Multi-cluster air computing method and system based on deep expansion graph neural network

Abstract

The invention discloses a multi-cluster air computing method and system based on a deep expansion graph neural network, and belongs to the technical field of air computing in a wireless network, and the method comprises the steps: firstly, defining an aggregation center node, a wireless device node and an edge for connecting the two nodes in a heterogeneous graph neural network; secondly, a multi-cluster control calculation optimization problem needing to be solved is decomposed into a neural network solving module of an unfolded structure; secondly, designing a feature extraction and feature embedding method in a neural network solving module; secondly, designing a heterogeneous graph neural network encoder to aggregate the features, and mapping the aggregated features into variables needing to be optimized through a designed multi-layer perceptron decoder; and finally, designing a parameter sharing strategy of the whole network and a learning method of the neural network, and training the neural network to obtain a universal model. The scheme provided by the invention is superior to an existing optimization scheme in the aspect of calculation rate, the solving speed is greatly improved, and the method has very strong generalization ability.

Description

Background Art

[0002] Due to its unique approach to data processing, air computing (AirComp) has shown great potential in multiple scenarios, such as perception data fusion, control information aggregation, and federated learning, covering a variety of cyber-physical system applications. Therefore, it is crucial to extend the existing single-task air computing to multi-cluster air computing in order to support concurrent computing instances of different services. From a networking perspective, this can be modeled as a multi-cluster air computing architecture, where each cluster is associated with an independent aggregation center. Under this architecture, different types of tasks, such as temperature measurement, fire monitoring, and neural network learning, can be performed simultaneously. However, the coexistence of multiple clusters brings interference issues, which becomes a major challenge because it affects the data aggregation within each cluster and further hinders the scalability to more diverse data. Therefore, in addition to the transmission and reception coordination issues in traditional air computing scenarios, effective interference management is also required to achieve efficient multi-cluster air computing.

[0003] In addition, the principle of air computing requires joint optimization at the sending and receiving ends to achieve effective computing. Therefore, efficient and accurate determination of the sending and receiving strategies becomes the basic support. However, dealing with this problem from an optimization perspective usually requires multiple rounds of iterations to reach a certain stable point strategy, which may not be the optimal solution and is time-consuming. At the same time, machine learning technology achieves fast execution through well-trained neural networks, but pure data-driven methods may lack sufficient adaptability in changing environments. In this context, the unfolding learning technology that combines knowledge-driven and data-driven design has become an attractive solution. Guided by theoretical derivation, the unfolding technology produces a neural network architecture inspired by problem-oriented iteration. Therefore, unfolding learning can efficiently interpret the problem structure through fast and executable neural networks, improving adaptability and scalability.

[0004] At the same time, for multi-cluster aerial computing systems, the policy decisions of one cluster will affect the policies of other clusters. From a learning perspective, graph neural networks (GNNs) are particularly suitable for solving such problems because they are naturally suitable for graph-structured data. Under the framework of graph neural networks, the sending and receiving behaviors in multi-cluster aerial computing can be embedded as features of graph elements, and the mutual influence between different clusters can also be embedded in the graph structure. In addition, the permutation equivalence property in graph neural networks is easy to generalize to networks of any size. Therefore, using graph neural networks in the unfolding learning architecture provides an effective tool for solving multi-cluster aerial computing problems, enabling well-trained neural networks to provide effective computing strategies and adapt to dynamic network environments. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a multi-cluster aerial computing method and system based on a deep unfolding graph neural network in response to the deficiencies in the above-mentioned prior art, which is used to solve the technical problem that traditional optimization methods are difficult to adapt to the flexible and changeable wireless environment and have low solution efficiency in multi-cluster aerial computing scenarios.

[0006] The present invention adopts the following technical solutions: A multi-cluster aerial computing method based on a deep unfolded graph neural network comprises the following steps: S1. Define the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network; S2, based on the aggregation center node, wireless device node and the edge connecting the two nodes defined in step S1, decomposing the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; S3, designing a feature extraction method in the neural network solution module obtained in step S2, and obtaining key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module; S4, designing a feature embedding method for a neural network, embedding the key features obtained in step S3, and mapping low-dimensional features to high-dimensional features; S5. Design a graph neural network encoder based on the key features embedded in step S4, and define methods for message generation and message aggregation in the graph neural network; S6. Design a multi-layer perceptron as a decoder to decode the output of the graph neural network obtained in step S5 into variables that need to be optimized; S7. Design a graph neural network parameter sharing strategy and a graph neural network learning method, train the graph neural network, and obtain a model with good generalization ability. Pass the channel information of the network in the new scenario as input to the model, output the transmission factor of each wireless device and the receiving beam of each aggregation center, and realize the improvement of multi-cluster air computing rate.

[0007] Preferably, in step S1, the air computing service coverage area includes N clusters, and the N clusters perform different air computing tasks. The set of all clusters is defined as ={1, 2, ..., K}; the center of each cluster is equipped with an aggregation center; For a given cluster k, it contains wireless device, represents the nth wireless device in the cluster, ={1, 2, ..., } represents the set of all wireless devices.

[0008] Wireless Devices The maximum transmission power is expressed as ; The kth aggregation center is equipped with Antenna, with Indicates wireless device The channel to the kth aggregation center; When l is equal to k, indicates a communication channel; otherwise, it indicates an interference channel; Wireless Devices The transfer factor is , the receiving beamforming at the aggregation center is ; Define all reflector devices as Node1={ }, , represents the first type of node in the first graph; all receiving end aggregation centers are defined as Node2={k}, , represents the second type of node in the graph; The channel between the first type of node and the second type of node is defined as the edge connecting the two types of nodes Edge={ }, .

[0009] Preferably, in step S2, the air computing problem is expanded into a plurality of cascaded neural network computing modules, and for module j, the current channel information { } and the transmission factor output by the previous module , output optimized transfer factor ; The solution module is encapsulated as an optimizer for the transmission factor, which includes an optimal receiving beam submodule, an optimal transmission factor phase submodule and a transmission factor modulus optimization submodule. The optimal receiving beam submodule adopts a cascade structure with the optimal transmission factor phase submodule and the transmission factor modulus optimization submodule respectively, and the optimal transmission factor phase submodule and the transmission factor modulus optimization submodule adopt a parallel structure, which are finally combined to form the optimized complex transmission factor.

[0010] Preferably, the cascaded multiple neural network computing modules are specifically:

[0011] in, For all channel information of the entire network, are the learnable parameters of the neural network, is the receiving beam of aggregation center k, For the receive beamforming optimization submodule, is the input transmission factor of the jth solution module, is the nth wireless device in the lth cluster, is the aggregation center set, is the phase of the transmission factor output by the jth solution module, is the transmission factor phase optimization submodule, For equipment The channel to the aggregation center k, is the nth device of the kth cluster, is the set of wireless devices in the kth cluster, is the output transmission factor modulus of the j-th solution module, is the transmission factor phase submodule, is the output transmission factor of the jth solution module, is the set of wireless devices in the lth cluster.

[0012] Preferably, in step S3, the feature extraction of wireless device nodes in the heterogeneous graph neural network adopts the following scheme:

[0013] in, Wireless device input for the jth solution module The items in brackets represent the characteristics of wireless devices. The old transmission factor, the equivalent channel of the corresponding aggregation center, the interference channel to other aggregation centers, the corresponding receiving power of the aggregation center, the interference power to other aggregation centers; The feature extraction of aggregation center nodes in heterogeneous graph neural networks adopts a similar scheme:

[0014] in, is the feature of the aggregation center k input to the j-th solution module. The items in the brackets represent the mean square error of the aggregation center k, the equivalent interference channel and from the effective device received by the aggregation center, the interference channel and from the interfering device received by the aggregation center, the effective power and received by the aggregation center, and the interference power and received by the aggregation center.

[0015] Preferably, in step S4, the device node The embedding operation of the aggregation center node k is expressed as:

[0016]

[0017] in, and Represents embedded wireless devices and the features of the aggregation center k, the embedding layer is a single-layer neural network, and the tanh function is used as the activation function. Embedding neural network for device features, Embedding neural network parameters for device features, Embedding neural network parameters for aggregation center features, is the embedding feature output by the embedding layer of the aggregation center k, For equipment The embedded features output by the feature embedding layer.

[0018] Preferably, in step S5, the encoder is composed of I layers of information transfer, each layer of the encoder includes, information transfer includes message generation, aggregation and combination, forming a cascaded linear layer, for message aggregation, each node aggregates the features of other types of nodes by weighting and aggregating, that is, the device node aggregates the features of the central node, and the node combines the previous features and the obtained aggregated message through a linear layer with a layer normalization function. This process is expressed as ; Introducing coefficients in the information transfer mechanism proposed in the heterogeneous graph neural network encoder , define information transmission so that information can be transmitted between nodes of different types, and information can not be transmitted between nodes of the same type.

[0019] Preferably, in step S6, the decoder is represented as:

[0020] in, Denotes a decoding operation, is the neural network parameter. SELU function is used as the activation function of the input layer and hidden layer, and Sigmoid is used as the output activation function.

[0021] Preferably, in step S7, the loss function for:

[0022] in, Denotes the channel random variable Find the mean, represents the air calculation rate function, represents the strategy of expanding heterogeneous graph neural network to generate transmission, represents the channel condition samples.

[0023] In a second aspect, an embodiment of the present invention provides a multi-cluster air computing system based on a deep unfolding graph neural network, comprising: The node module defines the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network; decomposes the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; Embedding module, designing the feature extraction method in the obtained neural network solution module, obtaining the key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module, designing the feature embedding method of the neural network, embedding the obtained key features, and mapping the low-dimensional features to high-dimensional features; The encoding module designs the graph neural network encoder based on the key features of the embedding and defines the methods of message generation and message aggregation in the graph neural network; Decoding module: a multi-layer perceptron is designed as a decoder to decode the output of the graph neural network into variables that need to be optimized. The computing module designs the graph neural network parameter sharing strategy and the graph neural network learning method, trains the graph neural network, obtains a model with good generalization ability, passes the channel information of the network in the new scenario as input to the model, outputs the transmission factor of each wireless device, and the receiving beam of each aggregation center, to achieve an increase in the multi-cluster air computing rate.

[0024] In a third aspect, a computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the multi-cluster aerial computing method based on a deep unfolded graph neural network when executing the computer program.

[0025] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned multi-cluster air computing method based on a deep unfolding graph neural network.

[0026] In a fifth aspect, a chip comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the multi-cluster air computing method based on a deep unfolded graph neural network are implemented.

[0027] In a sixth aspect, an embodiment of the present invention provides an electronic device, comprising a computer program, which, when executed by the electronic device, implements the steps of the above-mentioned multi-cluster air computing method based on deep unfolding graph neural network.

[0028] Compared with the prior art, the present invention has at least the following beneficial effects: A multi-cluster air computing method based on a deep unfolding graph neural network achieves significant air computing rate improvement under different clusters, devices and antenna numbers through an unfolding learning method, especially showing obvious advantages in multi-cluster and high interference scenarios. Compared with traditional optimization methods (such as alternating optimization), the present invention effectively avoids the problem of falling into suboptimal solutions in large-scale scenarios. Thanks to the introduction of heterogeneous graph neural network structure, the present invention has strong generalization ability and can maintain good performance in untrained scenarios, especially in the case of strong interference, through effective feature extraction to achieve better performance. This feature is particularly significant when the number of clusters and devices changes, indicating its adaptability in complex network environments. Compared with the large number of iterative calculations of traditional optimization methods, the unfolding learning method achieves nearly instantaneous calculation output through neural networks, greatly improving the computing efficiency. In addition, the present invention also adopts progressive learning and parameter sharing strategies to further optimize the training efficiency and reduce memory and computing resource consumption. Through the reasonable design of multiple receiving antennas, the receiving capacity and interference suppression effect of the aggregation center are improved, especially when the number of antennas is large, the receiving beamforming degree of freedom is enhanced, and a higher air computing rate is achieved. In the traditional optimization method, due to the limitation of suboptimal solution problem, the interference suppression effect is not as good as that of the present invention.

[0029] Furthermore, the use of heterogeneous graph neural networks is consistent with the architecture of multi-cluster aerial computing systems. In multi-cluster aerial computing, there are two different types of nodes: wireless devices and aggregation centers. The two different nodes are connected by channels. The two heterogeneous points in the heterogeneous graph neural network represent wireless devices and aggregation centers respectively, and the edges connecting the heterogeneous points represent channels.

[0030] Furthermore, the neural network structure required to directly solve the original problem is relatively complex, requiring a large number of samples for training, and is prone to falling into local optimal solutions. Decomposing the problem into an expanded structure is equivalent to the neural network only needing to output a better solution based on the suboptimal solution of the input. The tasks to be completed are relatively simple, and the design of shared parameters for each module can achieve the role of reusing training samples, and the distribution of training samples is more diverse, and it is not easy to fall into local optimal solutions.

[0031] Furthermore, there are a large number of wireless devices and aggregation centers in the network, and the aggregation center adopts a multi-antenna design, so the channel feature dimension is high, which also contains a lot of information redundancy, which is not conducive to the training of neural networks. Designing feature extraction methods can perform some equivalent processing on the channel, reduce the channel feature dimension, and extract key features that affect performance indicators in the channel, making subsequent training more targeted.

[0032] Furthermore, to achieve strong generalization capabilities, a single device and a single aggregation center are required as input, and there is a lack of representation of the overall characteristics of the network. Here, the extracted features are embedded, and low-dimensional features are mapped to high-dimensional features. High-dimensional features are equivalent to containers, which provide a basis for the subsequent use of graph neural networks to aggregate the overall network features.

[0033] Furthermore, the embedded features of each wireless device are aggregated to the aggregation center, and then the embedded features of the aggregation center are aggregated to each wireless device. Through multiple reciprocating operations, each aggregation center and each wireless device can obtain the characteristics of the entire network, which is conducive to a joint optimization of the entire network.

[0034] Furthermore, a multi-layer perceptron is used as a decoder to map the aggregated features of each device and obtain the corresponding transmission power allocation of each wireless device required by the multi-cluster air computing system.

[0035] Furthermore, each solution module has the same function, and sharing common parameters in each solution module can effectively reduce the number of parameters required for the model, making the model easier to train. At the same time, the input of each module can be used as a training sample, which expands the training evolution and makes the model more stable and reliable, and less prone to overfitting.

[0036] It can be understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0037] In summary, the present invention has superior air computing rate performance, strong generalization ability, high training efficiency, low computing overhead, and achieves the best interference suppression effect through the designed model.

[0038] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the present invention; Figure 2 It is a schematic diagram of the structure of the heterogeneous graph neural network of the present invention; Figure 3 A schematic diagram of the relationship between the weighted sum calculation rate and the number of clusters in the network according to the present invention; Figure 4 A comparison chart of the time consumed in solving the problem between the present invention and other solutions; Figure 5 A schematic diagram of a computer device provided by an embodiment of the present invention; Figure 6 The block diagram of a chip provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] In the description of the present invention, it should be understood that the terms “include” and “comprises” indicate the presence of described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0042] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0043] It should be further understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects are in an "or" relationship.

[0044] It should be understood that, although the terms first, second, third, etc. may be used to describe preset ranges, etc. in the embodiments of the present invention, these preset ranges should not be limited to these terms. These terms are only used to distinguish preset ranges from each other. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0045] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0046] Various structural schematic diagrams of the embodiments disclosed in the present invention are shown in the accompanying drawings. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0047] The present invention provides a multi-cluster air computing method based on a deep unfolding graph neural network. First, the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network are defined. Secondly, the multi-cluster control calculation optimization problem to be solved is decomposed into a neural network solving module with an unfolded structure. Next, a feature extraction and feature embedding method in the neural network solving module is designed. Then, a heterogeneous graph neural network encoder is designed to aggregate features, and the aggregated features are mapped to variables that need to be optimized through a designed multi-layer perceptron decoder. Finally, a parameter sharing strategy for the entire network and a learning method for the neural network are designed, and the neural network is trained to obtain a general model. The scheme of the present invention is superior to the existing optimization scheme in terms of computing rate, and the solution speed is greatly improved, and it has a strong generalization ability.

[0048] See also Figure 1 The present invention provides a multi-cluster aerial computing method based on a deep unfolding graph neural network, comprising the following steps: S1. Define the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network; See also Figure 2 ,Consider an air computing service coverage area, the entire area consists of N clusters, performing different air computing tasks, and the set of all clusters is defined as = {1, 2, ..., K}. The center of each cluster is equipped with a convergence center (FC).

[0049] For a given cluster k, it contains wireless device, represents the nth wireless device in the cluster, ={1, 2, ..., } represents the set of all wireless devices.

[0050] Wireless Devices The maximum transmission power is expressed as ; The kth FC is equipped with Antenna, with Indicates wireless device The channel to the kth FC.

[0051] When l is equal to k, Indicates a communication channel; otherwise, it indicates an interference channel. Wireless Device The transfer factor is expressed as , the receive beamforming of FC is expressed as .

[0052] Define all reflector devices in the network as Node1={ }, , represents the first type of node in the first graph, and defines all receiving end aggregation centers as Node2={k}, , represents the second type of node in the graph; the channel between the first type of node and the second type of node is defined as the edge connecting the two types of nodes Edge={ }, .

[0053] S2, decomposing the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; Expand the air computing problem into multiple cascaded neural network computing modules. For module j, input the current channel information { } and the transmission factor output by the previous module , output optimized transfer factor The solution module is encapsulated as an optimizer for the transmission factor, which consists of three submodules, namely the optimal receiving beam submodule, the optimal transmission factor phase submodule and the transmission factor modulus optimization submodule. The optimal receiving beam and the other two submodules adopt a cascade structure, while the optimal transmission factor phase and transmission factor modulus optimization submodules adopt a parallel structure. Finally, the two modules are combined to form the optimized complex transmission factor, which is expressed in the following mathematical form:

[0054] in, Represents all channel information of the entire network, Represents the learnable parameters of the neural network.

[0055] The j-th solution module can be abstracted as:

[0056] S3. Design a feature extraction method in the neural network solution module to obtain the key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module; For the transmission factor modulus optimization submodule in step S2, from the current channel information and optimal receiving beam and the output transfer factor of the previous solution module The key features reflecting the current graph neural network nodes are extracted. The feature extraction of wireless device nodes in heterogeneous graph neural networks adopts the following scheme:

[0057] The items in brackets represent wireless devices. The old transmission factor, the equivalent channel of the corresponding aggregation center, the interference channel to other aggregation centers, the corresponding receiving power of the aggregation center, and the interference power to other aggregation centers.

[0058] The feature extraction of aggregation center nodes in heterogeneous graph neural networks adopts a similar scheme:

[0059] Among them, the items in brackets represent the mean square error of the aggregation center k, the equivalent interference channel and from the effective device received by the aggregation center, the interference channel and from the interfering device received by the aggregation center, the effective power and received by the aggregation center, and the interference power and received by the aggregation center.

[0060] S4. Design a feature embedding method for the neural network to map low-dimensional features to high-dimensional features, so that other node features can be aggregated together later; Embed the features defined in step S3 and map them to high-dimensional vectors. Nodes of the same type share a neural network for embedding. and Represents the embedding function of the device node and the aggregation center node. The above two embedding functions are neural networks, and the parameters are defined as and .

[0061] Therefore, the device node The embedding operation of the aggregation center node k is expressed as:

[0062]

[0063] in, and Represents embedded wireless devices And the features of the aggregation center k, the embedding layer is a single-layer neural network, and the tanh function is used as the activation function.

[0064] S5. Design a graph neural network encoder and define the methods of message generation and message aggregation in the graph neural network. The heterogeneous graph neural network encoder processes the embedded features of the nodes by using the information passing mechanism so that the features can be fully aggregated to represent the network characteristics, which enables each node in the network to have the information of the entire network. The encoder consists of I layers of information passing, each layer of the encoder includes, information passing includes message generation, aggregation and combination, these operations constitute a cascade of linear layers, where the operation of the i-th layer is expressed as:

[0065]

[0066] in, and Respectively represent the information transmission related to the node characteristics of the device and the aggregation center, and are the shared neural network parameters generated by the information of the devices and aggregation centers of the i-th layer encoder in the j-th module.

[0067] Then, for message aggregation, each node aggregates the features of other types of nodes by weighted sum, that is, the device node aggregates the features of the central node, where the weight coefficient Defined as:

[0068] Finally, the node passes through a linear layer with layer normalization function to combine the previous features and the obtained aggregate message. This process is expressed as .

[0069] The information transfer mechanism proposed in the heterogeneous graph neural network encoder has two special designs. First, by introducing the coefficient ,The aggregation result is positively enhanced in terms of the expected signal, but negatively affected in terms of interference.,Secondly, the information transmission is defined so that information is transmitted between nodes of different,types. Nodes of the same type do not transmit information between each other, but,indirectly obtain information of nodes of the same type during multiple,information transmissions between nodes of different types, which matches the,physical model of communication.

[0070] S6. Design a multi-layer perceptron as a decoder to decode the output of the graph neural network into the variables that need to be optimized; A multi-layer perceptron (MLP) decoder is used to interpret the output features of the heterogeneous graph neural network encoding as the expected modulus value of the transfer factor of each device.

[0071] Specifically, the decoder is represented as:

[0072] in, Denotes a decoding operation, are neural network parameters.

[0073] In addition, the SELU function is used as the activation function of the input layer and the hidden layer, and the output activation function uses Sigmoid, which is further scaled to the interval , corresponds to the range of modulus values ​​of the transmission factor.

[0074] S7. Design the entire network parameter sharing strategy and the neural network learning method, train the neural network, and obtain a model with good generalization ability.

[0075] For the unfolded heterogeneous graph neural network introduced above, we need to train the network parameters to maximize the weighted and air computation rate.

[0076] Therefore, the loss function is defined as:

[0077] in, Represents the distribution of collected channel conditions.

[0078] Using unsupervised learning, given channel condition samples , Expand heterogeneous graph neural network generation transmission strategy , in order to obtain the highest average weighted and air computing rate, that is, to minimize the loss function. The stochastic gradient descent method is used for neural network training, and the calculation in the real number domain significantly simplifies the training process.

[0079] In addition, in order to improve training efficiency and performance, a progressive learning and parameter sharing strategy is adopted. The function is in the straight line region, the gradient is sparse, and it is difficult to train. Therefore, the progressive learning technique is used to divide the training into two stages. In the first stage, Relaxation .

[0080] Therefore, when the first phase of training is completed and some good parameters with differentiable loss are obtained, in the second phase, it is resumed Further enhance training.

[0081] In addition, the learning rate adopts an exponential decay schedule, which allows larger step sizes in the initial stage of training and more detailed step sizes when approaching the optimum to achieve better convergence.

[0082] In addition, for the proposed J-layer learning architecture, a parameter sharing strategy is adopted so that the blocks in the first half share the same parameters, and the same is true for the remaining blocks. In this way, the number of model parameters is significantly reduced, and the training process is completed in a more memory-saving and computationally efficient manner.

[0083] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as systems, methods or program products. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits", "modules" or "platforms".

[0084] In another embodiment of the present invention, a multi-cluster air computing system based on a deep expanded graph neural network is provided. The system can be used to implement the above-mentioned multi-cluster air computing method based on a deep expanded graph neural network. Specifically, the multi-cluster air computing system based on a deep expanded graph neural network includes a node module, an embedding module, an encoding module, a decoding module and a computing module.

[0085] Among them, the node module defines the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network; decomposes the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; Embedding module, designing the feature extraction method in the obtained neural network solution module, obtaining the key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module, designing the feature embedding method of the neural network, embedding the obtained key features, and mapping the low-dimensional features to high-dimensional features; The encoding module designs the graph neural network encoder based on the key features of the embedding and defines the methods of message generation and message aggregation in the graph neural network; Decoding module: a multi-layer perceptron is designed as a decoder to decode the output of the graph neural network into variables that need to be optimized. The computing module designs the graph neural network parameter sharing strategy and the graph neural network learning method, trains the graph neural network, obtains a model with good generalization ability, passes the channel information of the network in the new scenario as input to the model, outputs the transmission factor of each wireless device, and the receiving beam of each aggregation center, to achieve an increase in the multi-cluster air computing rate.

[0086] In another embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, graphics processors (GPU), tensor processors (TPU), digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding functions; the processor described in the embodiment of the present invention can be used for the operation of a multi-cluster air computing method based on a deep unfolding graph neural network, including: Define the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network; decompose the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; design the feature extraction method in the neural network solution module to obtain the key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module; design the feature embedding method of the neural network, embed the obtained key features, and map the low-dimensional features to high-dimensional features; design the graph neural network encoder based on the embedded key features, and define the message generation and message aggregation methods in the graph neural network; design a multi-layer perceptron as a decoder to decode the output of the graph neural network into the variables that need to be optimized; design the graph neural network parameter sharing strategy and the graph neural network learning method, train the graph neural network, and obtain a model with good generalization ability. The channel information of the network in the new scenario is passed as input to the model, and the transmission factor of each wireless device and the receiving beam of each aggregation center are output to achieve an increase in the multi-cluster air computing rate.

[0087] See also Figure 5, the terminal device is a computer device. The computer device 60 of this embodiment includes: a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the computer program 63 is executed by the processor 61, the multi-cluster aerial computing method based on the deep expanded graph neural network in the embodiment is implemented. To avoid repetition, it is not described one by one here. Alternatively, when the computer program 63 is executed by the processor 61, the functions of each model / unit in the multi-cluster aerial computing system based on the deep expanded graph neural network in the embodiment are implemented. To avoid repetition, it is not described one by one here.

[0088] The computer device 60 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will appreciate that Figure 5 This is only an example of the computer device 60 and does not constitute a limitation of the computer device 60. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.

[0089] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, graphics processing units (GPU), tensor processing units (TPU), digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0090] The memory 62 may be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 may also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the computer device 60.

[0091] Furthermore, the memory 62 may include both an internal storage unit of the computer device 60 and an external storage device. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 may also be used to temporarily store data that has been output or is to be output.

[0092] See also Figure 6 , the terminal device is a chip, and the chip 600 of this embodiment includes a processor 622, which may be one or more, and a memory 632 for storing a computer program executable by the processor 622. The computer program stored in the memory 632 may include one or more modules, each corresponding to a set of instructions. In addition, the processor 622 may be configured to execute the computer program to perform the above-mentioned multi-cluster air computing method based on the deep unfolding graph neural network.

[0093] In addition, the chip 600 may further include a power supply component 626 and a communication component 650, wherein the power supply component 626 may be configured to perform power management of the chip 600, and the communication component 650 may be configured to implement communication, for example, wired or wireless communication, of the chip 600. In addition, the chip 600 may further include an input / output interface 658. The chip 600 may operate based on an operating system stored in the memory 632.

[0094] In another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs. It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory (Non-VolatileMemory), such as at least one disk memory.

[0095] One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement the corresponding steps of the multi-cluster air computing method based on a deep unfolding graph neural network in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by the processor as follows: Define the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network; decompose the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; design the feature extraction method in the neural network solution module to obtain the key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module; design the feature embedding method of the neural network, embed the obtained key features, and map the low-dimensional features to high-dimensional features; design the graph neural network encoder based on the embedded key features, and define the message generation and message aggregation methods in the graph neural network; design a multi-layer perceptron as a decoder to decode the output of the graph neural network into the variables that need to be optimized; design the graph neural network parameter sharing strategy and the graph neural network learning method, train the graph neural network, and obtain a model with good generalization ability. The channel information of the network in the new scenario is passed as input to the model, and the transmission factor of each wireless device and the receiving beam of each aggregation center are output to achieve an increase in the multi-cluster air computing rate.

[0096] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings here can usually be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0097] See also Figure 3 , a schematic diagram of the relationship between the system weight and calculation rate and the number of clusters in the network is given.

[0098] The present invention shows the results of the air computing rate changing with the number of clusters under different schemes. In general, the air computing rate increases with the increase of the number of clusters, but gradually tends to saturation. In particular, when the number of clusters exceeds 4, the rates of the adaptive transmission and maximum power transmission schemes decrease instead. In contrast, the learning scheme of the present invention achieves better performance than the traditional alternating optimization scheme in a 5-cluster training scenario. In addition, when the number of clusters is larger, the scheme of the present invention has a strong generalization ability and its performance advantage is more obvious. The reason is that traditional optimization methods can only achieve suboptimal solutions, which becomes more obvious when the number of clusters increases and the interference increases; while the learning method of the present invention can extract the feature representation of the air computing network, bringing performance improvement. The performance of the multi-layer perceptron scheme is limited due to the lack of design for feature aggregation of different types of nodes.

[0099] See also Figure 4 , a comparison chart of the time consumed for solving the solution of the present invention and other solutions is given.

[0100] In general, alternating optimization methods require a large number of iterations, while adaptive power and maximum power schemes are almost closed-form solutions, and learning-based methods can be executed quickly. We show the average running time of the alternating optimization scheme in 5000 random deployments, as well as the total running time of the remaining schemes under the same settings. The results show that the neural network can output almost instantaneously, which is much more efficient than iterative optimization. It can be seen that the unfolding learning of the present invention has a significant performance advantage, which is further enhanced by fine-tuning.

[0101] In summary, the multi-cluster air computing method and system based on deep unfolding graph neural network of the present invention has better performance than traditional methods. In addition, the structure based on heterogeneous graph neural network has powerful generalization ability and can adapt to different network scenarios. The learned interference management ability significantly improves the performance of air computing.

[0102] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0103] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0104] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0105] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0106] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0107] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0108] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0109] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices, and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0110] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0112] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A multi-cluster aerial computing method based on deep unfolded graph neural network, characterized in that: The following steps are involved: S1. Define the aggregation center node, wireless device node and the edge connecting the two nodes in the heterogeneous graph neural network; S2, based on the aggregation center node, wireless device node and the edge connecting the two nodes defined in step S1, decomposing the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; S3, designing a feature extraction method in the neural network solution module obtained in step S2, and obtaining key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module; S4, designing a feature embedding method for a neural network, embedding the key features obtained in step S3, and mapping low-dimensional features to high-dimensional features; S5. Design a graph neural network encoder based on the key features embedded in step S4, and define methods for message generation and message aggregation in the graph neural network; S6. Design a multi-layer perceptron as a decoder to decode the output of the graph neural network obtained in step S5 into the variables to be optimized; S7. Design a graph neural network parameter sharing strategy and a graph neural network learning method, train the graph neural network, and obtain a model with good generalization ability. Pass the channel information of the network in the new scenario as input to the model, output the transmission factor of each wireless device and the receiving beam of each aggregation center, and realize the improvement of multi-cluster air computing rate.

2. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 1 is characterized in that: In step S1, the air computing service coverage area includes N clusters, and the N clusters perform different air computing tasks. The set of all clusters is defined as ={1, 2, ..., K}; the center of each cluster is equipped with an aggregation center; For a given cluster k, it contains wireless device, represents the nth wireless device in the cluster, ={1, 2, ..., } represents the set of all wireless devices; Wireless Devices The maximum transmission power is expressed as ; The kth aggregation center is equipped with Antenna, with Indicates wireless device The channel to the kth aggregation center; When l is equal to k, indicates a communication channel; otherwise, it indicates an interference channel; Wireless Devices The transfer factor is , the receiving beamforming at the aggregation center is ; Define all reflector devices as Node1={ }, , represents the first type of node in the first graph; all receiving end aggregation centers are defined as Node2={k}, , represents the second type of node in the graph; The channel between the first type of node and the second type of node is defined as the edge connecting the two types of nodes Edge={ }, .

3. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 1 is characterized in that: In step S2, the air computing problem is expanded into multiple cascaded neural network computing modules. For module j, the current channel information { } and the transmission factor output by the previous module , output optimized transfer factor ; The solution module is encapsulated as an optimizer for the transmission factor, which includes an optimal receiving beam submodule, an optimal transmission factor phase submodule and a transmission factor modulus optimization submodule. The optimal receiving beam submodule adopts a cascade structure with the optimal transmission factor phase submodule and the transmission factor modulus optimization submodule respectively, and the optimal transmission factor phase submodule and the transmission factor modulus optimization submodule adopt a parallel structure, which are finally combined to form the optimized complex transmission factor.

4. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 3 is characterized in that: The cascaded multiple neural network computing modules are specifically: in, For all channel information of the entire network, are the learnable parameters of the neural network, is the receiving beam of aggregation center k, For the receive beamforming optimization submodule, is the input transmission factor of the jth solution module, is the nth wireless device in the lth cluster, is the aggregation center set, is the phase of the transmission factor output by the jth solution module, is the transmission factor phase optimization submodule, For equipment The channel to the aggregation center k, is the nth device of the kth cluster, is the set of wireless devices in the kth cluster, is the output transmission factor modulus of the j-th solution module, is the transmission factor phase submodule, is the output transmission factor of the jth solution module, is the set of wireless devices in the lth cluster.

5. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 1 is characterized in that: In step S3, the feature extraction of wireless device nodes in the heterogeneous graph neural network adopts the following scheme: in, Wireless device input to the jth solution module The items in brackets represent the characteristics of wireless devices. The old transmission factor, the equivalent channel of the corresponding aggregation center, the interference channel to other aggregation centers, the corresponding receiving power of the aggregation center, the interference power to other aggregation centers; The feature extraction of aggregation center nodes in heterogeneous graph neural networks adopts a similar scheme: in, is the feature of the aggregation center k input to the j-th solution module. The items in the brackets represent the mean square error of the aggregation center k, the equivalent interference channel and from the effective device received by the aggregation center, the interference channel and from the interfering device received by the aggregation center, the effective power and received by the aggregation center, and the interference power and received by the aggregation center.

6. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 1 is characterized in that: In step S4, the device node The embedding operation of the aggregation center node k is expressed as: in, and Represents embedded wireless devices and the features of the aggregation center k, the embedding layer is a single-layer neural network, and the tanh function is used as the activation function. Embedding neural network for device features, Embedding neural network parameters for device features, Embedding neural network parameters for aggregation center features, is the embedding feature output by the embedding layer of the aggregation center k, For equipment Embedded features output by the feature embedding layer.

7. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 1 is characterized in that: In step S5, the encoder consists of I layers of information transfer, each layer of the encoder includes, information transfer includes message generation, aggregation and combination, forming a cascade of linear layers. For message aggregation, each node aggregates the features of other types of nodes by weighting and aggregating, that is, the device node aggregates the features of the central node. The node combines the previous features and the obtained aggregated message through a linear layer with layer normalization function. This process is expressed as ; Introducing coefficients in the proposed information transfer mechanism in the heterogeneous graph neural network encoder , define information transmission so that information can be transmitted between nodes of different types, and information can not be transmitted between nodes of the same type.

8. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 1, characterized in that: In step S6, the decoder is represented as: in, Denotes a decoding operation, is the neural network parameter. SELU function is used as the activation function of the input layer and hidden layer, and Sigmoid is used as the output activation function.

9. The multi-cluster aerial computing method based on deep unfolded graph neural network according to claim 1, characterized in that: In step S7, the loss function for: in, Denotes the channel random variable Find the mean, represents the air calculation rate function, represents the strategy of expanding heterogeneous graph neural network to generate transmission, represents the channel condition samples.

10. A multi-cluster air computing system based on deep unfolded graph neural network, characterized in that: include: The node module defines the aggregation center node, wireless device node, and edges connecting the two nodes in the heterogeneous graph neural network; Decompose the multi-cluster control calculation optimization problem to be solved into a neural network solution module with an unfolded structure; Embedding module, designing the feature extraction method in the obtained neural network solution module, obtaining the key features that determine the output of the current solution module from the input channel information and the transmission factor output by the previous solution module, designing the feature embedding method of the neural network, embedding the obtained key features, and mapping the low-dimensional features to high-dimensional features; The encoding module designs the graph neural network encoder based on the key features of the embedding and defines the methods of message generation and message aggregation in the graph neural network; Decoding module: a multi-layer perceptron is designed as a decoder to decode the output of the graph neural network into variables that need to be optimized. The computing module designs the graph neural network parameter sharing strategy and the graph neural network learning method, trains the graph neural network, obtains a model with good generalization ability, passes the channel information of the network in the new scenario as input to the model, outputs the transmission factor of each wireless device, and the receiving beam of each aggregation center, to achieve an increase in the multi-cluster air computing rate.

Images