Wireless network resource allocation method based on edge intelligence

Abstract

The invention relates to the technical field of wireless communication and edge computing fusion, in particular to a wireless network resource allocation method based on edge intelligence, which comprises the following steps: S1, collecting a multi-dimensional parameter set of network edge nodes in real time; s2, converting node position coordinates into a network topological structure map, and mapping service type labels and bandwidth requirements into service requirement feature vectors; s3, generating a domain division matrix based on weight distribution of the network topological structure atlas; s4, outputting a resource block configuration tensor according to the time-frequency feature of the business demand feature vector; s5, generating a resource allocation mapping relation table; and S6, generating and executing a node-level resource scheduling instruction set. According to the invention, by constructing a resource scheduling mechanism fusing a topological structure and service features, precise resource allocation and dynamic load balancing of a cooperative domain level and a node level are realized, and the efficiency and the resource utilization rate of multi-task concurrent processing in an edge computing environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fusion of wireless communication and edge computing, and in particular to a wireless network resource allocation method based on edge intelligence. Background Art

[0002] With the convergence of edge computing and wireless communication technologies, more and more delay-sensitive and compute-intensive services are being deployed to edge nodes close to users for real-time processing. In this context, edge computing nodes collaborate with terminal devices through wireless networks to perform data perception, computation, and feedback, and their performance is highly dependent on the rational allocation of network resources. Especially under the orthogonal frequency division multiplexing (OFDM) mechanism, precisely matching limited spectrum resources and computing resources to the edge collaboration domain and edge nodes where the tasks are located is a key prerequisite for efficient scheduling of edge computing.

[0003] However, existing resource allocation schemes in edge networks generally suffer from the following problems: First, they lack a unified modeling mechanism, failing to effectively correlate service demand characteristics, network topology, and resource time-frequency distribution. Second, the division of collaborative domains is static, failing to dynamically adjust the scope of resource collaboration based on node heterogeneity and topological complementarity. Third, resource scheduling is coarse-grained, with tasks statically matched based on node remaining resources, making it difficult to support dynamic multi-service concurrency and load balancing. Therefore, a wireless network resource allocation method based on edge intelligence is urgently needed to address these issues. Summary of the Invention

[0004] Based on the above objectives, the present invention provides a wireless network resource allocation method based on edge intelligence.

[0005] A wireless network resource allocation method based on edge intelligence includes the following steps:

[0006] S1: Real-time collection of multi-dimensional parameter sets of network edge nodes, including node location coordinates, available bandwidth, remaining computing units, and current service type labels;

[0007] S2: The topology modeling engine converts node location coordinates into a weighted network topology graph. A feature extractor maps service type labels and bandwidth requirements into a service requirement feature vector.

[0008] S3: Based on the weight distribution of the network topology graph, a dynamic complementary clustering algorithm is used to divide the edge collaboration domains and generate a domain partition matrix containing N collaboration domains;

[0009] S4: Generate a set of dynamic resource blocks in the orthogonal frequency division multiplexing resource pool based on the time-frequency characteristics of the service demand feature vector. Each dynamic resource block has corresponding time slot length and frequency bandwidth parameters, and output a resource block configuration tensor.

[0010] S5: Input the domain partition matrix and resource block configuration tensor into the edge intelligent allocator, calculate the adaptive mapping weight according to the resource gap degree of each collaborative domain, and generate a resource allocation mapping relationship table;

[0011] S6: Based on the resource allocation mapping table, a distributed load balancing algorithm is used in each collaborative domain, and the node-level resource scheduling instruction set is generated and executed in combination with the remaining computing units of each edge node.

[0012] Optionally, the S1 specifically includes:

[0013] S11: In the edge network architecture, each edge node is configured with a unique identification code and a positioning module. The two-dimensional or three-dimensional position coordinates of each edge node are obtained in real time using the global positioning system, and the position coordinates are recorded and updated at fixed time intervals.

[0014] S12: Use the bandwidth monitoring probe to monitor the instantaneous throughput rate of the node interface and calculate the average bandwidth usage within the unit time window. Calculate the available bandwidth value by subtracting the actual occupied bandwidth from the current physical bandwidth of the interface. Output this value as the node's available bandwidth parameter.

[0015] S13: Based on the computing power monitoring module embedded in the node control terminal, the difference between the total number of currently schedulable computing units of the node and the number of currently occupied computing units is counted to obtain the number of remaining computing units;

[0016] S14: After the service request is initiated, the service type is identified according to the application type field in the service access protocol, and the service type label processed by the current node is determined through the local service label mapping table.

[0017] Optionally, the S2 specifically includes:

[0018] S21: Receive the location coordinate data of each edge node, use the two-dimensional Euclidean distance formula to calculate the physical distance between any two nodes, define it as the initial weight value of the topological edge, and construct the node distance matrix;

[0019] S22: Based on the physical distance between nodes and the communication link strength threshold, select node pairs that meet the direct connection conditions, establish node connection edges, and attach communication channel quality parameters to the edges to form edge attribute weights;

[0020] S23: Fuse the distance matrix with the edge attribute matrix, and construct a network topology graph containing a node set, an edge set, and an edge weight set through a graph structure generation algorithm;

[0021] S24: Receive the service type label and available bandwidth parameter extracted in S1, call the preset service type-bandwidth weight mapping table based on the service label, and find the corresponding standard bandwidth requirement value;

[0022] S25: Use linear normalization processing to map the standard bandwidth requirement value to a unified vector space, and combine it with the service type code to form a service requirement feature vector.

[0023] Optionally, the S25 specifically includes:

[0024] S251: Convert the business type label number obtained in S14 into a one-hot encoding form to form a category vector T of the business type vec ;

[0025] S252: Normalize the standard bandwidth requirement value obtained by looking up the table with the maximum service bandwidth supported by the system to obtain the standard bandwidth intensity component R n ;

[0026] S253: Normalize the standard bandwidth intensity component R n Multiply it with the corresponding bandwidth weight coefficient B to form the final bandwidth characteristic component F;

[0027] S254: Encode the service type vector T vec Combined with the calculated bandwidth characteristic component F, a complete service demand characteristic vector V is constructed, which is expressed as: V = [T vec ,F].

[0028] Optionally, the S3 specifically includes:

[0029] S31: Receive the network topology graph constructed in S2, initialize the set of nodes to be clustered and set the threshold range of the number of collaboration domains to control the upper and lower limits of the number of collaboration domains finally generated;

[0030] S32: For each node, a complementarity index is calculated to measure its resource heterogeneity in terms of available bandwidth, remaining computing resources, and service type differences.

[0031] S33: Construct a similarity matrix between nodes based on the complementarity index and the topological distance between nodes;

[0032] S34: Based on the similarity matrix, a dynamic clustering algorithm is used to perform domain division. The node with the highest similarity is used as the cluster center. Adjacent nodes that meet the minimum number of nodes and the mean complementarity requirements are gradually absorbed into the same collaboration domain, eventually forming N edge collaboration domains that meet the constraints.

[0033] S35: Perform structured representation on the clustering results to form a domain partition matrix M.

[0034] Optionally, the S34 specifically includes:

[0035] S341: Initialize the candidate collaboration domain set to be empty, select the node with the highest average similarity from the similarity matrix as the initial cluster center, and record it as the seed node; each seed node is used as the starting point of a collaboration domain, and the process is carried out in sequence until all seed candidate nodes are traversed;

[0036] S342: For each seed node, traverse the set of directly connected adjacent nodes and select the nodes whose similarity with the seed node is greater than the threshold S in the similarity matrix. th The adjacent nodes of are the candidate members for this expansion;

[0037] S343: Calculate the average complementarity C of the selected nodes in the current cluster. Only when the absolute difference between the complementarity of the newly included node and this value does not exceed the set deviation threshold ΔC max Only when it is available can it be included in the collaborative domain;

[0038] S344: Repeat the node absorption process until the number of nodes in the current collaboration domain reaches the set minimum member number threshold N min , or until there are no nodes that meet the absorption conditions, the current collaboration domain is built and its member node numbers are recorded.

[0039] Optionally, the S4 specifically includes:

[0040] S41: Receive the service demand feature vector generated in S2, extract the bandwidth strength component in the vector as the frequency domain demand indicator, combine it with the transmission delay tolerance value corresponding to the service type, define the minimum resource block time slot length and minimum frequency bandwidth requirements of the service, and form a time-frequency feature demand pair;

[0041] S42: Accessing the OFDM resource pool and reading the currently available resource block index table, which contains the starting frequency, bandwidth, starting time slot, and time slot length parameters of all unoccupied resource blocks, for dynamically matching actual service needs;

[0042] S43: Based on the extracted time-frequency feature demand pairs, a matching strategy is executed in the orthogonal frequency division multiplexing resource pool. According to the principle of minimum satisfaction and priority for low interference, a set of resource blocks that can be allocated are screened to form a candidate list of resource blocks that can be allocated to the current service. The frequency domain interference level and time domain exclusivity index are marked for each candidate block.

[0043] S44: Selecting a resource block group from the resource block candidate list that satisfies the target constraint range in terms of bandwidth, time slot, and interference level, and arranging the allocation order according to the service priority to generate a final resource block set for the service;

[0044] S45: Summarize the final resource block sets of all services to be allocated, and construct a resource block configuration tensor based on the five-dimensional structure of service number, frequency starting point, bandwidth, time slot starting point and time slot length.

[0045] Optionally, the S5 specifically includes:

[0046] S51: Receive the collaborative domain partition matrix output by S3, obtain the node set of each collaborative domain, and based on the number of remaining computing units and available bandwidth of the nodes collected in S1, count the computing resource capacity and communication resource capacity currently available for allocation in each collaborative domain, which are recorded as and Where k represents the collaboration domain number;

[0047] S52: Receive the resource block configuration tensor output in S4 and analyze the minimum computing resource requirements for each business task in the tensor Minimum bandwidth requirement Where m represents the business task number;

[0048] S53: Calculate the resource gap degree Δ of collaboration domain k corresponding to business task m based on the candidate mapping relationship between business tasks and collaboration domains km ;

[0049] S54: For all candidate collaboration domain-business task combinations, calculate the adaptive mapping weight W based on the inverse relationship of the gap degree km ;

[0050] S55: Sort all candidate combinations of collaboration domains and business tasks from high to low according to mapping weights, and assign a collaboration domain with the highest weight to each business task as the target execution domain, and construct a resource allocation mapping relationship table.

[0051] Optionally, the S6 specifically includes:

[0052] S61: Receive the resource allocation mapping table output in S5, extract the target collaboration domain number corresponding to each business task, and build a node task candidate pool within the collaboration domain. The task candidate pool contains the numbers of all edge nodes in the domain that have not reached the computing load limit;

[0053] S62: Read the number of remaining computing units of each node in the collaboration domain, calculate its load carrying capacity based on the current resource status of the node, and record the number of idle computing units that can be allocated to each node in this scheduling cycle;

[0054] S63: Based on the computing resource requirements of the business task, a set of edge nodes that meet the requirements is screened from the candidate node pool, and the node with the most remaining computing units is preferentially selected from the set as the execution node for this task;

[0055] S64: For each selected execution node, construct a node-level resource scheduling instruction set, where the instruction set includes the task number to be executed by the corresponding node, the bound resource block number and its time-frequency configuration parameters;

[0056] S65: The constructed node-level scheduling instruction set is synchronously sent to the edge nodes in the collaboration domain, and parsed and executed by the local task scheduler of each node to complete the final edge execution operation.

[0057] Optionally, the S62 specifically includes:

[0058] S621: Receive the initial value of the remaining computing units of the edge node collected in S1, and in conjunction with the node operating system monitoring interface, obtain the number of computing units occupied by the running and pending tasks in the node processing queue at the current moment, and count the current total computing unit occupancy;

[0059] S622: Calculate the number of idle computing units available for task allocation in this round on the current node based on the maximum total number of schedulable computing units configured on the node and deducting the number of occupied computing units.

[0060] S623: Obtain the start and end time of the time window of the current scheduling cycle, and bind the number of idle computing units to the start and end time periods of the time window in combination with the resource scheduling cycle settings of the node operating system to form an idle resource status snapshot with a time tag;

[0061] S624: The number of idle computing units of each node and the node identifier are combined into a key-value pair, and recorded in a node load status mapping table within the collaboration domain.

[0062] Beneficial effects of the present invention:

[0063] The present invention achieves efficient fusion expression of edge node location, computing power, bandwidth status and time-frequency characteristics of business tasks by constructing a unified network topology map and business demand feature vector; dividing the collaboration domain based on the dynamic complementary clustering algorithm can maintain topological connectivity and task processing independence while ensuring the complementarity of resource heterogeneity between nodes, providing clear structural support for the accurate mapping of resource blocks.

[0064] The present invention, by introducing adaptive resource gap calculation and distributed load balancing scheduling mechanism, realizes the linkage allocation of resource blocks and node-level computing resources within the domain, ensuring that tasks are dynamically scheduled to the most appropriate execution node while meeting spectrum latency requirements, thereby improving resource utilization efficiency and task processing throughput. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 A schematic diagram of a wireless network resource allocation method according to an embodiment of the present invention;

[0067] Figure 2 The figure is a schematic diagram of a process of generating a domain partition matrix according to an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0069] It should be noted that references in the specification to "one embodiment," "an embodiment," "an exemplary embodiment," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes such specific features, structures, or characteristics. In addition, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0070] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0071] like Figure 1-Figure 2As shown, a wireless network resource allocation method based on edge intelligence includes the following steps:

[0072] S1: Real-time collection of multi-dimensional parameter sets of network edge nodes, including node location coordinates, available bandwidth, remaining computing units, and current service type labels;

[0073] S2: The topology modeling engine converts node location coordinates into a weighted network topology graph. A feature extractor maps service type labels and bandwidth requirements into a service requirement feature vector.

[0074] S3: Based on the weight distribution of the network topology graph, a dynamic complementary clustering algorithm is used to divide the edge cooperation domains and generate a domain partition matrix containing N cooperation domains, where each cooperation domain contains at least three edge nodes with complementary characteristics;

[0075] S4: Generate a set of dynamic resource blocks in the orthogonal frequency division multiplexing resource pool based on the time-frequency characteristics of the service demand feature vector. Each dynamic resource block has corresponding time slot length and frequency bandwidth parameters, and output a resource block configuration tensor.

[0076] S5: Input the domain partition matrix and resource block configuration tensor into the edge intelligent allocator, calculate the adaptive mapping weight according to the resource gap degree of each collaborative domain, and generate a resource allocation mapping relationship table;

[0077] S6: Based on the resource allocation mapping table, a distributed load balancing algorithm is used in each collaborative domain, and the node-level resource scheduling instruction set is generated and executed in combination with the remaining computing units of each edge node, thereby completing the dynamic allocation of wireless network resources.

[0078] S1 specifically includes:

[0079] S11: In the edge network architecture, each edge node is configured with a unique identification code and a positioning module. The two-dimensional or three-dimensional position coordinates of each edge node are obtained in real time using the Global Positioning System (GPS). The position coordinates are recorded and updated at fixed time intervals (less than 2 seconds).

[0080] S12: Use the bandwidth monitoring probe to monitor the instantaneous throughput rate of the node interface and calculate the average bandwidth usage within the unit time window. Calculate the available bandwidth value by subtracting the actual occupied bandwidth from the current physical bandwidth of the interface. Output this value as the node's available bandwidth parameter.

[0081] S13: Based on the computing power monitoring module embedded in the node control terminal, the difference between the total number of currently schedulable computing units of the node and the number of currently occupied computing units is counted to obtain the number of remaining computing units. The computing units are represented by a unified virtual core and reported to the resource management center as an integer value.

[0082] S14: After the service request is initiated, the service type is identified based on the application type field in the service access protocol, and the service type label processed by the current node is determined through the local service label mapping table. The service type label is represented by a fixed code, including but not limited to standard label values ​​such as video processing, intelligent recognition, and data caching. The above steps enhance the timeliness perception capability of resource status in the edge computing environment by uniformly collecting and formatting multi-dimensional parameters of node location, bandwidth, computing power, and service type, and provide an accurate and synchronized data support foundation for subsequent steps such as collaborative domain division and resource mapping.

[0083] S2 specifically includes:

[0084] S21: Receive the location coordinate data of each edge node, use the two-dimensional Euclidean distance formula to calculate the physical distance between any two nodes, define it as the initial weight value of the topological edge, and construct the distance matrix between nodes; the Euclidean distance calculation formula is: Among them, D ij Represents the distance between the i-th node and the j-th node, (x i ,y i ) and (x j ,y j ) are the position coordinates of the two nodes respectively;

[0085] S22: Based on the physical distance between nodes and the communication link strength threshold, select node pairs that meet the direct connection conditions, establish node connection edges, and attach communication channel quality parameters to the edges to form edge attribute weights. The edge weight calculation formula is: Among them, W ij represents the final weight of the edge between node i and node j, Q ij Indicates the channel quality factor measured based on the actual link;

[0086] S23: The distance matrix and the edge attribute matrix are integrated, and a network topology graph including a node set, an edge set, and an edge weight set is constructed through a graph structure generation algorithm, and output to the collaboration domain partitioning module; the network topology graph is defined as G = (U, E, W), where U is the set of edge nodes; E is the set of node-connecting edges; and W is the set of weights corresponding to each edge in the edge set;

[0087] S24: Receive the service type label and available bandwidth parameter extracted in S1, call the preset service type-bandwidth weight mapping table based on the service label, and find the corresponding standard bandwidth requirement value;

[0088] S25: Linear normalization is used to map the standard bandwidth demand value to a unified vector space, which is then combined with the service type code to form a service demand feature vector. This vector is then used as input for the resource block generation and scheduling model. The above steps ensure the dynamic visualization modeling of the network topology and the quantitative expression of service demands, enhancing the accuracy and adaptability of the collaborative domain formation and precise resource mapping in the subsequent resource scheduling process.

[0089] Table 1 Preset service type-bandwidth weight mapping

[0090]

[0091]

[0092] In Table 1 above, the service type label number (T) is the service task type label output by step S14, which represents the service type in a standardized number. The service type name corresponds to the specific service scenario indicated by the label number. The standard bandwidth requirement represents the average transmission bandwidth required to complete this type of service task, in Mbps, based on a preset service performance benchmark. The bandwidth weight coefficient (B) is used to participate in vector generation in conjunction with the normalization strategy when constructing the service demand feature vector. Before normalization, this coefficient serves as one of the feature strength weights.

[0093] S25 specifically includes:

[0094] S251: Convert the business type label number T obtained in S14 into a one-hot encoding form to form a category vector T of the business type vec For example, if the current service type is T3, the corresponding code in the 7 types of services is: T vec =[0,0,1,0,0,0,0];

[0095] S252: Normalize the standard bandwidth requirement value obtained by looking up the table with the maximum service bandwidth supported by the system to obtain the standard bandwidth intensity component R n , the formula is: Among them, R is the standard bandwidth requirement of the corresponding business, R max is the preset maximum reference bandwidth;

[0096] S253: Normalize the standard bandwidth intensity component R n Multiplying this with the corresponding bandwidth weight coefficient B forms the final bandwidth characteristic component F, which represents the relative demand intensity of the current service for bandwidth resources.

[0097] S254: Encode the service type vector T vecCombined with the calculated bandwidth characteristic component F, a complete service demand characteristic vector V is constructed, which is expressed as: V = [T vec ,F]; The service demand feature vector constructed in the above way has a unified structure, measurable bandwidth demand expression, and can be directly used as model input, which facilitates resource block generation and scheduling algorithms to achieve efficient matching between task types and communication resources.

[0098] The following are some examples:

[0099] If the service type is T3 (AR / VR real-time rendering), then R = 25Mbps; B = 1.25; R max =25Mbps;

[0100] Then we have:

[0101] R n =25 / 25=1.00;

[0102] F = 1.25 × 1.00 = 1.25;

[0103] T vec =[0,0,1,0,0,0,0];

[0104] V=[0, 0, 1, 0, 0, 0, 0, 1.25].

[0105] S3 specifically includes:

[0106] S31: Receive the network topology graph constructed in S2, initialize the set of nodes to be clustered and set the threshold range of the number of collaboration domains to control the upper and lower limits of the number of collaboration domains finally generated;

[0107] S32: For each node, a complementarity index is calculated to measure its resource heterogeneity in terms of available bandwidth, remaining computing resources, and service type differences. This complementarity index is a key feature basis for the subsequent clustering process. The complementarity calculation formula is as follows: Among them, C i is the complementarity index of node i; B i is the available bandwidth of node i; P i is the number of remaining computing units of node i; L i is the average difference between the business types of node i and its adjacent nodes; α, β, and γ are corresponding weight factors, which control the influence weight of different resource attributes in the complementarity evaluation;

[0108] S33: Based on the complementarity index and the topological distance between nodes, a similarity matrix between nodes is constructed. This matrix is ​​used to measure the suitability of any two nodes in the same collaboration domain, taking into account the dual effects of the physical distance between nodes and the difference in complementarity. The similarity calculation formula is as follows: Among them, S ij is the similarity between node i and node j; D ij is the Euclidean distance between nodes; δ is the distance decay coefficient; max(C) is the maximum complementarity among all nodes, which is used for normalization processing;

[0109] S34: Based on the similarity matrix, a dynamic clustering algorithm is used to divide the domain. The nodes with the highest similarity and outstanding complementarity are used as cluster centers. Adjacent nodes that meet the minimum number of nodes and the mean complementarity requirements are gradually absorbed into the same collaboration domain, eventually forming N edge collaboration domains that meet the constraints.

[0110] S35: The clustering results are structured to form a domain partition matrix M, which is a binary matrix of |U|×N, where each row corresponds to a node and each column corresponds to a collaborative domain. The above method implements a collaborative domain partitioning scheme with a clear structure and clear boundaries through resource complementarity modeling and node similarity quantification, providing a structural basis for partition optimization for subsequent resource block configuration and node-level scheduling, and effectively improving the system's resource allocation accuracy and scheduling efficiency.

[0111] S34 specifically includes:

[0112] S341: Initialize the candidate collaboration domain set to be empty, select the node with the highest average similarity from the similarity matrix as the initial cluster center, and record it as the seed node; each seed node is used as the starting point of a collaboration domain, and the process is carried out in sequence until all seed candidate nodes are traversed;

[0113] S342: For each seed node, traverse the set of directly connected adjacent nodes and select the nodes whose similarity with the seed node is greater than the threshold S in the similarity matrix. th The adjacent nodes of are the candidate members for this expansion;

[0114] S343: Calculate the average complementarity of the selected nodes in the current cluster Only when the absolute difference between the complementarity of the newly included node and this value does not exceed the set deviation threshold ΔC max Only when it is available can it be included in the collaborative domain;

[0115] S344: Repeat the node absorption process until the number of nodes in the current collaboration domain reaches the set minimum member number threshold N min , or there are no nodes that meet the absorption conditions, the current collaborative domain is built and its member node numbers are recorded; the above steps realize the construction of a high-quality collaborative domain division scheme on the basis of ensuring the compactness of node structure and resource heterogeneity, which significantly improves the accuracy of intra-domain scheduling and the overall load adaptability of the system.

[0116] S4 specifically includes:

[0117] S41: Receive the service demand feature vector generated in S2, extract the bandwidth strength component in the vector as the frequency domain demand indicator, combine it with the transmission delay tolerance value corresponding to the service type, define the minimum resource block time slot length and minimum frequency bandwidth requirements of the service, and form a time-frequency feature demand pair;

[0118] S42: Access the Orthogonal Frequency Division Multiplexing (OFDM) resource pool and read the currently available resource block index table. The resource block index table contains the starting frequency, bandwidth, starting time slot, and time slot length parameters of all unoccupied resource blocks, which is used to dynamically match the actual service needs.

[0119] S43: Based on the extracted time-frequency feature demand pairs, a matching strategy is executed in the orthogonal frequency division multiplexing resource pool. According to the principle of minimum satisfaction and priority for low interference, a set of resource blocks that can be allocated are screened to form a candidate list of resource blocks that can be allocated to the current service. The frequency domain interference level and time domain exclusivity index are marked for each candidate block.

[0120] Table 2 Resource block candidate list

[0121]

[0122] S44: Selecting a resource block group from the resource block candidate list that meets the target constraint range for bandwidth, time slot, and interference level, and arranging the allocation order according to the service priority to generate a final resource block set for the service, ensuring that resource allocation between services does not conflict with each other;

[0123] The process of selecting a resource block group that meets all constraints based on the candidate list is as follows:

[0124] S441: Extract the frequency domain demand value f in the service demand d and the time domain demand value t d , taking it as the minimum resource requirement threshold;

[0125] S442: Filter out a set of resource blocks that meet the following conditions from the candidate list:

[0126] 1. Bandwidth Δf ≥ f d ;

[0127] 2. Time slot length Δt ≥ t d ;

[0128] 3. Interference level I≤I max , where I max is the set maximum tolerable interference level;

[0129] 4. Exclusivity is yes;

[0130] S443: sorting the filtered resource blocks in ascending order of interference level and frequency starting point, and preferentially selecting a resource block combination with minimal resource interference and best frequency continuity;

[0131] S444: Select from the sorted resource blocks in sequence according to the service priority until the current service requirements are fully covered.

[0132] S45: Summarize the final resource block sets of all services to be allocated, and construct a resource block configuration tensor based on the five-dimensional structure of service number, frequency starting point, bandwidth, time slot starting point and time slot length. The output is then used in the subsequent collaborative domain mapping and node scheduling stages. The above steps achieve accurate dynamic generation of resource blocks within the OFDM resource pool by combining the bandwidth and latency characteristics of the service demand feature vector, and organize resource allocation results in a tensor structure to ensure resource isolation and scheduling compatibility under multi-service concurrency, providing a high-dimensional, highly adaptable input basis for edge resource allocation.

[0133] S5 specifically includes:

[0134] S51: Receive the collaborative domain partition matrix output by S3, obtain the node set of each collaborative domain, and based on the number of remaining computing units and available bandwidth of the nodes collected in S1, count the computing resource capacity and communication resource capacity currently available for allocation in each collaborative domain, which are recorded as and Where k represents the collaboration domain number;

[0135] S52: Receive the resource block configuration tensor output in S4 and analyze the minimum computing resource requirements for each business task in the tensor Minimum bandwidth requirement Where m represents the business task number;

[0136] S53: Calculate the resource gap degree Δ of collaboration domain k corresponding to business task m based on the candidate mapping relationship between the business task and the collaboration domain (i.e., the domain contains the business-related node). km , the expression is:

[0137] Among them, ω p ,ω b are the weight coefficients of computing resources and bandwidth resources respectively; if the collaborative domain resources are sufficient, the corresponding gap is 0; if they are insufficient, the excess is counted;

[0138] S54: For all candidate collaboration domain-business task combinations, calculate the adaptive mapping weight W based on the inverse relationship of the gap degree km , the expression is: The larger the weight value, the more capable the collaboration domain is in terms of resources to undertake the business task.

[0139] S55: Among all candidate combinations of collaboration domains and business tasks, sort them from high to low according to the mapping weight, and assign a collaboration domain with the highest weight to each business task as the target execution domain, and build a resource allocation mapping relationship table. The resource allocation mapping relationship table structure includes the business task number, the target collaboration domain number, the allocated resource block number and the corresponding time-frequency configuration parameters.

[0140] Table 3 Resource allocation mapping relationship

[0141]

[0142] By establishing a resource allocation mapping relationship table, the system realizes the three-dimensional binding between tasks and collaboration domains, and resource blocks, and the structured output of the scheduling plan list. This table not only clarifies the resource scheduling path, but also quantifies the adaptability basis of the allocation decision, providing a clear and executable data foundation for resource distribution and concurrent scheduling.

[0143] S6 specifically includes:

[0144] S61: Receive the resource allocation mapping table output in S5, extract the target collaboration domain number corresponding to each business task, and build a node task candidate pool within the collaboration domain. The task candidate pool contains the numbers of all edge nodes in the domain that have not reached the computing load limit;

[0145] S62: Read the number of remaining computing units of each node in the collaboration domain, calculate its load carrying capacity based on the current resource status of the node, and record the number of idle computing units that can be allocated to each node in this scheduling cycle;

[0146] S63: Based on the computing resource requirements of the business task, a set of edge nodes that meet the requirements is screened from the candidate node pool. The node with the most remaining computing units is preferentially selected from the set as the execution node for this task to ensure balanced load distribution within the domain.

[0147] S64: For each selected execution node, construct a node-level resource scheduling instruction set, where the instruction set includes the task number to be executed by the corresponding node, the bound resource block number and its time-frequency configuration parameters;

[0148] S65: The constructed node-level scheduling instruction set is synchronously sent to the edge nodes in the collaborative domain, and parsed and executed by the local task scheduler of each node. The execution content includes task loading, resource block binding and resource activation in the time-frequency domain, completing the final edge execution operation.

[0149] S62 specifically includes:

[0150] S621: Receive the initial value of the remaining computing units of the edge node collected in S1, and in conjunction with the node operating system monitoring interface, obtain the number of computing units occupied by the running and pending tasks in the node processing queue at the current moment, and count the current total computing unit occupancy;

[0151] S622: Calculate the number of idle computing units available for task allocation in this round on the current node based on the maximum total number of schedulable computing units configured for the node, deduct the number of occupied computing units, and record the number as the valid schedulable resources for the current scheduling period of the node.

[0152] S623: Obtain the start and end times of the time window of the current scheduling cycle. Combined with the resource scheduling cycle settings of the node operating system, bind the number of idle computing units to the start and end time periods of the time window to form a snapshot of the idle resource state with a time tag, ensuring that this state is used for allocation decisions within the current scheduling cycle.

[0153] S624: The number of idle computing units of each node and the node identifier are combined into a key-value pair and recorded in the node load status mapping table within the collaborative domain, providing a real-time availability reference for the subsequent task node selection in S63; the above steps realize the quantitative evaluation of resource availability before scheduling by periodically counting and structured recording the node computing resource status, providing a clear and currently valid basis for idle computing capacity for the subsequent task allocation process, thereby improving the scheduling accuracy and resource utilization efficiency within the domain.

[0154] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0155] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A wireless network resource allocation method based on edge intelligence, characterized in that: The following steps are involved: S1: Real-time collection of multi-dimensional parameter sets of network edge nodes, including node location coordinates, available bandwidth, remaining computing units, and current service type labels; S2: The topology modeling engine converts node location coordinates into a weighted network topology graph. A feature extractor maps service type labels and bandwidth requirements into a service requirement feature vector. S3: Based on the weight distribution of the network topology graph, a dynamic complementary clustering algorithm is used to divide the edge collaboration domains and generate a domain partition matrix containing N collaboration domains; S4: Generate a set of dynamic resource blocks in the orthogonal frequency division multiplexing resource pool based on the time-frequency characteristics of the service demand feature vector. Each dynamic resource block has corresponding time slot length and frequency bandwidth parameters, and output a resource block configuration tensor. S5: Input the domain partition matrix and resource block configuration tensor into the edge intelligent allocator, calculate the adaptive mapping weight according to the resource gap degree of each collaborative domain, and generate a resource allocation mapping relationship table; S6: Based on the resource allocation mapping table, a distributed load balancing algorithm is used in each collaborative domain, and the node-level resource scheduling instruction set is generated and executed in combination with the remaining computing units of each edge node.

2. The wireless network resource allocation method based on edge intelligence according to claim 1, characterized in that: Said S1 specifically includes: S11: In the edge network architecture, each edge node is configured with a unique identification code and a positioning module. The two-dimensional or three-dimensional position coordinates of each edge node are obtained in real time using the global positioning system, and the position coordinates are recorded and updated at fixed time intervals. S12: Use the bandwidth monitoring probe to monitor the instantaneous throughput rate of the node interface and calculate the average bandwidth usage within the unit time window. Calculate the available bandwidth value by subtracting the actual occupied bandwidth from the current physical bandwidth of the interface. Output this value as the node's available bandwidth parameter. S13: Based on the computing power monitoring module embedded in the node control terminal, the difference between the total number of currently schedulable computing units of the node and the number of currently occupied computing units is counted to obtain the number of remaining computing units; S14: After the service request is initiated, the service type is identified according to the application type field in the service access protocol, and the service type label processed by the current node is determined through the local service label mapping table.

3. The wireless network resource allocation method based on edge intelligence according to claim 1, characterized in that: The S2 specifically includes: S21: Receive the location coordinate data of each edge node, use the two-dimensional Euclidean distance formula to calculate the physical distance between any two nodes, define it as the initial weight value of the topological edge, and construct the node distance matrix; S22: Based on the physical distance between nodes and the communication link strength threshold, select node pairs that meet the direct connection conditions, establish node connection edges, and attach communication channel quality parameters to the edges to form edge attribute weights; S23: Fuse the distance matrix with the edge attribute matrix, and construct a network topology graph containing a node set, an edge set, and an edge weight set through a graph structure generation algorithm; S24: Receive the service type label and available bandwidth parameter extracted in S1, call the preset service type-bandwidth weight mapping table based on the service label, and find the corresponding standard bandwidth requirement value; S25: Use linear normalization processing to map the standard bandwidth requirement value to a unified vector space, and combine it with the service type code to form a service requirement feature vector.

4. The wireless network resource allocation method based on edge intelligence according to claim 3, characterized in that: The S25 specifically includes: S251: Convert the business type label number obtained in S14 into a one-hot encoding form to form a category vector T of the business type vec ; S252: Normalize the standard bandwidth requirement value obtained by looking up the table with the maximum service bandwidth supported by the system to obtain the standard bandwidth intensity component R n ; S253: Normalize the standard bandwidth intensity component R n Multiply it with the corresponding bandwidth weight coefficient B to form the final bandwidth characteristic component F; S254: Encode the service type vector T vec Combined with the calculated bandwidth characteristic component F, a complete service demand characteristic vector V is constructed, which is expressed as: V = [T vec ,F].

5. The wireless network resource allocation method based on edge intelligence according to claim 1, characterized in that: The S3 specifically includes: S31: Receive the network topology graph constructed in S2, initialize the set of nodes to be clustered and set the threshold range of the number of collaboration domains to control the upper and lower limits of the number of collaboration domains finally generated; S32: For each node, a complementarity index is calculated to measure its resource heterogeneity in terms of available bandwidth, remaining computing resources, and service type differences. S33: Construct a similarity matrix between nodes based on the complementarity index and the topological distance between nodes; S34: Based on the similarity matrix, a dynamic clustering algorithm is used to perform domain division. The node with the highest similarity is used as the cluster center. Adjacent nodes that meet the minimum number of nodes and the mean complementarity requirements are gradually absorbed into the same collaboration domain, eventually forming N edge collaboration domains that meet the constraints. S35: Perform structured representation on the clustering results to form a domain partition matrix M.

6. The wireless network resource allocation method based on edge intelligence according to claim 5, characterized in that: The S34 specifically includes: S341: Initialize the candidate collaboration domain set to be empty, select the node with the highest average similarity from the similarity matrix as the initial cluster center, and record it as the seed node; each seed node is used as the starting point of a collaboration domain, and the process is carried out in sequence until all seed candidate nodes are traversed; S342: For each seed node, traverse the set of directly connected adjacent nodes and select the nodes whose similarity with the seed node is greater than the threshold S in the similarity matrix. th The adjacent nodes of are the candidate members for this expansion; S343: Calculate the average complementarity of the selected nodes in the current cluster Only when the absolute difference between the complementarity of the newly included node and this value does not exceed the set deviation threshold ΔC max Only when it is available can it be included in the collaborative domain; S344: Repeat the node absorption process until the number of nodes in the current collaboration domain reaches the set minimum member number threshold N min , or until there are no nodes that meet the absorption conditions, the current collaboration domain is built and its member node numbers are recorded.

7. The wireless network resource allocation method based on edge intelligence according to claim 1, characterized in that: The S4 specifically includes: S41: Receive the service demand feature vector generated in S2, extract the bandwidth strength component in the vector as the frequency domain demand indicator, combine it with the transmission delay tolerance value corresponding to the service type, define the minimum resource block time slot length and minimum frequency bandwidth requirements of the service, and form a time-frequency feature demand pair; S42: Accessing the OFDM resource pool and reading the currently available resource block index table, which contains the starting frequency, bandwidth, starting time slot, and time slot length parameters of all unoccupied resource blocks, for dynamically matching actual service needs; S43: Based on the extracted time-frequency feature demand pairs, a matching strategy is executed in the orthogonal frequency division multiplexing resource pool. According to the principle of minimum satisfaction and priority for low interference, a set of resource blocks that can be allocated are screened to form a candidate list of resource blocks that can be allocated to the current service. The frequency domain interference level and time domain exclusivity index are marked for each candidate block. S44: Selecting a resource block group from the resource block candidate list that satisfies the target constraint range in terms of bandwidth, time slot, and interference level, and arranging the allocation order according to the service priority to generate a final resource block set for the service; S45: Summarize the final resource block set of all services to be allocated, and construct a resource block configuration tensor based on the five-dimensional structure of service number, frequency starting point, bandwidth, time slot starting point and time slot length.

8. The wireless network resource allocation method based on edge intelligence according to claim 1, characterized in that: The S5 specifically includes: S51: Receive the collaborative domain partition matrix output by S3, obtain the node set of each collaborative domain, and based on the number of remaining computing units and available bandwidth of the nodes collected in S1, count the computing resource capacity and communication resource capacity currently available for allocation in each collaborative domain, which are recorded as and Where k represents the collaboration domain number; S52: Receive the resource block configuration tensor output in S4 and analyze the minimum computing resource requirements for each business task in the tensor Minimum bandwidth requirement Where m represents the business task number; S53: Calculate the resource gap degree Δ of collaboration domain k corresponding to business task m based on the candidate mapping relationship between business tasks and collaboration domains km ; S54: For all candidate collaboration domain-business task combinations, calculate the adaptive mapping weight W based on the inverse relationship of the gap degree km ; S55: Sort all candidate combinations of collaboration domains and business tasks from high to low according to mapping weights, and assign a collaboration domain with the highest weight to each business task as the target execution domain, and construct a resource allocation mapping relationship table.

9. The wireless network resource allocation method based on edge intelligence according to claim 1, characterized in that: The S6 specifically includes: S61: Receive the resource allocation mapping table output in S5, extract the target collaboration domain number corresponding to each business task, and build a node task candidate pool within the collaboration domain. The task candidate pool contains the numbers of all edge nodes in the domain that have not reached the computing load limit; S62: Read the number of remaining computing units of each node in the collaboration domain, calculate its load carrying capacity based on the current resource status of the node, and record the number of idle computing units that can be allocated to each node in this scheduling cycle; S63: Based on the computing resource requirements of the business task, a set of edge nodes that meet the requirements is screened from the candidate node pool, and the node with the most remaining computing units is preferentially selected from the set as the execution node for this task; S64: For each selected execution node, construct a node-level resource scheduling instruction set, where the instruction set includes the task number to be executed by the corresponding node, the bound resource block number and its time-frequency configuration parameters; S65: The constructed node-level scheduling instruction set is synchronously sent to the edge nodes in the collaboration domain, and parsed and executed by the local task scheduler of each node to complete the final edge execution operation.

10. The wireless network resource allocation method based on edge intelligence according to claim 9, characterized in that: The S62 specifically includes: S621: Receive the initial value of the remaining computing units of the edge node collected in S1, and in conjunction with the node operating system monitoring interface, obtain the number of computing units occupied by the running and pending tasks in the node processing queue at the current moment, and count the current total computing unit occupancy; S622: Calculate the number of idle computing units available for task allocation in this round on the current node based on the maximum total number of schedulable computing units configured on the node and deducting the number of occupied computing units. S623: Obtain the start and end time of the time window of the current scheduling cycle, and bind the number of idle computing units to the start and end time periods of the time window in combination with the resource scheduling cycle settings of the node operating system to form an idle resource status snapshot with a time tag; S624: The number of idle computing units of each node and the node identifier are combined into a key-value pair, and recorded in a node load status mapping table within the collaboration domain.

Images