference, resulting from the reuse of cellular user resources, negatively impacts network communication quality and degrades the network environment. To improve system performance, previous research has focused on two key D2D technologies: mode selection and power allocation. Mode selection refers to the operating mode chosen by the D2D user; different modes occupy different frequency bands, effectively addressing the issue of frequency band allocation. Power control adjusts the transmit power of the user's transmitter based on the communication task and user needs, thereby reducing interference from resource reuse. These two technologies can be jointly modeled within an optimization framework; however, previous methods often involve excessive computation, limiting their practical application value.

[0004] The D2D communication mode is divided into a multiplexing mode and a dedicated mode. In the multiplexing mode, the D2D user and the cellular user share a spectrum, which can effectively save the system spectrum resource, but the communication interference exists between the cellular user and the D2D user; in the dedicated mode, the spectrum resource used by the D2D user is orthogonal to the spectrum resource of the cellular user, and there is no interference, but a large amount of spectrum resource is occupied. Therefore, how to maximize the system throughput in the case of balancing the spectrum resource and the communication interference becomes the core problem of mode selection, and the rationality of mode selection determines the performance of the network. At present, the academia is increasingly extensive in the research on the D2D mode selection technology. Yang Wujun et al. proposed a mode selection method based on user distance, which can effectively improve the system performance by distinguishing the users according to the communication interval and setting the priority according to the user channel state; Ma Bei et al. proposed an incomplete channel state information (CSI) based method, which obtains the minimum closed solution of the power of the D2D user and the cellular communication link by taking the quality of service of different users as the constraint condition, and derives the mode selection scheme; Liu Jinxin et al. proposed a mode selection scheme based on the signal to inference plus noise ratio (SINR), which sets the downlink as the interference area and uses the SINR as the constraint to quickly select the channel scheme by solving the interference suppression. In the solving algorithm, this kind of research mainly focuses on the mathematical model based on convex optimization, and uses traditional algorithms such as implicit enumeration method and Hungarian algorithm to solve, which is not conducive to practical application, mainly because of two reasons: first, the algorithm optimization needs to be performed once every time the channel allocation is performed, which leads to very high complexity of each optimization. Second, with the increase of the network size, the complexity of optimization increases exponentially. Therefore, it is necessary to further optimize the mode selection of the D2D communication to achieve better performance than the traditional algorithm. SUMMARY

[0005] In view of the above, the purpose of the present application is to provide a mode selection method and device for D2D assisted power Internet of Things, which innovatively introduces a graph neural network (GNN) architecture to solve the channel allocation problem, optimizes the mode selection of D2D communication by using the graph neural network, maximizes the system throughput, further reduces the computational complexity and improves the calculation speed, greatly improves the overall performance of the communication system, has high practical application value in solving the problems of spectrum resource shortage and communication interference, and provides an efficient solution for future power Internet of Things communication system.

[0006] To achieve the above-mentioned purpose of the application, the technical solutions provided by the present application are as follows:

[0007] In a first aspect, the embodiment of the present application provides a mode selection method for a D2D assisted power Internet of Things, comprising the following steps:

[0008] Based on the D2D user pairs and the cellular users in the power network, channel allocation label values are calculated, and a training data set is constructed;

[0009] A graph neural network model including a graph construction module, a graph convolution module and a classification decision module is constructed, in the graph construction module, the interference relationship between the D2D user pairs and the cellular users and their channels in the power network is modeled as a graph structure, and the node features and edge features in the graph structure are encoded into feature vectors, in the graph convolution module, the feature vectors are feature extracted by using a graph convolution neural network to obtain node hidden states including D2D user pair nodes and cellular user nodes, and in the classification decision module, the node hidden states are input into a classifier to calculate a channel selection variable to determine a selected channel mode;

[0010] The graph neural network model is trained by using the training data set and a loss function;

[0011] The trained graph neural network model is deployed to a base station for channel mode selection and is issued to perform communication.

[0012] Preferably, based on the D2D user pairs and the cellular users in the power network, channel allocation label values are calculated, and a training data set is constructed, comprising:

[0013] Different numbers of D2D user pairs and cellular users are randomly generated, a branch and bound method is used to calculate a channel allocation solution as a correct label value, and all D2D user pairs and cellular users and their corresponding channel allocation labels are constructed as a training data set.

[0014] Preferably, in the graph construction module, the interference relationship between the D2D user pairs and the cellular users and their channels in the power network is modeled as a graph structure, comprising:

[0015] By using a node modeling method, the D2D user pairs and the cellular users in the power network are modeled as nodes of a graph, the interference relationship between the user channels is modeled as edges of the graph, and finally the constructed graph structure is fully connected, that is, there is interference between each node, the node features are the transmission power of the D2D user pairs and the cellular users and the channel gain, and the edge features are the channel state of the interference link and the interference size.

[0016] Preferably, the node features and the edge features in the graph structure are encoded into feature vectors, comprising:

[0017] By using a structure2vec graph embedding algorithm, the feature vectors of each node are calculated based on the adjacency matrix of the graph structure, the node features and the edge features, and the specific nonlinear iterative representation is:

[0018]

[0019] wherein, denotes the feature vector of node v after the tth iteration, initialized as a zero vector, denotes a nonlinear operation, X v denotes the feature of node v, denotes the hidden state of its in-edge feature after the last iteration, denotes the hidden state of its predecessor neighbor after the last iteration, u denotes the u-th node in the set of all nodes N(v) in the graph structure, e denotes the in-edge feature, {D2D, CU} denotes the set of D2D user pair nodes and cellular user nodes, and the structure2vec algorithm finally obtains the feature vector h v of nodes after iteration.

[0020] Preferably, the feature extraction of the feature vector in the graph convolution module by the graph convolutional neural network obtains the hidden state of the nodes including D2D user pair nodes and cellular user nodes, comprising:

[0021] Let matrix H = [h v ] v∈{D2D,CU} denote the feature matrix of all nodes after feature encoding, wherein h v denotes the feature vector of node v, and {D2D, CU} denotes the set of D2D user pair nodes and cellular user nodes.

[0022] The graph convolutional neural network updates the representation of the node by aggregating the features of the neighbor nodes, and the formula is represented as:

[0023]

[0024] wherein, denotes the adjacency matrix with self-loop, denotes the degree matrix of H (l) denotes the node feature matrix of the l-th layer of the graph convolutional neural network, H (0) denotes the input feature matrix, i.e. the result of the last step of feature encoding, W (l) denotes the weight matrix of the l-th layer of the graph convolutional neural network and is updated constantly during the training process, and denotes a nonlinear operation. With the training, the graph convolutional neural network gradually learns the hidden state of the nodes including D2D user pair nodes and cellular user nodes h' v .

[0025] Preferably, the input of the hidden state of the nodes into the classifier in the classification decision module calculates the channel selection variable to determine the selected channel mode, comprising:

[0026] The N-ary classifier is constructed by using a deep neural network to calculate the channel selection probability, N being the total number of uplink orthogonal channels available to the cellular network, and is expressed by the following formula:

[0027]

[0028] where p(v) represents the probability of node v selecting channel n, h' v represents the hidden state of the node, represents the weight vector of channel n, represents the weight vector of any channel n' including channel n, and the superscript T represents transposition;

[0029] After obtaining the probability of node v selecting all channels, the channel corresponding to the maximum probability is selected as the channel selection result of node v, and the channel selection results of all nodes are integrated to finally construct the channel selection variables a k,n and a m,n of the D2D user pair and the cellular user, respectively. k,n and a m,n reflect the channel mode selection results of each D2D user pair and the cellular user.

[0030] Preferably, the loss function adopts a cross-entropy loss function which is expressed by the following formula:

[0031]

[0032] where p(v) represents the predicted value of the probability of node v selecting channel n, q(v) represents the label value of the probability of node v selecting channel n, {D2D, CU} represents the node set of the D2D user pair and the cellular user, and N represents the total number of uplink orthogonal channels available to the cellular network.

[0033] Preferably, the deployment of the trained graph neural network model to the base station for channel mode selection and the issuance of the execution of communication include:

[0034] Terminal information collection: the base station sends pilots to each terminal and collects the channel state information of all D2D user pairs and cellular users;

[0035] Base station processing: the base station deploys the trained graph neural network model, combines the channel state and interference information of the D2D user pairs and the cellular users, and optimizes the mode selection of the D2D user pairs and the cellular users, which is either a dedicated mode or a multiplexing mode;

[0036] Decision issuance: the base station issues the optimized communication mode to each terminal;

[0037] Execution of communication: each D2D user pair and the cellular user performs communication according to the allocated mode.

[0038] In a second aspect, to achieve the above-mentioned object, the embodiment of the present application further provides a mode selection device for D2D assisted power Internet of Things, which is implemented by using the mode selection method for D2D assisted power Internet of Things, and comprises a data set construction module, a model construction module, a model training module and a model deployment module.

[0039] The data set construction module is configured to calculate channel allocation label values based on D2D user pairs and cellular users in a power network, and construct the channel allocation label values into a training data set.

[0040] The model construction module is configured to construct a graph neural network model comprising a graph construction module, a graph convolution module and a classification decision module, model the interference relationship between D2D user pairs and cellular users and their channels in the power network into a graph structure in the graph construction module, encode the node features and edge features in the graph structure into feature vectors, use a graph convolution neural network to extract features from the feature vectors to obtain node hidden states of D2D user pair nodes and cellular user nodes in the graph convolution module, and input the node hidden states into a classifier to calculate a channel selection variable to determine a selected channel mode in the classification decision module.

[0041] The model training module is configured to train the graph neural network model by using the training data set and a loss function.

[0042] The model deployment module is configured to deploy the trained graph neural network model to a base station for channel mode selection and communication execution.

[0043] In a third aspect, to achieve the above-mentioned object, the embodiment of the present application further provides an electronic device comprising a memory and one or more processors, wherein the memory is configured to store a computer program, and the processor is configured to implement the mode selection method for D2D assisted power Internet of Things when executing the computer program.

[0044] Compared with the prior art, the present application has at least the following beneficial effects:

[0045] (1) Spectrum utilization is improved: by mode selection, a multiplexing mode can be selected in a suitable scenario, thereby significantly improving the spectrum utilization of the system and improving the overall throughput of the communication network.

[0046] (2) Reduce system interference: the graph neural network algorithm is used to optimize the D2D communication mode, which can effectively reduce the interference between D2D communication and cellular users, and ensure the quality of service (QoS) of the communication system.

[0047] (3) The complexity of calculation is reduced: compared with traditional optimization methods (such as implicit enumeration method and Hungarian algorithm), the pattern selection of the graph neural network model can be performed with lower complexity, which is especially suitable for large-scale network scenarios.

[0048] (4) The system throughput is improved: by combining the optimization selection of the dedicated mode and the multiplexing mode, the overall communication rate of the system can be significantly improved, so that the system has better throughput performance in a complex communication environment.

[0049] (5) The effect is more significant when the system scale is expanded: due to the particularity of the graph neural network training, when the system scale is increased, only the number of graph neural network nodes is increased, the parameters of small-scale training can be referred to to avoid the increase of complexity and reduce the training cost. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0051] Figure 1 is a flowchart of a mode selection method for a D2D assisted power Internet of Things provided by an embodiment of the present application;

[0052] Figure 2 is a schematic diagram of a communication system deployment environment provided by an embodiment of the present application;

[0053] Figure 3 is a graph structure schematic diagram of a communication system provided by an embodiment of the present application;

[0054] Figure 4 is a graph neural network model architecture schematic diagram provided by an embodiment of the present application;

[0055] Figure 5 is a structure schematic diagram of a mode selection device for a D2D assisted power Internet of Things provided by an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the protection scope of the present application.

[0057] The inventive concept of the present application is: in view of the high computational complexity of mode selection of D2D communication in the prior art, the embodiment of the present application provides a mode selection method and device for D2D assisted power Internet of Things, by constructing the D2D user pair and the interference relationship between the cellular users in the communication system into a graph structure, using a graph neural network model based on the graph structure to optimize the mode selection of D2D communication and maximize the system throughput, so as to further reduce the computational complexity and improve the calculation speed, and greatly improve the overall performance of the communication system.

[0058] The communication system of the power Internet of Things involved in the present application is as shown in Figure 2 There are two types of users in the system: cellular users and D2D user pairs. D2D user pairs (D2D-T represents the transmitter, and D2D-R represents the receiver) can communicate directly, and there are two modes to choose from: dedicated mode and multiplexing mode. In the dedicated mode, the D2D user pair exclusively uses the channel resources and does not multiplex with the cellular users, so the D2D communication link has no interference from the cellular users, and the communication quality is high, but the spectrum utilization is low. In the multiplexing mode, the D2D link can multiplex the uplink channel resources of the cellular network, i.e. the link of one of the cellular users, at this time, the D2D link will share the channel with the cellular users, causing bidirectional interference. The interference is divided into two parts: the interference of the D2D user pair to the uplink of the cellular user and the interference of the uplink of the cellular user to the D2D link. The size of this interference is related to the respective channel power gain, transmission power and other parameters, although it has a certain negative impact on the communication quality, but it can improve the utilization efficiency of the spectrum. The overall goal of the system is to maximize the overall system throughput by selecting appropriate D2D mode (dedicated or multiplexing) while ensuring the communication quality of the users.

[0059] Figure 1 is a flowchart of a mode selection method for D2D assisted power Internet of Things provided by the embodiment of the present application. As shown in Figure 1 The embodiment provides a mode selection method for D2D assisted power Internet of Things, comprising the following steps:

[0060] S1, based on the D2D user pair and the cellular user in the power network, the channel allocation label value is calculated and constructed as a training data set.

[0061] Construct a channel optimization problem: assume that the system has K D2D user pairs, defined as: Define the number of available uplink orthogonal channels of the cellular network as N, and the number of cellular users as M. Without loss of generality, assume that M≤N≤M+K, i.e. some D2D links must select the multiplexing mode to complete data transmission, and all cellular users can select different orthogonal channels for transmission without interference with each other. Define the indicator variable am,n denotes the channel selection of a cellular user, a k,n denotes the channel selection of a D2D user pair, both of which are {0, 1}, and the value of 1 represents selecting the corresponding orthogonal channel, and each cellular user or D2D user pair can only select one channel for transmission.

[0062] For the mthcellular user, its uplink transmission rate R m is represented as:

[0063]

[0064] where p m denotes the transmission power of the cellular user m, h m,B denotes the channel power gain between the cellular user m and the base station, N0denotes the white noise power, m' denotes any cellular user other than the cellular user m, p k denotes the transmission power of the D2D user pair, h k,B denotes the channel power gain of the D2D pair to the base station, and B denotes the bandwidth of the subchannel.

[0065] For the kthD2D user pair, its transmission rate R k is represented as:

[0066]

[0067] where p k denotes the transmission power of the kthD2D user pair, h k denotes the channel power gain of the kthD2D user pair, N0denotes the white noise power, k' denotes any D2D user pair other than the kthD2D user pair, p m and h m,k denote the transmission power and channel power gain of the corresponding cellular user m when the channel is multiplexed, h k′,k denotes the channel gain between the two D2D user pairs, and B denotes the bandwidth of the subchannel.

[0068] The goal of the system is to adjust the channel selection of the D2D user pairs and the cellular users so that the overall communication rate is maximized, i.e., represented as the following channel optimization problem:

[0069]

[0070] where R k denotes the transmission rate of the kthD2D user pair, R m denotes the uplink transmission rate of the mthcellular user, a k,n and a m,nrespectively, and a

[0071] The constraint conditions of the channel optimization problem are:

[0072]

[0073] SINR k SINR m ≥r th

[0074] SINR k SINR m respectively, and a th r k,n represents a given threshold value. The first two constraint conditions indicate that each cellular user or D2D user pair can only select one channel for transmission, and the third constraint condition indicates that the signal-to-interference-and-noise ratio (SINR) of the D2D user and the cellular user cannot be lower than a given value, which is to ensure the quality of service (QoS) of the communication system. In terms of optimization method, since the optimization problem is a 0-1 integer optimization problem, the branch and bound method, the cut plane method, the dynamic programming method, the Lagrangian relaxation method, and various heuristic algorithms can be used to solve the problem.

[0075] In order to train the graph neural network model, a data set needs to be constructed for the channel allocation problem under the coexistence of D2D user pairs and cellular users. Considering efficiency and computational complexity, the branch and bound method is used to solve the problem. The idea of solving the problem using the branch and bound method is to branch on the channel association variables a k,n and a m,n , such as the first D2D user reusing the channel of the first cellular user and the second D2D user reusing the channel of the second cellular user. For each branch, it is checked whether it satisfies the constraint condition. If it satisfies the constraint condition, an upper bound of the communication rate of the system is obtained when the remaining allocation is idealized. If the upper bound is lower than the optimal solution, the branch is pruned. Otherwise, the optimal solution is updated. Then backtracking is performed until all branches are traversed, and the optimal branch (optimal channel allocation) and the maximum rate of the system are obtained.

[0076] The branch and bound method has the advantage of not requiring offline calculation in advance and being able to directly obtain the optimal channel allocation scheme according to the model. However, the optimization problem needs to be solved once at each channel allocation, which leads to an increase in the complexity of the algorithm. When the size of the communication system increases, the complexity of the algorithm increases exponentially, so an algorithm with lower complexity needs to be used.

[0077] Randomly generate different numbers of D2D user pairs and cellular users, calculate the channel allocation solution as the correct label value, construct the required training data set, and then the base station is trained in an offline manner according to the graph neural network model framework.

[0078] S2, a graph neural network model including a graph construction module, a graph convolution module and a classification decision module is constructed, in the graph construction module, the interference relationship between the D2D user pairs and the cellular users in the power network and the channels is modeled as a graph structure, and the node features and edge features in the graph structure are encoded into feature vectors, in the graph convolution module, the feature vectors are extracted by using the graph convolution neural network to obtain the node hidden state including the D2D user pair nodes and the cellular user nodes, and in the classification decision module, the node hidden state is input into the classifier to calculate the channel selection variable to determine the selected channel mode.

[0079] The application of graph neural network (GNN) in communication system is a hot research direction in academia in recent years, which provides a new solution for mode selection optimization in large-scale communication system. The GNN model can quickly and efficiently perform mode selection and resource allocation by capturing the complex relationship between nodes (users and base stations) in the communication network, especially when dealing with multi-user interference and channel conditions. Compared with traditional optimization algorithms, the GNN model has the advantages of low running complexity and fast calculation speed when facing large-scale networks, which can greatly improve the overall performance of the system. Due to the natural compatibility of the topological properties of the communication system and the data structure of the graph, the previous research has achieved better results than traditional algorithms in link planning, privacy protection and resource allocation. The present application optimizes the mode selection of D2D communication based on graph neural network to achieve better performance than traditional algorithms.

[0080] The graph neural network model architecture and processing process constructed by the embodiment of the application are as shown in Figure 4 The main modules include the following three modules.

[0081] S2.1, the graph construction module.

[0082] The method based on the graph neural network first needs to topologically model the communication system as a graph structure. There are mainly two methods for graph modeling of the communication system: a node modeling method and a link modeling method. The node modeling method models all base stations and users of the system as communication nodes, models the transmission power and channel selection of the users as node features, models the existing communication links as edges, and models the corresponding edge features as channel gain and interference, etc. This modeling method is very intuitive, that is, the topological graph of the communication system is directly displayed in the form of a graph network structure; the link modeling method models the communication links and interference links of the system as nodes and edges, respectively, so that the graph modeled in this way is a directed graph, and the direction of the edge represents different interference links. This modeling method performs very well in a pure D2D communication system. Since the D2D link and the cellular user coexist in the power system involved in the present application, the present application selects an improved node modeling method: the D2D user pair and the cellular user are modeled as nodes of the graph, and the whole graph is fully connected, that is, there are interference links between each node, which is consistent with the modeling method of the rate calculation. In the graph structure modeled in the embodiment of the present application, the node features are the transmission power and channel gain of the D2D user pair or the cellular user, and the edge features are the channel state and the interference size of the interference link. Figure 3 A graph structure schematic diagram is shown, in which the positions of the base stations are only used as the reference for the relative positions of the nodes when modeling, and in the subsequent feature extraction process, the information of the relative positions will be extracted into the embedded features of the edges in the form of the gains of the communication links and the fading values of the interference links. The node features of the cellular node x CU include the gain of the communication link with the base station and the transmission power allocated by the base station; and the node features of the D2D node x D2D include the transmission power and the channel gain of the D2D link.

[0083] The first step of the graph neural network training is feature encoding, which encodes the features of the nodes and edges into feature vectors through an encoding algorithm for further calculation. In the selection of the encoding architecture, the present application adopts the structure2vec graph embedding algorithm to calculate the p-dimensional feature vector of each node based on the adjacency matrix of the graph structure (the adjacency matrix refers to the matrix composed of the connection relationship between the nodes), the node features and the edge features. The specific nonlinear iterative expression is:

[0084]

[0085] wherein the subscript v represents any node, which can be a D2D user pair node or a cellular user node; represents the feature vector of the node v after the tth iteration, which is initialized as a zero vector; Γ(·) represents a nonlinear operation, and in the embodiment, a rectified linear unit (ReLU) is selected; X vrepresents the feature of node v, if it is a D2D user pair node, it represents the D2D input power and the channel gain, if it is a cellular user, it is the base station power allocation and the channel gain of the cellular user; represents the hidden state of the node after the last iteration of the input edge feature, represents the hidden state of the node after the last iteration of the input edge feature, represents the hidden state of the node after the last iteration of the input edge feature, v .

[0086] S2.2, graph convolution module.

[0087] The node feature after completing the feature encoding needs to be input into a specific GNN architecture for feature extraction. In the selection of the GNN architecture, the embodiment of the application adopts a classical and simple graph convolutional neural network (GCN) architecture, which can efficiently solve the channel allocation problem and has a relatively low computational complexity. The GCN takes the input features of the cellular user nodes and the D2D user pair nodes and the input features of the edges as inputs, uses the features of the nodes themselves and the interference information of the neighbor nodes through the propagation mechanism of the graph neural network, extracts the high-dimensional feature representation of the nodes layer by layer, and finally outputs the embedding feature of each node, also known as the hidden state. The hidden state is the aggregation of the node feature, the edge feature and the graph feature, and the purpose of extracting the hidden state is to ensure that the feature of each node not only contains its own transmission power and channel gain information, but also can obtain the interference link information of its neighbor nodes through the message passing mechanism, and enrich its context information. The node hidden state output by the feature extraction will be used for further channel selection.

[0088] Specifically, let the matrix H = [h v ] v∈{D2D,CU} , represents the feature matrix of all nodes after feature encoding, where h v represents the feature vector of node v, and {D2D, CU} represents the node set of D2D user pairs and cellular users. The graph convolutional neural network uses the features of the neighbor nodes to update the representation of the node, and the formula is represented as:

[0089]

[0090] wherein, represents the adjacency matrix with self-loop, represents the degree matrix of H (l) represents the node feature matrix of the lth layer of the graph convolutional neural network, H(0) W represents the input feature matrix, which is the result of the feature encoding in the previous step. (l) This represents the weight matrix of the l-th layer of the graph convolutional neural network, which is continuously updated during training. σ represents a non-linear operation. As training progresses, the graph convolutional neural network gradually learns the hidden states h' of nodes, including D2D user-to-user nodes and cellular user nodes, in a semi-supervised learning manner. v .

[0091] S2.3, Classification Decision Module.

[0092] This invention proposes a solution method based on a graph neural network model, building upon traditional optimization algorithms. The hidden states of the outputs in the GCN architecture require semantic transformation to obtain the respective channel allocation results. In this invention, a classic deep neural network (DNN) is used to construct an N-gram classifier to output the results, i.e., the channel selection variable 'a'. k,n and a m,n The value of .

[0093] In this embodiment, based on the hidden states of nodes extracted by the graph neural network model, an N-gram classifier is constructed using a deep neural network to calculate the channel selection probability. Specifically, the SOFTMAX classifier is used, and the formula is expressed as follows:

[0094]

[0095] Where p(v) represents the probability that node v chooses channel n, and h' v This indicates the hidden state of a node. This represents the weight vector of channel n. Let T represent the weight vector of any channel n' including channel n, and the superscript T indicates transpose.

[0096] After calculating the probabilities of node v selecting all channels, the channel with the highest probability is selected as the channel selection result for node v. The channel selection results of all nodes are then combined to construct the final channel selection variable 'a' for the D2D user pair. k,n Channel selection variable a for cellular users m,n , through a k,n and a m,n This reflects the channel mode selection results for each D2D user and cellular user.

[0097] S3 trains the graph neural network model using the training dataset and loss function.

[0098] In this embodiment, the graph neural network model is trained using a training dataset and a loss function, wherein the loss function is the cross-entropy loss function. The formula is expressed as:

[0099]

[0100] wherein p(v) represents the probability prediction value of node v selecting channel n, q(v) represents the probability label value of node v selecting channel n, {D2D, CU} represents the node set of D2D user pairs and cellular users, and N represents the total number of uplink orthogonal channels available to the cellular network.

[0101] S4, deploying the trained graph neural network model to the base station for channel mode selection and issuing execution of communication.

[0102] During the system execution, information needs to be exchanged between the D2D terminal and the base station. Specifically, the terminal needs to send its communication mode selection requirement and channel condition information, and the base station will make a global resource allocation decision according to the state feedback of all users. The specific process is as follows.

[0103] S4.1, terminal information collection: the base station sends pilots to each terminal and collects channel state information (CSI) of all users (including cellular users and D2D user pairs).

[0104] S4.2, base station processing: the base station deploys the trained graph neural network model, combines the channel state and interference information of D2D user pairs and cellular users, and optimizes the mode selection of D2D user pairs and cellular users, which is either a dedicated mode or a multiplexing mode.

[0105] S4.3, decision issuance: the base station issues the optimized communication mode to each terminal.

[0106] S4.4, execution of communication: each D2D user pair and cellular user performs communication according to the allocated mode.

[0107] When the framework runs, the base station first trains the graph neural network model in an offline state, obtains the current user number information in the cell after sending the pilot in an online state, then fine-tunes the trained graph neural network model, and sends the channel selection result to each user in the cell according to the model output result. The advantage of the graph neural network model is that the model running complexity is low, and since the graph neural network model has migratability, the parameters of a small-scale network can be partially migrated to a large-scale network when the network scale is relatively large, which greatly reduces the time required for training, and the advantage is very significant compared with traditional optimization models when the network scale is relatively large.

[0108] In summary, the mode selection method for the D2D assisted power Internet of Things provided by the embodiment of the application optimizes the mode selection of D2D communication by using a graph neural network and maximizes the system throughput, can improve the spectrum utilization, reduce the system interference, reduce the computational complexity, improve the system throughput, and the effect is more significant when the system scale is expanded.

[0109] Based on the same inventive concept, as Figure 5 The embodiment of the application also provides a mode selection device 500 for the D2D assisted power Internet of Things, which comprises a data set construction module 510, a model construction module 520, a model training module 530 and a model deployment module 540.

[0110] The data set construction module 510 is used to calculate the channel allocation label value based on the D2D user pairs and the cellular users in the power network, and construct a training data set.

[0111] The model construction module 520 is used to construct a graph neural network model comprising a graph construction module, a graph convolution module and a classification decision module, the interference relationship between the D2D user pairs and the cellular users and their channels in the power network is modeled as a graph structure in the graph construction module, and the node features and edge features in the graph structure are encoded as feature vectors, the feature vectors are feature extracted by using a graph convolution neural network in the graph convolution module to obtain node hidden states comprising D2D user pair nodes and cellular user nodes, and the node hidden states are input into a classifier in the classification decision module to calculate a channel selection variable to determine the selected channel mode.

[0112] The model training module 530 is used to train the graph neural network model by using the training data set and a loss function.

[0113] The model deployment module 540 deploys the trained graph neural network model to a base station for channel mode selection and downlink communication execution.

[0114] Based on the same inventive concept, the embodiment of the application also provides an electronic device comprising a memory and one or more processors, the memory is used to store a computer program, and the processor is used to implement the mode selection method for the D2D assisted power Internet of Things when the computer program is executed.

[0115] It should be noted that the mode selection device for the D2D assisted power Internet of Things and the electronic device provided by the above-mentioned embodiment all belong to the same inventive concept as the mode selection method for the D2D assisted power Internet of Things, and the specific implementation process is detailed in the mode selection method for the D2D assisted power Internet of Things, which will not be repeated here.

[0116] The above detailed description of the specific embodiments of the present application has described the technical solutions and beneficial effects of the present application, and it should be understood that the above description is only the most preferred embodiment of the present application and is not intended to limit the present application. Any modifications, supplements and equivalent replacements made within the principle range of the present application shall be included in the protection range of the present application.

Claims

1. A mode selection method for D2D-assisted power internet of things, characterized in that, The method comprises the following steps: Based on the D2D user pair and the cellular user in the power network, the channel allocation label value is calculated, and the training data set is constructed; Construct the channel optimization problem: define the indicator variable a m,n denotes the channel selection of a cellular user, a k,n denotes the channel selection of a D2D user pair, each cellular user or D2D user pair can only select one channel for transmission; for the mthcellular user, its uplink transmission rate R m is denoted as: where p m denotes the transmission power of cellular user m, h m,B denotes the channel power gain between cellular user m and the base station, N0denotes the white noise power, m ′ denotes any cellular user other than cellular user m, p k denotes the transmission power of the kth D2D user pair, h k,B denotes the channel power gain of D2D to the base station, B denotes the bandwidth of the subchannel; for the kth D2D user pair, its transmission rate R k is given by: where h k represents the channel power gain of the kth D2D user pair, k' represents any D2D user pair other than the kth D2D user pair, h m,k represents the channel power gain between the cellular user m and the kth D2D user pair, h k′,k represents the channel gain between two D2D user pairs; the system calculates the channel allocation solution as a label value with the overall transmission rate and maximum as the target; A graph neural network model including a graph construction module, a graph convolution module and a classification decision module is constructed; in the graph construction module, the interference relationship between the D2D user pair and the cellular user and the channel thereof in the power network is modeled as a graph structure, and the node features and edge features in the graph structure are encoded into feature vectors, wherein the graph structure adopts a node modeling method, the D2D user pair and the cellular user in the power network are modeled as nodes of the graph, the interference relationship between the user channels is modeled as edges of the graph, and finally the constructed graph structure is fully connected, that is, there is interference between each node, the node features are the transmission power and channel gain of the D2D user pair or the cellular user, and the edge features are the channel state and interference size of the interference link; in the graph convolution module, the feature vectors are feature extracted by using a graph convolution neural network to obtain node hidden states including D2D user pair nodes and cellular user nodes; in the classification decision module, the node hidden states are input into a classifier to calculate a channel selection variable to determine a selected channel mode, including: a deep neural network is used to construct an N-ary classifier to calculate a channel selection probability, N is the total number of uplink orthogonal channels available for the cellular network, and the formula is as follows: where p n (v) represents the probability of node v selecting channel n, h v represents the hidden state of node, represents the weight vector of channel n, represents the weight vector of any channel n' including channel n, the superscript T represents transposition, after the probability of node v selecting all channels is obtained, the channel corresponding to the maximum probability is selected as the channel selection result of node v, and the channel selection results of all nodes are comprehensively constructed as the channel selection variable a k,n of the D2D user pair and the channel selection variable a m,n of the cellular user, through a k,n and a m,n , the channel mode selection results of each D2D user pair and the cellular user are reflected. The graph neural network model is trained by using the training data set and a loss function; The trained graph neural network model is deployed to the base station for channel mode selection and communication execution. 2.The mode selection method for D2D assisted power internet of things according to claim 1, wherein, The channel allocation label value is calculated based on the D2D user pair and the cellular user in the power network, and the training data set is constructed, including: Different numbers of D2D user pairs and cellular users are randomly generated, a channel allocation solution is calculated by using a branch and bound method as a correct label value, and all D2D user pairs and cellular users and corresponding channel allocation labels are constructed as the training data set. 3.The mode selection method for D2D assisted power internet of things according to claim 1, wherein, The node features and edge features in the graph structure are encoded into feature vectors, including: A structure2vec graph embedding algorithm is used to calculate the feature vectors of each node based on the adjacency matrix, node features and edge features of the graph structure, and a specific nonlinear iterative representation is as follows: wherein, denotes the feature vector of node v after the t-th iteration, initialized as a zero vector, denotes a nonlinear operation, X v denotes the feature of node v, denotes the hidden state of its in-edge features after the last iteration, denotes the hidden state of its predecessor neighbors after the last iteration, u denotes the u-th node in the set of all nodes N(v) in the graph structure, e denotes the in-edge feature, {D2D, CU} denotes the set of nodes of D2D user pairs and cellular users, and the structure2vec algorithm finally obtains the feature vector h v of the nodes after iterations. 4.The mode selection method for D2D assisted power internet of things according to claim 1, wherein, In the graph convolution module, the feature vectors are feature extracted by using a graph convolution neural network to obtain node hidden states including D2D user pair nodes and cellular user nodes, including: The matrix H = [h v ] v∈{D2D,CU} The matrix H = [h v The matrix H = [h {D2D, CU} represents the node set of D2D user pairs and cellular users. The graph convolution neural network uses the features of the aggregated neighbor nodes to update the representation of the node, and the formula is as follows: wherein, denotes an adjacency matrix with self-loops, denotes the degree matrix of H (l) denotes the node feature matrix of the l-th layer of the graph convolutional neural network, H (0) denotes the input feature matrix, i.e. the result of the previous feature encoding, W (l) denotes the weight matrix of the l-th layer of the graph convolutional neural network and is constantly updated during the training process, σ represents a non-linear operation, and the graph convolutional neural network gradually learns the node hidden states h ’ v including D2D user pairs nodes and cellular user nodes as the training proceeds. 5.The mode selection method for D2D assisted power internet of things according to claim 1, wherein, The loss function adopts a cross-entropy loss function The formula is expressed as: where p n (v) represents the probability prediction value of the node v selecting the channel n, q n (v) represents the probability label value of the node v selecting the channel n, {D2D, CU} represents the node set of the D2D user pair and the cellular user, and N represents the total number of uplink orthogonal channels available for the cellular network. 6.The mode selection method for D2D assisted power internet of things according to claim 1, wherein, The trained graph neural network model is deployed to the base station for channel mode selection and communication execution, including: Terminal information collection: the base station sends a pilot to each terminal, and collects the channel state information of all D2D user pairs and cellular users; Base station processing: the base station deploys the trained graph neural network model, combines the channel state and interference information of the D2D user pair and the cellular user, and optimizes the mode selection of the D2D user pair and the cellular user, and the mode selection is a dedicated mode or a multiplexing mode; Decision delivery: the base station delivers the optimized communication mode to each terminal; Communication execution: each D2D user pair and cellular user performs communication according to the allocated mode.

7. A mode selection apparatus for D2D assisted power internet of things, which realizes the mode selection method for D2D assisted power internet of things as claimed in any one of claims 1-6, characterized in that, The method comprises the following steps: The data set construction module, the model construction module, the model training module and the model deployment module; The data set construction module is configured to calculate channel allocation label values based on D2D user pairs and cellular users in the power network, and to construct a training data set; The model construction module is configured to construct a graph neural network model including a graph construction module, a graph convolution module and a classification decision module, model the interference relationship between D2D user pairs, cellular users and their channels in the power network as a graph structure in the graph construction module, encode the node features and edge features in the graph structure into feature vectors, use a graph convolution neural network to extract features from the feature vectors to obtain node hidden states including D2D user pair nodes and cellular user nodes in the graph convolution module, input the node hidden states into a classifier to calculate a channel selection variable to determine a selected channel mode in the classification decision module; The model training module is configured to train the graph neural network model using the training data set and a loss function; The model deployment module is configured to deploy the trained graph neural network model to a base station for channel mode selection and to issue an execution communication.

8. An electronic device comprising a memory and one or more processors, the memory to store a computer program, characterized in that, The processor is configured to implement the mode selection method for a D2D assisted power Internet of Things according to any one of claims 1-6 when executing the computer program.

Images